{"id":1869,"date":"2021-01-17T15:00:45","date_gmt":"2021-01-17T15:00:45","guid":{"rendered":"https:\/\/cog-ist.com\/?post_type=blog_content&#038;p=1869"},"modified":"2025-08-18T16:57:09","modified_gmt":"2025-08-18T16:57:09","slug":"zeka-nedir-egor-dezhic","status":"publish","type":"blog_content","link":"https:\/\/cog-ist.com\/en\/blog_content\/zeka-nedir-egor-dezhic\/","title":{"rendered":"Zeka Nedir? \u2014 Egor Dezhic"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\">\u00d6zg\u00fcn ad\u0131: \u201c<a href=\"https:\/\/towardsdatascience.com\/what-is-intelligence-a69cbd8bb1b4\" target=\"_blank\" rel=\"noreferrer noopener\">What Is Intelligence<\/a>?\u201d<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"76cd\"><em>Yakla\u015f\u0131k 7 y\u0131l \u00f6nce hen\u00fcz lisedeyken, bir web geli\u015ftiricisi olarak \u00e7al\u0131\u015f\u0131p hobi olarak psikoloji okurken, yapay n\u00f6ral a\u011flarla ilgili bir makaleye denk geldim. Heyecan vericiydi. Makaleyi okumay\u0131 bitirir bitirmez, insan zekas\u0131 hakk\u0131nda daha \u00f6nce \u00f6\u011frendiklerimi a\u00e7\u0131klayacak ve bunlar\u0131 bir \u015fekilde yapay kar\u015f\u0131l\u0131klar\u0131yla ba\u011flant\u0131land\u0131racak bir zeka teorisi aramaya ba\u015flad\u0131m. Psikoloji, n\u00f6robilim, sibernetik, bili\u015fsel bilim, bilgisayar bilimi, biyoloji, kimya, fizik, teoloji, sosyoloji ve bir\u00e7ok di\u011fer alana bakt\u0131m. Binlerce makaleye ve \u00e7al\u0131\u015fmaya, y\u00fczlerce kitaba ve d\u00fczinelerce derse ra\u011fmen hala beni tatmin edecek bir cevap bulamad\u0131m. B\u00f6ylece, bu makale ortaya \u00e7\u0131kt\u0131.<\/em><\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"dfd0\">\u0130\u015fe ilk olarak zeka teriminin tan\u0131m\u0131n\u0131 Vikipedi\u2019den okumakla ba\u015flayal\u0131m; i\u015fin asl\u0131, bu \u015fimdiye kadar g\u00f6rd\u00fc\u011f\u00fcm en iyi tan\u0131mlardan biri:<\/p>\n\n\n\n<blockquote class=\"wp-block-quote is-layout-flow wp-block-quote-is-layout-flow\">\n<p class=\"wp-block-paragraph\" id=\"1849\">Enformasyonu alg\u0131lama veya enformasyondan anlam \u00e7\u0131karabilme ile enformasyonu \u00e7evre ya da ba\u011flamdaki adaptif davran\u0131\u015flara(adaptive behaviors) uygulanacak bilgi (knowledge) \u015feklinde muhafaza etme yetene\u011fi.<\/p>\n<\/blockquote>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"ecd8\">Bu, istihbarat te\u015fkilatlar\u0131n\u0131n (CIA &amp; Co.) zeka i\u00e7in genellikle kulland\u0131\u011f\u0131ndan \u00e7ok farkl\u0131 bir tan\u0131m. Onlar\u0131n tan\u0131m\u0131&nbsp;<em>veri<\/em>&nbsp;(data) tan\u0131m\u0131na, \u00f6zellikle de verinin hassas ve de\u011ferli olanlar\u0131na daha yak\u0131n.<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn1\" target=\"_blank\" rel=\"noopener\">[1]<\/a>&nbsp;Ancak, hem yapay hem de biyolojik zekan\u0131n tan\u0131m\u0131 istihbarat te\u015fkilatlar\u0131n\u0131nkinden tamam\u0131yla farkl\u0131d\u0131r ve genelde bir t\u00fcr \u00f6znenin<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn2\" target=\"_blank\" rel=\"noopener\">[2]<\/a>&nbsp;(agent) \u00f6zelli\u011fi olarak tan\u0131mlan\u0131r. Dahas\u0131, bu \u00f6zne \u00e7evresinden tamamen izole bir \u015fekilde var olamaz ve \u00f6yle veya b\u00f6yle bedenlenmi\u015ftir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"d03d\">Yapay \u00f6znelerin (artificial agents) genel bir a\u00e7\u0131klamas\u0131 ile ba\u015flayaca\u011f\u0131m, \u00e7\u00fcnk\u00fc yapaylar biyolojik olanlardan \u00e7ok daha basit. Sonras\u0131nda biyolojik olanlara ge\u00e7i\u015f yapaca\u011f\u0131m ve aralar\u0131ndaki ortakl\u0131klara bakaca\u011f\u0131m.<\/p>\n\n\n\n<h1 class=\"wp-block-heading\" id=\"a48a\">Yapay Zeka<\/h1>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"2f25\">Yapay zekal\u0131 ilk \u00f6znenin ne oldu\u011funu belirlemek zor, ama Babbage\u2019nin Analitik Motor\u2019u (Analytical Engine) muhtemelen en iyi aday. Hi\u00e7bir \u00f6zel \u201cadaptif\u201d yetene\u011fi yok ama olduk\u00e7a esnek. Maalesef teorik olarak inan\u0131lmaz g\u00fczel olmas\u0131na ra\u011fmen tek bir prototipi bile tamamlanmam\u0131\u015f.<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1400\/0*MoDMy9HMREZgX0Js.jpg\" alt=\"\"\/><figcaption class=\"wp-element-caption\">Analitik Motor\u2019un 1840\u2019lardan kalma bir plan \u015femas\u0131. Kaynak: Vikipedi<\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"07ed\">\u0130lgin\u00e7 bir \u015fekilde, bu makine Turing do\u011fmadan daha 50 y\u0131l \u00f6ncesinde Turing-tamd\u0131 (Turing-complete).<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn3\" target=\"_blank\" rel=\"noopener\">[3]<\/a>&nbsp;Teorik olarak, verileri olan herhangi bir hesaplanabilir fonksiyonu girdi olarak alabilir ve tamamen mekanik bir \u015fekilde \u00e7\u0131kt\u0131 \u00fcretebilir \u2014 ak\u0131ll\u0131 telefonunuzdan yakla\u015f\u0131k 1.000.000.000.000 kat daha yava\u015f \u015fekilde.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"33a8\">Bunun sonucunda, algoritmik karma\u015f\u0131kl\u0131k teorisi (algorithmic complexity theory)&nbsp;<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn4\" target=\"_blank\" rel=\"noopener\">[4]<\/a>&nbsp;geli\u015ftirildi ve evrensel bir bilgisayar (universal computer) in\u015fa etmenin nispeten kolay bir i\u015f oldu\u011fu geni\u015f kitlelerce anla\u015f\u0131ld\u0131. Ayr\u0131ca algoritmalar birden fazla \u015fekilde uygulanabilirdi. Son 70 y\u0131lda, bu s\u00fcre\u00e7te ortaya \u00e7\u0131kan teknolojik zorluklara ra\u011fmen, ayn\u0131 fiyata sat\u0131n alabilece\u011finiz i\u015flemleme (computation) miktar\u0131 her iki y\u0131lda bir, yakla\u015f\u0131k olarak iki kat\u0131na \u00e7\u0131km\u0131\u015ft\u0131r.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"3762\">Ba\u015fka bir deyi\u015fle, potansiyel olarak herhangi bir hareket tarz\u0131n\u0131 (policy)<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn5\" target=\"_blank\" rel=\"noopener\">[5]<\/a>&nbsp;hesaplama (calculate) kabiliyetine sahip bir Yapay Zeka (Artifical Intelligence, AI) \u00f6znesi in\u015fa etmek olduk\u00e7a kolay. Ancak bu kabiliyet, sa\u011flanan veriler\/girdiler ve i\u015fleme (process) i\u00e7in gereken zaman taraf\u0131ndan hep s\u0131n\u0131rland\u0131r\u0131lacakt\u0131r. Bu, ak\u0131llara \u015fu ilgin\u00e7 soruyu getiriyor: Her bilgisayar\u0131n yetene\u011fi mevcut veri taraf\u0131ndan s\u0131n\u0131rland\u0131r\u0131l\u0131yorsa, bu bilgisayarlardan herhangi birini zeki olarak kabul edebilir miyiz? Okuru birazc\u0131k bu konu hakk\u0131nda d\u00fc\u015f\u00fcnmeye davet ediyorum.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"b976\">Ba\u011flam\u0131 anlayabilmek i\u00e7in, k\u0131saca AI ara\u015ft\u0131rmas\u0131n\u0131n tarihini hat\u0131rlayal\u0131m. \u0130lk g\u00fcnlerinden beri yapay zeka ara\u015ft\u0131rmas\u0131, kabaca iki ayr\u0131 ekole b\u00f6l\u00fcnm\u00fc\u015ft\u00fc:&nbsp;<em>sembolik<\/em>&nbsp;(symbolic) ve&nbsp;<em>ba\u011flant\u0131c\u0131&nbsp;<\/em>(connectionist)<em>.&nbsp;<\/em>Sembolik yakla\u015f\u0131m, formel mant\u0131k ve elle girilen bilgiye (hand-crafted knowledge)<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn6\" target=\"_blank\" rel=\"noopener\">[6]<\/a>&nbsp;\u00e7ok \u00f6nem verir; ba\u011flant\u0131c\u0131 yakla\u015f\u0131m ise yapay n\u00f6ral a\u011flar\u0131n (artificial neural networks, ANNler) varyasyonlar\u0131na ve istatistiksel \u00f6\u011frenmeye vurgu yapar. Ge\u00e7mi\u015fleri kabaca \u015f\u00f6yledir:<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1400\/1*ojJ7Cw58qI_xybLSeCGqiQ.png\" alt=\"\"\/><figcaption class=\"wp-element-caption\">Ba\u011flant\u0131c\u0131 ve Sembolik AI yakla\u015f\u0131mlar\u0131n\u0131n pop\u00fclerlik kar\u015f\u0131la\u015ft\u0131rmas\u0131. Kaynak:&nbsp;<a href=\"https:\/\/neurovenge.antonomase.fr\/\" rel=\"noreferrer noopener\" target=\"_blank\">Neurons spike back<\/a>&nbsp;| CC-BY-4.0<\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"0ee0\">Hala g\u00fc\u00e7l\u00fc AI<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn7\" target=\"_blank\" rel=\"noopener\">[7]<\/a>&nbsp;ya da yapay genel zekan\u0131n (artificial general intelligence, AGI)<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn8\" target=\"_blank\" rel=\"noopener\">[8]<\/a>&nbsp;tek ba\u015far\u0131l\u0131 uygulamas\u0131 olarak kabul edilen insan beyni ve n\u00f6robilimle ili\u015fkisinden dolay\u0131, \u201cTuruncular\u201d ilk zamanlarda liderdi. Ancak ilk yapay sinir a\u011flar\u0131, ger\u00e7ek d\u00fcnya problemleri ile ba\u015fa \u00e7\u0131kmak i\u00e7in \u00e7ok k\u0131s\u0131tl\u0131yd\u0131. Bu a\u011flar\u0131n \u00e7o\u011fu do\u011falar\u0131 gere\u011fi do\u011frusald\u0131 (linear) ve \u00e7o\u011funun becerileri Marvin Minsky\u2019nin derinlemesine ele\u015ftirdi\u011fi \u00fczere \u00e7ok k\u0131s\u0131tl\u0131yd\u0131. Bu s\u0131rada \u201cMaviler\u201d, titiz bir matematiksel teori geli\u015ftirmi\u015f ve daha kullan\u0131\u015fl\u0131 \u015feyler \u00fcretebilmi\u015fti.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"f6df\">Elle girilen bilgi miktar\u0131 artt\u0131k\u00e7a, bu sistemleri geli\u015ftirmek giderek daha zor hale geldi. Girdi ve \u00e7\u0131kt\u0131 uzaylar\u0131n\u0131n<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn9\" target=\"_blank\" rel=\"noopener\">[9]<\/a>&nbsp;birka\u00e7 bin se\u00e7enekten daha \u00e7ok oldu\u011fu senaryolarda s\u00fcrekli olarak ba\u015far\u0131s\u0131z oldular. Ayr\u0131ca, t\u0131pk\u0131 hukukta genel kanunlar\u0131n birbiriyle \u00e7at\u0131\u015fmas\u0131 gibi, uzman sistemlerdeki (expert systems)<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn10\" target=\"_blank\" rel=\"noopener\">[10]<\/a>&nbsp;kurallar da birbiriyle \u00e7at\u0131\u015ft\u0131. \u00c7at\u0131\u015fmalar artt\u0131k\u00e7a s\u00fcrekli daha fazla \u201chakim\u201d gerekir oldu. Geli\u015fim \u00e7\u0131kmaza girdi.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"475c\">Bu arada \u201cTuruncular\u201d, makul bir s\u00fcre i\u00e7erisinde a\u011flar\u0131n\u0131 \u201ce\u011fitmek\u201d i\u00e7in yeteri kadar etiketli veri (labelled data)<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn11\" target=\"_blank\" rel=\"noopener\">[11]<\/a>&nbsp;ve i\u015flemlemesel (computational) kaynaklar edindiler. Bu da d\u00fcnyan\u0131n her yerinden ara\u015ft\u0131rmac\u0131lar\u0131n deney yapmas\u0131na olanak sa\u011fl\u0131yordu. Ancak, yapay n\u00f6ral a\u011flar\u0131n (ANNs) ilk ba\u015far\u0131s\u0131zl\u0131klar\u0131ndan sonra kitlelerin g\u00fcvenini kazanmak epey zaman ald\u0131. Ayn\u0131 durum, geli\u015ftiricilerin Boolean ve Crisp<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn12\" target=\"_blank\" rel=\"noopener\">[12]<\/a>&nbsp;kar\u015f\u0131l\u0131klar\u0131na z\u0131t d\u00fc\u015fen bulan\u0131k mant\u0131k (fuzzy logic)<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn13\" target=\"_blank\" rel=\"noopener\">[13]<\/a>&nbsp;ve istatistik fikirlerine al\u0131\u015fmas\u0131 i\u00e7in de ge\u00e7erliydi.