{"id":2025,"date":"2022-10-12T15:00:35","date_gmt":"2022-10-12T15:00:35","guid":{"rendered":"https:\/\/cog-ist.com\/?post_type=blog_content&p=2025"},"modified":"2025-09-08T20:24:03","modified_gmt":"2025-09-08T20:24:03","slug":"filozoflar-icin-sade-ongormeli-isleme-ongormeli-isleme-uzerine-bir-basucu-metni-wanja-wiese-thomas-metzinger","status":"publish","type":"blog_content","link":"https:\/\/cog-ist.com\/en\/blog_content\/filozoflar-icin-sade-ongormeli-isleme-ongormeli-isleme-uzerine-bir-basucu-metni-wanja-wiese-thomas-metzinger\/","title":{"rendered":"Filozoflar i\u00e7in \u201cSade \u00d6ng\u00f6rmeli \u0130\u015fleme\u201d: \u00d6ng\u00f6rmeli \u0130\u015fleme \u00dczerine Bir Ba\u015fucu Metni \u2014\u00a0Wanja Wiese, Thomas Metzinger"},"content":{"rendered":"

\u00d6zg\u00fcn Ad\u0131:\u00a0Vanilla PP for Philosophers: A Primer on Predictive Processing<\/a><\/p>\n\n\n\n

Anahtar Kelimeler:<\/strong> aktif \u00e7\u0131karsama (active inference), dikkat (attention), Bayesci \u00e7\u0131karsama (Bayesian inference), \u00e7evresel tecrit (environmental seclusion), serbest enerji prensibi (free energy principle), hiyerar\u015fik i\u015fleme (hierarchical processing), ideomotor prensibi (ideomotor principle), alg\u0131 (perception), alg\u0131sal \u00e7\u0131karsama (perceptual inference), hassasiyet (precision), \u00f6ng\u00f6r\u00fc (prediction), \u00f6ng\u00f6r\u00fc hatas\u0131 (prediction error), minimizasyon (minimization), \u00f6ng\u00f6rmeli i\u015fleme (predictive processing), \u00f6ng\u00f6rmeli kontrol (predictive control), istatistiksel tahmin (statistical estimation), yukar\u0131dan a\u015fa\u011f\u0131 i\u015fleme (top-down processing)<\/p>\n\n\n\n

Felsefeci okurlar\u0131 hedef alan bu k\u0131sa b\u00f6l\u00fcm\u00fcn amac\u0131, \u00f6ng\u00f6rmeli i\u015flemede (predictive processing) yer alan ana konseptleri g\u00f6zden ge\u00e7irmek ve bu konseptlerin k\u0131sa ve \u00f6z a\u00e7\u0131klamalar\u0131n\u0131 sunmak. Ana terimlerin bu koleksiyonda nas\u0131l kullan\u0131ld\u0131klar\u0131n\u0131 g\u00f6rmek, kendini bu alanda uzman olarak g\u00f6renler i\u00e7in bile kullan\u0131\u015fl\u0131 olabilir. Ba\u015flang\u0131\u00e7ta i\u015fleri basit tutmak ad\u0131na, \u00f6ng\u00f6rmeli i\u015fleme (\u00d6\u0130) i\u00e7in \u00f6nem arz eden bir dizi \u00f6zellik tan\u0131mlay\u0131p bunlar\u0131 k\u0131sa a\u00e7\u0131klamalar ve bir alfabetik s\u00f6zl\u00fck ile destekleyece\u011fiz.<\/p>\n\n\n\n

Burada a\u00e7\u0131klanan \u00f6zellikler, yap\u0131lan \u00d6\u0130 tan\u0131mlar\u0131n\u0131n tamam\u0131nda yer alm\u0131yor. Bu \u00f6zelliklerin baz\u0131lar\u0131 belli bir \u00d6\u0130 modeli i\u00e7in gerekli de\u011filken, baz\u0131lar\u0131 \u00fczerinde de anla\u015fmazl\u0131klar var. Asl\u0131nda yaln\u0131zca bu koleksiyonun <\/em>yazarlar\u0131 bile bu \u00f6zelliklerin tamam\u0131n\u0131 kabul etmeyecektir. Biz de i\u015fi \u015feffafla\u015ft\u0131rmak i\u00e7in yazarlar\u0131m\u0131z\u0131, \u00f6ne s\u00fcrd\u00fckleri arg\u00fcmanlar\u0131 desteklemek ad\u0131na hangi \u00f6zelliklerin gerekli <\/em>oldu\u011funu ve (e\u011fer varsa) hangilerinin onlar\u0131n tan\u0131m\u0131yla uyumsuz <\/em>oldu\u011funu k\u0131saca a\u00e7\u0131klamalar\u0131 y\u00f6n\u00fcnde te\u015fvik ettik. A\u00e7\u0131kl\u0131\u011f\u0131 sa\u011flayabilmek i\u00e7in burada \u00f6zelliklerin tam listesini veriyor, \u00f6zellikleri \u201cSade \u00d6\u0130\u201d (yani \u00f6ng\u00f6rmeli i\u015flemenin, muhtemelen bu alanda \u00e7al\u0131\u015fan \u00e7o\u011fu ara\u015ft\u0131rmac\u0131 taraf\u0131ndan kabul g\u00f6recek bir form\u00fclasyonu) ad\u0131na bizce ne kadar merkezi olduklar\u0131na g\u00f6re kabaca s\u0131ral\u0131yoruz. Daha detayl\u0131 a\u00e7\u0131klamalar a\u015fa\u011f\u0131da verilecek. Bu \u00f6zelliklerin, \u201c\u00f6ng\u00f6rmeli i\u015fleme\u201d konseptinin uygulanmas\u0131 i\u00e7in her biri gerekli ve bir b\u00fct\u00fcn olarak yeterli ko\u015fullar\u0131 belirtmediklerini not edelim. \u015eu an i\u00e7in elimizdeki tek \u015fey anlamsal bir y\u0131\u011f\u0131n. Bu y\u0131\u011f\u0131n\u0131n, beraber al\u0131nd\u0131klar\u0131nda yeterli olan kriterlerden olu\u015fan alt k\u00fcmeleri olabilir ve belki bu alt k\u00fcmeler aras\u0131nda kesi\u015fimler de vard\u0131r. Bu hala geli\u015fmekte olan bir \u00e7er\u00e7eve; \u00d6\u0130\u2019yle ilgili t\u00fcm fikirler hakk\u0131nda teorilerden ba\u011f\u0131ms\u0131z (theory-neutral), arka planda g\u00fc\u00e7l\u00fc varsay\u0131mlara ba\u015fvurmadan a\u00e7\u0131klamalar getirmek \u00e7ok zor, belki de imk\u00e2ns\u0131z. Yine de a\u015fa\u011f\u0131daki \u00f6zelliklerden 1\u20137 aras\u0131ndakiler bu koleksiyonda \u00d6\u0130 olarak adland\u0131r\u0131lan \u015fey i\u00e7in gerekli \u00f6zellikler denebilir:<\/p>\n\n\n\n

1. Yukar\u0131dan A\u015fa\u011f\u0131 \u0130\u015fleme:<\/strong> Beyinde komp\u00fctasyon yap\u0131l\u0131rken yukar\u0131dan a\u015fa\u011f\u0131 ve a\u015fa\u011f\u0131dan yukar\u0131 i\u015fleme aras\u0131ndaki kar\u015f\u0131l\u0131kl\u0131 etkile\u015fim \u00f6nemli bir yer tutar. \u00d6\u0130, alg\u0131 ve eylemde yukar\u0131dan a\u015fa\u011f\u0131 ve a\u015fa\u011f\u0131dan yukar\u0131 sinyallerin birbirlerine g\u00f6re a\u011f\u0131rl\u0131kland\u0131rmas\u0131na vurgu yapar.<\/p>\n\n\n\n

2. \u0130statistiksel Tahmin: <\/strong>\u00d6\u0130 rastgele de\u011fi\u015fkenlere dair tahminlerin komp\u00fctasyonunu i\u00e7erir. Bu tahminler, duyusal sinyalleri a\u00e7\u0131klayabilmek i\u00e7in var\u0131lan istatistiksel hipotezler olarak d\u00fc\u015f\u00fcn\u00fclebilir.<\/p>\n\n\n\n

3. Hiyerar\u015fik \u0130\u015fleme: <\/strong>\u00d6\u0130, hiyerar\u015fik olarak organize olmu\u015f estimat\u00f6rler kullan\u0131r (bu estimat\u00f6rlerin her biri farkl\u0131 uzamsal ve zamansal \u00f6l\u00e7eklerdeki \u00f6zellikleri takip eder).<\/p>\n\n\n\n

4. \u00d6ng\u00f6r\u00fc:<\/strong> \u00d6\u0130, hiyerar\u015fideki bir\u00e7ok rastgele de\u011fi\u015fkenin birbirini \u00f6ng\u00f6rebilmeyi sa\u011fl\u0131yor olu\u015fundan faydalan\u0131r.<\/p>\n\n\n\n

5. \u00d6ng\u00f6r\u00fc Hatas\u0131 Minimizasyonu (\u00d6HM):<\/strong> \u00d6\u0130 \u00f6ng\u00f6r\u00fc hatalar\u0131n\u0131n komp\u00fctasyonunu i\u00e7erir, bu hata terimleri hassasiyet tahminleri taraf\u0131ndan a\u011f\u0131rl\u0131kland\u0131r\u0131l\u0131r ve \u00d6\u0130\u2019nin ana hedeflerinden biri bu a\u011f\u0131rl\u0131kland\u0131r\u0131lm\u0131\u015f \u00f6ng\u00f6r\u00fc hatalar\u0131n\u0131 minimize etmektir.<\/p>\n\n\n\n

6. Bayesci \u00c7\u0131karsama:<\/strong> \u00d6\u0130 Bayesci \u00e7\u0131karsaman\u0131n normlar\u0131yla \u00f6rt\u00fc\u015f\u00fcr: Hiyerar\u015fik modeldeki \u00f6ng\u00f6r\u00fc hatas\u0131 minimizasyonu tam olarak Bayesci \u00e7\u0131karsaman\u0131n sonucuna yak\u0131nsayacakt\u0131r.<\/p>\n\n\n\n

7. \u00d6ng\u00f6rmeli Kontrol: <\/strong>\u00d6\u0130\u2019ye g\u00f6re organizma ald\u0131\u011f\u0131 duyusal girdileri de\u011fi\u015ftirmek ve \u00f6ng\u00f6r\u00fcleriyle \u00f6rt\u00fc\u015ft\u00fcrmek ad\u0131na eyleme ge\u00e7ebilir ve b\u00f6ylece \u00f6ng\u00f6r\u00fc hatas\u0131n\u0131 minimize eder. Bu da organizman\u0131n hayati parametrelerini (kandaki oksijen seviyesi, kan \u015fekeri vb.) reg\u00fcle etmesini sa\u011flar.<\/p>\n\n\n\n

8. \u00c7evresel Tecrit:<\/strong> Organizman\u0131n, \u00e7evresine ve v\u00fccuduna dair durumlara (states) direkt eri\u015fimi yoktur (\u201cdirekt alg\u0131\u201d \u00fczerine kavramsal bir analiz i\u00e7in bkz. Snowdon 1992<\/a>), bu durumlar\u0131 -bedenden ve \u00e7evreden gelen duyusal sinyallerin gizli sebeplerini tahmin ederek- \u00e7\u0131karsayabilir. Bu, \u00d6\u0130\u2019nin baz\u0131 felsefi tan\u0131mlar\u0131na g\u00f6re temel \u00f6zelliklerden biri olsa da (Hohwy 2016<\/a>; Hohwy 2017<\/a>) hakk\u0131nda tart\u0131\u015fmalar bulunmaktad\u0131r (Anderson 2017<\/a>; Clark 2017<\/a>; Fabry 2017a<\/a>; Fabry 2017b<\/a>).<\/p>\n\n\n\n

9. \u0130deomotor Prensibi: \u201c<\/strong>\u0130deomotor\u201d tahminleri hesaplamak alg\u0131 ve eylemin dayana\u011f\u0131n\u0131 olu\u015fturmaktad\u0131r \u00e7\u00fcnk\u00fc bu tahminler, d\u00fcnyam\u0131zdaki -alg\u0131 taraf\u0131ndan kaydedilen ve eylemler taraf\u0131ndan olu\u015fturulabilen- de\u011fi\u015fimleri kodlamaktad\u0131r.<\/p>\n\n\n\n

10. Dikkat ve Hassasiyet: <\/strong>Dikkat, hassasiyet tahminlerini optimize etme s\u00fcreci olarak tan\u0131mlanabilir.<\/p>\n\n\n\n

11<\/strong>. Hipotez Testi:<\/strong> Alg\u0131, bili\u015f ve eylemin alt\u0131nda yatan komp\u00fctasyonel s\u00fcre\u00e7leri hipotez testi olarak tan\u0131mlayabiliriz (ya da i\u00e7sel model lehine delil biriktirme s\u00fcreci olarak). Kavramsal olarak pasif ve aktif hipotez testi aras\u0131nda bir ayr\u0131m yapabiliriz (ve aktif hipotez testini eylem ile, pasif hipotez testini ise alg\u0131 ile e\u015fle\u015ftirmeyi deneyebiliriz). Fakat beyindeki t\u00fcm hipotez testlerinin (e\u011fer b\u00f6yle bir \u015fey s\u00f6ylemek mant\u0131kl\u0131 ise) aktif hipotez testi oldu\u011fu ortaya \u00e7\u0131kabilir.<\/p>\n\n\n\n

12. Serbest Enerji Prensibi:<\/strong> Temelde \u00d6\u0130, yaln\u0131zca serbest enerjiyi (\u00e7o\u011fu \u00d6\u0130 tan\u0131m\u0131nda serbest enerji, \u00f6ng\u00f6r\u00fc hatas\u0131n\u0131n uzun vadedeki ortalamas\u0131d\u0131r) minimize etmenin bir yoludur.<\/p>\n\n\n\n

A\u015fa\u011f\u0131daki yaz\u0131da \u00d6\u0130 veya herhangi bir matematiksel temel hakk\u0131nda bilgi sahibi oldu\u011funuzu varsaym\u0131yoruz. Bu giri\u015f makalesi \u00e7o\u011funlukla \u00d6\u0130 \u00e7er\u00e7evesinin kavramsal temelleriyle s\u0131n\u0131rl\u0131 olacak. Ki\u015fi bu ba\u015fucu metnini okuduktan sonra, bu koleksiyonun di\u011fer makalelerindeki tart\u0131\u015fmalar\u0131 takip edebiliyor olmal\u0131. Fakat biz okuyucular\u0131m\u0131z\u0131 ayn\u0131 zamanda (Clark, 2016<\/a>) ve (Hohwy, 2013<\/a>) makalelerini -konu hakk\u0131nda m\u00fckemmel iki felsefi monografi- okuyarak anlay\u0131\u015flar\u0131n\u0131 derinle\u015ftirmeleri y\u00f6n\u00fcnde te\u015fvik ediyoruz.<\/p>\n\n\n\n

1. \u00d6ng\u00f6rmeli \u0130\u015fleme Nedir? Yedi Temel \u00d6zellik<\/strong><\/h1>\n\n\n\n

\u00d6ng\u00f6rmeli i\u015fleme (\u00d6\u0130); alg\u0131, eylem, bili\u015f ve bunlar aras\u0131ndaki ili\u015fkiyi kavramsal olarak tek bir \u00e7at\u0131 alt\u0131nda a\u00e7\u0131klayabilmek i\u00e7in kullan\u0131labilecek genel bir komp\u00fctasyon prensibini i\u00e7eren bir \u00e7er\u00e7evedir. Direkt olarak altta yatan n\u00f6ronal s\u00fcre\u00e7lere dair bir teori olmasa da (bu teori komp\u00fctasyoneldir, n\u00f6rofizyolojik de\u011fil) \u00f6ng\u00f6rmeli i\u015flemenin beyinde nas\u0131l y\u00fcr\u00fct\u00fclebilece\u011fine dair a\u015fa\u011f\u0131 yukar\u0131 spesifik \u00f6neriler var (bkz. Engel ve ark. 2001<\/a>; Friston 2005<\/a>; Wacongne ve ark. 2011<\/a>; Bastos ve ark. 2012<\/a>; Brodski ve ark. 2015<\/a>). Dahas\u0131, organizma-alt\u0131 (subpersonal) seviyedeki (\u00f6rn., komp\u00fctasyonel veya n\u00f6robiyolojik) analizlerin a\u00e7\u0131klamalar\u0131nda kullan\u0131labilen baz\u0131 prensipler, organizma seviyesindeki a\u00e7\u0131klamalara da (\u00f6rn., k\u0131lg\u0131n fenomenlere (agentive phenomena), mant\u0131kl\u0131 d\u00fc\u015f\u00fcnmenin yap\u0131s\u0131n\u0131 veya bilincin i\u00e7eriklerini a\u00e7\u0131klayan fenomenolojik raporlar) uygulanabilir gibi g\u00f6z\u00fck\u00fcyor. Bu da \u00d6\u0130\u2019nin felsefe i\u00e7in ilgin\u00e7 ve felsefeyle alakal\u0131 olmas\u0131n\u0131n sebeplerinden biri. E\u011fer \u00d6\u0130 teorisi do\u011fru yoldaysa:<\/p>\n\n\n\n

1. Bili\u015f ve bilin\u00e7 \u00fczerine yap\u0131lan teorik ve deneysel \u00e7al\u0131\u015fmalar aras\u0131nda kavramsal k\u00f6pr\u00fcler kurulmas\u0131n\u0131 sa\u011flayabilir;<\/p>\n\n\n\n

2. Birbirinden farkl\u0131 g\u00f6z\u00fcken fenomenler aras\u0131nda beklenmeyen ili\u015fkiler ortaya \u00e7\u0131karabilir;<\/p>\n\n\n\n

3. Farkl\u0131 teorik yakla\u015f\u0131mlar\u0131 belli \u00f6l\u00e7\u00fcde birle\u015ftirebilir.<\/p>\n\n\n\n

Fakat \u00f6ncelikle \u00d6\u0130 nedir? \u0130\u015fte teorinin anahtar fikirlerinden birinin g\u00f6rece eski form\u00fclasyonlar\u0131ndan biri:[1]<\/a><\/p>\n\n\n\n

\u201cWenn die Anschauung sich nach der Beschaffenheit der Gegenst\u00e4nde richten m\u00fc\u00dfte, so sehe ich nicht ein, wie man a priori von ihr etwas wissen k\u00f6nne; richtet sich aber der Gegenstand (als Ob\u00adjekt der Sinne) nach der Beschaffenheit unseres Anschauungsverm\u00f6gens, so kann ich mir diese M\u00f6glichkeit ganz wohl vorstellen.\u201d (Kant 1998[1781\/87], B XVII)[2]<\/a><\/p>\n\n\n\n

Saf Akl\u0131n Ele\u015ftirisi\u2019nin bu k\u0131sm\u0131nda Kant\u2019\u0131n vurgulad\u0131\u011f\u0131 \u015feylerden biri de \u00fczerinde sentezleme i\u015flemlerinin uyguland\u0131\u011f\u0131 duyusal materyali olu\u015fturan sezilerimizin[3]<\/a>, oldu\u011fu gibi verilen duyu-verileri olmad\u0131\u011f\u0131 (kar\u015f\u0131la\u015ft\u0131r\u0131n\u0131z: Brook 2013, \u00a7 3.2). Duyu verileri sadece oldu\u011fu gibi al\u0131nmaz, ayn\u0131 zamanda sezi yetimiz (Anschauungsverm\u00f6gen<\/em>) taraf\u0131ndan k\u0131smen \u015fekillendirilir. Bu fikir modern dile \u015f\u00f6yle d\u00f6k\u00fclebilir:<\/p>\n\n\n\n

Duyusal i\u015flemeye dair klasik teoriler beyni pasif ve uyaran (stimulus) taraf\u0131ndan y\u00f6netilen bir cihaz olarak g\u00f6r\u00fcyor. Bunun aksine daha yak\u0131n zamanl\u0131 yakla\u015f\u0131mlarsa alg\u0131n\u0131n yap\u0131c\u0131 do\u011fas\u0131n\u0131 \u00f6ne \u00e7\u0131kar\u0131yor, onu aktif ve olduk\u00e7a se\u00e7ici bir s\u00fcre\u00e7 olarak g\u00f6r\u00fcyor. Uyaranlar\u0131n i\u015flenmesi s\u00fcrecinin, talamokortikal a\u011flar\u0131 g\u00fc\u00e7l\u00fc bi\u00e7imde \u015fekillendiren ve gelecekteki duyusal olaylar hakk\u0131nda s\u00fcrekli \u00f6ng\u00f6r\u00fcler \u00fcreten \u201cyukar\u0131dan a\u015fa\u011f\u0131 etkiler\u201d taraf\u0131ndan kontrol edildi\u011fine dair bolca delil var (Engel ve ark., 2001<\/a>).<\/p>\n\n\n\n

