What is Generative Artificial Intelligence?
Philosophical Approaches to Artificial Intelligence, Özgüç Güven,Filiz Didem Çoban Sarı, Editör, IU Press, Bloomington (IN), USA , İstanbul, ss.107-125, 2026
- Yayın Türü: Kitapta Bölüm / Araştırma Kitabı
- Basım Tarihi: 2026
- Doi Numarası: 10.26650/bs/ah11t3.2026.002-5.08
- Yayınevi: IU Press, Bloomington (IN), USA
- Basıldığı Şehir: İstanbul
- Sayfa Sayıları: ss.107-125
- Editörler: Özgüç Güven,Filiz Didem Çoban Sarı, Editör
- İstanbul Üniversitesi Adresli: Evet
Özet
This article proposes the term “türetici yapay zekâ” (derivational/generative artificial intelligence) as a Turkish equivalent for “generative artificial intelligence,” and elaborates the conceptual, historical, and philosophical grounds for this proposal.
Historically, the article traces the development of artificial intelligence research from the 1943 McCulloch-Pitts paper through the 1956 Dartmouth Conference, the First and Second AI Winters, the rise and fall of expert systems and symbolic AI, and the emergence of big data, deep learning, and large language models in the twenty-first century. The structural failures of symbolic AI — including the frame problem, the qualification problem, and the limitations of knowledge representation projects such as Cyc — are examined in detail.
The article compares natural intelligence with generative AI across four dimensions: environmental adaptation, reasoning, learning, and mastery. It argues that large language models model grammar and logic not as internalized rule systems but merely as statistical by-products of training, and that these systems therefore neither think, understand, nor reason in any meaningful sense. The outputs of such systems are determined not by intelligence but by statistics; not by understanding but by computation.
After reviewing the practical capabilities of generative AI — text, code, image, and video generation — the article turns to its associated concerns: growing energy and hardware demands, data quality issues, corporate dependency, concentration of power, and the risk of instrumentalizing ethical discourse.
The article concludes by examining whether strong AI, or artificial general intelligence, is achievable. It argues that the theoretical foundations of computation have not changed since Turing, that reasoning cannot be substituted by statistics, and that the human capacities for discovery and original creation cannot be replicated through vector-based statistical computation. Given that brain, consciousness, and intelligence remain insufficiently defined concepts, the article concludes that strong AI has not been achieved — technically, theoretically, or conceptually — and that meaningful progress toward it will require paradigms that go beyond the current computational framework.