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"68e9\">Ancak yapay n\u00f6ral a\u011flar\u0131 detaylar\u0131yla tart\u0131\u015fmadan \u00f6nce, pek \u00e7ok alanda yararl\u0131 oldu\u011fu kan\u0131tlanan birka\u00e7 ba\u015fka metodun \u00fcst\u00fcnde durmak istiyorum. Ayr\u0131ca karma (mixed) metotlarla kar\u015f\u0131la\u015ft\u0131r\u0131ld\u0131\u011f\u0131nda, yaln\u0131zca sembolik veya yaln\u0131zca ba\u011flant\u0131c\u0131 yakla\u015f\u0131mlar\u0131n performanslar\u0131n\u0131n geride kald\u0131\u011f\u0131 netle\u015fti. Burada yaln\u0131zca ana algoritma s\u0131n\u0131flar\u0131n\u0131 tasvir edece\u011fimi belitmeliyim, \u00e7\u00fcnk\u00fc t\u00fcm AI (yapay zeka) metotlar\u0131n\u0131n k\u0131sa tan\u0131mlar\u0131 bile en az birka\u00e7 kitap tutar.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"462b\">Bence, en dikkate de\u011fer algoritmalar karar a\u011fa\u00e7lar\u0131 (Decision trees), olas\u0131l\u0131ksal modeller (Probabilistic models) ve evrimsel algoritmalar.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"861f\"><strong>Karar a\u011fa\u00e7lar\u0131&nbsp;<\/strong>en basit ve en etkili algoritmalar aras\u0131ndad\u0131r. K\u0131saca,\u201d, her veri \u00f6zelli\u011finin s\u0131ras\u0131yla incelenmesi ve hangisinin belli \u00e7\u0131kt\u0131lara y\u00f6nelik en fazla \u00f6ng\u00f6r\u00fc kabiliyetine (predictive power) sahip oldu\u011funun bulunmas\u0131yla\u201c\u00f6\u011frenirler.\u201d. Rassal orman (Random Forests) gibi daha geli\u015fmi\u015f varyasyonlar\u0131nda daha karma\u015f\u0131k \u00f6\u011frenme teknikleri kullan\u0131l\u0131r ve bunlar ayn\u0131 modeldeki \u00e7ok say\u0131da a\u011fac\u0131 birle\u015ftirirler; \u00e7\u0131kt\u0131lar\u0131 \u201coylama\u201dyla elde edilir. Ancak, altta yatan ilkeleri ve sezileri (intuition) ayn\u0131 kal\u0131r.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"93a4\"><strong>Olas\u0131l\u0131ksal modeller&nbsp;<\/strong>istatistiksel metotlar\u0131 temsil eder ve yapay n\u00f6ral a\u011flarla pek \u00e7ok ortak \u00f6zellik ta\u015f\u0131rlar. \u0130ki modelin \u00f6\u011frenme\/optimizasyon y\u00f6ntemleri ve hatta notasyonlar\u0131 s\u0131kl\u0131kla ayn\u0131d\u0131r; ama olas\u0131l\u0131ksal modeller \u00e7o\u011fu zaman olas\u0131l\u0131ksal mant\u0131k (genellikle Bayesci olas\u0131l\u0131k<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn14\" target=\"_blank\" rel=\"noopener\">[14]<\/a>) dolay\u0131s\u0131yla k\u0131s\u0131tl\u0131d\u0131r. Bunun aksine, yapay n\u00f6ral a\u011flar\u0131n bu t\u00fcr ba\u011flant\u0131lar\u0131 olmayabilir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"112c\"><strong>Evrimsel i\u015flemleme (evolutionary computation)&nbsp;<\/strong>ilk olarak biyolojik evrimden, \u00f6zellikle de rastgele mutasyon (random mutation) ve ya\u015fam \u015fartlar\u0131na uygunluk (fitness) konular\u0131ndan ilham alm\u0131\u015ft\u0131r. Modifikasyonlar\u0131n genellikle rastgele oldu\u011funu d\u00fc\u015f\u00fcn\u00fcrsek, s\u0131n\u0131rland\u0131r\u0131lm\u0131\u015f g\u00fcr\u00fclt\u00fcn\u00fcn (restricted noise)<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn15\" target=\"_blank\" rel=\"noopener\">[15]<\/a>&nbsp;ne kadar da etkili olabildi\u011fi \u015fa\u015f\u0131rt\u0131c\u0131d\u0131r. Bu s\u0131n\u0131f pek \u00e7ok y\u00f6nden \u0131s\u0131yla DNA gev\u015fetme (annealing)<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn16\" target=\"_blank\" rel=\"noopener\">[16]<\/a>&nbsp;s\u00fcrecine benzeyen bir y\u00f6nlendirilmi\u015f arama (guided search) \u00e7e\u015fididir.<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1400\/0*KcIiWL_stYNqKaIj.gif\" alt=\"\"\/><figcaption class=\"wp-element-caption\">\u201cY\u00fcr\u00fcyen canl\u0131lar\u0131n\u201d evrimi. Kaynak:&nbsp;<a href=\"https:\/\/www.alanzucconi.com\/2016\/04\/06\/evolutionary-coputation-1\/\" rel=\"noreferrer noopener\" target=\"_blank\">alanzucconi.com<\/a><\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"6b0b\">Bu metotlar\u0131n t\u00fcm\u00fcn\u00fcn ortak bir y\u00f6n\u00fc var: Bunlar genellikle \u00e7ok geli\u015fmemi\u015f bir hareket tarz\u0131 (policy) ile yola koyulurlar; ama yava\u015f yava\u015f, bir t\u00fcr performans de\u011ferlendirme fonksiyonunda daha iyi bir skora ula\u015fmak i\u00e7in geli\u015firler.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"946c\">Bu g\u00fcnlerde, makine \u00f6\u011frenmesi teknikleri ve \u00f6zellikle&nbsp;<strong>derin \u00f6\u011frenme<\/strong>,<strong>&nbsp;<\/strong>yapay zeka ara\u015ft\u0131rmas\u0131n\u0131 ve geli\u015fimini domine etmekte. 1 ya da 2 ara soyutlama katman\u0131 (intermediate layers of abstraction)<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn17\" target=\"_blank\" rel=\"noopener\">[17]<\/a>&nbsp;kullanan makine \u00f6\u011frenmesi modellerinin (s\u0131\u011f modeller [shallow models] diye bilinen modellerin) aksine; derin \u00f6\u011frenme, istiflenmi\u015f y\u00fczlerce ve hatta binlerce e\u011fitilebilir katmana sahip olabilir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"a715\">Deneyler ger\u00e7ekle\u015ftirilmeden \u00f6nce, bu t\u00fcr derin a\u011flar\u0131 e\u011fitmek i\u00e7in tamamen yeni optimizasyon prosed\u00fcrlerinin ke\u015ffinin gerekti\u011fine yayg\u0131n \u015fekilde inan\u0131l\u0131yordu. Ancak sonras\u0131nda, bilindik geri yay\u0131l\u0131m\u0131n<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn18\" target=\"_blank\" rel=\"noopener\">[18]<\/a>&nbsp;(backpropagation) \u2013di\u011fer bir deyi\u015fle zincir kural\u0131n\u0131n- dereceli azalma<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn19\" target=\"_blank\" rel=\"noopener\">[19]<\/a>&nbsp;(gradient descent) ile birlikte bu i\u015fi \u00e7ok kolay bir \u015fekilde yerine getirebildi\u011fi g\u00f6r\u00fcld\u00fc. Bu algoritmalarsa y\u00fczy\u0131llar \u00f6nce matematik\u00e7ilerce biliniyordu. Dereceli azalman\u0131n (gradient descent) problemlerini \u00e7\u00f6zmek i\u00e7in daha modern algoritmalar (Adam ya da RMSProp gibi) tasarland\u0131; ama bunlar\u0131n, \u00e7o\u011fu ger\u00e7ek d\u00fcnya senaryosunda gereksiz oldu\u011fu ortaya \u00e7\u0131kt\u0131.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"905b\">N\u00f6ral a\u011flar\u0131n \u00f6\u011frenmesi k\u0131saca a\u011fa\u011f\u0131daki gibidir:<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"abb5\">1. Bir grup m\u00fcmk\u00fcn (possible) girdiyi al<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"71a9\">2. Her birinin kar\u015f\u0131l\u0131\u011f\u0131 olan \u00e7\u0131kt\u0131lar\u0131 hesapla<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"79ae\">3. Performans\u0131 hesapla<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"122c\">4. Parametrelerini ayarlamak i\u00e7in, hatalar\u0131 bir \u00f6nceki katmana yolla (ve a\u011fdaki her katman i\u00e7in bunu tekrarla)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"9394\">5. Yeterince iyi bir performans elde edene kadar, t\u00fcm m\u00fcmk\u00fcn girdi gruplar\u0131 i\u00e7in bunu tekrar et<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"d213\">Dereceli azalma tek makine \u00f6\u011frenmesi algoritmas\u0131 de\u011fildir, fakat makine algoritmalar\u0131n\u0131n ezici bir \u00e7o\u011funlu\u011funun alt\u0131nda ayn\u0131 ilkeler yatar: Hareket tarz\u0131 (policy) e\u015fle\u015ftirmelerini kullanarak hatalar\u0131 geri al ve hata olu\u015fmas\u0131n\u0131 minimize etmek i\u00e7in parametreleri ayarla. Bu yakla\u015f\u0131mda duyulan ana kayg\u0131, yayg\u0131n bir kesimin a\u011flar\u0131n b\u00f6lgesel minimumda (local minimum) s\u0131k\u0131\u015ft\u0131\u011f\u0131na ve olas\u0131 en iyi ortamda olmad\u0131\u011f\u0131na inanmas\u0131ndan kaynaklan\u0131r. Ancak, son zamanlardaki teorik ilerlemeler bir\u00e7ok sinir a\u011f\u0131n\u0131n ger\u00e7ekten de global minimuma ula\u015fabilece\u011fini g\u00f6steriyor.<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn20\" target=\"_blank\" rel=\"noopener\">[20]<\/a><\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1400\/0*3j53zr4mzgqCutcr.png\" alt=\"\"\/><figcaption class=\"wp-element-caption\">Dereceli Azalma \u00f6rne\u011fi. Kaynak:&nbsp;<a href=\"https:\/\/distill.pub\/2017\/momentum\/\" rel=\"noreferrer noopener\" target=\"_blank\">distill.pub<\/a><\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"c869\">Derin \u00f6\u011frenmedeki ba\u015fka bir ilgin\u00e7 empirik sonu\u00e7 ise \u00f6\u011frenmenin ciddi \u015fekilde parallelle\u015ftirilebilmesidir \u2014 ki bu&nbsp;<strong>da\u011f\u0131t\u0131lm\u0131\u015f \u00f6\u011frenme&nbsp;<\/strong>(distributed learning)<strong>&nbsp;<\/strong>olarak bilinir.<strong>&nbsp;<\/strong>Ayn\u0131 mimariyi ayn\u0131 anda birden \u00e7ok makinede e\u011fitir ve arada s\u0131rada dereceleri (gradients) makineler aras\u0131nda de\u011fi\u015ftirirseniz<strong>,&nbsp;<\/strong>h\u0131z\u0131 1000 kat\u0131n\u0131n \u00fcst\u00fcne \u00e7\u0131karabilirsiniz. H\u0131zlanma oran\u0131 do\u011frudan mevcut bilgisayarlar\u0131n say\u0131s\u0131na ba\u011fl\u0131d\u0131r.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"0b45\">Ayr\u0131ca, e\u011fitilmi\u015f katmanlar benzer g\u00f6revler i\u00e7in tekrardan kullan\u0131labilir.&nbsp;<strong>Transfer \u00f6\u011frenmesi<\/strong>&nbsp;(transfer learning) olarak bilinen bu fenomen, yapay n\u00f6ral a\u011flar\u0131n bu kadar pop\u00fcler olmas\u0131ndan b\u00fcy\u00fck miktarda sorumludur. \u00d6rne\u011fin, g\u00f6r\u00fcnt\u00fc s\u0131n\u0131fland\u0131r\u0131lmas\u0131 i\u00e7in e\u011fitilmi\u015f bir a\u011f, ba\u015fka bir bilgisayarl\u0131 g\u00f6rme (computer vision)<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn21\" target=\"_blank\" rel=\"noopener\">[21]<\/a>&nbsp;g\u00f6revi i\u00e7in daha sonra kullan\u0131labilir. Ayn\u0131 ilke do\u011fal dil i\u015fleme ve ba\u015fka alanlar i\u00e7in de ge\u00e7erlidir. Dahas\u0131, ayn\u0131 a\u011f, farkl\u0131 modalitelerin problemlerini \u00e7\u00f6zmek amac\u0131yla da kullan\u0131labilir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"0502\">T\u00fcm bunlar,&nbsp;<strong>peki\u015ftirmeli \u00f6\u011frenme<\/strong>&nbsp;(Reinforced Learning, RL) alan\u0131nda bir araya gelirler. Peki\u015ftirmeli \u00f6\u011frenmenin arkas\u0131ndaki bu ilk fikirler davran\u0131\u015f\u00e7\u0131 psikolojiden \u00f6d\u00fcn\u00e7 al\u0131nm\u0131\u015ft\u0131r. Davran\u0131\u015f\u00e7\u0131 psikolojide ara\u015ft\u0131rmac\u0131lar, \u00f6d\u00fcl\u00fcn \u00f6\u011frenmeyi nas\u0131l etkiledi\u011fi ve hayvanlar\u0131n davran\u0131\u015f\u0131n\u0131 nas\u0131l \u015fekillendirdi\u011fi \u00fczerinde \u00e7al\u0131\u015fm\u0131\u015ft\u0131r.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"2588\">Yapay zeka ara\u015ft\u0131rmac\u0131lar\u0131 i\u00e7in, RL metotlar\u0131 \u00f6zellikle ilgi \u00e7ekicidir; \u00e7\u00fcnk\u00fc bu metotlar, \u00f6\u011frenme i\u00e7in tamam\u0131 do\u011fru olan bir \u00e7\u0131kt\u0131lar seti gerektirmez. \u00d6rne\u011fin, robota nas\u0131l hareket etmesi gerekti\u011fini tam olarak g\u00f6stermek yerine, yaln\u0131zca robotu ne kadar uza\u011fa gitti\u011fine ya da ne kadar h\u0131zl\u0131 giti\u011fine ba\u011fl\u0131 olarak peki\u015ftirmeli \u00f6\u011frenme teknikleriyle \u00f6d\u00fcllendirebilirsiniz. Kalan\u0131n\u0131 robot kendi anlayacakt\u0131r. Ancak bu, uygulamada en zorlu \u00f6\u011frenme y\u00f6ntemidir ve genellikle g\u00f6receli olarak daha basit olan g\u00f6revleri d\u00fczg\u00fcnce haz\u0131rlamak bile pek \u00e7ok \u00e7aba gerektirir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"4b8a\">Ger\u00e7ek d\u00fcnya problemlerinde \u00e7evredeki \u00f6d\u00fclleri belirlemenin genellikle zor oldu\u011funu ve bug\u00fcnlerde ara\u015ft\u0131rmac\u0131lar\u0131n daha ziyade i\u00e7sel \u00f6d\u00fcl modelleme (internal reward modelling) \u00fczerine odakland\u0131\u011f\u0131n\u0131 vurgulamak isterim.