\u0130\u015fte bu bizim \u00f6ng\u00f6rmeli i\u015flemenin ilk \u00f6zelli\u011fi olarak adland\u0131rd\u0131\u011f\u0131m\u0131z \u015fey: Yukar\u0131dan A\u015fa\u011f\u0131 \u0130\u015fleme<\/strong>. G\u00f6r\u00fcld\u00fc\u011f\u00fc \u00fczere alg\u0131n\u0131n k\u0131smen yukar\u0131dan a\u015fa\u011f\u0131 s\u00fcre\u00e7lerle y\u00f6netildi\u011fi fikri yeni de\u011fil (bunu s\u00f6yleyerek tabii ki ge\u00e7mi\u015fte alg\u0131ya dair bask\u0131n teorilerin, bu s\u00fcre\u00e7lerin rol\u00fcn\u00fc \u00f6nemsemedi\u011fi ger\u00e7e\u011fini reddetmiyoruz). \u00d6\u0130\u2019nin buradaki yeni katk\u0131s\u0131, yukar\u0131dan a\u015fa\u011f\u0131 i\u015flemleri ve halihaz\u0131rda var olan bilgileri alg\u0131n\u0131n her ko\u015ful ve zamanda -yani yaln\u0131zca duyusal girdiler g\u00fcr\u00fclt\u00fcl\u00fc veya belirsiz oldu\u011funda de\u011fil- mevcut bir bile\u015feni olarak g\u00f6stermesi.[4]<\/a> \u00d6\u0130\u2019ye g\u00f6re ki\u015finin beyni s\u00fcrekli istatistiksel tahminler olu\u015fturur ve bunlar d\u00fcnyada o an var olan \u015feylerin temsilleri[5]<\/a> olarak g\u00f6rev yaparlar (2. \u00f6zellik: \u0130statistiksel Tahmin<\/strong>). Bu tahminler ayn\u0131 zamanda hiyerar\u015fik bir organizasyona sahip (farkl\u0131 uzamsal ve zamansal \u00f6l\u00e7ekleri takip ediyorlar; yani 3. \u00f6zellik: Hiyerar\u015fik \u0130\u015fleme<\/strong>).[6]<\/a> Beyin bu temsilleri anl\u0131k (ve gelecekteki) duyusal girdileri ve bunlar\u0131n kayna\u011f\u0131n\u0131 \u00f6ng\u00f6rmek i\u00e7in kullan\u0131r. Bunun m\u00fcmk\u00fcn olmas\u0131n\u0131n sebebi de hiyerar\u015finin farkl\u0131 seviyelerindeki tahminlerinin birbirini \u00f6ng\u00f6r\u00fcyor<\/em> olmas\u0131d\u0131r (4. \u00f6zellik: \u00d6ng\u00f6r\u00fc<\/strong>). \u00d6ng\u00f6r\u00fclerle ger\u00e7ek duyusal girdiler aras\u0131ndaki uyumsuzluklar pasif \u015fekilde alg\u0131 olu\u015fumu i\u00e7in de\u011fil, yaln\u0131zca \u00f6nceden yarat\u0131lm\u0131\u015f olan temsillerin g\u00fcncellenmesi i\u00e7in kullan\u0131l\u0131r (b\u00f6ylece, m\u00fcmk\u00fcn oldu\u011fu kadar\u0131yla, gelmekte olan duyusal sinyali kestirmek i\u00e7in kullan\u0131l\u0131r). Bu g\u00fcncellemelerin amac\u0131, \u00f6ng\u00f6r\u00fcler sebebiyle olu\u015fan \u00f6ng\u00f6r\u00fc hatalar\u0131n\u0131, Bayesci \u00c7\u0131karsama<\/strong> (6. \u00f6zellik; bunun hakk\u0131nda a\u015fa\u011f\u0131da daha fazla konu\u015faca\u011f\u0131z) kurallar\u0131na uygun \u015fekilde minimize etmektir (5. \u00f6zellik: \u00d6ng\u00f6r\u00fc Hatas\u0131 Minimizasyonu<\/strong> (\u00d6HM)). \u00d6HM\u2019nin komp\u00fctasyonel prensibi, beyindeki t\u00fcm i\u015flemelerin (\u00d6\u0130\u2019deki t\u00fcm hiyerar\u015fi seviyelerinde) uydu\u011fu genel bir prensiptir. Bu noktada beyindeki i\u015flemenin tamam\u0131n\u0131 kontroll\u00fc faal (online) hal\u00fcsinasyon \u015feklinde tan\u0131mlamaya \u00e7ok yak\u0131n\u0131z:[7]<\/a><\/p>\n\n\n\n

\u201c\u0130nsan beynini incelemenin verimli bir yolu; ona s\u0131radan uyan\u0131k hallerde bile d\u00fcnyayla ilgili hal\u00fcsinasyonlar kuran, i\u00e7sel otonom sim\u00fclasyon dinamiklerini d\u0131\u015far\u0131dan gelmekte olan duyusal girdilerle \u201c\u00e7arp\u0131\u015ft\u0131ran\u201d, d\u00fcnyayla ilgili dinamik bir \u015fekilde hayal kurarak fenomenal deneyimleri olu\u015fturan bir sistem olarak bakmakt\u0131r.\u201d (Metzinger 2004[2003]<\/a>, syf. 52)<\/p>\n\n\n\n

Fenomenal deneyimlere dair i\u00e7eriklerin, \u00d6\u0130\u2019ye g\u00f6re, yaln\u0131zca hiyerar\u015fik organizasyona sahip \u00f6ng\u00f6r\u00fc hatas\u0131 minimizasyonu s\u00fcrecinin bir par\u00e7as\u0131 oldu\u011funa dikkat edelim (s\u00fcrece dahil olan i\u00e7eriklerin \u00e7o\u011fu bilin\u00e7d\u0131\u015f\u0131). Yukar\u0131da a\u00e7\u0131klanan alt\u0131 ana \u00f6zelli\u011fi toparlay\u0131p bir de 7. \u00f6zelli\u011fi ekledi\u011fimizde, bu koleksiyondaki anlam\u0131yla \u00f6ng\u00f6rmeli i\u015flemenin ne oldu\u011funa dair net bir ifade sunabiliyoruz (bu tan\u0131m\u0131 a\u015fa\u011f\u0131da 8\u201312. \u00f6zellikleri kullanarak zenginle\u015ftirece\u011fiz):<\/p>\n\n\n\n

\u00d6ng\u00f6rmeli i\u015fleme (\u00d6\u0130):<\/p>\n\n\n\n

\u00b7 hiyerar\u015fik <\/em>bir \u00f6ng\u00f6rmeli kodlamad\u0131r,<\/p>\n\n\n\n

\u00b7 hassasiyet \u00f6l\u00e7\u00fcmlerinin arac\u0131l\u0131k etti\u011fi<\/em>[8]<\/em><\/strong><\/a> (precision-mediated) <\/em>\u00f6ng\u00f6r\u00fc hatas\u0131 minimizasyonu i\u00e7erir[9]<\/a>,<\/p>\n\n\n\n

\u00b7 \u00f6ng\u00f6rmeli kontrole <\/em>imkan tan\u0131r.<\/p>\n\n\n\n

Bu tan\u0131m\u0131n, \u00e7o\u011funlukla \u00f6ng\u00f6rmeli kodlama <\/em>olarak adland\u0131r\u0131lan \u015feye g\u00f6re halihaz\u0131rda daha kapsaml\u0131 oldu\u011funa (\u00f6zellikle \u00f6ng\u00f6rmeli kodlama sadece veri s\u0131k\u0131\u015ft\u0131rma i\u00e7in kullan\u0131lan komp\u00fctasyonel bir stratejiyse; kar\u015f\u0131la\u015ft\u0131rma i\u00e7in Shi ve Sun 1999<\/a>; Clark 2013a<\/a>) dikkat edelim. Birincisi, \u00d6\u0130 hiyerar\u015fiktir. \u0130kincisi, hassasiyet tahminleri ge\u00e7mi\u015f varsay\u0131mlar ile anl\u0131k duyusal kan\u0131tlar aras\u0131ndaki dengeyi istatistiksel olarak ideal \u015fekilde sa\u011flaman\u0131n \u00f6tesinde i\u015flevsel g\u00f6revleri yerine getirebiliyor (bkz. Clark 2013b<\/a>). \u00dc\u00e7\u00fcnc\u00fcs\u00fc, \u00d6\u0130 \u00d6ng\u00f6rmeli Kontrole<\/strong> (7. \u00f6zellik; kar\u015f\u0131la\u015ft\u0131rma i\u00e7in Seth 2015<\/a>) olanak tan\u0131d\u0131\u011f\u0131 i\u00e7in s\u0131k s\u0131k eylem-odakl\u0131 olarak tan\u0131t\u0131l\u0131yor. Bu, baz\u0131lar\u0131 taraf\u0131ndan kabul g\u00f6ren \u201ceylem bir bak\u0131ma alg\u0131dan daha \u00f6nemlidir; alg\u0131 d\u00fcnya hakk\u0131nda bilgi edinme s\u00fcreci olarak tan\u0131mlansa da bu bilgiyi edinmenin as\u0131l i\u015flevi organizman\u0131n varl\u0131\u011f\u0131n\u0131 s\u00fcrd\u00fcrmesini sa\u011flayacak, verimli ve ba\u011flama duyarl\u0131 eylemlere imk\u00e2n tan\u0131mak\u201d \u015feklindeki varsay\u0131ma vurgu yap\u0131yor. \u00d6\u0130\u2019yi Friston\u2019\u0131n serbest enerji prensibi (SEP) ba\u011flam\u0131nda ele ald\u0131\u011f\u0131m\u0131zda bu daha belirginle\u015fiyor[10]<\/a>. Fakat bu konuyu a\u00e7madan \u00f6nce bir ad\u0131m geriye \u00e7ekilelim ve \u00f6ng\u00f6rmeli kodlama perspektifinden alg\u0131 problemine bakal\u0131m.<\/p>\n\n\n\n

2. \u00d6ng\u00f6rmeli \u0130\u015fleme ve \u00d6ng\u00f6rmeli Kodlama<\/strong><\/h1>\n\n\n\n

\u00d6\u0130\u2019nin neredeyse t\u00fcm \u00f6zelliklerinde oldu\u011fu gibi, \u00d6\u0130\u2019nin alg\u0131 hakk\u0131ndaki perspektifinin de \u00f6nc\u00fclleri mevcut. \u00d6rne\u011fin Helmholtz\u2019un \u015fu ifadesini ele alal\u0131m:<\/p>\n\n\n\n

\u201cDie psychischen Th\u00e4tigkeiten, durch welche wir zu dem Urtheile kommen, dass ein bestimmtes Object von bestimmter Beschaffenheit an einem bestimmten Orte ausser uns vorhanden sei, sind im Allgemeinen nicht bewusste Th\u00e4tigkeiten, sondern unbewusste. Sie sind in ihrem Resultate ei\u00adnem Schlusse gleich, insofern wir aus der beobachteten Wirkung auf unsere Sinne die Vorstellung von einer Ursache dieser Wirkung gewinnen, w\u00e4hrend wir in der That direct doch immer nur die Nervenerregungen, also die Wirkungen wahrnehmen k\u00f6nnen, niemals die \u00e4usseren Objecte.\u201d (Von Helmholtz 1867, p. 430)[11]<\/a><\/p>\n\n\n\n

Burada ele al\u0131nd\u0131\u011f\u0131 haliyle alg\u0131 probleminin iki boyutu var: 1) alg\u0131lar (percepts) bilin\u00e7d\u0131\u015f\u0131 \u00e7\u0131kar\u0131msal bir s\u00fcrecin sonucudur; 2) alg\u0131lar bize d\u0131\u015fsal nesnelerin \u00f6zelliklerini sunar, fakat asl\u0131nda d\u0131\u015fsal nesnelerin sadece etkilerini alg\u0131layabiliriz. Bu fikrin modern bir a\u00e7\u0131klamas\u0131n\u0131 Dennett\u2019in 2013\u2019te yay\u0131mlad\u0131\u011f\u0131 monografisinde (Sezgi Pompalar\u0131 ve Di\u011fer D\u00fc\u015f\u00fcnme Aletleri<\/em>) bulabiliriz. Dennett beynin kendini i\u00e7inde buldu\u011fu bu gizemli durumu, a\u015fa\u011f\u0131daki kurgusal senaryoya benzeterek karakterize ediyor:<\/p>\n\n\n\n

\u201cDev bir robotu kontrol edebildi\u011finiz bir odada (kontrol odas\u0131) hapsedilmi\u015f durumdas\u0131n\u0131z. [\u2026] Robot, bir\u00e7ok risk ve f\u0131rsat bar\u0131nd\u0131ran tehlikeli bir d\u00fcnyada ya\u015f\u0131yor. Robotun gelece\u011fi sizin ellerinizde, tabii kendi gelece\u011finiz de robotunuzu bu d\u00fcnyada ne kadar iyi y\u00f6netebildi\u011finize ba\u011fl\u0131. Robot yok edilirse odan\u0131zdaki elektrik kesilecek, art\u0131k buzdolab\u0131n\u0131zda yemek olmayacak ve \u00f6leceksiniz. \u0130yi \u015fanslar!\u201d (Dennett 2013<\/a>, syf. 102)<\/p>\n\n\n\n

Robotun i\u00e7indeki insan\u0131n, robotun sens\u00f6rleri arac\u0131l\u0131\u011f\u0131yla d\u00fcnyaya yaln\u0131zca dolayl\u0131 yoldan eri\u015fimi var ve yapt\u0131r\u0131lan eylemlerin etkilerini bilemez, bunlar\u0131 \u00e7\u0131karsamak zorunda. Bu, \u00c7evresel Tecrit<\/strong> olarak adland\u0131rd\u0131\u011f\u0131m\u0131z \u00f6zelli\u011fi \u00f6rneklendiriyor (8. \u00f6zellik). \u00c7evresel Tecrit, komp\u00fctasyonel de\u011fil, epistemolojik bir \u00f6zellik. Fakat \u00d6\u0130\u2019deki komp\u00fctasyonlar\u0131n \u00e7\u00f6zebildi\u011fi problemlerin tan\u0131mlar\u0131nda bu terimin ismi ge\u00e7iyor.[12]<\/a> \u0130\u00e7erideki insan\u0131n, robota ula\u015fan farkl\u0131 farkl\u0131 sinyallerin ne anlama geldi\u011fini bulabilmesi i\u00e7in bu sinyallerin gizli sebepleri hakk\u0131nda hipotezler \u00fcretmesi gerekiyor. Duyusal sinyallerin sebeplerini \u00e7\u0131karsamak bir ters problemdir (inverse problem) \u00e7\u00fcnk\u00fc d\u0131\u015fsal-gizli sebeplerden duyusal etkilere do\u011fru olan haritalamay\u0131 tersine \u00e7evirmeyi gerektirir. Bu en hafif tabiriyle zor bir problem \u00e7\u00fcnk\u00fc ayn\u0131 etkinin birden fazla sebebi olabilir.[13]<\/a> Yani sebepler (S) ve etkiler (E) aras\u0131ndaki ili\u015fki deterministik bir haritalamayla a\u00e7\u0131klanabiliyor bile olsa (f<\/em>: C -> E), bu haritalaman\u0131n tersi (f<\/em>-1: E -> C) genellikle var olmayacakt\u0131r. Peki beyin bu problemi nas\u0131l \u00e7\u00f6z\u00fcyor?<\/p>\n\n\n\n

\u00d6ncelikle duyusal bir etkinin sebebi, etki sayesinde tam olarak belirlenemez; dolay\u0131s\u0131yla bu gizli sebep hakk\u0131nda iyi bir tahmin yapabilmek i\u00e7in ge\u00e7mi\u015ften gelen bilgilerin kullan\u0131lmas\u0131 gerekiyor. Dahas\u0131, duyusal bir ayg\u0131t\u0131n d\u0131\u015fsal sebepler taraf\u0131ndan nas\u0131l etkilendi\u011fini bilirsek bu gizli sebepler hakk\u0131nda bilgimiz oldu\u011funda duyusal etkileri \u00e7\u0131karsamak, tam tersini yapmaya g\u00f6re daha kolay. Yani gizli sebepler hakk\u0131nda bilgimiz olursa bunlar\u0131n duyusal etkileri hakk\u0131nda bir \u00f6ng\u00f6r\u00fc<\/em> olu\u015fturabiliriz. Bu \u00f6ng\u00f6r\u00fc ger\u00e7ekte olan duyusal sinyalle k\u0131yaslanabilir ve ikisi aras\u0131ndaki fark\u0131n -yani \u00f6ng\u00f6r\u00fc hatas\u0131n\u0131n<\/em>– miktar\u0131 bize gizli sebep hakk\u0131nda yapt\u0131\u011f\u0131m\u0131z tahminin kalitesi hakk\u0131nda bir ipucu verir. Bu tahmini g\u00fcncelleyebilir, yeni bir \u00f6ng\u00f6r\u00fc hesaplayabilir, yeniden anl\u0131k duyusal sinyallerle kar\u015f\u0131la\u015ft\u0131rabiliriz ve b\u00f6ylece \u00f6ng\u00f6r\u00fc hatas\u0131n\u0131 (umuyoruz ki) minimize edebiliriz. Gizli sebebe dair yapt\u0131\u011f\u0131m\u0131z ilk tahminin \u00e7ok k\u00f6t\u00fc olmas\u0131 pek m\u00fchim de\u011fil \u00e7\u00fcnk\u00fc s\u00fcrekli \u00f6ng\u00f6r\u00fcler ve \u00f6ng\u00f6r\u00fc hatalar\u0131 hesaplayarak -ve tahminimizi bunlara g\u00f6re g\u00fcncelleyerek- gizli sebebin iyi bir temsilini buldu\u011fumuza dair g\u00fcvenimizi giderek art\u0131rabiliriz.<\/p>\n\n\n\n

Bu stratejiyi basit bir \u00f6rnekle a\u00e7\u0131klayal\u0131m. Bir \u00f6\u011fretmen s\u0131n\u0131fa girer ve masas\u0131n\u0131n \u00fczerinde \u201c\u00d6\u011fretmen bir sahtek\u00e2r. Ger\u00e7ekte b\u00f6yle biri yok bile.\u201d yazan bir k\u00e2\u011f\u0131t bulur. Mesaj\u0131n mavi bir dolmakalemle yaz\u0131lm\u0131\u015f olmas\u0131 (\u00f6\u011fretmenin bildi\u011fi kadar\u0131yla) bir\u00e7ok \u00f6\u011frencisini adaylar\u0131n d\u0131\u015f\u0131nda b\u0131rak\u0131r. Su\u00e7luyu bulmak i\u00e7in \u00f6\u011fretmen mavi dolmakalem kullanan t\u00fcm \u00f6\u011frencilerin tahtaya \u00e7\u0131kmas\u0131n\u0131 ve kendi kalemlerini kullanarak bir k\u00e2\u011f\u0131da bir \u015feyler yazmas\u0131n\u0131 ister. \u00dc\u00e7 \u00f6\u011frenci (A, B ve C) \u00f6ne \u00e7\u0131kar ve hepsinin farkl\u0131 markalarda m\u00fcrekkep kulland\u0131klar\u0131 anla\u015f\u0131l\u0131r (bu da onlar\u0131n birbirinden ayr\u0131\u015ft\u0131r\u0131labilmelerini sa\u011flar). \u00d6\u011fretmen art\u0131k mesaj\u0131n gizli sebebi hakk\u0131nda bilgiye dayal\u0131 bir tahmin yapabilir: \u0130lk tahmininde A\u2019n\u0131n su\u00e7lu oldu\u011funu d\u00fc\u015f\u00fcn\u00fcr ve ondan mesaj\u0131n ayn\u0131s\u0131n\u0131 yazmas\u0131n\u0131 ister. Bu, as\u0131l mesaj\u0131n bir \u00f6ng\u00f6r\u00fcs\u00fc olarak g\u00f6r\u00fclebilir ve \u00f6\u011fretmen as\u0131l mesaj ile mesaj\u0131n A taraf\u0131ndan tekrar yaz\u0131ld\u0131\u011f\u0131 versiyonunu kar\u015f\u0131la\u015ft\u0131rarak gizli sebep hakk\u0131nda yapt\u0131\u011f\u0131 tahmini de\u011ferlendirir. E\u011fer m\u00fcrekkep ayn\u0131 g\u00f6z\u00fck\u00fcyorsa \u00f6ng\u00f6r\u00fc hatas\u0131 yoktur ve gizli sebebe dair yap\u0131lan tahminin de\u011fi\u015fmesine gerek olmaz \u2014 su\u00e7lu bulunmu\u015ftur. E\u011fer bir farkl\u0131l\u0131k varsa \u00f6\u011fretmen tahminini \u00f6rne\u011fin B\u2019nin su\u00e7lu oldu\u011fu varsay\u0131m\u0131yla g\u00fcncelleyebilir. \u00d6\u011fretmen s\u00fcrekli \u00f6ng\u00f6r\u00fcler (tahtaya \u00e7\u0131kar\u0131lan \u00f6\u011frencilerin yazd\u0131\u011f\u0131 mesajlar) olu\u015fturup bunlar\u0131 as\u0131l duyusal sinyallerle (masadaki as\u0131l mesaj) kar\u015f\u0131la\u015ft\u0131rarak \u00f6ng\u00f6r\u00fc hatas\u0131n\u0131 sonunda minimize eder ve ger\u00e7ek su\u00e7luyu bulur.<\/p>\n\n\n\n

Bu kurgusal senaryo ile beynin ger\u00e7ekte kendisini buldu\u011fu durum aras\u0131nda bir\u00e7ok farkl\u0131l\u0131k var. Bir tanesi bu \u00f6rne\u011fin ki\u015fi seviyesinde bir irade i\u00e7ermesi (Dennett\u2019in \u201cdev robot\u201d d\u00fc\u015f\u00fcnce deneyindeki gibi): Yani \u00f6\u011fretmen A\u2019n\u0131n su\u00e7lu oldu\u011fu hipotezini A\u2019ya o mesaj\u0131 yeniden yazd\u0131rarak test ediyor. Dahas\u0131, olas\u0131 gizli sebeplerin say\u0131s\u0131 sonlu ve \u00f6ng\u00f6r\u00fc hatas\u0131 \u00f6\u011fretmene sadece belirli bir \u00f6\u011frencinin olaya dahil olmad\u0131\u011f\u0131n\u0131 g\u00f6steriyor. Su\u00e7lu hakk\u0131nda daha fazla bir bilgi vermiyor; sadece bir tane \u015f\u00fcpheliyi eliyor. Beyin t\u00fcm olas\u0131 hipotezler \u00fczerinden tek tek ge\u00e7emez \u00e7\u00fcnk\u00fc d\u00fcnyadaki olas\u0131 gizli sebeplerin say\u0131s\u0131 sonsuz olabilir. Dahas\u0131, d\u00fcnya dinamiktir, yani gizli sebeplerin temsillerinin de dinamik olmas\u0131; \u00e7evredeki t\u00fcm alakal\u0131 ve \u00f6ng\u00f6r\u00fclebilir de\u011fi\u015fimlere adapte olup onlar\u0131 \u00f6ng\u00f6rmesi gerekiyor. Son olarak \u00f6rne\u011fi daha ger\u00e7ek\u00e7i k\u0131lmak i\u00e7in \u00f6\u011fretmenin kendisine ula\u015fan her <\/em>duyusal girdi i\u00e7in her zaman <\/em>\u00f6ng\u00f6r\u00fcler olu\u015fturdu\u011funu d\u00fc\u015f\u00fcnelim. \u00d6\u011fretmen mesaja yol a\u00e7an nedensel etkile\u015fimleri \u00e7\u0131karsayabilece\u011fi gibi \u00e7evresinde nedensellik \u00f6rg\u00fcs\u00fc ba\u011flam\u0131nda olan biten her \u015feyi de (duyusal yola\u011fa kendi etkisini de i\u00e7ererek) \u00e7\u0131karsayabilir.<\/p>\n\n\n\n