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"2891\">RL ile paralel olarak ters peki\u015ftirmeli \u00f6\u011frenme (Inverse Reinforcement Learning) metotlar\u0131 geli\u015ftirilmi\u015ftir. Ters peki\u015ftirilmeli \u00f6\u011frenmede uzman taraf\u0131ndan \u00fcretilmi\u015f girdileri ve \u00e7\u0131kt\u0131lar\u0131 alan \u00f6zne (agent), davran\u0131\u015f\u0131n\u0131 y\u00f6nlendirecek bir \u00f6d\u00fcl fonksiyonuna yakla\u015fmaya \u00e7al\u0131\u015f\u0131r.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"2855\">Yukar\u0131da bahsedilen metotlar\u0131n yan\u0131 s\u0131ra, yapay genel zekaya y\u00f6nelik baz\u0131 ara\u015ft\u0131rmalar \u00f6nemli derecede farkl\u0131 temellere sahiptir. Bu yap\u0131lardan baz\u0131lar\u0131 kat\u0131 matematiksel teorilerden gelmekte, baz\u0131lar\u0131 da n\u00f6ronal devrelerden (neuronal circuits) etkilenmektedir. Baz\u0131lar\u0131ysa psikolojik modellere dayan\u0131r. Ancak \u00e7o\u011funun ortak \u00f6zelli\u011fi, pop\u00fcler muadillerinin ba\u015far\u0131s\u0131z oldu\u011fu y\u00f6nlere odaklanmalar\u0131. Benim alt\u0131n\u0131 \u00e7izmek istedi\u011fim yap\u0131lar HTM, AIXI, ACT-R ve SOAR.<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn22\" target=\"_blank\" rel=\"noopener\">[22]<\/a><\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"bbe7\"><strong>Hiyerar\u015fik Ge\u00e7ici Bellek&nbsp;<\/strong>(Hierarchical Temporal Memory, HTM) ile ba\u015flayal\u0131m. Ba\u015flang\u0131\u00e7ta neokorteks devre sisteminden (neocortex circuitry) esinlenen fikirlere dayan\u0131yordu; ama bu devrelerin hala yeterince anla\u015f\u0131lmad\u0131\u011f\u0131n\u0131 ve HTM\u2019nin yaln\u0131zca bu devre sisteminin kabaca bir benzeri g\u00f6revi g\u00f6rebilece\u011fini unutmamak gerekir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"70ef\">Ancak, HTM teorisinin merkezinde \u00f6zellikle \u00f6nemli bir kavram yatar \u2014&nbsp;<strong>Seyrek Da\u011f\u0131t\u0131lm\u0131\u015f Temsil<\/strong>&nbsp;(Sparse Distributed Representation) ya da k\u0131saca SDR. SDR, pratikte yakla\u015f\u0131k birka\u00e7 bin \u00f6geden olu\u015fan k\u00fc\u00e7\u00fck bir diziden ibarettir ve bu \u00f6geler, semantik olarak ili\u015fkili girdilerin, \u00e7ok say\u0131da \u00f6rt\u00fc\u015fen bit<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn23\" target=\"_blank\" rel=\"noopener\">[23]<\/a>&nbsp;ile SDR\u2019lere e\u015flenece\u011fi \u015fekilde olu\u015fturulmu\u015ftur. Bu, kavramsal olarak, n\u00f6ral a\u011flardan t\u00fcretilmi\u015f olan vekt\u00f6rle\u015ftirilmi\u015f temsillere (vectorized representations) benzer, ancak SDR\u2019nin vekt\u00f6rle\u015ftirilmi\u015f temsillerden temel fark\u0131 seyreklik ve a\u015f\u0131r\u0131 kapasitedir. Derin n\u00f6ral a\u011f (DNN) yak\u0131nsakl\u0131\u011f\u0131<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn24\" target=\"_blank\" rel=\"noopener\">[24]<\/a>&nbsp;ispat\u0131n\u0131n arkas\u0131ndaki temel varsay\u0131mlardan biri a\u011f\u0131n a\u015f\u0131r\u0131 parametrelenmesi (overparameterization) oldu\u011fu i\u00e7in, bu farklar \u00f6zellikle \u00f6ne \u00e7\u0131kmaktad\u0131r.<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1400\/0*Y_8v_k4feKv_DDWu.png\" alt=\"\"\/><figcaption class=\"wp-element-caption\">G\u00fcr\u00fclt\u00fc varl\u0131\u011f\u0131nda \u00f6rt\u00fc\u015fen SDR \u00f6rne\u011fi. Kaynak:&nbsp;<a href=\"https:\/\/numenta.com\/htmschool\/\" rel=\"noreferrer noopener\" target=\"_blank\">numenta.com<\/a><\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"e559\">HTM teorisinin sundu\u011fu di\u011fer fikirler bence o kadar da ilgin\u00e7 de\u011fil. \u0130nhibisyon (inhibition), y\u0131\u011f\u0131n normalle\u015ftirmeye (batch normalization)<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn25\" target=\"_blank\" rel=\"noopener\">[25]<\/a>&nbsp;ve baz\u0131 d\u00fczenleme (regularization) tekniklerine benziyor. Artt\u0131rma (boosting) ise makine \u00f6\u011frenmesinde g\u00f6rece eski bir kavram. Neokorteks daha karma\u015f\u0131k ba\u011flant\u0131sall\u0131k (connectivity) \u00f6r\u00fcnt\u00fclerine sahipken, hiyerar\u015fik yap\u0131 \u00e7ok kat\u0131 g\u00f6r\u00fcnmekte. Topoloji, n\u00f6ral a\u011flar mimarisinin yaln\u0131zca e\u015f anlaml\u0131s\u0131 gibi g\u00f6r\u00fcn\u00fcyor ve genel olarak bu teori, objeler aras\u0131ndaki ili\u015fkiye \u00e7ok daha az vurgu yaparken objelere daha \u00e7ok vurgu yap\u0131yor ve hatta SDR\u2019ler bile, aktivasyonlara ceza de\u011ferleri verilirken (penalizing) \u00e7ok say\u0131da n\u00f6ron kullan\u0131larak s\u0131radan yapay n\u00f6ral a\u011flar ile olu\u015fturulabiliyorlar. B\u00fct\u00fcn\u00fcyle, HTM hala di\u011fer makine \u00f6\u011frenmesi rakipleriyle uyumlu performansa ula\u015fmak i\u00e7in \u00e7ok fazla ince ayar gerektiriyor. Her neyse, Numenta\u2019n\u0131n (HTM\u2019yi geli\u015ftiren \u015firket) bu fikirlerin basit ve kolay anla\u015f\u0131l\u0131r a\u00e7\u0131klamalar\u0131 i\u00e7in sayg\u0131y\u0131 hak etti\u011fine inan\u0131yorum.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"0ac6\">S\u0131radaki \u201ckonu\u011fumuz\u201d&nbsp;<strong>AIXI<\/strong>.<strong>&nbsp;<\/strong>AIXI<strong>&nbsp;<\/strong>basitlik a\u00e7\u0131s\u0131ndan HTM\u2019ye g\u00f6re geri kalm\u0131\u015f; ama ona nazaran \u00e7ok daha sa\u011flam matematiksel temellere sahip.<strong>&nbsp;<\/strong>Ancak \u00f6nemli bir dezavantaj\u0131 var \u2014 i\u015flemlenebilir de\u011fil (uncomputable). Asl\u0131nda, \u00e7o\u011fu makine \u00f6\u011frenmesi algoritmas\u0131n\u0131n tam olarak i\u015flemlenebilmesi imk\u00e2ns\u0131z ve yak\u0131nsamalar (approximations) kullanmak zorunday\u0131z. Her neyse, bu yak\u0131nsamalar, uygulamada s\u0131kl\u0131kla i\u015fimize yar\u0131yor. AIXI bir sat\u0131rda \u015f\u00f6yle a\u00e7\u0131klanabilir:<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1400\/0*ZRevgtlOCdvbKxe_.png\" alt=\"\"\/><figcaption class=\"wp-element-caption\">Modelin, eylemleri (\u00e7\u0131kt\u0131lar\u0131), g\u00f6zlemleri (girdileri) ve \u00f6d\u00fclleri (girdinin spesifik bir b\u00f6l\u00fcm\u00fc olarak tasvir edilebilir) kullanarak etkile\u015fimde bulundu\u011fu bir \u00f6znesi (agent) ve \u00e7evresi (environment) var. \u00d6zne, d\u0131\u015far\u0131 bir eylem (a) g\u00f6nderiyor ve sonras\u0131nda ise \u00e7evre hem g\u00f6zlem (o) hem de bir \u00f6d\u00fcl (r) g\u00f6nderiyor; l(q) ifadesiyse \u00e7evrenin karma\u015f\u0131kl\u0131\u011f\u0131n\u0131 (complexity) g\u00f6stermekte. Bu s\u00fcre\u00e7 bu \u015fekilde hep devam ediyor. Kaynak:&nbsp;<a href=\"https:\/\/www.lesswrong.com\/posts\/i3BTagvt3HbPMx6PN\/embedded-agency-full-text-version\" rel=\"noreferrer noopener\" target=\"_blank\">lesswrong.com<\/a><\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"89a4\">AIXI pek \u00e7ok y\u00f6nden optimum bir yap\u0131 oldu\u011funu kan\u0131tlad\u0131 ve bence, AGI\u2019n\u0131n matematiksel olarak neye benzeyebilece\u011fi konusunda bug\u00fcn elimizdeki en iyi betimleme bu. Ayr\u0131ca, AIXI genel ama\u00e7l\u0131 (general purpose) bir peki\u015ftirmeli \u00f6\u011frenme \u00f6znesi (agent) ve bir\u00e7ok y\u00f6nden Schmidhuber taraf\u0131ndan geli\u015ftirilen&nbsp;<strong>G\u00f6del Makinesi<\/strong>\u2019ne benziyor. Ancak, her ikisi de yapay genel zekay\u0131 tasvir etmekle kalan modeller; onu ortaya \u00e7\u0131karacak form\u00fcller de\u011filler. Her neyse, yapay zeka ara\u015ft\u0131rmac\u0131lar\u0131 i\u00e7in b\u00fcy\u00fck ilham kaynaklar\u0131.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"be25\">Aksine ACT-R ya da (D\u00fc\u015f\u00fcncenin Adaptif Kontrol\u00fc-Rasyonel, Adaptive Control of Thought-Rational)<a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftn26\" target=\"_blank\" rel=\"noopener\">[26]<\/a>&nbsp;sadece bir teori de\u011fildir; ayn\u0131 zamanda LISP programlama dilinde yaz\u0131lm\u0131\u015f bir yaz\u0131l\u0131m \u00e7er\u00e7evesidir. Geli\u015fimi, di\u011fer programlama dillerindeki yan \u00fcr\u00fcnleri (spin-off) ve orijinal modelin de\u011fi\u015ftirilmi\u015f versiyonlar\u0131yla birlikte onlarca y\u0131ll\u0131k bir ge\u00e7mi\u015fe sahiptir.<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1400\/0*-TpZm64uuv8KiZ1T.jpg\" alt=\"\"\/><figcaption class=\"wp-element-caption\">Kaynak:&nbsp;<a href=\"https:\/\/towardsdatascience.com\/what-is-intelligence-a69cbd8bb1b4\" rel=\"noreferrer noopener\" target=\"_blank\">towardsdatascience.com<\/a><\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"e7b0\">ACT-R \u00e7o\u011funlukla farkl\u0131 t\u00fcrdeki bellekler \u00fczerine odaklan\u0131r; i\u00e7indeki verinin d\u00f6n\u00fc\u015f\u00fcmlerine ise daha az odaklanmaktad\u0131r. \u0130nsan zihninin i\u015flemlemesel bir modeli olarak geli\u015ftirilmi\u015ftir ve belli bir dereceye kadar ba\u015far\u0131l\u0131 olmu\u015ftur. fMRI g\u00f6r\u00fcnt\u00fcleme sonu\u00e7lar\u0131n\u0131 tahmin etmek i\u00e7in uygulanm\u0131\u015f ve buna ek olarak bellek \u00fczerine ger\u00e7ekle\u015fen baz\u0131 psikolojik deneylerde kullan\u0131lm\u0131\u015ft\u0131r. Ancak, prati\u011fe y\u00f6nelik uygulamalarda s\u00fcrekli olarak ba\u015far\u0131s\u0131z olmu\u015ftur ve sadece ara\u015ft\u0131rmac\u0131lar\u0131n kulland\u0131\u011f\u0131 bir ara\u00e7 olmaktan ileri ge\u00e7ememi\u015ftir. D\u00f6rt bili\u015fsel mimariden biri olan&nbsp;<strong>SOAR<\/strong>&nbsp;da benzer k\u00f6klere ve temel hipotezlere sahip, ama insan bili\u015finin modellenmesinden daha \u00e7ok yapay genel zekan\u0131n ba\u015far\u0131l\u0131 olmas\u0131 konusuna odaklanm\u0131\u015ft\u0131r.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"9627\">ACT-R ve SOAR yapay zekaya sembolik yakla\u015f\u0131m\u0131n klasik temsilcileridir ve her ikisi de ba\u011flant\u0131c\u0131 (connectionist) yakla\u015f\u0131mlarla ili\u015fkili olarak pop\u00fclerliklerini yava\u015f\u00e7a kaybediyor. Bili\u015fsel bilimin geli\u015fiminde \u00f6nemli bir rol oynad\u0131lar ama modern ba\u011flant\u0131c\u0131 makine \u00f6\u011frenmesindeki muadilleri ile kar\u015f\u0131la\u015ft\u0131r\u0131ld\u0131\u011f\u0131nda, uygulamalar\u0131 \u00e7ok daha fazla konfig\u00fcrasyon ve \u00f6n bilgi gerektiriyor. Dahas\u0131, n\u00f6ro-g\u00f6r\u00fcnt\u00fcleme ve zihni \u00e7al\u0131\u015fmak i\u00e7in kullan\u0131lan ba\u015fka ara\u00e7lar daha detayl\u0131 ve do\u011fru bir hale gelirken ACT-R ve SOAR geride kal\u0131yor; ayr\u0131ca baz\u0131 a\u00e7\u0131lardan konuya dahil olamayacak kadar kat\u0131 kal\u0131yorlar.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"6d5a\">Ancak bana g\u00f6re yapay zekan\u0131n gelece\u011fi, en az\u0131ndan AI \u00f6znelerinin insan dostu sembollerle olu\u015fmu\u015f kurallar\u0131 takip edip anlad\u0131\u011f\u0131 kadar sembolik olmal\u0131d\u0131r.