3. \u00d6ng\u00f6rmeli \u0130\u015fleme ve Bayesci \u00c7\u0131karsama<\/strong><\/h1>\n\n\n\n

Bayesci \u00c7\u0131karsama <\/strong>(6. \u00f6zellik), var olan fakat hakk\u0131nda belirsizlik bulunan bilgiyi mant\u0131\u011fa uygun \u015fekilde[14]<\/a> yeni delillerle birle\u015ftirmeyi sa\u011flayan komp\u00fctasyonel bir metottur. Burada belirsizlik, bilginin olas\u0131l\u0131ksal bir formatta -yani olas\u0131l\u0131k da\u011f\u0131l\u0131m\u0131 kullan\u0131larak- a\u00e7\u0131klanabilmesi anlam\u0131na geliyor. \u00c7ok basit bir \u00f6rnek olarak bir \u00f6znenin -yukar\u0131da bahsetti\u011fimiz \u00f6\u011fretmen gibi- sonlu say\u0131daki hipotezlerden hangisinin do\u011fru oldu\u011fu hususunda belirsizlik ya\u015famas\u0131n\u0131 verebiliriz. Bu durumda belirsizlik, \u00f6znenin hipotezlere 1\u2019den k\u00fc\u00e7\u00fck olas\u0131l\u0131klar atamas\u0131 (hi\u00e7bir hipotezin do\u011fruluk olas\u0131l\u0131\u011f\u0131na 1 vermemesi) \u015feklinde yans\u0131t\u0131l\u0131r. Fakat baz\u0131 durumlarda -\u00f6rne\u011fin \u00f6zne s\u00fcrekli bir aral\u0131kta herhangi bir de\u011fere sahip olabilecek bir nicelik hakk\u0131nda g\u00fcr\u00fclt\u00fcl\u00fc (noisy) bir \u00f6l\u00e7\u00fcm yapt\u0131\u011f\u0131nda- \u00f6znenin sahip oldu\u011fu bilgi en iyi, sonsuz say\u0131da ihtimal (\u201chipotez\u201d) olarak modellenebilir. B\u00f6yle bir durumda bilgi, alt b\u00f6lgelerine farkl\u0131 olas\u0131l\u0131klar atayan bir olas\u0131l\u0131k yo\u011funluk fonksiyonu (probability density function) -yani bir model- kullan\u0131larak kodlanabilir. Bayesci \u00e7\u0131karsama, Bayes kural\u0131n\u0131 kullanarak \u015fu soruya mant\u0131kl\u0131 bir cevap sa\u011flar: Yeni bir bilgi edindi\u011fimde modelimi nas\u0131l g\u00fcncellemeliyim? Yeni bilgiye bir \u00f6rnek, \u00f6znenin bir \u00f6l\u00e7\u00fcm yaparak elde etti\u011fi bilgi olabilir (\u00f6znenin \u00f6l\u00e7t\u00fc\u011f\u00fc nicelik hakk\u0131nda \u00f6nceden de belirsiz bir bilgiye sahip oldu\u011funu varsay\u0131yoruz).<\/p>\n\n\n\n

Bu g\u00fcncelleme formel olarak, a posteriyori <\/em>(\u201cdeneyden sonra\u201d) da\u011f\u0131l\u0131m\u0131 <\/em>(bu terim ayn\u0131 zamanda sadece sonraki da\u011f\u0131l\u0131m <\/em>olarak da adland\u0131r\u0131l\u0131yor) <\/em>hesaplamay\u0131 i\u00e7eriyor. Sonraki da\u011f\u0131l\u0131m, a priori <\/em>(\u201cdeneyden \u00f6nce\u201d) da\u011f\u0131l\u0131m <\/em>(ya da \u00f6nceki da\u011f\u0131l\u0131m<\/em>) ile olabilirlik (likelihood) <\/em>de\u011fi\u015fkenlerini birle\u015ftirerek elde ediliyor. \u00d6nceki da\u011f\u0131l\u0131m, \u00f6znenin halihaz\u0131rda sahip oldu\u011fu bilgiyi; olabilirlik ise \u00f6znenin bilgi sahibi oldu\u011fu alan ile yeni gelen bilginin birbiriyle nas\u0131l bir ili\u015fkide oldu\u011funu kodluyor. Bayesci \u00e7\u0131karsaman\u0131n g\u00fczel bir \u00f6zelli\u011fi, belirsizli\u011fi azaltabilmesi. Bu formel olarak sonraki da\u011f\u0131l\u0131m\u0131n, \u00f6nceki da\u011f\u0131l\u0131ma g\u00f6re daha d\u00fc\u015f\u00fck bir varyansa sahip oldu\u011fu -yani daha hassas (precise) oldu\u011fu- anlam\u0131na geliyor.<\/p>\n\n\n\n

Y\u00fczeysel olarak bakt\u0131\u011f\u0131m\u0131zda \u00f6ng\u00f6r\u00fc hatas\u0131 minimizasyonu (\u00d6HM) ile Bayesci \u00e7\u0131karsama aras\u0131nda belirgin bir ba\u011flant\u0131 yok. Asl\u0131nda Bayesci \u00e7\u0131karsamay\u0131 \u00d6HM kullanarak uygulamam\u0131z\u0131n neden yararl\u0131 olabilece\u011fi de bariz de\u011fil. Fakat bir iyi nedenimiz var. Alg\u0131yla ilgili yukar\u0131da bahsetti\u011fimiz ters problemin k\u00f6t\u00fc tan\u0131mlanm\u0131\u015f bir problem oldu\u011funu hat\u0131rlayal\u0131m: D\u0131\u015fsal olaylar\u0131n etkileri olarak g\u00f6r\u00fclen duyusal sinyaller, \u00e7evremizdeki gizli sebeplere haritalanamaz \u00e7\u00fcnk\u00fc her bir duyusal sinyal i\u00e7in birden fazla olas\u0131 d\u0131\u015fsal sebep var. Di\u011fer bir deyi\u015fle, gizli sebepler hakk\u0131nda belirsizlik mevcut. Bu sebepler hakk\u0131nda \u00f6nceki varsay\u0131mlar verildi\u011finde ve \u00f6l\u00e7t\u00fc\u011f\u00fcm\u00fcz duyusal sinyalleri yeni deliller olarak d\u00fc\u015f\u00fcnd\u00fc\u011f\u00fcm\u00fczde, Bayesci \u00e7\u0131karsama bize gizli sebepler hakk\u0131ndaki ge\u00e7mi\u015f varsay\u0131mlar\u0131m\u0131z\u0131 nas\u0131l g\u00fcncellememiz gerekti\u011fine dair mant\u0131kl\u0131 bir \u00e7\u00f6z\u00fcm sunuyor. Yani Bayesci \u00e7\u0131karsaman\u0131n bize verdi\u011fi \u015fey (en az\u0131ndan teoride) \u201colas\u0131l\u0131ksal ters haritalama\u201d gibi bir \u015fey. Bu fonksiyon \u00f6l\u00e7\u00fclen bir duyusal etkiyi, farkl\u0131 olas\u0131 gizli sebeplere harital\u0131yor ve hangi olas\u0131 sebeplerin duyusal etkiye ger\u00e7ekten sebep olma ihtimalinin daha fazla oldu\u011funu bildiriyor.<\/p>\n\n\n\n

Fakat Bayesci \u00e7\u0131karsamaya sahipsek neden \u00d6HM\u2019ye ihtiyac\u0131m\u0131z var? \u00c7\u00fcnk\u00fc Bayesci \u00e7\u0131karsama komp\u00fctasyonel olarak kompleks ve \u00e7ok zor olabiliyor. Basit durumlarda sonraki da\u011f\u0131l\u0131m\u0131 analitik olarak hesaplamak m\u00fcmk\u00fcnken di\u011fer durumlarda bunu anca yakla\u015f\u0131k bir de\u011fer olarak bulabiliyoruz. Di\u011fer baz\u0131 durumlarda ise sonraki da\u011f\u0131l\u0131m\u0131 hesaplamak m\u00fcmk\u00fcn olsa da as\u0131l bu da\u011f\u0131l\u0131m\u0131n maksimize eden <\/em>k\u0131sm\u0131n\u0131 bulmay\u0131 istiyoruz (\u00f6rne\u011fin yeni delilleri hesaba katt\u0131\u011f\u0131m\u0131zda do\u011fruluk olas\u0131l\u0131\u011f\u0131 en y\u00fcksek olan tek bir hipotez). Bu maksimize eden<\/em>\u2019i bulmak da komp\u00fctasyonel olarak u\u011fra\u015ft\u0131r\u0131c\u0131 olabilir ve yakla\u015f\u0131k tahminler bulduran metotlar\u0131 gerektirebilir. Bu metotlardan baz\u0131lar\u0131 \u00f6ng\u00f6r\u00fc hatas\u0131 minimizasyonunu i\u00e7erir. Bayesci \u00e7\u0131karsaman\u0131n amac\u0131 \u00f6ng\u00f6r\u00fc hatas\u0131 minimizasyonununkinden farkl\u0131 da olsa; Bayesci \u00e7\u0131karsamay\u0131 alg\u0131 probleminin bir \u00e7\u00f6z\u00fcm\u00fc olarak ald\u0131\u011f\u0131m\u0131zda, \u00f6ng\u00f6r\u00fc hatas\u0131 minimizasyonu da Bayesci g\u00fcncellemelerin hesaplanmas\u0131 i\u00e7in bir \u00e7\u00f6z\u00fcm sunuyor.<\/p>\n\n\n\n

Bayesci \u00e7\u0131karsaman\u0131n ayn\u0131 zamanda hiyerar\u015fik modellerde de \u00e7al\u0131\u015ft\u0131\u011f\u0131na dikkat edelim. Birbirine kom\u015fu olmayan seviyelerdeki de\u011fi\u015fkenlerin birbirinden ko\u015fullu ba\u011f\u0131ms\u0131z oldu\u011funu varsayd\u0131\u011f\u0131m\u0131zda, farkl\u0131 seviyelerdeki tahminler paralel olarak[15]<\/a> g\u00fcncellenebilir (kar\u015f\u0131la\u015ft\u0131rma i\u00e7in Friston 2003<\/a>, syf. 1342), bu da ideal bir d\u00fcnyada evrensel olarak tutarl\u0131 bir tahminler seti ortaya \u00e7\u0131kar\u0131r (Lee ve Mumford 2003<\/a>, syf. 1437\u2019de de\u011finildi\u011fi \u00fczere pratikte i\u015fler daha karma\u015f\u0131k). Buradaki ana fikir \u015fu: D\u00fcnyadaki bir\u00e7ok nesne birbirini direkt nedensel olarak etkilemiyor olsa da bunlar ayn\u0131 d\u00fcnyan\u0131n nesneleri; bu da iki rastgele nesne aras\u0131ndaki nedensel ili\u015fkilere genellikle di\u011fer nesnelerin arac\u0131l\u0131k etti\u011fi <\/em>anlam\u0131na geliyor. Benzer \u015fekilde, tek bir nesnenin farkl\u0131 \u00f6zellikleri de birbirinden tamamen ba\u011f\u0131ms\u0131z de\u011fil \u00e7\u00fcnk\u00fc bunlar ayn\u0131 nesnenin \u00f6zellikleri. Fakat bu durum, bu \u00f6zelliklerin beyinde her zaman birlikte i\u015flenmesi gerekti\u011fi anlam\u0131na gelmiyor. \u00d6rne\u011fin mavi bir diskin temsili, bir lokasyonda belli bir rengin (mavi) ve ayn\u0131 lokasyonda belli bir \u015feklin (disk) temsili sayesinde olu\u015fturulabilir. Rengin lokasyonu hakk\u0131ndaki bilgi bana \u015feklin lokasyonu hakk\u0131nda bilgi veriyor. Fakat diskin lokasyonu hakk\u0131nda ayr\u0131 bir temsile sahipsem, renk ve \u015fekil birbirinden (ko\u015fullu) ba\u011f\u0131ms\u0131zm\u0131\u015f gibi davranabilirim; yani diskin lokasyonu verildi\u011finde, renk hakk\u0131ndaki bilgi bana \u015fekille ilgili yeni bir bilgi vermiyor[16]<\/a>. Bu, komp\u00fctasyonel a\u00e7\u0131dan daha az yo\u011fun temsillere olanak tan\u0131yor. Beyindeki i\u015flevsel da\u011f\u0131l\u0131m (functional segregation) da bu durumu yans\u0131t\u0131yor olabilir (kar\u015f\u0131la\u015ft\u0131rma i\u00e7in, bahsetti\u011fimiz durumu zamansal boyuta odaklanarak irdeleyen Friston ve Buzsaki 2016<\/a>).<\/p>\n\n\n\n

4. \u00d6ng\u00f6rmeli \u0130\u015fleme ve \u0130deomotor Prensibi<\/strong><\/h1>\n\n\n\n

\u015eu ana kadar \u00f6ng\u00f6r\u00fc hatas\u0131 minimizasyonundan sadece \u201cduyusal girdiler \u0131\u015f\u0131\u011f\u0131nda alg\u0131lar yaratma y\u00f6ntemi\u201d olarak bahsettik. Fakat \u00d6HM\u2019nin as\u0131l rol\u00fc d\u00fcnyadaki gizli sebepleri \u00e7\u0131karsamak de\u011fil, d\u00fcnyada \u00f6znenin hayatta kalmas\u0131n\u0131 sa\u011flayacak de\u011fi\u015fimleri hayata ge\u00e7irmek olabilir (a\u015fa\u011f\u0131da 7. k\u0131s\u0131ma bkz.). Ayn\u0131 zamanda bu de\u011fi\u015fimlerin hedefi d\u0131\u015fsal \u00e7evre de\u011fil, \u00f6znenin i\u00e7sel do\u011fas\u0131, yani v\u00fccudu olabilir. Biyolojik sistemlerde organizma b\u00fct\u00fcnl\u00fc\u011f\u00fc \u00fcst d\u00fczey \u00f6neme sahip \u00e7\u00fcnk\u00fc stabil (i\u00e7sel durumlar\u0131n\u0131 kontrol edebilen) bir organizma farkl\u0131 \u00e7evrelerde hayatta kalabilirken stabil olmayan bir organizma uygun \u00e7evrelerde bile hayatta kalamayabilir. Bu durum Anil Seth taraf\u0131ndan \u015f\u00f6yle dile getiriliyor:<\/p>\n\n\n\n

\u201c\u00d6\u0130, d\u0131\u015f duyumdansa (d\u0131\u015f \u00e7evreden do\u011fan duyusal sinyalleri ta\u015f\u0131yan klasik duyular) i\u00e7 duyumda (v\u00fccudun i\u00e7sel fizyolojik durumuyla ilgili duyu) daha do\u011fal olarak uygulanabilir. \u00c7\u00fcnk\u00fc bir organizma i\u00e7in i\u00e7 duyuma dair beklenmedik durumlardan ka\u00e7\u0131nmak, d\u0131\u015f durumlara dair beklenmedik durumlardan ka\u00e7\u0131nmaktan daha \u00f6nemlidir. Kandaki oksijen ya da \u015feker miktar\u0131n\u0131n beklenmedik seviyelerde olmas\u0131 organizma i\u00e7in muhtemelen k\u00f6t\u00fc \u015feylerin habercisiyken d\u0131\u015f duyuma dair beklenmedik duyumsamalar (\u00f6rne\u011fin yeni g\u00f6rsel girdiler) daha az zararl\u0131 olmakla birlikte baz\u0131 durumlarda arzulanan bir \u015fey bile olabilir [\u2026]\u201d (Seth 2015<\/a>, syf.9)<\/p>\n\n\n\n

\u0130\u00e7-duyusal (interoceptive) \u00e7\u0131karsama yaln\u0131zca v\u00fccudun i\u00e7sel durumunu \u00e7\u0131karsamak de\u011fil, kandaki oksijen seviyesi ve kan \u015fekeri gibi ya\u015famsal parametrelerin \u00f6ng\u00f6rmeli kontrol\u00fcn\u00fc <\/em>(7. \u00f6zellik) sa\u011flamakt\u0131r. Seth bu konuda a\u015fa\u011f\u0131daki \u00f6rne\u011fi veriyor.<\/p>\n\n\n\n

\u201cBeyin i\u00e7-duyusal \u00e7\u0131karsama arac\u0131l\u0131\u011f\u0131yla kan \u015fekerinde bir d\u00fc\u015f\u00fc\u015f fark etti\u011finde olu\u015fan alg\u0131 (\u015fekerli bir \u015feyler a\u015ferme), hiyerar\u015fik olarak daha yukar\u0131 seviyelerde \u00f6ng\u00f6r\u00fc hatalar\u0131na yol a\u00e7ar. Bu modeller birbiriyle e\u015fle\u015fen i\u00e7-duyusal ve d\u0131\u015f-duyusal girdilere dair \u00f6ng\u00f6r\u00fcler olu\u015fturmaya ba\u015flar ve bu \u00f6ng\u00f6r\u00fcler hiyerar\u015fi boyunca a\u015fa\u011f\u0131 y\u00f6nde ilerler. B\u00f6ylece olu\u015fan \u00f6ng\u00f6r\u00fc hatalar\u0131 dizisi ya v\u00fccuttaki ya\u011f deposunu kullanmak gibi bir otonom kontrol mekanizmas\u0131yla (aktif \u00e7\u0131karsama) ya da d\u0131\u015f \u00e7evreyi i\u00e7eren allostatik[17]<\/a> eylemlerle (\u015fekerli \u015feyler bulup yemek gibi) \u00e7\u00f6z\u00fcl\u00fcr.\u201d (Seth 2015<\/a>, syf. 10)<\/p>\n\n\n\n

Bu y\u00fczden i\u00e7-duyusal \u00f6ng\u00f6r\u00fc hatas\u0131 minimizasyonu, \u00d6\u0130\u2019ye g\u00f6re alg\u0131 ve eylem aras\u0131nda nas\u0131l bir ba\u011f oldu\u011funa dair ayd\u0131nlat\u0131c\u0131 bir \u00f6rnek. \u0130\u00e7-duyusal \u00d6HM\u2019nin ama\u00e7lar\u0131ndan biri organizman\u0131n ya\u015famsal parametrelerini uygun s\u0131n\u0131rlar i\u00e7inde tutabilmek; bu da hem bu parametrelerin g\u00fcncel durumunu isabetli bi\u00e7imde \u00e7\u0131karsayabilmeyi hem de gerekti\u011finde onlar\u0131 de\u011fi\u015ftirebilmeyi i\u00e7eriyor. Friston bu durumu \u015f\u00f6yle a\u00e7\u0131kl\u0131yor (asl\u0131nda serbest enerjiyi minimize etmek ba\u011flam\u0131nda a\u00e7\u0131kl\u0131yor fakat bunu yapabilmek -belli varsay\u0131mlar e\u015fli\u011finde- \u00f6ng\u00f6r\u00fc hatas\u0131n\u0131 minimize etmeyi beraberinde getiriyor):<\/p>\n\n\n\n

\u201c\u00d6zneler serbest enerjiyi bask\u0131lamak i\u00e7in onun ba\u011fl\u0131 oldu\u011fu iki \u015feyi de\u011fi\u015ftirebilir: Gelen duyusal girdileri \u00e7evrelerini etkilemek suretiyle de\u011fi\u015ftirebilir veya i\u00e7sel durumlar\u0131 \u00fczerinde oynayarak tan\u0131ma yo\u011funluklar\u0131n\u0131 (recognition density) de\u011fi\u015ftirebilirler. Bu ayr\u0131m eylem ve alg\u0131 aras\u0131ndaki farkla da g\u00fczelce e\u015fle\u015fiyor [\u2026]\u201d (Friston 2010<\/a>, syf. 129)<\/p>\n\n\n\n

K\u0131sacas\u0131 duyusal sinyaller ile bu sinyallere dair (i\u00e7sel tahminler sayesinde edinilen) \u00f6ng\u00f6r\u00fcler aras\u0131ndaki hata miktar\u0131, i\u00e7sel tahminleri ya da duyusal sinyalleri (eylem arac\u0131l\u0131\u011f\u0131yla) de\u011fi\u015ftirerek minimize edilebilir. Bu durum, alg\u0131 s\u0131ras\u0131nda aktifle\u015fen i\u00e7sel temsillerin ayn\u0131lar\u0131n\u0131n eylemlere olanak tan\u0131ma esnas\u0131nda da kullan\u0131labilece\u011fini \u00f6neriyor. Bu da sadece ortak bir veri format\u0131 oldu\u011fu anlam\u0131na de\u011fil, ayn\u0131 zamanda alg\u0131n\u0131n alt\u0131nda yatan temsillerin en az\u0131ndan baz\u0131lar\u0131n\u0131n eylemlerin alt\u0131nda yatan temsillerle n\u00fcmerik olarak ayn\u0131 oldu\u011fu anlam\u0131na geliyor.<\/p>\n\n\n\n

Bu varsay\u0131m James\u2019in \u00f6z\u00fcn\u00fc \u015f\u00f6yle a\u00e7\u0131klad\u0131\u011f\u0131 ideomotor teorisinde <\/em>zaten var (James 1890<\/a>)[18]<\/a>: \u201cEylem E\u2019nin duyusal etkilerini temsil eden fikir, hareketin \u00fcretilmesi i\u00e7in an\u0131nda bir \u00f6nc\u00fcl te\u015fkil edecektir\u201d (James 1890<\/a>, syf. 586; italik yaz\u0131m ihmal edildi). Yak\u0131n zamanda bu fikir eylem temsilinin ortak kodlama <\/em>ba\u011flam\u0131nda a\u00e7\u0131klanmas\u0131 \u015feklinde ele al\u0131nd\u0131 (bkz. Hommel ve ark. 2001<\/a>; Hommel 2015<\/a>; Prinz 1990<\/a>).[19]<\/a> Temel fikir hep ayn\u0131: D\u00fcnyadaki gizli sebeplere dair n\u00f6ral temsiller, eylem haz\u0131rl\u0131klar\u0131na dair n\u00f6ral temsillerle kesi\u015fiyor (yani bu temsillerin baz\u0131 k\u0131s\u0131mlar\u0131 n\u00fcmerik olarak e\u015fit). Di\u011fer bir deyi\u015fle, hem alg\u0131lar\u0131m\u0131z olarak hem de motor emirler olarak i\u015flev g\u00f6rebilen baz\u0131 \u201cideomotor\u201d temsiller mevcut.[20]<\/a><\/p>\n\n\n\n

Ideomotor Prensibi <\/strong>(9. \u00f6zellik), komp\u00fctasyonel olarak eylem ve alg\u0131 aras\u0131ndaki formel bir ikilikten faydalan\u0131yor. \u0130kilik \u015fu: E\u011fer ben alg\u0131sal olarak i\u015flerin geldi\u011fi nokta (state of affairs) olarak adland\u0131rabilece\u011fimiz bir p noktas\u0131na eri\u015febiliyorsam, p\u2019nin c olarak adland\u0131rabilece\u011fimiz alg\u0131lanmas\u0131 m\u00fcmk\u00fcn sonu\u00e7lar\u0131 (ya da bile\u015fenleri) vard\u0131r. Eylem hedef odakl\u0131 oldu\u011fu i\u00e7in bir eylemde bulunarak i\u015flerin geldi\u011fi bir p noktas\u0131na eri\u015fmek istiyorum. Bu da demek oluyor ki eylemi, p\u2019nin alg\u0131lanmas\u0131 m\u00fcmk\u00fcn sonu\u00e7lar\u0131 c\u2019nin ortaya \u00e7\u0131kar\u0131ld\u0131\u011f\u0131 bir s\u00fcre\u00e7 olarak tan\u0131mlayabiliriz. Alg\u0131y\u0131 da farazi bir eylemin etkilerinin (bu etkiler p\u2019yi, dolay\u0131s\u0131yla c\u2019yi ortaya \u00e7\u0131kar\u0131yor) \u00e7\u0131karsand\u0131\u011f\u0131 bir s\u00fcre\u00e7 olarak tan\u0131mlamak m\u00fcmk\u00fcn (bu fikrin detayl\u0131 bir a\u00e7\u0131klamas\u0131 i\u00e7in bkz. Todorov 2009<\/a>). Bu ikili perspektifin komp\u00fctasyonel faydalar\u0131, Friston ve meslekta\u015flar\u0131 taraf\u0131ndan geli\u015ftirilen aktif \u00e7\u0131karsama <\/em>kavram\u0131yla a\u00e7\u0131\u011fa \u00e7\u0131k\u0131yor:<\/p>\n\n\n\n

\u201cBeyni bu \u015fekilde ele ald\u0131\u011f\u0131m\u0131zda n\u00f6ronlar hem sebep hem de sonucu temsil ediyor: D\u00fcnyadaki gizli durumlar hakk\u0131nda ko\u015fullu beklentiler kodluyor, ayn\u0131 zamanda da eylemler vas\u0131tas\u0131yla dolayl\u0131 yoldan bu durumlara sebep oluyorlar. [\u2026] K\u0131sacas\u0131 aktif \u00e7\u0131karsama, duyusal (sonu\u00e7) ve motor (sebep) temsiller aras\u0131ndaki klasik ayr\u0131mlar\u0131 y\u0131k\u0131yor ve dairesel bir nedensellik ili\u015fkisi do\u011furuyor. Bu da temsilleri optimize etmenin alg\u0131 veya niyete (intention), yani alg\u0131 veya niyet olu\u015fturmaya denk d\u00fc\u015fmesi anlam\u0131na geliyor.\u201d (Friston ve ark. 2011<\/a>, syf. 138)[21]<\/a><\/p>\n\n\n\n

Aktif <\/em>\u00e7\u0131karsama \u00e7o\u011funlukla alg\u0131sal <\/em>\u00e7\u0131karsamadan ayr\u0131l\u0131r. Fakat hem ikisi de \u00f6ng\u00f6r\u00fc hatas\u0131n\u0131n minimize edilmesiyle ger\u00e7ekle\u015ftirildi\u011fi, hem de uygulan\u0131\u015flar\u0131 birbirinden d\u00fczg\u00fcnce ayr\u0131\u015famad\u0131\u011f\u0131 i\u00e7in aktif <\/em>\u00e7\u0131karsama \u00f6zellikle Friston ve meslekta\u015flar\u0131 taraf\u0131ndan ayn\u0131 zamanda daha genel bir terim olarak da kullan\u0131l\u0131yor. Bu terim serbest enerji prensibi ba\u011flam\u0131nda serbest enerjiyi minimize eden ve hem eylemin hem de alg\u0131n\u0131n alt\u0131nda yatan komp\u00fctasyonel s\u00fcre\u00e7leri ifade ediyor: \u201cAktif \u00e7\u0131karsama \u2014 yani d\u00fcnyay\u0131 etkilemek yoluyla (eylem) i\u00e7sel durumlar\u0131 (alg\u0131) ve duyusal durumlar\u0131 de\u011fi\u015ftirerek serbest enerjiyi minimize etme.\u201d (Friston ve ark. 2012a<\/a>, syf. 539)[22]<\/a><\/p>\n\n\n\n

Eylem ve alg\u0131n\u0131n ortak yan\u0131, (bilin\u00e7d\u0131\u015f\u0131, yakla\u015f\u0131k olarak Bayesci) \u00e7\u0131karsamad\u0131r. Eylem ve alg\u0131n\u0131n alt\u0131nda yatan n\u00f6ral yap\u0131lar\u0131n birbiriyle kesi\u015fti\u011fi varsay\u0131ld\u0131\u011f\u0131 i\u00e7in aktif ve alg\u0131sal \u00e7\u0131karsama birbiriyle uyum i\u00e7erisinde \u00e7al\u0131\u015f\u0131yor.[23]<\/a> Dolay\u0131s\u0131yla \u0130deomotor Prensibinin <\/strong>bu g\u00fcncellenmi\u015f ve zenginle\u015ftirilmi\u015f versiyonu, prensibin derin \u00f6nerileri daha yeni yeni ke\u015ffedilmeye ba\u015flansa da eylem ve alg\u0131 \u00fczerine birle\u015ftirici bir perspektif sunuyor.[24]<\/a><\/p>\n\n\n\n

5. Dikkat ve Hassasiyet<\/strong><\/h1>\n\n\n\n

\u00d6\u0130 perspektifinde olu\u015fturulmu\u015f bir\u00e7ok kullan\u0131\u015fl\u0131 fikirden bir tanesi de dikkatimizi y\u00f6neltme eyleminin asl\u0131nda hassasiyet tahminlerini optimize etme olarak analiz edilebilece\u011fi fikri (10. \u00f6zellik). Fikir ilk olarak Karl Friston ve Klaas Stephan taraf\u0131ndan ortaya at\u0131ld\u0131 (Friston ve Stephan, 2007<\/a>) (Feldman ve Friston 2010<\/a> ile Hohwy 2012<\/a> makaleleri de bu fikri daha da a\u00e7\u0131yor). Hassasiyet tahminleri \u00f6ng\u00f6r\u00fc hatas\u0131 terimlerini a\u011f\u0131rl\u0131kland\u0131rmaya yarad\u0131\u011f\u0131 i\u00e7in, belli bir \u00f6ng\u00f6r\u00fc hatas\u0131yla e\u015fle\u015fen hassasiyet bu hatan\u0131n di\u011fer seviyelerdeki i\u015flemelere etkisini belirliyor. Yani tahmin edilen hassasiyetin artmas\u0131, bir uyaran\u0131n daha derin bi\u00e7imde i\u015flenmesini sa\u011fl\u0131yor. Dahas\u0131, hassasiyet tahminleri a\u015fa\u011f\u0131dan yukar\u0131ya ve yukar\u0131dan a\u015fa\u011f\u0131ya olmak \u00fczere iki farkl\u0131 \u015fekilde de\u011fi\u015ftirilebiliyor: A\u015fa\u011f\u0131dan yukar\u0131ya, hassasiyet eldeki nesneye dair bir fonksiyon olarak tahmin edilebilir (yani nesnenin varyans\u0131n\u0131n tersi); yukar\u0131dan a\u015fa\u011f\u0131ya y\u00f6nde ise hassasiyet tahminleri hassasiyetin artmas\u0131 ya da azalmas\u0131 beklenen ba\u011flamlarda ayarlanabilir (mod\u00fcle edilebilir) veya zihinsel eylemler i\u00e7in amac\u0131 temsil eden \u015feyler olarak i\u015flev g\u00f6rebilir (Metzinger 2017<\/a>). Hassasiyet tahminlerinin a\u015fa\u011f\u0131dan yukar\u0131ya ve yukar\u0131dan a\u015fa\u011f\u0131ya de\u011fi\u015fimleri aras\u0131ndaki fark, i\u00e7sel ve d\u0131\u015fsal dikkat aras\u0131ndaki fark ile e\u015fle\u015ftirilebilir (detaylar i\u00e7in, bkz. Feldman ve Friston 2010<\/a> ve Hohwy 2012<\/a>).<\/p>\n\n\n\n

Dikkatin hassasiyet-optimizasyonu olarak a\u00e7\u0131klanmas\u0131n\u0131 kullanarak eylem ve dikkat aras\u0131nda bir ba\u011flant\u0131 da olu\u015fturabiliriz. \u0130deomotor prensibine g\u00f6re baz\u0131 n\u00f6ral yap\u0131lar\u0131n hem eylem hem de alg\u0131n\u0131n alt\u0131nda yatan n\u00f6ral devrelerin birer par\u00e7as\u0131 olduklar\u0131n\u0131 hat\u0131rlayal\u0131m. N ad\u0131nda bir n\u00f6ral yap\u0131n\u0131n, birinin \u00e7enesini ka\u015f\u0131mak \u00fczere oldu\u011funu alg\u0131lad\u0131\u011f\u0131mda ve kendi \u00e7enemi ka\u015f\u0131mak \u00fczere oldu\u011fumda aktifle\u015fti\u011fini varsayal\u0131m. Friston ve ark. (Friston ve ark. 2011<\/a>, syf. 138) makalesini takiben, N hem alg\u0131 hem de niyeti temsilen i\u015flev g\u00f6rebilir fakat \u00e7o\u011funlukla bunlar\u0131n yaln\u0131zca bir tanesi olarak i\u015flev g\u00f6r\u00fcr. Yani, e\u011fer ekopraksi hastas\u0131 de\u011filsem, bir hareketi alg\u0131lamak benim ayn\u0131 \u015fekilde hareket etmeme sebep olmayacak (tabii insanlar\u0131n birbirini belli \u00f6l\u00e7\u00fclerde taklit ettikleri durumlar da var, bkz. Quadt 2017<\/a>). \u00c7enemi ka\u015f\u0131mak \u00fczere oldu\u011fum hipotezinin proprioseptif[25]<\/a> duyu ve di\u011fer duyulara dair \u00f6ng\u00f6r\u00fclere (\u00f6rne\u011fin kolum hareket ederken kaslar\u0131m\u0131n durumunu a\u00e7\u0131klayan \u00f6ng\u00f6r\u00fcler) sebep olaca\u011f\u0131na dikkat \u00e7ekerek yukar\u0131da bahsetti\u011fimiz durumu \u00d6\u0130 \u00e7er\u00e7evesinden a\u00e7\u0131klayabiliriz. Ger\u00e7ekten \u00e7enemi ka\u015f\u0131m\u0131yorsam, bu hipotez duyusal sinyallerle \u00e7eli\u015fecek ve b\u00fcy\u00fck bir \u00f6ng\u00f6r\u00fc hatas\u0131 olu\u015facak; bu da \u00e7enemi ka\u015f\u0131d\u0131\u011f\u0131m hipotezinin g\u00fcncellenmesine yol a\u00e7acak. Di\u011fer bir deyi\u015fle, bu \u00f6ng\u00f6r\u00fc hatalar\u0131n\u0131n varl\u0131\u011f\u0131nda hipotez s\u00fcrd\u00fcr\u00fclemez. Yani hareketi sa\u011flayabilmek ad\u0131na duyusal \u00f6ng\u00f6r\u00fc hatalar\u0131yla e\u015fle\u015fen hassasiyet tahminlerinin, yukar\u0131dan a\u015fa\u011f\u0131ya mod\u00fclasyon ile ortadan kald\u0131r\u0131lmas\u0131 laz\u0131m. Bunu dikkatin hassasiyet tahminlerini g\u00fc\u00e7lendirdi\u011fi hipoteziyle birle\u015ftirdi\u011fimizde, bahsetti\u011fimiz mod\u00fclasyonu dikkatimizi somatosensoriyel<\/em>[26]<\/em><\/strong><\/a> sinyallerden uzakla\u015ft\u0131rma <\/em>olarak tan\u0131mlayabiliriz. Tersine, dikkatimizi duyusal uyaranlara y\u00f6neltmek de normal hareketi engellemeli (bkz. Limanowski 2017<\/a>).<\/p>\n\n\n\n

Eylem ve dikkat aras\u0131ndaki bu ba\u011flant\u0131dan ki\u015finin kendisini neden g\u0131d\u0131klayamad\u0131\u011f\u0131n\u0131 a\u00e7\u0131klama ba\u011flam\u0131nda da yararlan\u0131lm\u0131\u015f (bkz. Van Doorn ve ark. 2014<\/a> ve Van Doorn ve ark. 2015<\/a>). Hassasiyet tahminlerindeki sapmalar dikkat ve motor bozukluklar\u0131yla ba\u011fda\u015ft\u0131r\u0131lm\u0131\u015f ve otizm ile \u015fizofreni ba\u011flam\u0131nda dile getirilmi\u015f (bkz. Gonzalez-Gadea ve ark. 2015<\/a>; Palmer ve ark. 2015<\/a>, Van de Cruys ve ark. 2014<\/a>; Friston ve ark. 2014<\/a>, Adams ve ark. 2016<\/a>). Bu, \u00d6\u0130\u2019nin bili\u015fsel n\u00f6ropsikiyatri ve ili\u015fkili alanlarda yeni bulgular\u0131n ortaya \u00e7\u0131kabilmesi i\u00e7in ne \u00f6l\u00e7\u00fcde bir verimlili\u011fe sahip oldu\u011funa ve bulgular\u0131 a\u00e7\u0131klama g\u00fcc\u00fcne dair yaln\u0131zca bir \u00f6rnek.<\/p>\n\n\n\n

6. Bir \u201cHipotez Test Edici\u201d Olarak Beyin<\/strong><\/h1>\n\n\n\n

Yukar\u0131da bahsetti\u011fimiz \u00fczere dev bir robotun i\u00e7ine hapsolan bir insan (Dennett\u2019in d\u00fc\u015f\u00fcnce deneyini hat\u0131rlayal\u0131m) \u00e7evresi hakk\u0131nda hipotezler kurmak zorunda. Baz\u0131lar\u0131 \u00d6\u0130\u2019ye \u015f\u00fcpheyle yakla\u015fsa da baz\u0131lar\u0131na bu teorinin \u00e7ekici gelme sebebi, bunun gibi birey seviyesinde a\u00e7\u0131klamalar\u0131 komp\u00fctasyon seviyesindeki a\u00e7\u0131klamalara da uygulayabilmesi (kar\u015f\u0131la\u015ft\u0131rma i\u00e7in; 11. \u00f6zellik, Hohwy 2010<\/a>: \u201cBir Hipotez Test Edici Olarak Beyin (The hypothesis testing brain)\u201d adl\u0131 makale). Fakat bu tarz a\u00e7\u0131klamalar d\u00fcnyay\u0131 neden bu \u015fekliyle alg\u0131lad\u0131\u011f\u0131m\u0131z ve bilhassa alg\u0131sal organizasyonun formel prensiplerinin neler oldu\u011fu sorular\u0131na yan\u0131t verebilmemiz i\u00e7in en az\u0131ndan bulgusal olarak faydal\u0131 olabilir. Richard Gregory\u2019nin \u201cHipotezler Olarak Alg\u0131lar (Perceptions as Hypotheses)\u201d adl\u0131 klasik makalesi, alg\u0131lar\u0131n duyusal sinyalleri a\u00e7\u0131klad\u0131\u011f\u0131 <\/em>ve \u00f6ng\u00f6rme g\u00fcc\u00fcne <\/em>sahip oldu\u011fu fikrini inceliyor (Gregory 1980, syf. 182, 186<\/a>). Helmholtz bu fikrin daha da detayland\u0131r\u0131lm\u0131\u015f halini, yani hareketlerin (movement) deneyler olarak g\u00f6r\u00fclebilece\u011fi fikrini halihaz\u0131rda \u00f6nermi\u015f:<\/p>\n\n\n\n

\u201c[W]ir beobachten unter fortdauernder eigener Th\u00e4tigkeit, und gelangen dadurch zur Kenntniss des Bestehens eines gesetzlichen Verh\u00e4ltnisses zwischen unseren Innervationen und dem Pr\u00e4sent\u00adwerden der verschiedenen Eindr\u00fccke aus dem Kreise der zeitweiligen Pr\u00e4sentabilien. Jede unserer willk\u00fchrlichen Bewegungen, durch die wir die Erscheinungsweise der Objecte ab\u00e4ndern, ist als ein Experiment zu betrachten, durch welches wir pr\u00fcfen, ob wir das gesetzliche Verhalten der vorliegenden Erscheinung, d.h. ihr vorausgesetztes Bestehen in bestimmter Raumordnung, richtig aufgefasst haben.\u201d (Von Helmholtz 1959[1879\/1887], p. 39)[27]<\/a><\/p>\n\n\n\n

G\u00fcn\u00fcm\u00fczde ilgili tart\u0131\u015fmalara konu olan bir uygulama da \u015fu anda hipotez test etmenin bedenlenmi\u015f (embodied) bir versiyonu olarak kavramla\u015ft\u0131r\u0131lan sakkadik (s\u0131\u00e7ramal\u0131) g\u00f6z hareketleri (Friston ve ark. 2012<\/a>). Bunun gibi bulgusal d\u00fc\u015f\u00fcncelere ek olarak, e\u011fer \u00d6\u0130 bir teori olarak do\u011fru yoldaysa beynin ger\u00e7ekten<\/em> \u00e7\u0131karsama ile u\u011fra\u015f\u0131p u\u011fra\u015fmad\u0131\u011f\u0131n\u0131 sorabiliriz. Bu soru Alex Kiefer taraf\u0131ndan bu koleksiyondaki bir makalede olumlu cevaplan\u0131yor (bkz. Kiefer 2017<\/a>). Jelle Bruineberg ise daha \u015f\u00fcpheci bir pozisyonu benimsiyor (bkz. Bruineberg 2017<\/a> ve Bruineberg ve ark. 2016<\/a>). Halk psikolojisi (folk psychology) ve \u00d6\u0130\u2019nin birbiriyle nas\u0131l bir ba\u011flant\u0131 i\u00e7erisinde olduklar\u0131 ve halk psikolojisine dair kavramlar\u0131n bilimsel ba\u011flamda kullan\u0131mlar\u0131n\u0131n ne \u00f6l\u00e7\u00fcde yeniden g\u00f6zden ge\u00e7irilmesi gerekti\u011fine dair daha genel sorunlar Joe Dewhurst taraf\u0131ndan tart\u0131\u015f\u0131l\u0131yor (bkz. Dewhurst 2017<\/a>).<\/p>\n\n\n\n

7. \u00d6ng\u00f6rmeli \u0130\u015fleme ve Karl Friston\u2019\u0131n Serbest Enerji Prensibi<\/strong><\/h1>\n\n\n\n

\u015eu totolojiyi d\u00fc\u015f\u00fcnelim: Belli bir s\u00fcre boyunca canl\u0131 kalmay\u0131 ba\u015faran her organizma bu s\u00fcre zarf\u0131nda \u00f6lmemi\u015f olur. Dahas\u0131, canl\u0131 kalmak \u00f6lme riskini beraberinde getirir. Bu, ya\u015fam konseptiyle ilgili derin bir i\u00e7g\u00f6r\u00fc de\u011fil, sadece ya\u015fayan organizmalarla ilgili y\u00fczeysel bir yorum. Fakat bu yorum ilgin\u00e7 \u015feyler ima ediyor. Ya\u015fayan her organizman\u0131n hayatta kalmak i\u00e7in ka\u00e7\u0131nmas\u0131 gereken \u00f6l\u00fcmc\u00fcl durumlar var ve belki de bir organizma ne kadar kompleks ise bu durumlar\u0131n say\u0131s\u0131 o kadar fazla oluyor. Bunu anlamak i\u00e7in bir bakterinin hayatta kalabilece\u011fi \u00e7evreleri d\u00fc\u015f\u00fcnelim ve bunu insan\u0131nkilerle kar\u015f\u0131la\u015ft\u0131ral\u0131m. Bir organizma bir s\u00fcre hayatta kalm\u0131\u015fsa, o s\u00fcrenin sonuna kadar \u00f6l\u00fcmc\u00fcl durumlardan ka\u00e7\u0131nabilmi\u015f demektir.<\/p>\n\n\n\n

Bir organizman\u0131n kendisini i\u00e7inde bulabilece\u011fi durumlar\u0131n listesini yap\u0131p bunu organizman\u0131n hayatta kalabilece\u011fi durumlar\u0131n listesiyle kar\u015f\u0131la\u015ft\u0131r\u0131rsak iki \u015fey buluruz:<\/p>\n\n\n\n

1. \u00c7o\u011fu organizma i\u00e7in (\u00f6rne\u011fin insanlar) ikinci liste birinci listeden \u00e7ok daha k\u0131sa olacakt\u0131r (\u00e7\u00fcnk\u00fc \u00e7ok fazla \u00f6l\u00fcmc\u00fcl durum var).<\/p>\n\n\n\n