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"5630\"><strong>Vah\u015fi Do\u011fada Yapay Zeka<\/strong><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"4dce\">Yukar\u0131da a\u011f\u0131rl\u0131kl\u0131 olarak g\u00fcn\u00fcm\u00fczde var olan AI \u00f6znelerinin hareket tarz\u0131n\u0131 (policy) tan\u0131mlayan algoritmalar\u0131 tasvir ettim; ama her biri ayn\u0131 zamanda bir t\u00fcr bedene sahip: bilgisayarlar, robotlar ya da sunucular (servers) ve i\u015flem yapt\u0131klar\u0131 (operate) \u00e7evre \u2014 ki \u00e7evre genelde internet hizmetleri taraf\u0131ndan algoritmalar\u0131n ba\u011flant\u0131 kurdu\u011fu alan olarak tan\u0131mlan\u0131r.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"6364\">Ki\u015fisel bilgisayar, ak\u0131ll\u0131 telefon ve di\u011fer ayg\u0131tlar\u0131n \u00e7o\u011fu \u00e7ok benzer donan\u0131m performanslar\u0131na sahip. Hareket tarzlar\u0131 i\u015fletim sistemleri taraf\u0131ndan tan\u0131mlan\u0131r ve harici bir yaz\u0131l\u0131m y\u00fckleyerek \u201c\u00f6\u011frenirler\u201d. Eski bilgisayarlarda \u00f6\u011frenim sadece insan etkile\u015fimine dayan\u0131yordu; ama g\u00fcn\u00fcm\u00fczde \u00e7o\u011fu, internet arac\u0131l\u0131\u011f\u0131yla g\u00fcncelle\u015ftirme al\u0131yor.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"7f4e\">Sunucu \u00f6znelerinin rol\u00fc gittik\u00e7e b\u00fcy\u00fcyor ve daha fazla veri bulutlara (clouds) ta\u015f\u0131n\u0131yor. Bu \u00f6znelerin \u00e7o\u011fu yo\u011fun i\u015flemlemeselli\u011fe sahip g\u00f6revlerden sorumlu ve k\u0131smen merkezi sinir sistemine benziyor. Bunun aksine, t\u00fcketiciye y\u00f6nelik aletler girdi\/\u00e7\u0131kt\u0131 becerilerini geli\u015ftirerek periferik sinirlere benziyor.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"90f2\">Bunun en u\u00e7 durumu genel olarak nesnelerin interneti (Internet of Things) olarak bilinmekte. Nesnelerin internetinde d\u00fczinelerce k\u00fc\u00e7\u00fck ve olduk\u00e7a \u00f6zelle\u015fmi\u015f ayg\u0131ttan her biri sadece bir ya da birka\u00e7 fonksiyonu yerine getirir; bulut tabanl\u0131 \u201cbeyin\u201d, evleri, fabrikalar\u0131 ve hatta t\u00fcm alanlar\u0131 kontrol etmek i\u00e7in bunlar\u0131n hepsini d\u00fczenler.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"2b52\">Robotik ise genellikle daha otonom \u00f6znelere odaklan\u0131r. Bu robotlar, karma\u015f\u0131k olan ger\u00e7ek d\u00fcnya girdi\/\u00e7\u0131kt\u0131 kanallar\u0131yla ger\u00e7ek zamanl\u0131 olarak kendileri ilgilenmek zorundad\u0131r. Otonom ara\u00e7lar muhtemelen bunun en \u00fcnl\u00fc \u00f6rneklerindendir:<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1400\/0*kZJkBSxsUYTbGkBb.jpeg\" alt=\"\"\/><figcaption class=\"wp-element-caption\">Voyage adl\u0131 otonom taksinin sistem-seviyesindeki g\u00f6r\u00fcn\u00fcm\u00fc. Kaynak:&nbsp;<a href=\"https:\/\/towardsdatascience.com\/what-is-intelligence-a69cbd8bb1b4\" rel=\"noreferrer noopener\" target=\"_blank\">towardsdatascience.com<\/a><\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"c629\">Bu, Voyage\u2019\u0131n sadece basitle\u015ftirilmi\u015f bir resmi. Ger\u00e7ek sistemler genellikle s\u00fcrekli bir girdi ak\u0131\u015f\u0131 olan y\u00fczden fazla sens\u00f6re sahipken, \u00e7\u0131kt\u0131lar\u0131 ise hayat ve \u00f6l\u00fcm aras\u0131ndaki fark\u0131 yaratabilir. Bu t\u00fcr \u00f6znelerin m\u00fchendisli\u011fi, yapay zeka ara\u015ft\u0131rmas\u0131nda g\u00fcn\u00fcm\u00fcz\u00fcn en zor alanlar\u0131ndan biridir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"2727\">Buna ek olarak, t\u00fcketiciye y\u00f6nelik robotlar otonom \u00f6znelerin yaln\u0131zca k\u00fc\u00e7\u00fck bir par\u00e7as\u0131d\u0131r ve nispeten yeni bir geli\u015fmedir. Otonom \u00f6znelerin \u00e7o\u011fu, end\u00fcstriyel ve askeri ama\u00e7larla tasarlanmaktad\u0131r. B\u00f6yle bak\u0131ld\u0131\u011f\u0131nda otonom bir taksinin yanl\u0131\u015f davran\u0131\u015f\u0131, bir silahl\u0131 insans\u0131z hava arac\u0131n\u0131n ya da n\u00fckleer santral kontrol birimlerinin bir ar\u0131zas\u0131 ile kar\u015f\u0131la\u015ft\u0131r\u0131ld\u0131\u011f\u0131nda k\u00fc\u00e7\u00fck bir kaza gibi g\u00f6r\u00fcnebilir. Bu t\u00fcr sistemlerin hareket tarzlar\u0131 i\u00e7in programlama prensipleri, \u201ckarakutu\u201d \u00f6\u011frenme algoritmalar\u0131na dayand\u0131r\u0131lamaz; genelde i\u015flerinin her y\u00f6n\u00fc i\u00e7in kat\u0131 matematikel spesifikasyonlar i\u00e7erirler.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"c704\">K\u0131sacas\u0131, AI \u00f6zneleri t\u00fcm \u015fekillerde ve renklerde bulunabilir, ama g\u00fcncel ak\u0131m, periferik ayg\u0131tlar k\u00fc\u00e7\u00fcl\u00fcrken veri merkezlerinin giderek b\u00fcy\u00fcmesi \u015feklinde.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"411d\"><strong>Kuantum D\u00fcnyas\u0131<\/strong><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"2e95\">Bu b\u00f6l\u00fcm zeka konusundan ba\u011f\u0131ms\u0131zm\u0131\u015f gibi dursa da, fizi\u011fin \u2014 \u00f6zellikle de kuantum fizi\u011finin pek \u00e7ok nedenden \u00f6t\u00fcr\u00fc \u00f6zel bir dikkat hak etti\u011fine inan\u0131yorum.<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1400\/1*-5gw2isr07kmTmagJlNlKw.png\" alt=\"\"\/><figcaption class=\"wp-element-caption\">Kaynak:&nbsp;<a href=\"https:\/\/physics.stackexchange.com\/\" rel=\"noreferrer noopener\" target=\"_blank\">physics.stackexchange.com<\/a><\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"6261\">Her \u015feyden \u00f6nce, kuantum mekani\u011fi (Quantum Mechanics, QM) t\u00fcm yapay ve biyolojik \u00f6zneler (agents) i\u00e7in ortak paydad\u0131r. Hem yar\u0131 iletken hem de biyokimyasal \u00f6znelerin \u00e7al\u0131\u015fmalar\u0131 esas olarak kuantumun etkilerinden temel al\u0131r. Zeka hakk\u0131nda atom ve atomla ilgili bir seviyede konu\u015fmak \u00e7ok mant\u0131kl\u0131 gelmese de, her t\u00fcrl\u00fc materyalden evrensel bilgisayarlar in\u015fa etmek m\u00fcmk\u00fcnd\u00fcr.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"e9fe\">\u0130kinci olarak, 300 y\u0131l \u00f6nce gezegen hareketlerini hesaplamak i\u00e7in geli\u015ftirilmi\u015f matematiksel ara\u00e7lar, geri yay\u0131l\u0131m (backpropagation) ve dereceli azalma (gradient descent) i\u00e7in gereken temel yap\u0131y\u0131 olu\u015fturmu\u015ftur. Dahas\u0131 olas\u0131l\u0131k teorisi, istatistiksel mekanikler ve matris mekanikleri QM i\u00e7in temeldir ve modern yapay zekan\u0131n akrabalar\u0131na daha yak\u0131nlard\u0131r. \u015eu an derin \u00f6\u011frenme simya ilmi gibidir, ama inan\u0131yorum ki fizik onu daha iyi anlamam\u0131z\u0131 sa\u011flabilir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"b3f8\">\u00dc\u00e7\u00fcnc\u00fcs\u00fc ise kuantum i\u015flemlemenin (quantum computing) y\u00fckseli\u015fi. Kuantum bilgisayarlar hala geli\u015fme d\u00f6neminde, ama g\u00fcn\u00fcm\u00fcz deneyleri belli t\u00fcr optimizasyon problemleri i\u00e7in \u00f6nemli bir h\u0131zlanma potansiyelini \u00e7oktand\u0131r g\u00f6steriyor. \u00d6rne\u011fin, Boltzmann Makinesi (Boltzmann Machine, BM) \u00e7o\u011fu uygulama durumunda \u00e7\u00f6z\u00fcmlenemeyen (intractable) bir t\u00fcr yapay n\u00f6ral a\u011fd\u0131r; bu nedenle bu a\u011f\u0131 \u00fcretenler, ilk derin n\u00f6ral a\u011flardan biri olan k\u0131s\u0131tl\u0131 bir varyasyonunu \u00fcrettiler. Ancak belki de kuantum bilgisayarlar, BM\u2019lerin oldu\u011fu kadar bir\u00e7ok olas\u0131l\u0131ksal modelin tam g\u00fcc\u00fcnden faydalanmam\u0131za olanak sa\u011flayacaklar.<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1400\/1*sK8gxw-dyzMvhzrtjd_3-Q.png\" alt=\"\"\/><figcaption class=\"wp-element-caption\">Kaynak:&nbsp;<a href=\"https:\/\/towardsdatascience.com\/what-is-intelligence-a69cbd8bb1b4\" rel=\"noreferrer noopener\" target=\"_blank\">towardsdatascience.com<\/a><\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"c7cd\">Son olarak, Kuantum Mekani\u011fi yukar\u0131da bahsedilenlerden \u00e7ok daha zor anla\u015f\u0131labilecek bir konu. Olas\u0131l\u0131k genli\u011fi (probability amplitude), klasik olas\u0131l\u0131ksal mant\u0131\u011f\u0131n ihlali ve atomla ilgili seviyede ger\u00e7ekle\u015fen her \u015feyin belirsiz resmi yaln\u0131zca buzda\u011f\u0131n\u0131n g\u00f6r\u00fcnen k\u0131sm\u0131. \u0130ronik olarak, pek \u00e7ok insan yapay n\u00f6ral a\u011flar\u0131 zay\u0131f yorumlanabilirli\u011fi i\u00e7in ele\u015ftirse de insanlar bile kuantum fizi\u011fini sezgisel terimlerle a\u00e7\u0131klamakta ba\u015far\u0131s\u0131z oluyor.<\/p>\n\n\n\n<h1 class=\"wp-block-heading\" id=\"1f1c\"><strong>Biyolojik \u00d6zneler<\/strong><\/h1>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"209d\">Yakla\u015f\u0131k 100 y\u0131ld\u0131r var olan yapay zeka \u00f6znelerinin aksine, biyolojik \u00f6zneler 3 milyar y\u0131ld\u0131r varlar. D\u00fcnyada milyonlarca t\u00fcr var ve hepsi ortak bir \u015feye sahip:&nbsp;<strong>DNA<\/strong>.<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1400\/0*DrGHJL2O-WJDr93q.png\" alt=\"\"\/><figcaption class=\"wp-element-caption\">Source:&nbsp;<a href=\"https:\/\/www.evogeneao.com\/learn\/tree-of-life\" rel=\"noreferrer noopener\" target=\"_blank\">evogeneao.com<\/a><\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"4175\">DNA neden bu kadar \u00f6nemli? Genellikle DNA, h\u00fccrelerin \u201cmerkezi sinir sistemi\u201ddir. Ayr\u0131ca, DNA-bazl\u0131 bir ya\u015fam olmadan \u00f6nce RNA-bazl\u0131 organizmalar\u0131n bulundu\u011fu, ama her ikisinin de i\u015flevsel ve yap\u0131sal olarak \u00e7ok benzer oldu\u011fu \u00e7o\u011funlukla kabul edilmektedir.<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1400\/0*eTSv4tf1cEkyVSiD.png\" alt=\"\"\/><figcaption class=\"wp-element-caption\">Kaynak: Wikipedia<\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"d6d7\">DNA\u2019n\u0131n \u00e7o\u011fu -bu oran insanlarda yakla\u015f\u0131k y\u00fczde 98\u2019dir- proteinleri kodlamaz ve bu y\u00fczden uzun bir zaman boyunca bu k\u0131sm\u0131n kullan\u0131\u015fs\u0131z oldu\u011fu d\u00fc\u015f\u00fcn\u00fcld\u00fc. Ancak, bu b\u00f6l\u00fcm\u00fcn \u00f6nemli bir par\u00e7as\u0131, \u00e7evreye ba\u011fl\u0131 olarak DNA\u2019n\u0131n hangi \u015fifreleme b\u00f6l\u00fcm\u00fcn\u00fcn aktive olaca\u011f\u0131n\u0131 kontrol etmede \u00f6nemli bir rol oynar. Ayr\u0131ca, DNA\u2019n\u0131n par\u00e7alar\u0131 metilleme ile geri d\u00f6nd\u00fcr\u00fclebilir \u015fekilde deaktive edilir ve bu, ya\u015fam d\u00f6ng\u00fcs\u00fc boyunca birka\u00e7 kez meydana gelebilir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"96ac\">Bunlar\u0131n tamam\u0131, konak h\u00fccrenin (host cell) hangi rolde uzmanla\u015fmas\u0131 ve nas\u0131l aktive olmas\u0131 gerekti\u011fine karar vererek, genomun farkl\u0131 veri birle\u015fimlerine farkl\u0131 \u015fekillerde tepki vermesine olanak sa\u011flar. Ayr\u0131ca, DNA asl\u0131nda var olabilmek i\u00e7in bir konak h\u00fccreye ihtiya\u00e7 duymaz. H\u00fccred\u0131\u015f\u0131 (extracellular) DNA \u00e7\u00f6z\u00fcnmeye u\u011frar, ama daha k\u00fc\u00e7\u00fck par\u00e7alar\u0131 \u00e7ok\u00e7a y\u0131l hayatta kalabilir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"a49d\">Bu arada, modern biyoteknoloji DNA\u2019y\u0131 istedi\u011fimiz gibi sentezlememize ve d\u00fczenlememize olanak sa\u011flar. Bu nedenle bu noktada yapay ve biyolojik \u00f6zneler aras\u0131ndaki ayr\u0131m esasen ortadan kaybolur.