2. Hayatta kalmay\u0131 ba\u015farabilen bir organizmay\u0131 hat\u0131r\u0131 say\u0131l\u0131r bir s\u00fcre boyunca g\u00f6zlemlersek, rastgele bir anda onu muhtemelen ikinci listeden bir durumun i\u00e7erisinde buluruz (bu yukar\u0131da bahsetti\u011fimiz totolojinin bir yans\u0131mas\u0131).<\/p>\n\n\n\n

Bu iki g\u00f6zlemi biraz daha teknik bir anlat\u0131mla yeniden ifade edebiliriz. Organizman\u0131n i\u00e7inde bulunmas\u0131 m\u00fcmk\u00fcn olan t\u00fcm durumlar\u0131n olu\u015fturdu\u011fu sete bu organizman\u0131n durum uzay\u0131 <\/em>diyelim. Buradaki \u201cdurum\u201d, organizman\u0131n duyu sisteminin ald\u0131\u011f\u0131 mevcut duyusal sinyaller olarak tan\u0131mlan\u0131yor. Prensipte, bu uzay\u0131n farkl\u0131 k\u0131s\u0131mlar\u0131na olas\u0131l\u0131klar atayan ve organizmay\u0131 ya\u015fad\u0131\u011f\u0131 s\u00fcre boyunca bu k\u0131s\u0131mlarda bulman\u0131n ne \u00f6l\u00e7\u00fcde olas\u0131 oldu\u011funu belirten bir olas\u0131l\u0131k da\u011f\u0131l\u0131m\u0131 tan\u0131mlayabiliriz. Bu durumda baz\u0131 b\u00f6lgeler y\u00fcksek, baz\u0131 b\u00f6lgeler d\u00fc\u015f\u00fck olas\u0131l\u0131klara sahip olacak (\u00f6rne\u011fin bir bal\u0131\u011f\u0131 denizde bulmak olduk\u00e7a olas\u0131, karada bulmak ise hi\u00e7 olas\u0131 de\u011fil). Dahas\u0131, durum uzay\u0131n\u0131n \u00e7o\u011fu k\u0131sm\u0131 d\u00fc\u015f\u00fck olas\u0131l\u0131\u011fa sahip olacak (\u00e7\u00fcnk\u00fc \u00e7ok fazla \u00f6l\u00fcmc\u00fcl durum var). Bu formel olarak olas\u0131l\u0131k da\u011f\u0131l\u0131m\u0131n\u0131n entropisinin<\/em> d\u00fc\u015f\u00fck oldu\u011fu anlam\u0131na geliyor (durum uzay\u0131n\u0131n t\u00fcm b\u00f6lgeleri e\u015fit olas\u0131l\u0131\u011fa sahip olsayd\u0131 entropi[28]<\/a> en fazla olurdu; basit bir formel \u00f6rnek i\u00e7in paragraf\u0131n devam\u0131na bkz.). Elimizdeki bu olas\u0131l\u0131k da\u011f\u0131l\u0131m\u0131yla ya\u015fam\u0131 boyunca rastgele bir anda g\u00f6zlemleyece\u011fimiz bir organizman\u0131n, durum uzay\u0131n\u0131n hangi b\u00f6lgesinde bulunaca\u011f\u0131 \u00fczerine bir iddiada bulunabiliriz. Bahsetti\u011fimiz da\u011f\u0131l\u0131m durum uzay\u0131n\u0131n \u00e7o\u011fu b\u00f6lgesine \u00e7ok d\u00fc\u015f\u00fck olas\u0131l\u0131klar atayaca\u011f\u0131 i\u00e7in olduk\u00e7a hassas bir tahminde bulunabiliriz (\u00f6rne\u011fin bir bal\u0131\u011f\u0131n suda, bir tatl\u0131 su bal\u0131\u011f\u0131n\u0131n da tatl\u0131 suda olaca\u011f\u0131n\u0131 tahmin edebiliriz).<\/p>\n\n\n\n

\u015eimdi de ya\u015fam\u0131 boyunca bir bal\u0131\u011f\u0131n i\u00e7inde bulundu\u011fu durumlar\u0131n tekrarl\u0131 \u015fekilde \u00f6rneklerini toplad\u0131\u011f\u0131m\u0131z\u0131 ve bu \u00f6rnekleri kullanarak bir deneysel da\u011f\u0131l\u0131m <\/em>olu\u015fturdu\u011fumuzu d\u00fc\u015f\u00fcnelim. Deneysel da\u011f\u0131l\u0131m, \u00f6rneklerin her bir b\u00f6lgeden al\u0131nma frekanslar\u0131n\u0131 <\/em>yans\u0131tan olas\u0131l\u0131klar atar. Basit bir \u00f6rnek olarak bir butona bas\u0131ld\u0131\u011f\u0131nda 0 ya da 1 rakamlar\u0131ndan birini g\u00f6steren bir cihaz d\u00fc\u015f\u00fcnelim; bu numaralar butona basan ki\u015fi taraf\u0131ndan bilinmeyen olas\u0131l\u0131klarla g\u00f6steriliyor olsun. \u0130ki numara birbiriyle e\u015fit olas\u0131l\u0131klarla g\u00f6sterilebilir (0.5) ya da biri di\u011ferinden \u00e7ok daha y\u00fcksek olas\u0131l\u0131kla g\u00f6sterilebilir (\u00f6rne\u011fin 0.9 olas\u0131l\u0131kla 0 rakam\u0131, 0.1 olas\u0131l\u0131kla da 1 rakam\u0131 g\u00f6sterilebilir). Birinin butona bast\u0131\u011f\u0131 her anda, ba\u015fka biri hangi rakam\u0131n g\u00f6sterildi\u011fini not etsin (bir kere not edildi\u011finde bir \u00f6rnek elde edilmi\u015f olur). B\u00f6ylece her bir rakam\u0131n hangi s\u0131kl\u0131kta g\u00f6sterildi\u011fini sayarak, yani g\u00f6receli frekanslar\u0131 kullanarak, deneysel bir da\u011f\u0131l\u0131m olu\u015fturabiliriz. Mesela 100 \u00f6rnekten 14\u2019\u00fc 0 ise ve 86\u2019s\u0131 1 ise deneysel da\u011f\u0131l\u0131m 0 rakam\u0131na 0.14, 1 rakam\u0131naysa 0.86 olas\u0131l\u0131\u011f\u0131n\u0131 atar. Bu da\u011f\u0131l\u0131m\u0131n entropisi yakla\u015f\u0131k 0.58 olur.<\/p>\n\n\n\n

Fakat \u00f6ncelikle entropi nedir? Entropi, bu \u00f6rnekte cihaz\u0131n g\u00f6sterdi\u011fi farkl\u0131 \u00e7\u0131kt\u0131lar\u0131n yol a\u00e7t\u0131\u011f\u0131 ortalama s\u00fcrpriz. Burada s\u00fcrpriz, bir olay\u0131n olas\u0131l\u0131\u011f\u0131n\u0131n negatif logaritmas\u0131na kar\u015f\u0131l\u0131k gelen teknik bir kavram. Ortalama s\u00fcrpriz (yani entropi) \u015fu \u015fekilde hesaplan\u0131r: H = -[0.14*log(0.14) + 0.86*log(0.86)]. Bu say\u0131 d\u00fc\u015f\u00fck ise sebebi tekil \u00e7\u0131kt\u0131lar\u0131n (en az\u0131ndan bu \u00e7\u0131kt\u0131lar\u0131n \u00e7o\u011funun) s\u00fcrpriz de\u011ferlerinin d\u00fc\u015f\u00fck olmas\u0131 olur. Yani entropinin d\u00fc\u015f\u00fck olmas\u0131 i\u00e7in, herhangi bir anda (en az\u0131ndan \u00e7o\u011fu zaman) durumlar\u0131n yaratt\u0131\u011f\u0131 s\u00fcrprizlerin d\u00fc\u015f\u00fck olmas\u0131 gerekir.<\/p>\n\n\n\n

Bunu yine bal\u0131k \u00f6rne\u011fine uygulayabiliriz. Bal\u0131k \u00e7o\u011funlukla s\u00fcrpriz olmayan durumlarda bulunacak. Bal\u0131klar hakk\u0131nda say\u0131s\u0131z bilgimize dayanarak durum uzay\u0131nda bal\u0131\u011f\u0131n bulunmas\u0131 olas\u0131 b\u00f6lgeleri tan\u0131mlayabilir ve bu bilgiyi yans\u0131tan bir \u201ckoltuk\u201d da\u011f\u0131l\u0131m\u0131 olu\u015fturabiliriz. Veya bal\u0131\u011f\u0131 g\u00f6zlemleyip durum uzay\u0131n\u0131n farkl\u0131 b\u00f6lgelerinde g\u00f6receli bulunma s\u0131kl\u0131klar\u0131n\u0131 not edebiliriz. Uzun vadede deneysel da\u011f\u0131l\u0131m, \u201ckoltuk\u201d da\u011f\u0131l\u0131m\u0131na giderek daha \u00e7ok benzemeli. (Bu, ergodiklik[29]<\/a> varsay\u0131m\u0131n\u0131n, yani belirli rastgele s\u00fcre\u00e7lerin formel olarak tan\u0131mlanm\u0131\u015f bir \u00f6zelli\u011finin, formel olmayan bir a\u00e7\u0131klamas\u0131; bkz. Friston 2009<\/a>, syf. 293.)<\/p>\n\n\n\n

\u015eu ana kadar bal\u0131\u011f\u0131 d\u0131\u015far\u0131dan, g\u00f6zlemcinin perspektifinden g\u00f6zlemledik. Peki bak\u0131\u015f a\u00e7\u0131m\u0131z\u0131 de\u011fi\u015ftirip bal\u0131\u011f\u0131 \u201chayvan\u0131n perspektifinden (Eliasmith 2000<\/a>, syf. 25\u2019te adland\u0131r\u0131ld\u0131\u011f\u0131 gibi)\u201d g\u00f6zlemlersek ne olur? Buradaki kilit fark bal\u0131\u011f\u0131n \u015fu anki durumunu bile bilmememiz. Bir organizma o anki durumu hakk\u0131ndaki bilgiyi duyusal \u00f6l\u00e7\u00fcmlerden al\u0131r fakat bu \u00f6l\u00e7\u00fcmler ona ancak k\u0131smi ve belki de g\u00fcr\u00fclt\u00fcl\u00fc bilgiler verir. Dahas\u0131, organizman\u0131n i\u00e7inde bulundu\u011fu durumlar\u0131n s\u00fcrpriz miktar\u0131n\u0131 hesaplayabilmesine yarayacak olas\u0131l\u0131k da\u011f\u0131l\u0131m\u0131na eri\u015fimi yok. \u0130\u015fte burada serbest enerji prensibi (SEP) bir \u00e7\u00f6z\u00fcm sunuyor (12. \u00f6zellik).<\/p>\n\n\n\n

SEP\u2019in genel stratejisi iki ad\u0131mdan olu\u015fuyor. Birincisi i\u00e7sel olarak kodlanan bir olas\u0131l\u0131k da\u011f\u0131l\u0131m\u0131n\u0131 (bir tan\u0131ma da\u011f\u0131l\u0131m\u0131), gizli durumlar\u0131n duyusal sinyaller verildi\u011findeki ger\u00e7ek sonraki (posteriyor) da\u011f\u0131l\u0131m\u0131yla e\u015fle\u015ftirmeye \u00e7al\u0131\u015fmak. \u0130kincisi duyusal sinyalleri, duyusal ve gizli durumlar\u0131n s\u00fcrpriz miktarlar\u0131n\u0131n herhangi bir anda d\u00fc\u015f\u00fck olaca\u011f\u0131 \u015fekilde de\u011fi\u015ftirmeye \u00e7al\u0131\u015fmak. Bu, i\u015fleri daha da zorla\u015ft\u0131rabilir \u00e7\u00fcnk\u00fc \u015fimdi \u00f6n\u00fcm\u00fczde iki problem var: Tan\u0131ma da\u011f\u0131l\u0131m\u0131, bilinmeyen bir posteriyor da\u011f\u0131l\u0131mla nas\u0131l e\u015fle\u015ftirilebilir ve s\u00fcrpriz miktar\u0131n\u0131 hesaplamay\u0131 sa\u011flayan da\u011f\u0131l\u0131m bilinmiyorsa duyusal sinyallerin s\u00fcrpriz miktarlar\u0131 nas\u0131l minimize edilebilir? SEP\u2019in marifeti iki problemi de serbest enerjiyi minimize ederek \u00e7\u00f6zmesinde yat\u0131yor. Burada serbest enerji, bilgi teorisiyle ili\u015fkili bir nicelik ve bunu minimize etmek hayvan\u0131n perspektifinden de m\u00fcmk\u00fcn (detaylar i\u00e7in, bkz. Friston 2008<\/a>; Friston 2009<\/a>; Friston 2010<\/a>).<\/p>\n\n\n\n

Bunu anlatmak i\u00e7in biraz daha formel bir a\u00e7\u0131klamaya ihtiyac\u0131m\u0131z var (burada i\u015fleri olduk\u00e7a basitle\u015ftiriyoruz fakat serbest enerji prensibine dair bundan \u00e7ok daha detayl\u0131, yine de okunabilir, bir a\u00e7\u0131klama Bogacz 2015<\/a> makalesinde bulunabilir). \u00d6ncelikle tan\u0131ma da\u011f\u0131l\u0131m\u0131n\u0131 bilinmeyen bir posteriyor da\u011f\u0131l\u0131mla \u201ce\u015fle\u015ftirmek\u201d, yaln\u0131zca tan\u0131ma da\u011f\u0131l\u0131m\u0131n\u0131n belirli bir forma (\u00f6rne\u011fin Gauss da\u011f\u0131l\u0131m\u0131na) sahip oldu\u011funu varsayd\u0131\u011f\u0131m\u0131z Bayesci \u00e7\u0131karsamaya bir yak\u0131nsamadan ibaret. Bu da hesaplamalar\u0131 basitle\u015ftiriyor. \u0130kincisi, do\u011fru modele yak\u0131nsamay\u0131 hesaplad\u0131ktan sonra, serbest enerji duyusal sinyallerin s\u00fcrpriz miktar\u0131na kat\u0131 bir s\u0131n\u0131rlama getiriyor. B\u00f6ylece duyusal sinyalleri de\u011fi\u015ftirme yoluyla serbest enerjiyi minimize etmek, dolayl\u0131 yoldan s\u00fcrpriz miktar\u0131n\u0131 da minimize ediyor.<\/p>\n\n\n\n

SEP ile Bayesci \u00e7\u0131karsaman\u0131n aras\u0131ndaki ba\u011flant\u0131 a\u00e7\u0131k\u00e7a ortada: Serbest enerjiyi minimize etmek, tan\u0131ma da\u011f\u0131l\u0131m\u0131n\u0131 kullanarak posteriyor da\u011f\u0131l\u0131ma bir yak\u0131nsama yapmay\u0131 i\u00e7eriyor. Tan\u0131ma da\u011f\u0131l\u0131m\u0131n\u0131n bir Gauss da\u011f\u0131l\u0131m\u0131 (\u00e7an \u015fekline sahip \u00fcnl\u00fc olas\u0131l\u0131k yo\u011funluk fonksiyonu) oldu\u011funu varsayarsak, serbest enerjiyi minimize etmek ayn\u0131 zamanda hassasiyet tahminleriyle a\u011f\u0131rl\u0131kland\u0131r\u0131lan \u00f6ng\u00f6r\u00fc hatalar\u0131n\u0131 minimize etmeyi i\u00e7eriyor. Yani en az\u0131ndan bu varsay\u0131m alt\u0131nda (Laplace varsay\u0131m\u0131) SEP ile \u00f6ng\u00f6r\u00fc hatas\u0131 minimizasyonu aras\u0131nda da bir ba\u011flant\u0131 oluyor. Asl\u0131nda SEP\u2019e, \u00f6ng\u00f6rmeli i\u015flemenin yukar\u0131da anlat\u0131lan farkl\u0131 \u00f6zelliklerini tekil ve formel olarak titizce tan\u0131mlanm\u0131\u015f bir \u00e7er\u00e7evede birle\u015ftiren temel bir teori g\u00f6z\u00fcyle bak\u0131labilir. Fakat bu \u00f6zelliklerinin hangilerinin asl\u0131nda SEP\u2019in kapsam\u0131na girdi\u011fi tart\u0131\u015fma konusu. Yukar\u0131da ifade edildi\u011fi gibi \u00c7evresel Tecrit<\/strong>, tart\u0131\u015fmal\u0131 \u00f6zelliklerden bir tanesi (bkz. Fabry 2017a<\/a>; Clark 2017<\/a>). Dolay\u0131s\u0131yla bu yeni \u00f6nerinin \u00e7e\u015fitli boyutlar\u0131na sadece empirik de\u011fil, ayn\u0131 zamanda kavramsal ve meta-teorik bir perspektiften bakmak faydal\u0131 olabilir. Bu metin koleksiyonunun olu\u015fturulmas\u0131ndaki ana motivasyonlar\u0131m\u0131zdan bir tanesi de buydu.<\/p>\n\n\n\n

\"\"\/<\/figure>\n\n\n\n

\u015eekil 1. <\/strong>Serbest enerjiyi minimize etmenin dolayl\u0131 yoldan s\u00fcrpriz miktar\u0131n\u0131 da nas\u0131l minimize edebilece\u011fine dair \u015fematik bir g\u00f6sterim. Tan\u0131ma da\u011f\u0131l\u0131m\u0131, ba\u015flang\u0131\u00e7ta ger\u00e7ek posteriyor da\u011f\u0131l\u0131mla (gizli sebeplerin duyusal sinyaller verildi\u011findeki da\u011f\u0131l\u0131m\u0131) tam olarak e\u015fle\u015fmeyecek. Tan\u0131ma da\u011f\u0131l\u0131m\u0131n\u0131 geli\u015ftirmek i\u00e7in, onu \u00f6yle bir de\u011fi\u015ftirebiliriz ki \u00f6l\u00e7\u00fclen duyusal sinyaller bu da\u011f\u0131l\u0131m verildi\u011finde daha olas\u0131 olur (yani modele dair deliller artar). Bunu prati\u011fe d\u00f6kmenin bir yolu \u00f6ng\u00f6r\u00fc hatas\u0131n\u0131 minimize etmektir. Buradaki varsay\u0131m duyusal sinyallerin s\u00fcrpriz miktar\u0131n\u0131n d\u00fc\u015f\u00fck olmas\u0131; tan\u0131ma da\u011f\u0131l\u0131m\u0131 da bu durumu yans\u0131tmal\u0131 (yani tan\u0131ma da\u011f\u0131l\u0131m\u0131 \u00f6yle bir de\u011fi\u015fecek ki, bu da\u011f\u0131l\u0131ma g\u00f6re duyusal sinyallerin s\u00fcrpriz miktar\u0131 d\u00fc\u015f\u00fck olacak). Tabii ki bu durumda h\u00e2l\u00e2 duyusal sinyaller ger\u00e7ek posteriyor da\u011f\u0131l\u0131ma g\u00f6re s\u00fcrpriz verici olabilir. Bu y\u00fczden tan\u0131ma da\u011f\u0131l\u0131m\u0131n\u0131n test edilmesi gerekir. Bu da \u00e7evremizi de\u011fi\u015ftirerek dolayl\u0131 yoldan yap\u0131l\u0131r: Bu durumda e\u011fer uygun bir tan\u0131ma da\u011f\u0131l\u0131m\u0131m\u0131z varsa duyusal sinyallerin s\u00fcrpriz miktar\u0131 d\u00fc\u015f\u00fck olur. Buna aktif \u00f6rnekleme denir. Duyusal sinyaller her zaman belli bir \u00f6l\u00e7\u00fcde \u015fa\u015f\u0131rt\u0131c\u0131 olacakt\u0131r, dolay\u0131s\u0131yla tan\u0131ma da\u011f\u0131l\u0131m\u0131n\u0131n s\u00fcrekli ayarlanmas\u0131 gerekir. Bunu aktif \u00f6rnekleme takip eder, sonras\u0131nda tan\u0131ma da\u011f\u0131l\u0131m\u0131 yeniden ayarlan\u0131r vs. Yani bu \u00f6ny\u00fckleme (bootstrapping) s\u00fcreci s\u00fcrekli bir deneme yan\u0131lma protokol\u00fcyle \u00e7al\u0131\u015f\u0131r ve \u00f6zne ile \u00e7evresi aras\u0131ndaki s\u0131k\u0131 nedensellik ili\u015fkisine dayan\u0131r. \u015eekildeki siyah oklar zamansal bir sekansa i\u015faret etmek amac\u0131yla eklenmi\u015f olsa da alg\u0131sal \u00e7\u0131karsama ile aktif \u00e7\u0131karsama aras\u0131nda net bir zamansal ayr\u0131m olmak zorunda de\u011fil: \u00d6ny\u00fckleme s\u00fcreci bedensel hareketler ile de ba\u015flayabilir.<\/p>\n\n\n\n

S\u00f6zl\u00fck<\/strong><\/h1>\n\n\n\n

Aktif \u00e7\u0131karsama: <\/strong>1) \u00d6ng\u00f6r\u00fc hatas\u0131 minimizasyonunu, i\u00e7sel modelimizi de\u011fi\u015ftirerek (alg\u0131sal \u00e7\u0131karsama: \u201cmodeli d\u00fcnyaya uydurma\u201d) de\u011fil de d\u00fcnyam\u0131z \u00fczerinde eylemde bulunarak ger\u00e7ekle\u015ftirdi\u011fimiz (\u201cd\u00fcnyay\u0131 modele uydurma\u201d) komp\u00fctasyonel s\u00fcre\u00e7. 2) Ayn\u0131 zamanda hem eylem hem de alg\u0131n\u0131n alt\u0131nda yatan hesaplamalar\u0131n tamam\u0131 i\u00e7in kullan\u0131lan daha genel bir terim. SEP ba\u011flam\u0131nda da serbest enerjiyi minimize eden t\u00fcm komp\u00fctasyonel s\u00fcre\u00e7ler i\u00e7in kullan\u0131l\u0131yor.<\/p>\n\n\n\n