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"1c4a\"><strong>H\u00fccreler<\/strong><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"caa9\">Basit i\u015flevsel h\u00fccreler,&nbsp;<strong>protosel&nbsp;<\/strong>olarak bilinir:<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1400\/0*HmFlfIZZZyZ98Niv.png\" alt=\"\"\/><figcaption class=\"wp-element-caption\">Kaynak:&nbsp;<a href=\"https:\/\/xabier.barandiaran.net\/research\/origins-of-life\/\" rel=\"noreferrer noopener\" target=\"_blank\">xabier.barandiaran.net<\/a>&nbsp;| CC-BY-SA-3.0<\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"dd1d\">Protoseller ilk canl\u0131 organizmalar\u0131n neye benziyor olabilece\u011fini g\u00f6sterir. Yakla\u015f\u0131k olarak 3 ile 4 milyar y\u0131l \u00f6ncesinde yery\u00fcz\u00fcndeki \u00e7evre modelleri, lipit baloncuklar\u0131n\u0131n tesad\u00fcfen ilk genomlar\u0131 \u00fcretmeye yetecek kadar n\u00fckleotidi i\u00e7lerine alm\u0131\u015f olabilece\u011fini; bunun da \u00e7evrelerinden besinleri alarak \u00e7o\u011falmaya ba\u015flam\u0131\u015f olabilece\u011fini ileri s\u00fcr\u00fcyor. Belli bir e\u015fi\u011fi a\u015fan miktardaki genleri ve di\u011fer kimyasallar\u0131 biriktirdikten sonra, bu baloncuklar i\u00e7 bas\u0131n\u00e7lar\u0131n\u0131n etkisiyle b\u00f6l\u00fcnd\u00fcler.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"127f\">Bir ba\u015fka basit \u00f6rnek ise&nbsp;<strong>vir\u00fcslerdir<\/strong>. Protoseller ile aralar\u0131ndaki en temel fark, vir\u00fcslerin i\u00e7sel bir metabolizma s\u00fcrd\u00fcrmemesidir ve \u00e7o\u011falmak i\u00e7in ba\u015fka biyolojik \u00f6zneye ihtiya\u00e7lar\u0131 vard\u0131r. Genomlar\u0131 genellikle \u00e7ok k\u0131sad\u0131r ve 1 ya da 2 proteini kodlayabilir. Ancak, vir\u00fcsler&nbsp;<strong>yatay gen transferi<\/strong>&nbsp;(horizontal gene transfer) olarak bilinen s\u00fcre\u00e7te konaklar\u0131 ile DNA de\u011fi\u015f toku\u015fu yaparak \u201cileti\u015fim kurabilir\u201d. \u00c7o\u011fu tek h\u00fccreli organizma bunu yapabilir ve bu mekanizman\u0131n evrimde \u00f6nemli bir rol\u00fc vard\u0131r.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"d2b2\">Bakterilerse bunun aksine farkl\u0131 kimyasallar, \u0131\u015f\u0131k, bas\u0131n\u00e7, s\u0131cakl\u0131k vb. i\u00e7in \u00e7ok say\u0131da sens\u00f6re sahip olabilir. \u00c7o\u011funun, molek\u00fcler seviyede i\u00e7ten yanmal\u0131 motorlara benzeyen hareket mekanizmalar\u0131 vard\u0131r.<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:880\/0*zZz2f6SlI1gowcXM.gif\" alt=\"\"\/><figcaption class=\"wp-element-caption\">Bacteria E. Coli. Source:&nbsp;<a href=\"https:\/\/gfycat.com\/\" rel=\"noreferrer noopener\" target=\"_blank\">gfycat.com<\/a><\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"e3d1\">Ayr\u0131ca, olduk\u00e7a geli\u015fmi\u015f bir ileti\u015fimleri vard\u0131r ve s\u00fcr\u00fc halinde bir araya gelebilirler. \u00c7\u0131kt\u0131lar\u0131 art\u0131k sadece at\u0131k de\u011fildir. Genomlar\u0131 ve \u00e7evresindeki t\u00fcm protein tipleri onlar\u0131n geni\u015f bir besin yelpazesini sindirmelerine ve olduk\u00e7a karma\u015f\u0131k davran\u0131\u015flar\u0131 sergilemelerine olanak sa\u011flar. Ancak genelde yap\u0131lar\u0131 protosellere ve arkealara \u00e7ok benzer.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"47f8\"><strong>\u00d6karyot<\/strong>&nbsp;h\u00fccrelerde ise pek \u00e7ok organel bulunur. Mitokondri ve kloroplast gibi baz\u0131lar\u0131n\u0131n kendi DNA par\u00e7alar\u0131 vard\u0131r ve bunlar\u0131n ge\u00e7mi\u015fte ayr\u0131 organizmalar olmas\u0131 m\u00fcmk\u00fcnd\u00fcr. Ayr\u0131ca, mitokondri, metabolizma i\u00e7in elzem olanKrebs d\u00f6ng\u00fcs\u00fcnde \u00f6nemli bir rol oynar.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"cddc\">Tipik \u00f6karyot h\u00fccrelerinin daha karma\u015f\u0131k bir kimyasal aksam\u0131 vard\u0131r, ama kendi ba\u015flar\u0131na hareket etmekten mahrumdurlar. Dahas\u0131, hayvan h\u00fccreleri de kloroplastlardan ve h\u00fccre duvarlar\u0131ndan mahrumdur, bu da \u00f6zerkliklerini daha da tehlikeye atar. Genel olarak, yukar\u0131daki evrim a\u011fac\u0131nda g\u00f6sterilen organizmalar\u0131n h\u00fccreleri, g\u00f6rselde soldan so\u011fa do\u011fru yava\u015f yava\u015f kendi ba\u015flar\u0131na hayatta kalma yeteneklerini kaybederken daha karma\u015f\u0131k \u201csosyal\u201d hareket tarzlar\u0131 ve \u00f6zel i\u015flevler kazan\u0131rlar.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"fad3\">H\u00fccrelerin \u00e7evrelerindeki de\u011fi\u015fikliklere tepki verebilmesi i\u00e7in en h\u0131zl\u0131 yollardan biri&nbsp;<strong>aksiyon potansiyelidir<\/strong>. Sens\u00f6rlerin baz\u0131lar\u0131 kimyasallar\u0131, bas\u0131nc\u0131 ve di\u011fer uyaranlar\u0131 alg\u0131lad\u0131\u011f\u0131nda, h\u00fccre zar\u0131nda elektriksel potansiyelde h\u0131zl\u0131 bir de\u011fi\u015fikli\u011fe yol a\u00e7abilirler. Bu da her t\u00fcrl\u00fc \u00e7\u0131kt\u0131ya yol a\u00e7an kimyasal reaksiyon dizilerinin kademeli olarak ortaya \u00e7\u0131kmas\u0131na neden olur.<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1178\/0*wq2n2fvuljK3uFyU.gif\" alt=\"\"\/><figcaption class=\"wp-element-caption\">Venus Flytrap plant. Source:&nbsp;<a href=\"https:\/\/giphy.com\/explore\/venus-flytrap\" rel=\"noreferrer noopener\" target=\"_blank\">giphy.com<\/a><\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"8f17\">Ancak, aksiyon potansiyeli sinyali kaynak h\u00fccre ve onunla do\u011frudan do\u011fruya membran- membran ba\u011flant\u0131lar\u0131 olan h\u00fccrelerle s\u0131n\u0131rl\u0131d\u0131r. Sinyal, sinyalizasyon molek\u00fclleri arac\u0131l\u0131\u011f\u0131yla di\u011fer h\u00fccrelere iletilebilir ama bu s\u00fcre\u00e7 \u00e7ok \u00e7ok daha yava\u015ft\u0131r. Bu darbo\u011fazdan ka\u00e7\u0131nmak i\u00e7in \u00e7o\u011fu hayvan<strong>&nbsp;n\u00f6ron<\/strong>&nbsp;denilen \u00f6zelle\u015fmi\u015f h\u00fccrelere sahiptir.<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1400\/0*RUXLCOeLuqrBvM_s.png\" alt=\"\"\/><figcaption class=\"wp-element-caption\">Bir n\u00f6ronun \u015fematik g\u00f6r\u00fcnt\u00fcs\u00fc. Kaynak: Wikipedia<\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"e921\">N\u00f6ronlar farkl\u0131 farkl\u0131 \u015fekillerdedir ve hayatlar\u0131 boyunca yeni sinapslar \u00fcretip eskilerinden kurtulabilirler. Ara (intermediate) n\u00f6ronlarda 10 binden daha fazla ba\u011flant\u0131 varken, periferal n\u00f6ronlar genellikle sadece birka\u00e7 y\u00fcz ba\u011flant\u0131ya sahiptir. T\u00fcm bu aksam, onlar\u0131n sinyalleri \u00e7abucak ta\u015f\u0131mas\u0131na ve sinaptik g\u00fc\u00e7leri (synaptic strengths) ayarlayarak d\u00f6n\u00fc\u015ft\u00fcrmelerine (transform) izin verir. Ayr\u0131ca, omurgal\u0131lardaki bir\u00e7ok akson daha az membran kanal\u0131n\u0131 aktive ederken ve enerji tasarrufu sa\u011flarken elektrik potansiyellerinin daha da h\u0131zl\u0131 hareket etmesini sa\u011flayan miyelin k\u0131l\u0131f\u0131na sahiptir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"78f6\">Ancak n\u00f6ronlar birbirleriyle \u00e7ok say\u0131da ba\u011flant\u0131 kurmu\u015f sistemlerdir ve makro \u00f6l\u00e7ekte ne yapt\u0131klar\u0131n\u0131 anlamak i\u00e7in t\u00fcm ba\u011flant\u0131lar\u0131 g\u00f6z \u00f6n\u00fcnde bulundurman\u0131z gerekir. \u015eimdiye kadar \u00fczerinde en iyi \u00e7al\u0131\u015f\u0131lan sinir sistemlerinden biri,&nbsp;<em>C. elegans<\/em>&nbsp;solucan\u0131n\u0131n sinir sistemidir:<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1400\/0*IOTCj4-_ry8l1ACC\" alt=\"\"\/><figcaption class=\"wp-element-caption\"><em>C. elegans<\/em>&nbsp;sinir sistemine bir bak\u0131\u015f. Pek \u00e7ok n\u00f6ron, n\u00f6ron halkas\u0131n\u0131n yak\u0131n\u0131ndaki gangliyoda bulunur. Kaynak:&nbsp;<a href=\"http:\/\/rstb.royalsocietypublishing.org\/content\/370\/1677\/20140212\" rel=\"noreferrer noopener\" target=\"_blank\">stb.royalsocietypublishing.org<\/a><\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"89ab\">Bu sinir sistemi 50 y\u0131l\u0131 a\u015fk\u0131n s\u00fcredir \u00e7al\u0131\u015f\u0131l\u0131yor ve 5000\u2019den fazla sinaps i\u00e7eren 302 n\u00f6ronunun detayl\u0131 yap\u0131s\u0131n\u0131 \u00e7oktan biliyoruz:<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1400\/0*3TrLbpmWb6_GYZf_\" alt=\"\"\/><figcaption class=\"wp-element-caption\"><em>C. elegans<\/em>&nbsp;somatik sinir sisteminin ve kas sisteminin k\u0131smi devri \u015femas\u0131. Duyusal n\u00f6ronlar \u00fc\u00e7genlerle, intern\u00f6ronlar alt\u0131genlerle, motor n\u00f6ronlar dairelerle, kaslar da d\u00f6rtgenlerle temsil edilmi\u015ftir. Oklar uyar\u0131c\u0131 ya da inhibe edici olabilen kimyasal sinapslarla ba\u011flant\u0131lar\u0131 temsil ediyor. Kesikli \u00e7izgiler elektriksel sinapslar yoluyla ba\u011flant\u0131lar\u0131 temsil ediyor. VNC: ventral sinir kordonu. Kaynak:&nbsp;<a href=\"http:\/\/rstb.royalsocietypublishing.org\/content\/370\/1677\/20140212\" rel=\"noreferrer noopener\" target=\"_blank\">rstb.royalsocietypublishing.org<\/a><\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"cdf3\">G\u00f6rebilece\u011finiz \u00fczere, 302 n\u00f6ronun bile her birini anlamak ger\u00e7ek bir zorluk te\u015fkil ediyor. Bu, onlar\u0131n \u201c\u00f6\u011freniyor olmalar\u0131\u201d ile ve i\u015flevlerinin ger\u00e7ek zamanl\u0131 olarak de\u011fi\u015fmesi ile daha da karma\u015f\u0131k hale geliyor. \u015eimdi insan beynindeki milyarlarca h\u00fccreyle neler oldu\u011funu hayal etmeye \u00e7al\u0131\u015f\u0131n.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"91d6\">T\u00fcm bu karma\u015f\u0131kl\u0131k g\u00f6z \u00f6n\u00fcne al\u0131nd\u0131\u011f\u0131nda, n\u00f6robilimdeki \u00e7o\u011fu ara\u015ft\u0131rma belli b\u00f6lgelere, yolaklara (pathways) ve h\u00fccre tiplerine odaklan\u0131r. Evrimsel ge\u00e7mi\u015fi eski yap\u0131lar\u0131n \u00e7o\u011fu solunum, kalp at\u0131\u015f\u0131, uyku\/uyanma d\u00f6ng\u00fcleri, a\u00e7l\u0131k ve ba\u015fka hayati \u00f6nemi olan i\u015flevlerden sorumludur. Ancak,&nbsp;<strong>serebral korteks&nbsp;<\/strong>her \u015feyden daha fazla dikkat \u00e7ekiyor.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"402a\">Yap\u0131sal olarak korteks yakla\u015f\u0131k 2\u20133 milimetre kal\u0131nl\u0131\u011f\u0131nda katlanm\u0131\u015f katmanl\u0131 bir tabakad\u0131r ve beynin di\u011fer k\u0131s\u0131mlar\u0131n\u0131 \u00e7evreleyen, yakla\u015f\u0131k olarak bir kuma\u015f yemek pe\u00e7etesi boyutunda bir aland\u0131r.