Bayesci \u00e7\u0131karsama: <\/strong>Modeli Bayes kural\u0131na g\u00f6re g\u00fcncelleme (yani posteriyor da\u011f\u0131l\u0131m\u0131 hesaplama: p(c|s) = p(s|c)*p(c)\/p(s). \u00d6rnek i\u00e7in (bkz. Harkness ve Keshava 2017<\/a>).<\/p>\n\n\n\n

Kar\u015f\u0131olgusal (Counterfactual) model<\/strong>[30]<\/strong><\/a>: <\/strong>Olas\u0131 eylemleri gelecekteki olas\u0131 durumlara ba\u011flayan bir ko\u015fullu olas\u0131l\u0131k da\u011f\u0131l\u0131m\u0131 (tan\u0131m Friston ve ark. 2012b<\/a> makalesi izlenerek yap\u0131lm\u0131\u015ft\u0131r).<\/p>\n\n\n\n

Tahmin edici: <\/strong>Bir istatistiksel tahmin edici, \u00f6rneklem olarak al\u0131nan bir grup rastgele de\u011fi\u015fkenin bir fonksiyonudur. Yani bir tahmin edici, g\u00f6zlemlenen verilerden bir tahminin nas\u0131l hesaplanaca\u011f\u0131n\u0131 belirler. Tahmin, bir tahmin edicinin ortaya \u00e7\u0131kard\u0131\u011f\u0131 belirli bir de\u011ferdir (bu de\u011fer belirli \u00f6rneklemler -yani rastgele de\u011ferlerin uygulamada kullan\u0131lmalar\u0131- elde edildi\u011finde hesaplan\u0131r).<\/p>\n\n\n\n

\u201c\u00d6rtbas etmek (Explaining away)\u201d: <\/strong>Bu metindeki ba\u011flam\u0131yla \u201c\u00f6rtbas etmek\u201d kavram\u0131 belirsiz bir kavram. 1) Baz\u0131 yazarlar yukar\u0131dan a\u015fa\u011f\u0131ya \u00f6ng\u00f6r\u00fclerle duyusal sinyallerin \u201c\u00f6rtbas edildi\u011fi\u201dni savunuyor (kar\u015f\u0131la\u015ft\u0131rma i\u00e7in Clark 2013a<\/a>, syf. 187). 2) Terimin kullan\u0131ld\u0131\u011f\u0131 bir di\u011fer ba\u011flam da birbiriyle yar\u0131\u015fan hipotez veya modellerin birbirini \u201c\u00f6rtbas etmesi\u201d (kar\u015f\u0131la\u015ft\u0131rma i\u00e7in Hohwy 2010<\/a>, syf. 137). 3) \u00dc\u00e7\u00fcnc\u00fc bir anlam ise \u00f6ng\u00f6r\u00fc hatas\u0131n\u0131n \u00f6rtbas edilmesi<\/em> (kar\u015f\u0131la\u015ft\u0131rma i\u00e7in Clark 2013a<\/a>, syf. 187).<\/p>\n\n\n\n

Serbest enerji: <\/strong>Friston\u2019\u0131n SEP teorisi ba\u011flam\u0131nda serbest enerji, termodinamik de\u011fil, bilgi teorisine ba\u011fl\u0131 bir niceliktir ve s\u00fcrpriz miktar\u0131n\u0131n yukar\u0131 limitini belirler. Bu s\u0131k\u0131 bir limitse, d\u00fcnyada de\u011fi\u015fiklik yaratmak suretiyle serbest enerjiyi minimize etti\u011fimizde duyusal sinyallerin s\u00fcrpriz miktarlar\u0131n\u0131 da azaltabiliriz.<\/p>\n\n\n\n

Gauss da\u011f\u0131l\u0131m\u0131: <\/strong>\u00c7an \u015feklindeki \u00fcnl\u00fc olas\u0131l\u0131k da\u011f\u0131l\u0131m\u0131 (normal da\u011f\u0131l\u0131m olarak da adland\u0131r\u0131l\u0131r). \u00d6nemi, bir\u00e7ok da\u011f\u0131l\u0131m\u0131n toplu olarak Gauss da\u011f\u0131l\u0131m\u0131yla temsil edilebilece\u011fini belirten merkezi limit teoreminden geliyor.<\/p>\n\n\n\n

Generatif model: <\/strong>\u0130ki ya da daha fazla rastgele de\u011fi\u015fkenin olu\u015fturdu\u011fu birle\u015fik olas\u0131l\u0131k da\u011f\u0131l\u0131m\u0131 (\u00c7o\u011funlukla \u00f6nceki (prior) ve olabilirlik (likelihood) terimleri cinsinden ifade ediliyor: p(s,c) = p(s|c)*p(c). Bazen olabilirlik (p(s|c)) terimi tek ba\u015f\u0131na \u201cgeneratif model\u201d olarak adland\u0131r\u0131l\u0131yor). Modele, duyusal sinyallerin (s) gizli sebepler \u00a9 taraf\u0131ndan nas\u0131l olu\u015fturuldu\u011funu modelliyor olu\u015fu sebebiyle generatif deniyor. Dahas\u0131, bu modeller gizli sebeplerin bir tahmini verildi\u011finde sahte duyusal sinyaller de olu\u015fturabiliyor.<\/p>\n\n\n\n

Hiyerar\u015fi: <\/strong>\u00d6\u0130, farkl\u0131 uzay ve zamansal \u00f6l\u00e7eklerde \u00e7al\u0131\u015fan bir tahmin ediciler hiyerar\u015fisini bar\u0131nd\u0131r\u0131r (b\u00f6ylece bu tahmin ediciler farkl\u0131 \u00f6l\u00e7eklerdeki \u00f6zellikleri takip eder). Hiyerar\u015finin illa bir tepe noktas\u0131 olmak zorunda de\u011fil fakat bir merkezi olabilir \u2014 hiyerar\u015fideki seviyeleri bir disk veya k\u00fcre \u00fczerindeki halkalar olarak d\u00fc\u015f\u00fcnebiliriz.<\/p>\n\n\n\n

Ters Problem: <\/strong>\u00d6ng\u00f6rmeli i\u015fleme perspektifinden bakt\u0131\u011f\u0131m\u0131zda alg\u0131 problemi, gizli sebeplerden duyusal sinyallere do\u011fru olan haritalamay\u0131 tersine \u00e7evirmeyi gerektirir. Problemin birden fazla \u00e7\u00f6z\u00fcm\u00fc oldu\u011fu ve duyusal sinyaller \u00e7o\u011funlukla g\u00fcr\u00fclt\u00fcl\u00fc oldu\u011fu i\u00e7in bu zor bir problemdir (yani gizli sebeplerden duyusal sinyallere do\u011fru olan haritalama deterministik de\u011fildir).<\/p>\n\n\n\n

\u00d6ng\u00f6r\u00fc: <\/strong>\u00d6ng\u00f6r\u00fc, bir tahminin, ba\u015fka bir tahmin (\u00f6ng\u00f6r\u00fclen tahmin) ile de kar\u015f\u0131la\u015ft\u0131r\u0131labilen, deterministik bir fonksiyonudur. \u00d6ng\u00f6r\u00fcler mutlaka gelecek hakk\u0131nda de\u011fildir (bir de\u011fi\u015fken di\u011fer bir de\u011fi\u015fken hakk\u0131nda bilgi ta\u015f\u0131yorsa, yani aralar\u0131nda bir korelasyon varsa, onu \u00f6ng\u00f6rebilir; kar\u015f\u0131la\u015ft\u0131rma i\u00e7in Anderson ve Chemero 2013<\/a>, syf. 204). Yine de \u00d6\u0130\u2019deki bir\u00e7ok tahmin, zamansal anlamda da \u00f6ng\u00f6rme i\u015flevi \u00fcstlenir (kar\u015f\u0131la\u015ft\u0131rma i\u00e7in Butz 2017<\/a>; Clark 2013c<\/a>, syf. 236).<\/p>\n\n\n\n

Hassasiyet: <\/strong>Rastgele bir de\u011fi\u015fkenin hassasiyeti, varyans\u0131n\u0131n tersidir. Di\u011fer bir deyi\u015fle ortalamadan sapmalar artt\u0131k\u00e7a, o rastgele de\u011fi\u015fkenin hassasiyeti d\u00fc\u015fer.<\/p>\n\n\n\n

Rastgele De\u011fi\u015fken: <\/strong>Olas\u0131l\u0131k uzay\u0131 ile \u00f6l\u00e7\u00fclebilir uzay aras\u0131nda \u00f6l\u00e7\u00fclebilen bir fonksiyon. \u00d6rne\u011fin alt\u0131 y\u00fczl\u00fc bir zar, alt\u0131 tane e\u015fit olas\u0131l\u0131kl\u0131 olay\u0131 {1,2,3,4,5,6} setindeki rakamlar\u0131n bir tanesine haritalayan bir rastgele de\u011fi\u015fken olarak modellenebilir.<\/p>\n\n\n\n

S\u00fcrpriz: <\/strong>Bir model verildi\u011finde bir olay\u0131n ne kadar beklenmedik oldu\u011funu belirten, bilgi teorisine ait bir kavram. Tam olarak bir olay\u0131n olas\u0131l\u0131\u011f\u0131n\u0131n negatif logaritmas\u0131na denk geliyor. Ki\u015fi seviyesinin alt\u0131nda (subpersonal), bilgi teorisi ba\u011flam\u0131nda bahsedilen \u201cs\u00fcrpriz\u201d konseptiyle ki\u015fi seviyesindeki fenomenolojik \u201cs\u00fcrpriz\u201d kavram\u0131n\u0131 kar\u0131\u015ft\u0131rmamak \u00f6nemli.<\/p>\n\n\n\n

Notlar<\/strong><\/h1>\n\n\n\n

[1]<\/a> Burada Helmholtz\u2019un, alg\u0131n\u0131n bilin\u00e7d\u0131\u015f\u0131 \u00e7\u0131karsamalar\u0131n bir sonucu oldu\u011fu y\u00f6n\u00fcndeki \u00fcnl\u00fc fikrine bir g\u00f6nderme beklenebilir \u2014 a\u015fa\u011f\u0131da bu kesite de\u011finece\u011fiz. Helmholtz\u2019un alg\u0131 hakk\u0131ndaki g\u00f6r\u00fc\u015f\u00fc \u00f6nemli \u00f6l\u00e7\u00fcde Kant\u2019tan etkilenmi\u015ftir (her ne kadar Helmholtz \u00f6\u011frenme ve deneyimin rol\u00fcn\u00fc Kant\u2019a g\u00f6re daha fazla vurgulam\u0131\u015f gibi g\u00f6z\u00fckse de, bkz. Lenoir 2006<\/a>, syf. 201&203). \u201cDass die Art unserer Wahrnehmungen ebenso sehr durch die Natur unserer Sinne, wie durch die \u00e4usseren Dinge bedingt sei, wird durch die angef\u00fchrten Thatsachen sehr augenscheinlich an das Licht gestellt, und ist f\u00fcr die Theorie unseres Erkenntnissverm\u00f6gens von der h\u00f6chsten Wichtigkeit. Gerade dasselbe, was in neuerer Zeit die Physiologie der Sinne auf dem Wege der Erfahrung nachgewiesen hat, suchte Kant schon fr\u00fcher f\u00fcr die Vorstellungen des menschlichen Geistes \u00fcberhaupt zu thun, indem er den Antheil darlegte, welchen die besonderen eingeborenen Gesetze des Geistes, gleichsam die Organisation des Geistes, an unseren Vorstellungen haben.\u201d (Von Helmholtz 1855, p. 19). (Bizim \u00e7evirimiz: Bu ger\u00e7ekler, alg\u0131m\u0131z\u0131n do\u011fas\u0131n\u0131n d\u0131\u015fsal objeler kadar bizim duyular\u0131m\u0131z\u0131n do\u011fas\u0131yla da s\u0131n\u0131rland\u0131r\u0131ld\u0131\u011f\u0131n\u0131 g\u00f6steriyor. Bu, epistemik yetilerimiz hakk\u0131ndaki bir teori i\u00e7in azami \u00f6neme sahiptir. Kant\u2019\u0131n daha \u00f6nce zihnimizin do\u011fu\u015ftan var olan kurallar\u0131n\u0131n -bir bak\u0131ma zihnimizin organizasyonunun- fikirlerimize verdi\u011fi katk\u0131y\u0131 a\u00e7\u0131klamak suretiyle insan zihninin \u00fcretti\u011fi fikirler hakk\u0131nda de\u011findi\u011fi noktay\u0131, duyular\u0131m\u0131z\u0131n fizyolojisi de deneyim yoluyla g\u00f6sterdi. \u00d6\u0130\u2019nin Kant\u00e7\u0131 temellerine dair genel bir de\u011ferlendirme i\u00e7in bkz. Swanson 2016<\/a>.<\/p>\n\n\n\n

[2]<\/a> \u201cSeziler nesnelerin yap\u0131s\u0131na g\u00f6re \u015fekillenmek zorundaysa, nesneler hakk\u0131nda apriori olarak herhangi bir \u015fey bilmemizin nas\u0131l m\u00fcmk\u00fcn olabilece\u011fini g\u00f6remiyorum; fakat nesne (duyular\u0131m\u0131z\u0131n nesnesi) sezi yetimizin yap\u0131s\u0131na g\u00f6re \u015fekilleniyorsa, bu olas\u0131l\u0131\u011f\u0131 kendime pek\u00e2l\u00e2 a\u00e7\u0131klayabilirim.\u201d (Kant 1998<\/a>, B xvii)<\/p>\n\n\n\n

[3]<\/a> (\u00c7.N) Burada Kant\u2019\u0131n \u201csezi\u201d (intuition) kelimesiyle ne kastetti\u011fine dair bir a\u00e7\u0131klama sunmak ad\u0131na \u015fu al\u0131nt\u0131y\u0131 payla\u015fmay\u0131 faydal\u0131 buluyorum: \u201cSezi; bili\u015fin hangi yolla ve ara\u00e7larla olursa olsun nesneyle kurdu\u011fu ili\u015fki, nesneyle direkt olarak ba\u011flant\u0131 kurmak ad\u0131na kulland\u0131\u011f\u0131 yol ve \u00fczerine t\u00fcm d\u00fc\u015f\u00fcncelerin birer ara\u00e7 olarak y\u00f6neltildi\u011fi \u015feydir. Fakat bu yaln\u0131zca nesnenin bize verildi\u011fi \u00f6l\u00e7\u00fcde ger\u00e7ekle\u015fir, nesnenin bize verilmesi de nesne zihni belli bir \u015fekilde etkileyebiliyorsa m\u00fcmk\u00fcnd\u00fcr. Nesnelerden etkilenme yolumuz sayesinde temsiller edinebilme kapasitemiz (al\u0131c\u0131l\u0131k) duyarl\u0131k (sensibility) olarak adland\u0131r\u0131l\u0131r. Dolay\u0131s\u0131yla nesneler bize duyarl\u0131k vas\u0131tas\u0131yla verilir ve bu sezi yetimizin var olmas\u0131 i\u00e7in yeterlidir. Fakat seziler hakk\u0131nda anlay\u0131\u015f (understanding) vas\u0131tas\u0131yla d\u00fc\u015f\u00fcn\u00fcr\u00fcz ve bu s\u00fcre\u00e7 sonucunda ortaya kavramlar \u00e7\u0131kar.\u201d (A 19\/B 33, ilk paragraf)<\/p>\n\n\n\n

[4]<\/a> Tabii ki Kant\u2019\u0131n kendisinin seziler (Anschauungen) \u00fczerindeki aktif (yukar\u0131dan a\u015fa\u011f\u0131) etkilerin ne derecede \u201cher zaman mevcut olan\u201d etkiler oldu\u011funu d\u00fc\u015f\u00fcnd\u00fc\u011f\u00fc sorulmaya de\u011fer bir soru. En az\u0131ndan Saf Akl\u0131n Ele\u015ftirisi\u2019nden baz\u0131 kesitler, Kant\u2019\u0131n kavray\u0131\u015f yetimiz (cognizing) (kavramlar\u0131n kendili\u011findenli\u011fi) taraf\u0131ndan uygulanan yukar\u0131dan a\u015fa\u011f\u0131 etkilere daha fazla \u00f6nem verdi\u011fini g\u00f6steriyor:<\/p>\n\n\n\n

\u201cUnsere Erkenntnis entspringt aus zwei Grundquellen des Gem\u00fcts, deren die erste ist, die Vorstellungen zu empfangen (die Rezeptivit\u00e4t der Eindr\u00fccke), die zweite das Verm\u00f6gen, durch diese Vorstellungen einen Gegenstand zu erkennen (Spontaneit\u00e4t der Begriffe); durch die erstere wird uns ein Gegen\u00adstand gegeben, durch die zweite wird dieser im Verh\u00e4ltnis auf jene Vorstellung (als blo\u00dfe Bestimmung des Gem\u00fcts) gedacht.\u201d (Kant 1998[1781\/87], B 74).\u201d<\/p>\n\n\n\n

\u201cBili\u015fimiz zihindeki iki temel kaynaktan do\u011far, birincisi temsillerin al\u0131nmas\u0131 (izlenimlerin al\u0131nabilirli\u011fi), ikincisi de bir nesneyi onun temsilleri yoluyla kavrayabilme yetisi (kavramlar\u0131n kendili\u011findenli\u011fi). \u0130lki arac\u0131l\u0131\u011f\u0131yla nesneler bize verilir, ikincisi arac\u0131l\u0131\u011f\u0131yla da bu temsillerle ili\u015fkisi \u00fczerinden d\u00fc\u015f\u00fcn\u00fcl\u00fcr (yaln\u0131zca zihnin bir saptamas\u0131 olarak). (Kant 1998<\/a>, B 74). Fakat bu sorunun ciddi bir incelemesi i\u00e7in, sezilerin olu\u015fumunda bilin\u00e7d\u0131\u015f\u0131 temsillerin etkisine odaklan\u0131lmal\u0131 (bkz. Giordanetti ve ark. 2012<\/a>).<\/p>\n\n\n\n

[5]<\/a> \u201cTemsil\u201d kelimesinin buradaki kullan\u0131m\u0131 tart\u0131\u015fmaya tamamen kapal\u0131 de\u011fil. \u00d6\u0130\u2019nin temsillere dair bir \u015fey \u00f6nerip \u00f6nermedi\u011fi ve e\u011fer \u00f6neriyorsa onlar\u0131 en iyi nas\u0131l tan\u0131mlayabilece\u011fimize dair tart\u0131\u015fmalar var (bkz. G\u0142adziejewski 2016<\/a>; Downey 2017<\/a>; Dolega 2017<\/a>). Fakat en az\u0131ndan \u00d6\u0130\u2019nin \u00f6nermelerine dair temsilci perspektiften yap\u0131lan a\u00e7\u0131klamalara temsilci (veya y\u00f6nelimsel) bir terminolojiyle yakla\u015fabiliriz (kar\u015f\u0131la\u015ft\u0131rma i\u00e7in Egan 2014<\/a>; Anderson 2017<\/a>). Sonu\u00e7 olarak baz\u0131lar\u0131 kar\u015f\u0131 \u00e7\u0131kacak olsa da biz \u00d6\u0130\u2019de yap\u0131lan tahminleri temsiller olarak tan\u0131mlamay\u0131 en az\u0131ndan bu ba\u015fucu metni \u00f6zelinde kullan\u0131\u015fl\u0131 buluyoruz (baz\u0131 yazarlar bu \u00f6nermelerin tam anlam\u0131yla temsiller olmad\u0131klar\u0131n\u0131 savunsa da).<\/p>\n\n\n\n

[6]<\/a> Buradaki tahminlerin hiyerar\u015fisi, bir hiyerar\u015fik \u00fcretken (generatif) model gerektiriyor. Generatif model, bir dizi rastgele de\u011fi\u015fkenin ortak da\u011f\u0131l\u0131m\u0131d\u0131r (s\u00f6zl\u00fck k\u0131sm\u0131na bak\u0131n\u0131z). Bir hiyerar\u015fik generatif model, -kom\u015fu olmayan seviyelerdeki de\u011fi\u015fkenlerin ko\u015fullu ba\u011f\u0131ms\u0131z (conditionally independent) oldu\u011fu- rastgele de\u011fi\u015fkenlerin hiyerar\u015fisine denk gelir (bu, \u00f6rne\u011fin nedensel ili\u015fkiye sahip nesne veya olaylar\u0131n hiyerar\u015fisini temsil edebilir, bkz. Drayson 2017<\/a>). \u00d6\u0130\u2019deki tahminler hiyerar\u015fisi, rastgele de\u011fi\u015fkenler hiyerar\u015fisindeki de\u011ferleri takip eder. Generatif modellerin bulgusal bir anlat\u0131s\u0131 (Clark, 2016<\/a>) kitab\u0131n\u0131n giri\u015f k\u0131sm\u0131ndan okunabilir. Bize generatif modellerden bahsetmeyi hat\u0131rlatan Chris Burr\u2019e minnettar\u0131z.<\/p>\n\n\n\n

[7]<\/a> Horn (Horn 1980, syf. 373), \u201cg\u00f6rmenin kontroll\u00fc bir hal\u00fcsinasyon oldu\u011fu\u201d fikrini Clowes\u2019a (Clowes 1971) atfediyor. Fakat Clowes\u2019un bu form\u00fclasyona en \u00e7ok yakla\u015ft\u0131\u011f\u0131 ifadesi \u201c\u0130nsanlar ne g\u00f6rmeyi bekliyorlarsa onu g\u00f6r\u00fcr\u201d (Clowes 1969, syf. 379; kar\u015f\u0131la\u015ft\u0131rma i\u00e7in Sloman 1984) gibi duruyor. Daha yak\u0131n zamanda benzer bir fikir Grush taraf\u0131ndan g\u00fcndeme getiriliyor (Grush 2004, syf. 395; o bu fikrin ortaya \u00e7\u0131k\u0131\u015f\u0131n\u0131 Ramesh Jain\u2019e atfediyor): \u201cDuyunun oynad\u0131\u011f\u0131 rol, bu temsilin \u015feklini ve evrimini s\u0131n\u0131rlamak. Bir parolaya indirgeyecek olursak, alg\u0131 kontroll\u00fc bir hal\u00fcsinasyon s\u00fcrecidir.\u201d<\/p>\n\n\n\n