<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1400\/0*7SRhIXuyaKza1CVb.png\" alt=\"\"\/><figcaption class=\"wp-element-caption\">Korteks kesiti. Kaynak:&nbsp;<a href=\"https:\/\/etc.usf.edu\/clipart\/55800\/55800\/55800_cortex.htm\" rel=\"noreferrer noopener\" target=\"_blank\">etc.usf.edu<\/a><\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"8e85\">Korteks dil, bilin\u00e7, planlama vb. gibi y\u00fcksek bili\u015fsel beceri olarak kabul etti\u011fimiz her \u015feyde m\u00fcdahildir. \u0130nsanlarda korteksin yakla\u015f\u0131k y\u00fczde 90\u2019\u0131, beyne dair en yeni evrimsel geli\u015fmelerden biri olan&nbsp;<strong>neokorteks<\/strong>&nbsp;ile temsil edilir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"a2ad\">\u0130yi bir \u015fekilde incelenen ba\u015fka bir b\u00f6lge ise&nbsp;<strong>hipokamp\u00fcst\u00fcr<\/strong>:<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1400\/0*LEBnEfrI3vvJm9QN.jpg\" alt=\"\"\/><figcaption class=\"wp-element-caption\">Kaynak:&nbsp;<a href=\"http:\/\/www.gregadunn.com\/\" rel=\"noreferrer noopener\" target=\"_blank\">gregadunn.com<\/a><\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"1a08\">T\u00fcm omurgal\u0131lar palyum (pallium) ad\u0131 verilen benzer bir yap\u0131ya sahiptir ama sadece memeliler yukar\u0131da resmedilen yap\u0131n\u0131n daha geli\u015fmi\u015f bir versiyonunu bar\u0131nd\u0131r\u0131yor. Bu yap\u0131 uzamsal (spatial) ve epizodik bellekte \u00f6nemli bir rol oynuyor. Basit\u00e7e ifade etmek gerekirse bili\u015fsel bir uzay-zamansal bir harita olarak i\u015flev g\u00f6r\u00fcyor. Bu haritayla beyin g\u00f6rsel, i\u015fitsel ve di\u011fer t\u00fcr temsillerde \u00f6zelle\u015fmi\u015f b\u00f6l\u00fcmlerde karma\u015f\u0131k an\u0131lar depolayabilir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"1d28\">Beyne dair ilk \u00e7al\u0131\u015fmalar yaralanmalara ve lezyonlara odaklanm\u0131\u015ft\u0131r. Ancak eksik beyin b\u00f6lgeleri ve eksik bili\u015fsel fonksiyonlar aras\u0131ndaki korelasyonun korteks i\u00e7in nispeten daha zay\u0131f oldu\u011fu anla\u015f\u0131ld\u0131. An\u0131lar\u0131n korteks boyunca da\u011f\u0131ld\u0131\u011f\u0131 ve hatta kom\u015fu n\u00f6ronlar\u0131n bir par\u00e7as\u0131n\u0131n cerrahi olarak \u00e7\u0131kar\u0131lmas\u0131ndan sonra bile eksik fonksiyonlar\u0131 yeniden \u00f6\u011frenebildi\u011fi ortaya \u00e7\u0131kt\u0131. Ayr\u0131ca b\u00f6yle bir operasyonun sonucunda olu\u015fan yaralanman\u0131n s\u0131n\u0131rlar\u0131n\u0131 kesin olarak belirlemek de genellikle zor. Bu \u00e7al\u0131\u015fmalar \u015f\u00f6yle bir tablo ortaya \u00e7\u0131kar\u0131yor:<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1400\/1*3j6uaY5amgudlex_n2KInA.png\" alt=\"\"\/><figcaption class=\"wp-element-caption\">Kaynak:&nbsp;<a href=\"https:\/\/www.pinterest.fr\/pin\/526287906430361181\/?lp=true\" rel=\"noreferrer noopener\" target=\"_blank\">pinterest.fr<\/a><\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"b1e7\">Bu tablolardaki temel sorun pratik ve teorik olarak her iki u\u00e7ta da kesinlik eksikli\u011fidir. Deneysel ortamda beynin k\u00fc\u00e7\u00fck k\u0131s\u0131mlar\u0131n\u0131 uyarabilir ve tepkiyi izleyebilirsiniz; ama bu birincil duyusal ve motor alanlar d\u0131\u015f\u0131nda, genellikle bulan\u0131k sonu\u00e7lar do\u011furur. \u00d6te yandan g\u00fcn\u00fcm\u00fczde denekler g\u00f6revlerini yerine getirirken beynin hangi b\u00f6l\u00fcmlerinin aktif oldu\u011funu izlemek i\u00e7in fonsksiyonel manyetik rezonans g\u00f6r\u00fcnt\u00fclemeyi (fMRI) kullanabilirsiniz, ama alanlar yaln\u0131zca birka\u00e7 g\u00f6revde uzmanla\u015fmad\u0131\u011f\u0131ndan sonu\u00e7lar genelde bulan\u0131kt\u0131r. Ayr\u0131ca fMRI asl\u0131nda oksijen tedarik seviyelerini \u00f6l\u00e7\u00fcyor, bu y\u00fczden de tek tek n\u00f6ron seviyesindeki aktiviteyi \u00f6l\u00e7mek yeterli de\u011fildir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"00fe\">N\u00f6robilim ara\u015ft\u0131rmalar\u0131ndaki en umut verici g\u00fcncel y\u00f6nlerden biri optogenetiktiktir. N\u00f6ronlar i\u00e7in \u0131\u015f\u0131k sens\u00f6rleri sa\u011flayan genleri kullanarak n\u00f6ronlar\u0131n aktivitesini ayr\u0131 ayr\u0131 \u00e7ok daha y\u00fcksek bir netlikle kontrol etmemizi sa\u011flar; ancak insanl\u0131 deneylerde kullan\u0131lamayan genetik manip\u00fclasyonlar gerektirir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"05e8\">Beyin aktivitesinin bir ba\u015fka \u00f6zelli\u011fi de dalgalar halinde olmas\u0131d\u0131r:<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1400\/0*xFNAHAFwQU4YOFin.jpg\" alt=\"\"\/><figcaption class=\"wp-element-caption\">EEG kay\u0131tlar\u0131n\u0131n y\u00fcksek seviyeli yorumu (cps: saniyedeki d\u00f6ng\u00fc). Kaynak:&nbsp;<a href=\"https:\/\/towardsdatascience.com\/what-is-intelligence-a69cbd8bb1b4\" rel=\"noreferrer noopener\" target=\"_blank\">towardsdatascience.com<\/a><\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"f27e\">T\u00fcm bu \u00e7al\u0131\u015fmalar n\u00f6rolojik hastal\u0131klar\u0131 tedavi etmemizi ve anlamam\u0131z\u0131 sa\u011flar; ancak ki\u015finin ne yapt\u0131\u011f\u0131na ya da ne d\u00fc\u015f\u00fcnd\u00fc\u011f\u00fcne dair olduk\u00e7a belirsiz tan\u0131mlamalarla, baz\u0131 b\u00f6l\u00fcmlerdeki aktivitelerin korelasyonlar\u0131 haricinde insanlar\u0131n davran\u0131\u015flar\u0131n\u0131 a\u00e7\u0131klamaktan olduk\u00e7a uzakt\u0131r. Her neyse, zihin ara\u015ft\u0131rmalar\u0131na y\u00f6nelik bu a\u015fa\u011f\u0131dan yukar\u0131ya (bottom-up) yakla\u015f\u0131m, bir ki\u015finin yapaca\u011f\u0131 se\u00e7imin n\u00f6ral aktiviteye g\u00f6re \u00f6ng\u00f6r\u00fclmesi olas\u0131l\u0131\u011f\u0131n\u0131n \u00f6l\u00e7\u00fclebilmesi ve \u201cmerkezi\u201d bir beyin b\u00f6l\u00fcm\u00fcn\u00fcn olmad\u0131\u011f\u0131 gibi bir\u00e7ok \u00f6nemli ke\u015ffe yol a\u00e7t\u0131.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"e0d4\">\u00d6te yandan, psikolojik perspektiften gelen davran\u0131\u015fsal \u00e7al\u0131\u015fmalar genetik, k\u00fclt\u00fcrel ve \u00e7evresel fakt\u00f6rlerden \u00e7ok\u00e7a etkilenmektedir. Bu ara\u015ft\u0131rman\u0131n en iyi bilinen sonu\u00e7lar\u0131ndan biri&nbsp;<strong>zeka katsay\u0131s\u0131<\/strong>&nbsp;(Intelligence Quotient, IQ)\u2019n\u0131n yan\u0131 s\u0131ra onu \u00f6l\u00e7mek i\u00e7in kullan\u0131lan testlerdir. \u00c7oklu zeka teorisi, \u00fc\u00e7l\u00fc (triarchic) zeka teorisi ve di\u011ferleri gibi zekay\u0131 a\u00e7\u0131klamaya \u00e7al\u0131\u015fan bir\u00e7ok teori vard\u0131r; ancak hi\u00e7birisi bu zamana dek yayg\u0131n bir \u015fekilde kabul edilmedi.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"3c7e\">Psikolojik teorilerin temel sorunu, tan\u0131mlay\u0131c\u0131 do\u011falar\u0131n\u0131n onlar\u0131 nicel olarak kan\u0131tlaman\u0131n bir yolunu sa\u011flamamas\u0131d\u0131r. Y\u00fcr\u00fcmek ve \u201cmerhaba\u201d demek gibi basit eylemlerin alt\u0131nda yatan n\u00f6ron d\u00fczeyindeki s\u00fcre\u00e7lerin miktar\u0131 son derece b\u00fcy\u00fckt\u00fcr, ek olarak her bir h\u00fccrede i\u00e7indeki DNA ve di\u011fer biyo-aksamlar\u0131n karma\u015f\u0131kl\u0131\u011f\u0131n\u0131 da d\u00fc\u015f\u00fcn\u00fcrsek, n\u00f6robilimsel ara\u015ft\u0131rmalar\u0131n psikolojik yorumlamalar\u0131 genellikle deneylerin kendisinden daha karma\u015f\u0131kt\u0131r. Ancak baz\u0131 insan bili\u015f modelleri, davran\u0131\u015fsal ve n\u00f6ral aktivite aras\u0131nda g\u00fc\u00e7l\u00fc ba\u011flant\u0131lar kurmaktad\u0131r.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"ab0b\">Bana g\u00f6re en ilgin\u00e7 olan\u0131 bu aksiyomlara dayanan&nbsp;<strong>B\u00fct\u00fcnle\u015fik Bilgi Teorisi<\/strong>\u2019dir (Integrated Information Theory, IIT):<\/p>\n\n\n\n<figure class=\"wp-block-image\"><img decoding=\"async\" src=\"https:\/\/miro.medium.com\/v2\/resize:fit:1400\/1*OzASq2N-5ietsxAQMPQRZA.png\" alt=\"\"\/><figcaption class=\"wp-element-caption\">ITT\u2019nin aksiyomlar\u0131 ve postulatlar\u0131. Kaynak:&nbsp;<a href=\"https:\/\/en.wikipedia.org\/wiki\/Integrated_information_theory#\/media\/File:Axioms_and_postulates_of_integrated_information_theory.jpg\" rel=\"noreferrer noopener\" target=\"_blank\">wikipedia.org<\/a><\/figcaption><\/figure>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"9be9\">Di\u011fer teoriler peki\u015ftirmeli \u00f6\u011frenmeyi ve onun beyinde nas\u0131l uyguland\u0131\u011f\u0131n\u0131, \u00e7ok say\u0131da haf\u0131za, g\u00f6rme, i\u015fitme, dil ve di\u011ferleri \u00fczerine modeller i\u00e7eririr. Ancak bana g\u00f6re IIT, hepsinin aras\u0131nda en geni\u015f teorik \u00e7er\u00e7eveyi \u00f6neren kuramd\u0131r.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"1183\">Yukar\u0131da bahsedilen modeller \u00e7o\u011funlukla bireylerin davran\u0131\u015f\u0131 \u00fczerine odaklansa da, \u201csosyal psikoloji\u201d \u00e7o\u011fu canl\u0131 organizma i\u00e7in \u00e7ok \u00f6nemlidir. Ba\u011f\u0131rsa\u011f\u0131n\u0131zdaki bakteri kolonisinden ba\u015flayarak bal\u0131klara, kar\u0131ncalara, ar\u0131lara, ku\u015flara ve insan topluluklar\u0131na her \u015fey sosyal etkile\u015fimlerden do\u011far. Kar\u0131ncalar\u0131n kimyasal dili ve ar\u0131lar\u0131n \u201cdans\u201d yoluyla nas\u0131l ileti\u015fim kurdu\u011fu hakk\u0131nda zaten \u00e7ok \u015fey biliyoruz, ancak insan duygular\u0131n\u0131 anlamak b\u00fcy\u00fck bir zorluk te\u015fkil eder. Sahip oldu\u011fumuz t\u00fcm diller, kurallar ve dinler gittk\u00e7e daha karma\u015f\u0131k hale gidiyor.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\" id=\"3e5d\"><strong>O halde zeka nedir?<\/strong><\/h2>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"b316\">Bu sorunun pek \u00e7ok yan\u0131t\u0131 var, ancak hen\u00fcz geni\u015f \u00e7apta kabul edilen birle\u015fik bir biyolojik ve yapay zeka teorisi yok. Ancak bir AIXI ve IIT hibritinin bizi bu teoriye yakla\u015ft\u0131rabilece\u011fine inan\u0131yorum. Bunlar\u0131 birle\u015ftirmek i\u00e7in t\u0131p ve ekonomiden t\u00fcretilebilecek, her yapay ve biyolojik \u00f6zneye uygulanabilecek fiziksel bir \u00f6d\u00fcl\/fayda (reward\/utility) kavram\u0131na ihtiyac\u0131m\u0131z olacak ki bu da kendi ba\u015f\u0131na \u00e7ok b\u00fcy\u00fck bir sorun.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"01c8\">Hemen hemen t\u00fcm mevcut zeka \u00f6l\u00e7\u00fcmleri baz\u0131 g\u00f6revlerdeki performanslara dayan\u0131yor ve bu, \u00e7evrenin de \u00f6znenin kar\u015f\u0131la\u015fabilece\u011fi g\u00f6revlerin de s\u00fcrekli de\u011fi\u015fti\u011fi ger\u00e7ek d\u00fcnyada sorun te\u015fkil ediyor. Di\u011fer yandan bilincin \u201cherhangi bir muhtemel deneyim\u201d olarak tan\u0131mlanmas\u0131 ve AIXI\u2019nin arkas\u0131ndaki zeka \u00e7er\u00e7evesine ili\u015fkin IIT \u00e7er\u00e7evesi, bili\u015fsel performans\u0131n daha geni\u015f bir resmini sa\u011flayabilir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"ce85\">\u0130\u00e7eriden bak\u0131ld\u0131\u011f\u0131nda, herhangi bir \u00f6znenin \u00e7al\u0131\u015fmalar\u0131 bir kuantum sisteminin bir dalga fonksiyonu olarak tan\u0131mlanabilir; ama bu, hemen hemen t\u00fcm durumlarda i\u015flemlemesel a\u00e7\u0131dan \u00e7\u00f6z\u00fcmlenemez olacakt\u0131r. Ayr\u0131ca, \u00f6\u011frenilmi\u015f ara (intermediate) temsillerin yorumlanmas\u0131 hem biyolojik hem de yapay \u00f6zneler i\u00e7in b\u00fcy\u00fck bir zorluk te\u015fkil eder.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"2d0c\">En \u00f6nemlisi, ben tek bir algoritman\u0131n ya da mekanizman\u0131n nihayetinde zekadan sorumlu olaca\u011f\u0131na inanm\u0131yorum; ama bu mekanizma, \u00f6znenin \u00e7evresi ile nas\u0131l etkile\u015fim kurdu\u011funu niteleyen bir \u00f6zelliktir.<\/p>\n\n\n\n<h1 class=\"wp-block-heading\" id=\"1907\"><strong>S\u0131rada ne var?