[8]<\/a> (\u00c7.N) Bu terimin a\u00e7\u0131klamas\u0131 metnin ilerleyen k\u0131s\u0131mlar\u0131nda yap\u0131lacak. (bkz. 5. Dikkat ve Hassasiyet<\/strong>)<\/p>\n\n\n\n

[9]<\/a> Bu tan\u0131m\u0131n ilk iki maddesi a\u015fa\u011f\u0131 yukar\u0131 Clark\u2019\u0131n \u00f6nerdi\u011fi tan\u0131ma denk d\u00fc\u015f\u00fcyor (Clark 2013a<\/a>, syf. 202; Clark 2015<\/a>, syf. 5). Clark ayr\u0131ca (Clark 2013a<\/a>)\u2019da eylem-odakl\u0131 \u00d6\u0130\u2019yi tan\u0131t\u0131yor (bu da bizim tan\u0131m\u0131m\u0131z\u0131n son maddesini kaps\u0131yor). Bu \u00fc\u00e7 \u00f6zellik ayn\u0131 zamanda Hohwy\u2019nin \u00f6ng\u00f6r\u00fc hatas\u0131 minimizasyonu anlat\u0131s\u0131 i\u00e7in de merkezi bir konumda (bkz. Hohwy 2013<\/a>\u2019\u00fcn ilk d\u00f6rt b\u00f6l\u00fcm\u00fc).<\/p>\n\n\n\n

[10]<\/a> Buna a\u015fa\u011f\u0131da daha detayl\u0131 de\u011finece\u011fiz. SEP olmadan da \u00d6\u0130 tan\u0131mlar\u0131 geli\u015ftirebilece\u011fimize dikkat edelim (\u00d6\u0130, bir bak\u0131ma SEP\u2019ten ba\u011f\u0131ms\u0131z), fakat \u00d6\u0130, \u00f6ng\u00f6r\u00fc hatas\u0131 minimizasyonunu serbest enerjiyi minimize etmenin bir yolu olarak in\u015fa etmek suretiyle (bu da SEP\u2019in \u00f6zel bir durumu olur) SEP\u2019in i\u00e7erisine yerle\u015ftirilebilir (bkz. Friston ve Kiebel 2009<\/a>).<\/p>\n\n\n\n

[11]<\/a> \u201c\u00d6n\u00fcm\u00fczde belli bir yerde belli bir karaktere sahip belli bir nesne oldu\u011funu \u00e7\u0131karsamam\u0131z\u0131 sa\u011flayan zihinsel aktiviteler, genellikle bilin\u00e7li de\u011fil bilin\u00e7d\u0131\u015f\u0131 aktivitelerdir. Bu aktivitelerin sonu\u00e7lar\u0131; ancak duyular\u0131m\u0131z taraf\u0131ndan g\u00f6zlemlenen eylemler bize bu eylemlerin olas\u0131 sebepleri hakk\u0131nda bir fikir sahibi olma imk\u00e2n\u0131 tan\u0131d\u0131\u011f\u0131 \u00f6l\u00e7\u00fcde bir nihailik ta\u015f\u0131r. Asl\u0131nda direkt olarak alg\u0131lad\u0131\u011f\u0131m\u0131z \u015fey her zaman sinirsel uyar\u0131mlard\u0131r fakat bu uyar\u0131mlar d\u0131\u015fsal nesnelerin kendileri de\u011fil, eylemlerdir.\u201d (Von Helmholtz 1985[1925]<\/a>, syf. 4)<\/p>\n\n\n\n

[12]<\/a> \u0130\u015fte baz\u0131 \u00f6rnekler: \u201cG\u00f6rsel alg\u0131 s\u0131ras\u0131nda beynimizin, g\u00f6ze gelen \u0131\u015f\u0131\u011f\u0131n \u015fiddeti ve dalga boyunun uzamsal da\u011f\u0131l\u0131m\u0131 hakk\u0131ndaki bilgiye eri\u015fimi var. Beyin, bu bilgiden yola \u00e7\u0131karak alg\u0131lanan g\u00f6r\u00fcnt\u00fcy\u00fc (yani g\u00f6r\u00fcnt\u00fc olu\u015fturma s\u00fcrecinin sonucunu) olu\u015fturan nesnelerin dizili\u015fine (yani sebebe) dair \u00e7\u0131kar\u0131m yapmak zorunda.\u201d (Spratling 2016<\/a>, syf. 1 \u00f6nbask\u0131)<\/p>\n\n\n\n

\u201cBunlar\u0131n ilki (olas\u0131l\u0131ksal generatif modellerin alg\u0131 ve eylemde yukar\u0131dan a\u015fa\u011f\u0131 y\u00f6nde kullan\u0131l\u0131\u015f\u0131) olduk\u00e7a \u00f6nemli fakat ayn\u0131 zamanda soyut bir \u00f6neriyi i\u00e7eriyor: Alg\u0131 ve eylemin her ikisi de \u2018sentez arac\u0131l\u0131\u011f\u0131yla analiz\u2019 yapman\u0131n belli bir formuna dayan\u0131yor; g\u00f6zlemlenen duyusal veriler, bu verileri olu\u015fturabilecek gizli sebepler i\u00e7erisinden en iyi adaylar\u0131n bulunmas\u0131 sayesinde a\u00e7\u0131klan\u0131yor.\u201d (Clark 2013a<\/a>, syf. 234; fakat nitelikli bir g\u00f6r\u00fc\u015f i\u00e7in ayn\u0131 zamanda bkz. Clark (bas\u0131m a\u015famas\u0131nda)<\/a>)<\/p>\n\n\n\n

\u201cBenzer \u015fekilde, bilincin b\u00fct\u00fcnselli\u011fi (unity) kavram\u0131na \u00f6ng\u00f6r\u00fc hatas\u0131 \u00e7er\u00e7evesinden yakla\u015f\u0131rken kullan\u0131lan ba\u015flang\u0131\u00e7 noktas\u0131 da bahsetti\u011fimiz dolayl\u0131l\u0131k ile ili\u015fkili: Beyin, kafatas\u0131n\u0131n i\u00e7erisinden, kendisine ula\u015fan duyusal girdilerin gizli sebeplerini \u00e7\u0131karsamak durumunda.\u201d (Hohwy 2013<\/a>, syf. 220)<\/p>\n\n\n\n

[13]<\/a> Bu sebeple, bahsetti\u011fimiz problem k\u00f6t\u00fc tan\u0131mlanm\u0131\u015f bir problem olarak ele al\u0131nabilir (bkz. Spratling 2016<\/a>) fakat baz\u0131 yazarlar \u201cbu problemi nas\u0131l \u00e7\u00f6zece\u011fimizi bulma problemi\u201dni k\u00f6t\u00fc tan\u0131mlanm\u0131\u015f olarak ele al\u0131yorlar (bkz. Anderson 2017<\/a>).<\/p>\n\n\n\n

[14]<\/a> Burada \u201cmant\u0131\u011fa uygun \u015fekilde\u201d demek, olas\u0131l\u0131k aksiyomlar\u0131yla uyumlu olmak demek. Ko\u015fullu olas\u0131l\u0131\u011f\u0131n tan\u0131m\u0131 kullan\u0131larak, Bayesci \u00e7\u0131karsaman\u0131n ayn\u0131 zamanda enformasyon teorisi ba\u011flam\u0131nda da ideal oldu\u011fu g\u00f6sterilebilir (bkz. Zellner 1988<\/a>).<\/p>\n\n\n\n

[15]<\/a> (\u00c7.N) Birbiriyle e\u015f zamanl\u0131 da denebilir.<\/p>\n\n\n\n

[16]<\/a> (\u00c7.N) Renk ve \u015fekil birbirinden ko\u015fullu ba\u011f\u0131ms\u0131z de\u011filken; renk hakk\u0131ndaki bir bilgi (rengin lokasyonu) bana \u015fekil hakk\u0131nda yeni bir bilgi veriyordu (\u015feklin lokasyonu).<\/p>\n\n\n\n

[17]<\/a> (\u00c7.N) V\u00fccudun \u00e7evredeki stres unsurlar\u0131na kar\u015f\u0131 aksiyon alarak fizyolojik dengeyi korumas\u0131 olarak tan\u0131mlanabilir.<\/p>\n\n\n\n

[18]<\/a> Bu fikrin bir di\u011fer \u00f6nc\u00fcl\u00fc Herbart\u2019\u0131n (Herbart 1825<\/a>, syf. 464 f.) ve Lotze\u2019nin (Lotze 1852<\/a>, syf. 313 f.) \u00e7al\u0131\u015fmalar\u0131ndan okunabilir.<\/p>\n\n\n\n

[19]<\/a> \u0130deomotor yakla\u015f\u0131mlara dair bir derleme (Badets ve ark. 2014<\/a>) makalesinden okunabilir. Tarihsel bir genel bak\u0131\u015f da (Stock ve Stock 2004<\/a>) makalesinde bulunabilir.<\/p>\n\n\n\n

[20]<\/a> Asl\u0131nda ideomotor temsiller bazen alg\u0131ya hiyerar\u015finin y\u00fcksek k\u0131s\u0131mlar\u0131ndan gelen (ge\u00e7) katk\u0131lar ve eylemin (erken) \u00f6nc\u00fclleri olarak an\u0131l\u0131yor (\u015fimdiki al\u0131nt\u0131da \u201cEKT\u201d, eylem kodlama teorisini (EKT) temsil edecek): \u201cEKT, eylemlere yol a\u00e7acak olan \u2018erken\u2019 duyusal i\u015flemelerin kompleks mekanizmalar\u0131yla u\u011fra\u015fm\u0131yor. Di\u011fer taraftan eylemler konusundaki oda\u011f\u0131 da -\u00e7evrede olu\u015fturulacak olan olaylar\u0131n (=eylemler) belirli \u00f6zelliklerini temsil eden- eylemin \u2018erken\u2019 bili\u015fsel \u00f6nc\u00fcllerinde. EKT, eylemlerin ger\u00e7ekle\u015ftirilmesine hizmet eden \u2018ge\u00e7\u2019 motor s\u00fcre\u00e7lerin (eylemlerin kontrol ve koordinasyonu gibi) kompleks mekanizmalar\u0131yla da u\u011fra\u015fm\u0131yor. Dolay\u0131s\u0131yla EKT, (ge\u00e7) alg\u0131 ve (erken) eylem veya eylem planlama (action planning) aras\u0131ndaki ba\u011flant\u0131lar\u0131 anlamak i\u00e7in bir \u00e7er\u00e7eve sunuyor\u201d (Hommel ve ark. 2001<\/a>, syf. 849)<\/p>\n\n\n\n

[21]<\/a> Bu ayn\u0131 zamanda Holst ve Mittelstaedt\u2019in (Von Holst ve Mittelstaedt 1950<\/a>) \u201cyeniden i\u00e7eri iletme (reafference) prensibi\u201dyle de \u00f6rt\u00fc\u015f\u00fcyor. Bu prensibe g\u00f6re de alg\u0131ya yol a\u00e7an n\u00f6ral olaylar duyusal sinyallerin sadece etkilerinden de\u011fil, ayn\u0131 zamanda onlar\u0131n sebeplerinden de olu\u015fuyor \u00e7\u00fcnk\u00fc bu sebepler duyusal sinyalleri eylemler vas\u0131tas\u0131yla etkileyebiliyor.<\/p>\n\n\n\n

[22]<\/a> Ayr\u0131ca bkz. (Clark 2016<\/a>, syf. 181) ve (Burr 2017<\/a>).<\/p>\n\n\n\n

[23]<\/a> Ayn\u0131 fikir; Lake, Salakhutdinov ve Tenenbaum\u2019un konsept \u00f6\u011frenimi \u00fczerine yak\u0131n zamanl\u0131 \u00e7al\u0131\u015fmalar\u0131nda da kullan\u0131l\u0131yor: Sistem g\u00f6rsel karakterleri bir \u201colas\u0131l\u0131ksal program\u201d ile \u00e7\u0131kars\u0131yor ve bu program bir generatif model olup g\u00f6rsel girdinin olu\u015fturulmas\u0131nda da kullan\u0131labiliyor (kar\u015f\u0131la\u015ft\u0131rma i\u00e7in Lake ve ark. 2015<\/a>, syf. 1333)<\/p>\n\n\n\n

[24]<\/a> \u00d6rne\u011fin (Wiese 2016<\/a>), ideomotor teorisinin \u00d6\u0130 versiyonu do\u011fru yolda ise sistemsel yanl\u0131\u015f-temsillerin (misrepresentations) eyleme olanak tan\u0131mas\u0131 gerekti\u011fini savunuyor. (Colombo 2017<\/a>) ise \u00d6\u0130\u2019ye g\u00f6re sosyal motivasyonda arzular ve de\u011ferlerin \u00e7ekicili\u011fi de\u011fil de sosyal belirsizli\u011fi minimize etmenin \u00f6nemli olaca\u011f\u0131 sebebiyle \u00d6\u0130\u2019nin Hume\u2019cu motivasyon teorisiyle \u00e7at\u0131\u015ft\u0131\u011f\u0131n\u0131 savunuyor.<\/p>\n\n\n\n

[25]<\/a> (\u00c7.N) V\u00fccudumuzun pozisyonu ve hareketlerine dair duyu sistemi.<\/p>\n\n\n\n

[26]<\/a> (\u00c7.N) v\u00fccudumuzun herhangi bir yerinden gelen, makaledeki ba\u011flam\u0131yla kaslar\u0131m\u0131z\u0131n durumuna dair ald\u0131\u011f\u0131m\u0131z t\u00fcm duyusal girdiler olarak d\u00fc\u015f\u00fcn\u00fclebilir.<\/p>\n\n\n\n

[27]<\/a> S\u00fcrekli devam eden aktivitelerimiz aras\u0131nda sinir sistemimizi olu\u015fturan ba\u011flant\u0131lar\u0131m\u0131z ve ge\u00e7ici olarak bize sunulan farkl\u0131 izlenimler [Prasentabilien] aras\u0131nda belli kurallara ba\u011fl\u0131 bir ili\u015fki olup olmad\u0131\u011f\u0131 hakk\u0131nda g\u00f6zlemler yap\u0131p bilgi ediniyoruz. \u015eeylerin g\u00f6r\u00fcn\u00fcmlerini de\u011fi\u015ftirmemize arac\u0131l\u0131k eden t\u00fcm g\u00f6n\u00fcll\u00fc hareketlerimiz birer deney olarak g\u00f6r\u00fclmeli. Bu deneyler sayesinde elimizdeki \u015feyin g\u00f6r\u00fcn\u00fcm\u00fcn\u00fcn kurallara ba\u011fl\u0131 davran\u0131\u015flar\u0131n\u0131 -yani onun belirli uzamsal yap\u0131lar aras\u0131ndaki s\u00f6zde varl\u0131\u011f\u0131n\u0131- do\u011fru anlay\u0131p anlayamad\u0131\u011f\u0131m\u0131z\u0131 test ediyoruz.<\/p>\n\n\n\n

[28]<\/a> (\u00c7.N) Fizik alan\u0131ndaki anlam\u0131yla entropi: Evrendeki t\u00fcm madde ve enerjinin d\u00fczensizlik e\u011filimi.<\/p>\n\n\n\n

[29]<\/a> (\u00c7.N) Ergodiklik, bir sistemin davran\u0131\u015f\u0131n\u0131 ya da uzayda bulundu\u011fu t\u00fcm noktalar\u0131, o sistemin herhangi bir noktas\u0131n\u0131 se\u00e7erek a\u00e7\u0131klayabilmek olarak tan\u0131mlanabilir.<\/p>\n\n\n\n

[30]<\/a> Bu ve bunun gibi baz\u0131 kavramlar, s\u00f6zl\u00fckte tan\u0131mlanmalar\u0131na ra\u011fmen metinde ge\u00e7miyor. Fakat metinde bu kelimelerin ayn\u0131s\u0131 olmasa da benzer kelimelere yer verilmi\u015f.<\/p>\n\n\n\n

Kaynak\u00e7a<\/strong><\/h1>\n\n\n\n

Adams, R. A., Huys, Q. J. & Roiser, J. P. (2016). Compu- tational psychiatry: Towards a mathematically informed understanding of mental illness. J Neurol Neurosurg Psy- chiatry, 87 <\/em>(1), 53\u201363. https:\/\/dx.doi.org\/10.1136\/jnnp-2015\u2013310737.<\/a><\/p>\n\n\n\n

Anderson, M. L. (2017). Of Bayes and bullets: An embod- ied, situated, targeting-based account of predictive pro- cessing. In T. Metzinger & W. Wiese (Eds.) Philosophy and predictive processing<\/em>. Frankfurt am Main: MIND Group.<\/p>\n\n\n\n

Anderson, M. L. & Chemero, T. (2013). The problem with brain GUTs: Conflation of different senses of \u2018\u2019predic- tion\u2019\u2019 threatens metaphysical disaster. Behavioral and Brain Sciences, 36 <\/em>(3), 204\u2013205.<\/p>\n\n\n\n

Badets, A., Koch, I. & Philipp, A. M. (2014). A review of ideomotor approaches to perception, cognition, action, and language: Advancing a cultural recycling hypothe- sis. Psychological Research, 80 <\/em>(1), 1\u201315. https:\/\/dx.doi.<\/a> org\/10.1007\/s00426\u2013014\u20130643\u20138.<\/a><\/p>\n\n\n\n

Bastos, A. M., Usrey, W. M., Adams, R. A., Mangun, G. R., Fries, P. & Friston, K. J. (2012). Canonical microcircuits for predictive coding. Neuron, 76 <\/em>(4), 695\u2013711. https:\/\/<\/a> dx.doi.org\/10.1016\/j.neuron.2012.10.038.<\/a><\/p>\n\n\n\n

Bogacz, R. (2015). A tutorial on the free-energy frame- work for modelling perception and learning. Journal of Mathematical Psychology<\/em>. https:\/\/dx.doi.org\/10.1016\/j.<\/a> jmp.2015.11.003.<\/a><\/p>\n\n\n\n

Brodski, A., Paasch, G.-F., Helbling, S. & Wibral, M. (2015). The faces of predictive coding. The Journal of Neuros- cience, 35 <\/em>(24), 8997\u20139006. https:\/\/dx.doi.org\/10.1523\/<\/a> jneurosci.1529\u201314.2015.<\/a><\/p>\n\n\n\n

Brook, A. (2013). Kant\u2019s view of the mind and conscious- ness of self. In E. N. Zalta (Ed.) The Stanford encyclopedia of philosophy<\/em>.<\/p>\n\n\n\n

Bruineberg, J. (2017). Active inference and the primacy of the \u2018I can\u2019. In T. Metzinger & W. Wiese (Eds.) Philosophy and predictive processing<\/em>. Frankfurt am Main: MIND Group.<\/p>\n\n\n\n

Bruineberg, J., Kiverstein, J. & Rietveld, E. (2016). The an- ticipating brain is not a scientist: The free-energy prin- ciple from an ecological-enactive perspective. Synthese<\/em>, 1\u201328. https:\/\/dx.doi.org\/10.1007\/s11229-016-1239-1.<\/a><\/p>\n\n\n\n

Burr, C. (2017). Embodied decisions and the predictive brain. In T. Metzinger & W. Wiese (Eds.) Philosophy and predictive processing<\/em>. Frankfurt am Main: MIND Group.<\/p>\n\n\n\n

Butz, M. V. (2017). Which structures are out there? Learn- ing predictive compositional concepts based on social sensorimotor explorations. In T. Metzinger & W. Wiese (Eds.) Philosophy and predictive processing<\/em>. Frankfurt am Main: MIND Group.<\/p>\n\n\n\n

Clark, A. (2013a). Whatever next? Predictive brains, situ- ated agents, and the future of cognitive science. Behavi- oral and Brain Sciences, 36 <\/em>(3), 181\u2013204. https:\/\/dx.doi.<\/a> org\/10.1017\/S0140525X12000477.<\/a><\/p>\n\n\n\n

\u2014 \u2014 \u2014 <\/em>(2013b). The many faces of precision (Replies to com- mentaries on \u2018\u2019Whatever next? Neural prediction, situ- ated agents, and the future of cognitive science\u2019\u2019). Fron- tiers in Psychology, 4<\/em>, 270. https:\/\/dx.doi.org\/10.3389\/fpsyg.2013.00270.<\/a><\/p>\n\n\n\n

\u2014 \u2014 \u2014 <\/em>(2013c). Are we predictive engines? Perils, pros- pects, and the puzzle of the porous perceiver. Behavio- ral and Brain Sciences, 36 <\/em>(3), 233\u2013253. https:\/\/dx.doi.org\/10.1017\/S0140525X12002440.<\/a><\/p>\n\n\n\n

\u2014 \u2014 \u2014 <\/em>(2015). Radical predictive processing. The South- ern Journal of Philosophy<\/em>, 53, 3\u201327. https:\/\/dx.doi.org\/10.1111\/sjp.12120<\/a>.<\/p>\n\n\n\n

\u2014 \u2014 \u2014 <\/em>(2016). Surfing uncertainty: Prediction, action, and the embodied mind. <\/em>New York: Oxford University Press.<\/p>\n\n\n\n

\u2014 \u2014 \u2014 <\/em>(2017). How to knit your own Markov blanket: Re- sisting the second law with metamorphic minds. In T. Metzinger & W. Wiese (Eds.) Philosophy and predictive processing<\/em>. Frankfurt am Main: MIND Group.<\/p>\n\n\n\n

\u2014 \u2014 \u2014 <\/em>(in press). Busting out: Predictive brains, embod- ied minds, and the puzzle of the evidentiary veil. No\u00fbs<\/em>.<\/a> https:\/\/dx.doi.org\/10.1111\/nous.12140.<\/a><\/p>\n\n\n\n

Clowes, M. B. (1969). Pictorial relationships \u2014 A syntactic approach. In B. Meltzer & D. Michie (Eds.) (pp. 361\u2013 383). Edinburgh, UK: Edinburgh University Press.<\/p>\n\n\n\n

Colombo, M. (2017). Social motivation in computational neuroscience: Or if brains are prediction machines then the Humean theory of motivation is false. In J. Kiever- stein (Ed.) Routledge handbook of philosophy of the social mind. <\/em>Abingdon, OX \/ New York, NY: Routledge.<\/p>\n\n\n\n