<\/strong><\/h1>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"c756\">Yapay zekadaki geli\u015fmeler ve insan zekas\u0131n\u0131n derinlemesine kavranmas\u0131 bir\u00e7ok avantaja ve say\u0131s\u0131z pratik uygulamaya sahipken, ayn\u0131 zamanda ba\u015fa \u00e7\u0131kmam\u0131z gereken bir\u00e7ok zorlu\u011fu da ortaya koyuyorlar ve \u00e7o\u011fu \u015fu kategorilerden birine giriyor:<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"3774\">\u2022 Gizlilik: \u00d6nceden sizin veriniz size ve belli bir \u00f6l\u00e7\u00fcde h\u00fck\u00fcmete aitti. Verilerin ak\u0131\u015f\u0131n\u0131 d\u00fczenleyen kat\u0131 kurallar mevcuttu. \u015eimdi ise y\u00fczlerce izleme servisi, sosyal a\u011flar ve bu verilerin nas\u0131l kullan\u0131ld\u0131\u011f\u0131na dair neredeyse hi\u00e7 a\u00e7\u0131klama yapmayan \u015firketler mevcut.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"e793\">\u2022 \u00d6nyarg\u0131 (Bias): Yapay olarak d\u00fczenlenenler hari\u00e7, her \u00f6\u011frenme veri k\u00fcmesinin \u00f6nyarg\u0131lar\u0131 vard\u0131r ve \u00f6neri motorlar\u0131 (recommendation engines) gibi kapal\u0131 d\u00f6ng\u00fc sistemlerinde g\u00fc\u00e7lenme e\u011filimindedirler.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"e0e7\">\u2022 Ahlaki Tutum (Alignment): Yapay zeka e\u011fitimlerinin \u00e7o\u011fu fayday\u0131 en \u00fcst d\u00fczeye \u00e7\u0131karmaya veya hatay\u0131 en alt d\u00fczeye indirgemeye dayan\u0131r ve bu objektif i\u015flevler t\u00fcm insan de\u011ferlerini ve ahlak\u0131n\u0131 temsil etmez.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"842e\">\u2022 Yerini Alma: Teknolojiler zaten bir s\u00fcredir bir\u00e7ok g\u00f6revde insanlar\u0131n yerini al\u0131yor ve insan\u0131n evrimi yapay zekan\u0131n evriminden daha yava\u015f. Sadece birka\u00e7 y\u0131l \u00f6nce bilgisayarlar profesyonellerin kulland\u0131\u011f\u0131 ve nadir g\u00f6r\u00fclen ara\u00e7lard\u0131, ama g\u00fcn\u00fcm\u00fczde onlar\u0131 her g\u00fcn kullanmadan g\u00fcncel kalmak zor.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"52a2\">\u2022 Siber Sald\u0131r\u0131lar: Eskiden siber sald\u0131r\u0131lar tek bir ki\u015fiyi hedeflerken \u00e7ok fazla haz\u0131rl\u0131k gerektiriyordu; ama modern yapay zeka bilgi toplayabilir, \u015fifreleri tahmin edebilir, kimlik av\u0131 i\u00e7eri\u011fi olu\u015fturabilir ve s\u00fcre\u00e7 i\u00e7inde kendini geli\u015ftirirken insanlardan \u00e7ok daha h\u0131zl\u0131 bir \u015fekilde bir ba\u015fkas\u0131 gibi davranabilir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"451c\">\u2022 Psikom\u00fchendislik (Psycho-engineering): Tarihten al\u0131nan bir\u00e7ok ders ve psikolojik deney, \u00f6nceden herhangi bir \u015fiddet e\u011filimi olmayan insanlar\u0131n bile d\u00fczg\u00fcn bir \u015fekilde manip\u00fcle edildi\u011finde zarar verebilece\u011fini ortaya \u00e7\u0131kard\u0131. Facebook, Google ve di\u011fer b\u00fcy\u00fck \u015firketler muhtemelen temelde bizim hakk\u0131m\u0131zdaki her \u015feyi hedeflemek, taramak ve bizi herhangi bir \u015fey yapmaya zorlamak i\u00e7in yeteri kadar bilgiye sahip.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"1e43\">K\u00fcresel ekonominin b\u00fcy\u00fck k\u0131s\u0131mlar\u0131n\u0131 kontrol eden ticaret botlar\u0131 s\u00f6z konusu oldu\u011funda \u00f6nyarg\u0131 ve ahlaki tutum problemlerini nas\u0131l \u00e7\u00f6zebiliriz? Yeterince e\u011fitilmedikleri senaryolarda yapay zeka \u00f6znelerinin hatalar\u0131ndan kim sorumlu? Zihnimiz \u00fczerinde tahakk\u00fcm kurmayacak, hataya dayan\u0131kl\u0131 beyin-bilgisayar aray\u00fczlerini (brain-computer interfaces) nas\u0131l \u00fcretebiliriz? Ayr\u0131ca bu sorunlar\u0131n \u00e7o\u011fu, insanlarla oldu\u011fu kadar AI ile de ilgilidir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"8c13\">5, 10 ya da 20 y\u0131l i\u00e7erisinde nerede olaca\u011f\u0131z? Bilmiyorum ve yapay zekay\u0131 ilgilendiren herhangi bir tahmin konusunda da sizi \u015f\u00fcpheci olmaya davet ediyorum. Tarih, \u00f6nde gelen yapay zeka ara\u015ft\u0131rmac\u0131lar\u0131n\u0131n bile \u00e7o\u011fu tahmininin yanl\u0131\u015f \u00e7\u0131kt\u0131\u011f\u0131n\u0131 g\u00f6steriyor (bazen de b\u00fcy\u00fck bir farkla). Ancak, yapay ve biyolojik zekan\u0131n ortak ya\u015fam\u0131n\u0131n ka\u00e7\u0131n\u0131lmaz oldu\u011funa inan\u0131yorum. Bu durum e\u011fer ilgili problemleri tan\u0131rsak ve onlarla ba\u015fa \u00e7\u0131karsak bizler i\u00e7in yararl\u0131 bile olabilir.<\/p>\n\n\n\n<h1 class=\"wp-block-heading\" id=\"71aa\"><strong>Kaynak\u00e7a<\/strong><\/h1>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"600c\">\u2022 arxiv.org\/cs\/0309048 \u2014 \u201cGoedel Machines: Self-Referential Universal Problem Solvers Making Provably Optimal Self-Improvements\u201d by Juergen Schmidhuber<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"b787\">\u2022 numenta.com\/hierarchical-temporal-memory-white-paper \u2014 \u201cHierarchical Temporal Memory (HTM)\u201d by Jeff Hawkins<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"f36a\">\u2022 amazon.com\/Soar-Cognitive-Architecture-John-Laird \u2014 \u201cThe SOAR Cognitive Architecture\u201d by John E. Laird<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"3df4\">\u2022 act-r.psy.cmu.edu\/526FSQUERY.pdf \u2014 \u201c An Integrated Theory of the Mind\u201d by John R. Anderson, Daniel Bothell, Michael D. Byrne, Scott Douglass, Christian Lebiere, Yulin Qin<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"5337\">\u2022 arxiv.org\/1812.06162 \u2014 \u201cAn Empirical Model of Large-Batch Training\u201d by Sam McCandlish, Jared Kaplan, Dario Amodei, OpenAI Dota Team<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"dc87\">\u2022 arxiv.org\/1606.06565 \u2014 \u201cConcrete Problems in AI Safety\u201d by Dario Amodei, Chris Olah, Jacob Steinhardt, Paul Christiano, John Schulman, Dan Man\u00e9<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"7ecc\">\u2022 arxiv.org\/1811.03962 \u2014 \u201cA Convergence Theory for Deep Learning via Over-Parameterization\u201d by Zeyuan Allen-Zhu, Yuanzhi Li, Zhao Song<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"43bc\">\u2022 arxiv.org\/1805.08974 \u2014 \u201cDo Better ImageNet Models Transfer Better?\u201d by Simon Kornblith, Jonathon Shlens, Quoc V. Le<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"9652\">\u2022 apps.dtic.mil\/708563.pdf \u2014 \u201cUtility Theory for Decision Making\u201d by Peter C. Fishburn<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"4932\">\u2022 arxiv.org\/1812.08342 \u2014 \u201cSafety and Trustworthiness of Deep Neural Networks: A Survey\u201d by Xiaowei Huang, Daniel Kroening, Marta Kwiatkowska, Wenjie Ruan, Youcheng Sun, Emese Thamo, Min Wu, Xinping Yi<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"3373\">\u2022 archive.org\/CerebralMechanismsInBehavior \u2014 \u201cCerebral Mechanisms In Behavior\u201d by Lloyd A. Jeffress<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"1744\">\u2022 cognitivemap.net \u2014 \u201cThe Hippocampus as a Cognitive Map\u201d by John O\u2019Keefe, Lynn Nadel<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"b1ee\">\u2022 mitpress.mit.edu\/spikes \u2014 \u201cSpikes: Exploring the Neural Code\u201d by William Bialek, Rob de Ruyter van Steveninck, Fred Rieke, David Warland<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"abfd\">\u2022 psyarxiv.com\/d6qhu\/ \u2014 \u201cThe Origins of WEIRD Psychology\u201d by Jonathan Schulz, Duman Barahmi-Rad, Jonathan Beauchamp, Joseph Henrich<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"0356\">\u2022 nature.com\/articles\/d41586\u2013018\u201305097-x \u2014 \u201cWhat Is Consciousness?\u201d by Christof Koch<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"7b27\">\u2022 arxiv.org\/0706.3639 \u2014 \u201cA Collection of Definitions of Intelligence\u201d by Shane Legg, Marcus Hutter<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"0822\">\u2022 amazon.com\/Frames-Mind-Theory-Multiple-Intelligences \u2014 \u201cFrames of Mind: The Theory of Multiple Intelligences\u201d by Howard Gardner<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"68b7\">\u2022 amazon.com\/Beyond-IQ-Triarchic-Theory-Intelligence \u2014 \u201cBeyond IQ: A Triarchic Theory of Human Intelligence\u201d by Robert Sternberg<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"24df\">\u2022 archive.org\/tom_bingham_the_rule_of_law \u2014 \u201cThe Rule Of Law\u201d by Tom Bingham<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"dd20\">\u2022 link.medium.com\/cDzwHQm0YR \u2014 \u201cFacebook and the 2018 Midterms: A Look at the Data\u201d by Jonathan Albright<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"fce5\">\u2022 intelligence.org\/all-publications<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"072c\">\u2022 integratedinformationtheory.org<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"4126\">\u2022 portal.brain-map.org<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"8530\">\u2022 github.com\/OpenWorm<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"ee1a\">Bunlara ek olarak&nbsp;<a href=\"https:\/\/www.coursera.org\/\" rel=\"noreferrer noopener\" target=\"_blank\">coursera.org<\/a>,&nbsp;<a href=\"https:\/\/www.edx.org\/\" rel=\"noreferrer noopener\" target=\"_blank\">edx.org<\/a>&nbsp;ve pek \u00e7ok ba\u015fka a\u00e7\u0131k e\u011fitim platformundan da yararland\u0131m. T\u00fcm bu kaynaklardan \u00e7al\u0131\u015fmaya ba\u015flad\u0131\u011f\u0131mda, bir yaz\u0131 yay\u0131nlamak gibi bir plan\u0131m yoktu; bu nedenle bir kaynak\u00e7a haz\u0131rlamam\u0131\u015ft\u0131m. E\u011fer \u00e7al\u0131\u015fman\u0131z bu yaz\u0131da yer al\u0131yorsa ve kaynak g\u00f6sterilmemi\u015fse \u00f6z\u00fcr dilerim (l\u00fctfen bana twitter\u2019da @eDezhic adresinden veya e-posta ile edezhic@gmail.com\u2019dan ula\u015fmaktan \u00e7ekinmeyin).<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"45ec\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref1\" target=\"_blank\" rel=\"noopener\">[1]<\/a>&nbsp;\u0130ngilizce\u2019de \u201cintelligence\u201d kelimesi, zeka haricinde \u201cistihbarat\u201d anlam\u0131nda da kullan\u0131l\u0131yor. (E.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"64f5\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref2\" target=\"_blank\" rel=\"noopener\">[2]<\/a>&nbsp;Burada \u201cagent\u201d ile kastedilen, eylemleri ger\u00e7ekle\u015ftiren fail, fakat kolay takip edilebilmesi i\u00e7in yaz\u0131 boyunca \u201cagent\u201d s\u00f6zc\u00fc\u011f\u00fcn\u00fcn kar\u015f\u0131l\u0131\u011f\u0131 olarak \u201c\u00f6zne\u201d s\u00f6zc\u00fc\u011f\u00fcn\u00fc kullanaca\u011f\u0131z. (E.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"1319\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref3\" target=\"_blank\" rel=\"noopener\">[3]<\/a>&nbsp;Turing-b\u00fct\u00fcnl\u00fc\u011f\u00fc: Bu kavram \u0130ngiliz matematik\u00e7i ve bilgisayar bilimci Alan Turing\u2019in soyad\u0131ndan gelir. T\u00fcm i\u015flemlemesel problemleri yeterli zaman ve bellek sa\u011fland\u0131\u011f\u0131 takdirde \u00e7\u00f6zebilece\u011fi belirtilen, evrensel Turing makinesiyle sim\u00fcle edilebilir sistemlerin sahip oldu\u011fu bir niteliktir. Kimi zaman Turing-b\u00fct\u00fcnl\u00fc\u011f\u00fc olarak \u00e7evrilse de, ayn\u0131 literat\u00fcr\u00fcn NP-Tam \u00e7evirisi ile tutarl\u0131l\u0131\u011f\u0131 korumak i\u00e7in bu \u015fekilde \u00e7evirmeyi tercih ettik. (\u00c7.