Dennett, D. C. (2013). Intuition pumps and other tools for thinking. <\/em>New York, N.Y., and London, UK: W.W. Nor- ton & Company.<\/p>\n\n\n\n

Dewhurst, J. (2017). Folk psychology and the Bayesian brain. In T. Metzinger & W. Wiese (Eds.) Philosophy and predictive processing<\/em>. Frankfurt am Main: MIND Group.<\/p>\n\n\n\n

Downey, A. (2017). Radical sensorimotor enactivism & predictive processing. Providing a conceptual frame- work for the scientific study of conscious perception. In T. Metzinger & W. Wiese (Eds.) Philosophy and predictive processing<\/em>. Frankfurt am Main: MIND Group.<\/p>\n\n\n\n

Do\u0142ega, K. (2017). Moderate predictive processing. In T. Metzinger & W. Wiese (Eds.) Philosophy and predictive processing<\/em>. Frankfurt am Main: MIND Group.<\/p>\n\n\n\n

Drayson, Z. (2017). Modularity and the predictive mind. In T. Metzinger & W. Wiese (Eds.) Philosophy and predicti- ve processing<\/em>. Frankfurt am Main: MIND Group.<\/p>\n\n\n\n

Egan, F. (2014). How to think about mental content. Phi- losophical Studies, 170 <\/em>(1), 115\u2013135. https:\/\/dx.doi.<\/a> org\/10.1007\/s11098\u2013013\u20130172\u20130.<\/a><\/p>\n\n\n\n

Eliasmith, C. (2000). How neurons mean: A neurocomputa- tional theory of representational content. <\/em>PhD dissertation, Washington University in St. Louis. Department of Philosophy.<\/p>\n\n\n\n

Engel, A. K., Fries, P. & Singer, W. (2001). Dynamic predic- tions: Oscillations and synchrony in top-down processing. Nat Rev Neurosci, 2 <\/em>(10), 704\u2013716.<\/p>\n\n\n\n

Fabry, R. E. (2017a). Predictive processing and cognitive development. In T. Metzinger & W. Wiese (Eds.) Philosophy and predictive processing<\/em>. Frankfurt am Main: MIND Group.<\/p>\n\n\n\n

\u2014 \u2014 \u2014 <\/em>(2017b). Transcending the evidentiary boundary: Prediction error minimization, embodied interaction, and explanatory pluralism. Philosophical Psychology<\/em>, 1\u201320. https:\/\/dx.doi.org\/10.1080\/09515089.2016.1272674.<\/a><\/p>\n\n\n\n

Feldman, H. & Friston, K. J. (2010). Attention, uncertain- ty, and free-energy. Frontiers in Human Neuroscience, 4<\/em>. https:\/\/dx.doi.org\/10.3389\/fnhum.2010.00215.<\/a><\/p>\n\n\n\n

Friston, K. (2003). Learning and inference in the brain. Neural Networks, 16 <\/em>(9), 1325\u20131352.<\/p>\n\n\n\n

\u2014 \u2014 \u2014 <\/em>(2005). A theory of cortical responses. Philosophi- cal Transactions of the Royal Society B: Biological Scien- ces, 360 <\/em>(1456), 815\u2013836. https:\/\/dx.doi.org\/10.1098\/<\/a> rstb.2005.1622.<\/a><\/p>\n\n\n\n

\u2014 \u2014 \u2014 <\/em>(2008). Hierarchical models in the brain. PLoS Com- putational Biology, 4 <\/em>(11), e1000211. https:\/\/dx.doi.<\/a> org\/10.1371\/journal.pcbi.1000211.<\/a><\/p>\n\n\n\n

\u2014 \u2014 \u2014 <\/em>(2009). The free-energy principle: A rough guide to the brain? Trends in Cognitive Sciences, 13 <\/em>(7), 293\u2013301. https:\/\/dx.doi.org\/10.1016\/j.tics.2009.04.005.<\/a><\/p>\n\n\n\n

\u2014 \u2014 \u2014 <\/em>(2010). The free-energy principle: A unified brain theory? Nature Reviews Neuroscience, 11 <\/em>(2), 127\u2013138.<\/a> https:\/\/dx.doi.org\/10.1038\/nrn2787.<\/a><\/p>\n\n\n\n

Friston, K. & Buzs\u00e1ki, G. (2016). The functional anatomy of time: What and when in the brain. Trends in Cognitive Sciences, 20 <\/em>(7), 500\u2013511. https:\/\/dx.doi.org\/10.1016\/j.tics.2016.05.001.<\/a><\/p>\n\n\n\n

Friston, K. & Kiebel, S. (2009). Predictive coding under the free-energy principle. Philosophical Transactions of the Royal Society B: Biological Sciences, 364 <\/em>(1521), 1211\u20131221. https:\/\/dx.doi.org\/10.1098\/rstb.2008.0300.<\/a><\/p>\n\n\n\n

Friston, K. J. & Stephan, K. E. (2007). Free-energy and the brain. Synthese, 159 <\/em>(3), 417\u2013458. https:\/\/dx.doi.<\/a> org\/10.1007\/s11229\u2013007\u20139237-y.<\/a><\/p>\n\n\n\n

Friston, K., Mattout, J. & Kilner, J. (2011). Action under- standing and active inference. Biological Cybernetics, 104 <\/em>(1\u20132), 137\u2013160. https:\/\/dx.doi.org\/10.1007\/s00422-011\u20130424-z.<\/a><\/p>\n\n\n\n

Friston, K., Samothrakis, S. & Montague, R. (2012a). Ac- tive inference and agency: Optimal control without cost functions. Biological Cybernetics, 106 <\/em>(8), 523\u2013541. https:\/\/dx.doi.org\/10.1007\/s00422-012-0512-8.<\/a><\/p>\n\n\n\n

Friston, K., Adams, R., Perrinet, L. & Breakspear, M. (2012b). Perceptions as hypotheses: Saccades as exper- iments. Frontiers in Psychology, 3 <\/em>(151). https:\/\/dx.doi.<\/a> org\/10.3389\/fpsyg.2012.00151.<\/a><\/p>\n\n\n\n

Friston, K. J., Stephan, K. E., Montague, R. & Dolan, R. J. (2014). Computational psychiatry: The brain as a phantastic organ. The Lancet Psychiatry, 1 <\/em>(2), 148\u2013158. https:\/\/dx.doi.org\/10.1016\/S2215-0366(14)70275-5.<\/a><\/p>\n\n\n\n

Giordanetti, P., Pozzo, R. & Sgarbi, M. (2012). Kant\u2018s phi- losophy of the unconscious. <\/em>Berlin, Boston: De Gruyter.<\/p>\n\n\n\n

Gonzalez-Gadea, M. L., Chennu, S., Bekinschtein, T. A., Rattazzi, A., Beraudi, A., Tripicchio, P., Moyano, B., Soffita, Y., Steinberg, L., Adolfi, F., Sigman, M., Marino, J., Manes, F. & Ibanez, A. (2015). Predictive coding in autism spectrum disorder and attention deficit hyperactivity disorder. Journal of Neurophysiology, 114 <\/em>(5), 2625-2636. https:\/\/dx.doi.org\/10.1152\/jn.00543.2015.<\/a><\/p>\n\n\n\n

Gregory, R. L. (1980). Perceptions as hypotheses. Philoso- phical Transactions of the Royal Society of London. B, Bio- logical Sciences, 290 <\/em>(1038), 181\u2013197.<\/p>\n\n\n\n

Grush, R. (2004). The emulation theory of representation: Motor control, imagery, and perception. Behavioral and Brain Sciences, 27 <\/em>(3), 377\u2013396.<\/p>\n\n\n\n

G\u0142adziejewski, P. (2016). Predictive coding and repre- sentationalism. Synthese<\/em>, 559\u2013582. https:\/\/dx.doi.org\/10.1007\/s11229\u2013015\u20130762\u20139.<\/a><\/p>\n\n\n\n

Harkness, D. L. & Keshava, A. (2017). Moving from the what to the how and where \u2014 Bayesian models and pre- dictive processing. In T. Metzinger & W. Wiese (Eds.) Philosophy and predictive processing<\/em>. Frankfurt am Main: MIND Group.<\/p>\n\n\n\n

Herbart, J. F. (1825). Psychologie als Wissenschaft neu ge- gr\u00fcndet auf Erfahrung, Metaphysik und Mathematik. Zweiter, analytischer Teil. <\/em>K\u00f6nigsberg: Unzer.<\/p>\n\n\n\n

Hohwy, J. (2010). The hypothesis testing brain: Some phil- osophical applications. In W. Christensen, E. Schier & J. Sutton (Eds.) Proceedings of the 9th conference of the Australasian society for cognitive science <\/em>(pp. 135\u2013144). Macquarie Centre for Cognitive Science. https:\/\/dx.doi.org\/10.5096\/ASCS200922.<\/a><\/p>\n\n\n\n

\u2014 \u2014 \u2014 <\/em>(2012). Attention and conscious perception in the hy- pothesis testing brain. Frontiers in Psychology, 3<\/em>. https:\/\/dx.doi.org\/10.3389\/fpsyg.2012.00096.<\/a><\/p>\n\n\n\n

\u2014 \u2014 \u2014 <\/em>(2013). The predictive mind. <\/em>Oxford: Oxford Univer- sity Press.<\/p>\n\n\n\n

\u2014 \u2014 \u2014 <\/em>(2016). The self-evidencing brain. No\u00fbs, 50 <\/em>(2), 259\u2013285. https:\/\/dx.doi.org\/10.1111\/nous.12062.<\/a><\/p>\n\n\n\n

\u2014 \u2014 \u2014 <\/em>(2017). How to entrain your evil demon. In T. Metzinger & W. Wiese (Eds.) Philosophy and predictive processing<\/em>. Frankfurt am Main: MIND Group.<\/p>\n\n\n\n

Hommel, B. (2015). The theory of event coding (TEC) as embodied-cognition framework. Frontiers in Psychology, 6<\/em>. https:\/\/dx.doi.org\/10.3389\/fpsyg.2015.01318.<\/a><\/p>\n\n\n\n

Hommel, B., M\u00fcsseler, J., Aschersleben, G. & Prinz, W. (2001). The theory of event coding (TEC): A framework for perception and action planning. Behavioral and Brain Sciences, 24<\/em>, 849\u2013878. https:\/\/dx.doi.org\/10.1017\/<\/a> S0140525X01000103.<\/a><\/p>\n\n\n\n

Horn, B. K. P. (1980). Derivation of invariant scene characteristics from images <\/em>(pp. 371\u2013376). https:\/\/dx.doi.<\/a> org\/10.1145\/1500518.1500579.<\/a><\/p>\n\n\n\n

James, W. (1890). The principles of psychology. <\/em>New York: Henry Holt.<\/p>\n\n\n\n

Kant, I. (1998). Critique of pure reason. <\/em>Cambridge, MA: Cambridge University Press.<\/p>\n\n\n\n

\u2014 \u2014 \u2014 <\/em>(1998[1781\/87]). Kritik der reinen Vernunft. <\/em>Hamburg: Meiner.<\/p>\n\n\n\n

Kiefer, A. (2017). Literal perceptual inference. In T. Metzinger & W. Wiese (Eds.) Philosophy and predictive processing<\/em>. Frankfurt am Main: MIND Group.<\/p>\n\n\n\n

Lake, B. M., Salakhutdinov, R. & Tenenbaum, J. B. (2015). Human-level concept learning through probabilistic program induction. Science, 350 <\/em>(6266), 1332\u20131338. https:\/\/dx.doi.org\/10.1126\/science.aab3050.<\/a><\/p>\n\n\n\n

Lee, T. S. & Mumford, D. (2003). Hierarchical Bayesian inference in the visual cortex. J. Opt. Soc. Am. A, 20 <\/em>(7), 1434\u20131448. https:\/\/dx.doi.org\/10.1364\/<\/a> JOSAA.20.001434.<\/a><\/p>\n\n\n\n

Lenoir, T. (2006). Operationalizing Kant: Manifolds, mod- els, and mathematics in Helmholtz\u2019s theories of percep- tion. In M. Friedman & A. Nordmann (Eds.) The Kantian legacy in nineteenth-century science <\/em>(pp. 141\u2013210). Cambridge, MA: MIT Press.<\/p>\n\n\n\n

Limanowski, J. (2017). (Dis-)attending to the body. Action and self-experience in the active inference framework. In T. Metzinger & W. Wiese (Eds.) Philosophy and predictive processing<\/em>. Frankfurt am Main: MIND Group.<\/p>\n\n\n\n

Lotze, R. H. (1852). Medicinische Psychologie oder Physiol- ogie der Seele. <\/em>Leipzig: Weidmann\u2019sche Buchhandlung.<\/p>\n\n\n\n

Metzinger, T. (2004[2003]). Being no one: The self-model theory of subjectivity. <\/em>Cambridge, MA: MIT Press.<\/p>\n\n\n\n

\u2014 \u2014 \u2014 <\/em>(2017). The problem of mental action. Predictive con- trol without sensory sheets. In T. Metzinger & W. Wiese (Eds.) Philosophy and predictive processing<\/em>. Frankfurt am Main: MIND Group.<\/p>\n\n\n\n

Palmer, C. J., Paton, B., Kirkovski, M., Enticott, P. G. & Hohwy, J. (2015). Context sensitivity in action decreases along the autism spectrum: A predictive processing perspective. Proceedings of the Royal Society of Lon- don B: Biological Sciences, 282 <\/em>(1802). https:\/\/dx.doi.<\/a> org\/10.1098\/rspb.2014.1557.<\/a><\/p>\n\n\n\n

Prinz, W. (1990). A common coding approach to percep- tion and action. In O. Neumann & W. Prinz (Eds.) Re- lationships between perception and action <\/em>(pp. 167\u2013201). Berlin; Heidelberg: Springer.<\/p>\n\n\n\n

Quadt, L. (2017). Action-oriented predictive processing and social cognition. In T. Metzinger & W. Wiese (Eds.) Philosophy and predictive processing<\/em>. Frankfurt am Main: MIND Group.<\/p>\n\n\n\n

Seth, A. K. (2015). The cybernetic Bayesian brain: From interoceptive inference to sensorimotor contingen- cies. In T. Metzinger & J. M. Windt (Eds.) Open MIND<\/em>. Frankfurt am Main: MIND Group. https:\/\/dx.doi.<\/a> org\/10.15502\/9783958570108.<\/a><\/p>\n\n\n\n

Shi, Y. Q. & Sun, H. (1999). Image and video compression for multimedia engineering: fundamentals, algorithms, and standards. <\/em>Boca Raton, FL: CRC Press.<\/p>\n\n\n\n

Sloman, A. (1984). Experiencing omputation: A tribute to Max Clowes. In M. Yazdani (Ed.) New horizons in educa- tional computing <\/em>(pp. 207\u2013219). Chichester: John Wiley & Sons.<\/p>\n\n\n\n

Snowdon, P. (1992). How to interpret \u2018direct perception\u2019. In T. Crane (Ed.) The contents of experience <\/em>(pp. 48\u201378). Cambridge: Cambridge University Press.<\/p>\n\n\n\n

Spratling, M. W. (2016). A review of predictive cod- ing algorithms. Brain and Cognition<\/em>. https:\/\/dx.doi.<\/a> org\/10.1016\/j.bandc.2015.11.003.<\/a><\/p>\n\n\n\n

Stock, A. & Stock, C. (2004). A short history of ideo-mo- tor action. Psychological Research, 68<\/em>, 176\u2013188. https:\/\/<\/a> dx.doi.org\/10.1007\/s00426\u2013003\u20130154\u20135.<\/a><\/p>\n\n\n\n

Swanson, L. R. (2016). The predictive processing paradigm has roots in Kant. Frontiers in Systems Neuroscience, 10<\/em>, 79. https:\/\/dx.doi.org\/10.3389\/fnsys.2016.00079.<\/a><\/p>\n\n\n\n

Todorov, E. (2009). Parallels between sensory and motor information processing. In M. S. Gazzaniga (Ed.) The cognitive neurosciences. 4th edition <\/em>(pp. 613\u2013623). Cam- bridge, MA \/ London, UK: MIT Press.<\/p>\n\n\n\n

Van de Cruys, S., Evers, K., Van der Hallen, R., van Eylen, L., Boets, B., de-Wit, L. & Wagemans, J. (2014). Precise minds in uncertain worlds: Predictive coding in autism. Psychological Review, 121 <\/em>(4), 649\u2013675. https:\/\/dx.doi.org\/10.1037\/a0037665.<\/a><\/p>\n\n\n\n

Van Doorn, G., Hohwy, J. & Symmons, M. (2014). Can you tickle yourself if you swap bodies with someone else? Consciousness and Cognition, 23<\/em>, 1\u201311. http:\/\/dx.doi.org\/10.1016\/j.concog.2013.10.009<\/a>.<\/p>\n\n\n\n

Van Doorn, G., Paton, B., Howell, J. & Hohwy, J. (2015). Attenuated self-tickle sensation even under trajectory perturbation. Consciousness and Cognition, 36<\/em>, 147\u2013153. https:\/\/dx.doi.org\/10.1016\/j.concog.2015.06.016.<\/a><\/p>\n\n\n\n

Von Helmholtz, H. (1855). Ueber das Sehen des Menschen. <\/em>Leipzig: Leopold Voss.<\/p>\n\n\n\n

\u2014 \u2014 \u2014 <\/em>(1867). Handbuch der physiologischen Optik. <\/em>Leipzig: Leopold Voss.<\/p>\n\n\n\n

\u2014 \u2014 \u2014 <\/em>(1959[1879\/1887]). Die Tatsachen in der Wahrneh- mung. Z\u00e4hlen und Messen. <\/em>Darmstadt: Wissenschaftli- che Buchgesellschaft.<\/p>\n\n\n\n

\u2014 \u2014 \u2014 <\/em>(1985[1925]). Helmholtz\u2018s treatise on physiological op- tics. <\/em>Birmingham, AL: Gryphon Editions.<\/p>\n\n\n\n

Von Holst, E. & Mittelstaedt, H. (1950). Das Reafferenz- prinzip. Die Naturwissenschaften, 37 <\/em>(20), 464\u2013476.<\/p>\n\n\n\n

Wacongne, C., Labyt, E., van Wassenhove, V., Bekinschtein, T., Naccache, L. & Dehaene, S. (2011). Evidence for a hi- erarchy of predictions and prediction errors in human cortex. Proc Natl Acad Sci U S A, 108 <\/em>(51), 20754\u20139. https:\/\/dx.doi.org\/10.1073\/pnas.1117807108.<\/a><\/p>\n\n\n\n

Wiese, W. (2016). Action is enabled by systematic mis- representations. Erkenntnis<\/em>. https:\/\/dx.doi.org\/10.1007\/s10670\u2013016\u20139867-x.<\/a><\/p>\n\n\n\n

Zellner, A. (1988). Optimal information processing and Bayes\u2019s theorem. The American Statistician, 42 <\/em>(4), 278\u2013280. https:\/\/dx.doi.org\/10.2307\/2685143.<\/a><\/p>","protected":false},"featured_media":2026,"template":"","meta":{"_acf_changed":false},"event_publishing_tags":[83,280,286,94,93,378,518,82,691,64,377,323,221,217,281,285,74,220,219,741,282,218,740,294,287,77,66,65,467,960,959,482,75,177,957,958,466,471,96,176],"kategori":[305],"class_list":["post-2025","blog_content","type-blog_content","status-publish","has-post-thumbnail","hentry","event_publishing_tags-beyin","event_publishing_tags-bilgisayar","event_publishing_tags-bilgisayar-bilimi","event_publishing_tags-bilis","event_publishing_tags-bilissel-bilim","event_publishing_tags-bilissel-norobilim","event_publishing_tags-bilissel-sinirbilim","event_publishing_tags-brain","event_publishing_tags-cogist","event_publishing_tags-cognition","event_publishing_tags-cognitive-neuroscience","event_publishing_tags-cogsci","event_publishing_tags-computation","event_publishing_tags-computational-neuroscience","event_publishing_tags-computer","event_publishing_tags-computer-science","event_publishing_tags-felsefe","event_publishing_tags-hesaplama","event_publishing_tags-hesaplamali-norobilim","event_publishing_tags-hesaplamali-sinirbilim","event_publishing_tags-islemleme","event_publishing_tags-islemlemesel-norobilim","event_publishing_tags-islemlemesel-sinirbilim","event_publishing_tags-matematik","event_publishing_tags-mathematics","event_publishing_tags-mental","event_publishing_tags-neuroscience","event_publishing_tags-norobilim","event_publishing_tags-olasilik","event_publishing_tags-ongormeli-isleme","event_publishing_tags-ongormeli-kodlama","event_publishing_tags-philmind","event_publishing_tags-philosophy","event_publishing_tags-philosophy-of-mind","event_publishing_tags-predictive-coding","event_publishing_tags-predictive-processing","event_publishing_tags-probability","event_publishing_tags-sinirbilim","event_publishing_tags-zihin","event_publishing_tags-zihin-felsefesi","kategori-ceviri"],"acf":[],"_links":{"self":[{"href":"https:\/\/cog-ist.com\/en\/wp-json\/wp\/v2\/blog_content\/2025","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/cog-ist.com\/en\/wp-json\/wp\/v2\/blog_content"}],"about":[{"href":"https:\/\/cog-ist.com\/en\/wp-json\/wp\/v2\/types\/blog_content"}],"version-history":[{"count":0,"href":"https:\/\/cog-ist.com\/en\/wp-json\/wp\/v2\/blog_content\/2025\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/cog-ist.com\/en\/wp-json\/wp\/v2\/media\/2026"}],"wp:attachment":[{"href":"https:\/\/cog-ist.com\/en\/wp-json\/wp\/v2\/media?parent=2025"}],"wp:term":[{"taxonomy":"event_publishing_tags","embeddable":true,"href":"https:\/\/cog-ist.com\/en\/wp-json\/wp\/v2\/event_publishing_tags?post=2025"},{"taxonomy":"kategori","embeddable":true,"href":"https:\/\/cog-ist.com\/en\/wp-json\/wp\/v2\/kategori?post=2025"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}