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"45e2\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref4\" target=\"_blank\" rel=\"noopener\">[4]<\/a>&nbsp;Algoritmik karma\u015f\u0131kl\u0131k teorisi: Di\u011fer ad\u0131 \u201ci\u015flemlemesel karma\u015f\u0131kl\u0131k teorisi\u201d olan bu teori, bilgisayar biliminde \u00f6nemli yer tutan i\u015flemleme teorisinin (theory of computation) bir dal\u0131d\u0131r. Algoritmalar\u0131n farkl\u0131 boyutlardaki girdilere g\u00f6re ihtiya\u00e7 duyduklar\u0131 \u00e7al\u0131\u015fma zaman\u0131 ve bellek miktarlar\u0131ndaki de\u011fi\u015fimleri inceler. (\u00c7.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"0693\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref5\" target=\"_blank\" rel=\"noopener\">[5]<\/a>&nbsp;Hareket tarz\u0131 (policy), makine \u00f6\u011frenmesi (machine learning) algoritmalar\u0131nda, yapay zeka \u00f6znesinin eylem se\u00e7imini modelleyen eylem-evre (action-state) e\u015fle\u015ftirmesidir. (E.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"f195\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref6\" target=\"_blank\" rel=\"noopener\">[6]<\/a>&nbsp;Elle girilen bilgiler, \u00f6r\u00fcnt\u00fclerin (patterns) ve \u201cise-o halde\u201d (if-then) olarak ge\u00e7en korrelasyonlar\u0131n yapay zekaya manuel yolla girildi\u011fi bilgi t\u00fcr\u00fcd\u00fcr. (E.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"34b2\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref7\" target=\"_blank\" rel=\"noopener\">[7]<\/a>&nbsp;G\u00fc\u00e7l\u00fc AI: 1980 y\u0131l\u0131nda John Searle taraf\u0131ndan icat edilen bir terim. Bir AI sistemi d\u00fc\u015f\u00fcnebilir ve zihin sahibi olabilir ise AI hipotezine g\u00f6re g\u00fc\u00e7l\u00fc birimdir. (\u00c7.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"0344\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref8\" target=\"_blank\" rel=\"noopener\">[8]<\/a>&nbsp;Yapay genel zeka, bir insan\u0131n yapabilece\u011fi herhangi bir zihinsel g\u00f6revi ba\u015far\u0131yla ger\u00e7ekle\u015ftirebilecek bir makinenin zekas\u0131d\u0131r. (\u00c7.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"3084\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref9\" target=\"_blank\" rel=\"noopener\">[9]<\/a>&nbsp;Girdi uzay\u0131 (input space) bir sisteme verilmesi m\u00fcmk\u00fcn olan t\u00fcm girdileri, \u00e7\u0131kt\u0131 uzay\u0131 (output space) ise bir sistemden al\u0131nmas\u0131 m\u00fcmk\u00fcn olan t\u00fcm \u00e7\u0131kt\u0131lar\u0131 ifade etmektedir. (E.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"ae53\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref10\" target=\"_blank\" rel=\"noopener\">[10]<\/a>&nbsp;Bir yapay zeka terimi olan uzman sistemler, bilgi birikimlerinden yararlanarak yapt\u0131klar\u0131 ak\u0131l y\u00fcr\u00fctmelerle karma\u015f\u0131k problemleri \u00e7\u00f6zmek \u00fczere tasarlanm\u0131\u015flard\u0131r ve uzman insanlar\u0131n karar alma becerilerini taklit (emulating) ederler. (E.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"9e2c\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref11\" target=\"_blank\" rel=\"noopener\">[11]<\/a>&nbsp;Etiketli veri, bir ya da daha fazla etiketle sunulmu\u015f bir grup \u00f6rnektir. Denetlenmi\u015f \u00f6\u011frenme (supervised learning) s\u0131ras\u0131nda etiketli veriler kullan\u0131l\u0131r ve yapay zekan\u0131n kendisine verilmi\u015f etiket s\u0131n\u0131flar\u0131n\u0131 daha sonradan etiketlenmemi\u015f verilere de do\u011fru bi\u00e7imde uygulamas\u0131 hedeflenir. (E.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"5ec1\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref12\" target=\"_blank\" rel=\"noopener\">[12]<\/a>&nbsp;Boolean ve Crisp mant\u0131\u011f\u0131 (logic), bulan\u0131k mant\u0131ktan farkl\u0131 olarak yaln\u0131zca iki evre halinde, 1 ya da 0 de\u011ferlerinden birine sahip olan de\u011fi\u015fkenler bar\u0131nd\u0131ran bir mant\u0131k \u00e7e\u015fididir. (E.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"92d3\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref13\" target=\"_blank\" rel=\"noopener\">[13]<\/a>&nbsp;Bulan\u0131k mant\u0131\u011f\u0131n fark\u0131, de\u011fi\u015fkenlerinin yaln\u0131z 1 ve 0 de\u011fil, ikisi aras\u0131ndaki herhangi bir reel say\u0131 de\u011ferini de alabilmesidir. (E.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"e090\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref14\" target=\"_blank\" rel=\"noopener\">[14]<\/a>&nbsp;Bir olgunun ger\u00e7ekle\u015fme s\u0131kl\u0131\u011f\u0131 ya da e\u011filiminden de\u011fil, bir bilgiye veya say\u0131salla\u015ft\u0131r\u0131lm\u0131\u015f ki\u015fisel kan\u0131ya dayanarak bir olas\u0131l\u0131k beklentisi kurulan hesap \u00e7e\u015fididir. (E.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"5bc2\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref15\" target=\"_blank\" rel=\"noopener\">[15]<\/a>&nbsp;\u0130statistiksel g\u00fcr\u00fclt\u00fc (statistical noise), ger\u00e7ek hayattan al\u0131nan verilerde bulunan rastgele d\u00fczensizliklerdir. Bir \u00f6r\u00fcnt\u00fcye sahip de\u011fillerdir. (E.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"c619\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref16\" target=\"_blank\" rel=\"noopener\">[16]<\/a>&nbsp;Annealing (ba\u011flanma): PCR i\u015flemi s\u0131ras\u0131nda \u0131s\u0131t\u0131larak birbirinden ayr\u0131lm\u0131\u015f olan DNA zinciri \u00e7iftlerinin, so\u011futmayla birlikte kendilerini tamamlayan zincirlere ba\u011flanmas\u0131. Adli t\u0131pta, birden fazla DNA bulundu\u011funda aranan DNA zincirinin varl\u0131\u011f\u0131n\u0131 tespit etmek i\u00e7in kullan\u0131l\u0131r. (\u00c7.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"ad77\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref17\" target=\"_blank\" rel=\"noopener\">[17]<\/a>&nbsp;\u0130\u015flemlemede (computing), bir soyutlama katman\u0131, bir altsistemin (subsystem) \u00e7al\u0131\u015fma detaylar\u0131n\u0131 gizleme ve arai\u015flerlikle (interoperability) platform ba\u011f\u0131ms\u0131zl\u0131\u011f\u0131n\u0131 (platform independence) sa\u011flama y\u00f6ntemidir. Her katman \u00fcst\u00fcndeki katmanlar olmadan var olabilir, ancak altlar\u0131ndaki katman olmadan i\u015flem yapamazlar. (E.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"6bfb\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref18\" target=\"_blank\" rel=\"noopener\">[18]<\/a>&nbsp;Zincir kural\u0131 olarak da bilinir. (\u00c7.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"c8f2\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref19\" target=\"_blank\" rel=\"noopener\">[19]<\/a>&nbsp;Dereceli azalma (Gradient descent): Dereceli azalma, bir fonksiyondaki parametre de\u011ferlerinin minimum maliyeti (cost) sa\u011flad\u0131\u011f\u0131 noktay\u0131 bulmakta kullan\u0131lan bir optimizasyon algoritmas\u0131d\u0131r. (\u00c7.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"bc1f\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref20\" target=\"_blank\" rel=\"noopener\">[20]<\/a>&nbsp;B\u00f6lgesel minimum, bir fonksiyon de\u011ferinin kom\u015fu noktalar\u0131ndan k\u00fc\u00e7\u00fck olmas\u0131d\u0131r; ancak global minimum t\u00fcm fonksiyon de\u011ferlerinin i\u00e7indeki en d\u00fc\u015f\u00fck de\u011ferdir. (E.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"6063\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref21\" target=\"_blank\" rel=\"noopener\">[21]<\/a>&nbsp;Bilgisayarl\u0131 g\u00f6rme, bilgisayarlar\u0131n dijital g\u00f6rseller (images) ve videolar \u00fcst\u00fcne \u00fcst d\u00fczey bir anlay\u0131\u015f edinebilece\u011fini \u00e7\u00f6zmeye \u00e7al\u0131\u015fan interdisipliner bir aland\u0131r. (E.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"64e0\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref22\" target=\"_blank\" rel=\"noopener\">[22]<\/a>&nbsp;HTM, AIXI, ACT-R ve SOAR, bili\u015fsel mimarilerden d\u00f6rd\u00fcn\u00fcn ad\u0131d\u0131r. Bir bili\u015fsel mimari ise hem insan zihninin yap\u0131s\u0131na ili\u015fkin bir teoriyi, hem de b\u00f6yle bir teorinin yapay zeka ve i\u015flemlemesel bili\u015fsel bilim (computational cognitive science) alanlar\u0131nda kullan\u0131lan i\u015flemlemesel bir \u00f6rneklemesidir. (E.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"20ee\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref23\" target=\"_blank\" rel=\"noopener\">[23]<\/a>&nbsp;Bilgisayar bilimindeki temel veri birimi (E.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"7749\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref24\" target=\"_blank\" rel=\"noopener\">[24]<\/a>&nbsp;Yak\u0131nsama&nbsp;<em>approximation<\/em>\u2019\u0131 kar\u015f\u0131lad\u0131\u011f\u0131ndan,&nbsp;<em>convergence<\/em>\u2019\u0131 b\u00f6yle \u00e7evirmeyi tercih ettik. (E.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"7a7b\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref25\" target=\"_blank\" rel=\"noopener\">[25]<\/a>&nbsp;Y\u0131\u011f\u0131n normalle\u015ftirme, girdi katman\u0131n\u0131 (input layer) yeniden merkezleme (re-centering) ve yeniden \u00f6l\u00e7eklendirme (re-scaling) ile normalle\u015ftirerek, yapay n\u00f6ral a\u011flar\u0131 daha h\u0131zl\u0131 ve daha stabil hale getiren bir y\u00f6ntemdir. (E.N.)<\/p>\n\n\n\n<p class=\"wp-block-paragraph\" id=\"2a96\"><a href=\"https:\/\/medium.com\/cogist\/zeka-nedir-egor-dezhic-eb9c18635410#_ftnref26\" target=\"_blank\" rel=\"noopener\">[26]<\/a>&nbsp;ACT-R, ACT\u2019nin (Adaptive Control of Thought) daha g\u00fcncel bir s\u00fcr\u00fcm\u00fcd\u00fcr. (E.N.)<\/p>\n","protected":false},"featured_media":1870,"template":"","meta":{"_acf_changed":false},"event_publishing_tags":[84,280,286,94,93,691,64,92,323,281,285,227,222,348,621,228,611,610,61,349],"kategori":[305],"class_list":["post-1869","blog_content","type-blog_content","status-publish","has-post-thumbnail","hentry","event_publishing_tags-artificial-intelligence","event_publishing_tags-bilgisayar","event_publishing_tags-bilgisayar-bilimi","event_publishing_tags-bilis","event_publishing_tags-bilissel-bilim","event_publishing_tags-cogist","event_publishing_tags-cognition","event_publishing_tags-cognitive-science","event_publishing_tags-cogsci","event_publishing_tags-computer","event_publishing_tags-computer-science","event_publishing_tags-deep-learning","event_publishing_tags-derin-ogrenme","event_publishing_tags-intelligence","event_publishing_tags-learning","event_publishing_tags-machine-learning","event_publishing_tags-makine-ogrenimi","event_publishing_tags-ogrenme","event_publishing_tags-yapay-zeka","event_publishing_tags-zeka","kategori-ceviri"],"acf":[],"_links":{"self":[{"href":"https:\/\/cog-ist.com\/en\/wp-json\/wp\/v2\/blog_content\/1869","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\/1869\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/cog-ist.com\/en\/wp-json\/wp\/v2\/media\/1870"}],"wp:attachment":[{"href":"https:\/\/cog-ist.com\/en\/wp-json\/wp\/v2\/media?parent=1869"}],"wp:term":[{"taxonomy":"event_publishing_tags","embeddable":true,"href":"https:\/\/cog-ist.com\/en\/wp-json\/wp\/v2\/event_publishing_tags?post=1869"},{"taxonomy":"kategori","embeddable":true,"href":"https:\/\/cog-ist.com\/en\/wp-json\/wp\/v2\/kategori?post=1869"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}