The Society and Technology Podcast hosted Selçuk Bayraktar and Aykut Fırat

In recent years, the increasing number of large language models and other artificial intelligence tools has become widely known. Their widespread adoption naturally brings with it debates about the nature and future of large language models. So, what processes have our current artificial intelligence tools gone through to reach us? Do they think like humans, or will they be able to do so in the future? Is artificial intelligence an existential threat capable of bringing about humanity’s end, or merely a tool to make our tasks easier?

The latest episode of the Society and Technology Podcast features Selçuk Bayraktar and Aykut Fırat as guests, addressing these questions. If you are curious about what artificial intelligence can and cannot do today, you must listen to this episode.

The episode begins with the most fundamental question in the field: “What is artificial intelligence?” Today, it is a common error to label all computations that can be described as advanced algorithms as artificial intelligence. Selçuk Bayraktar emphasizes that under this umbrella term, there are various algorithms and approaches, and that large language models specifically refer to models that operate via artificial neural network based algorithms. Similarly, Aykut Fırat notes that the term has been in use since the beginnings of computer engineering, gaining momentum with early expert system systems and later developments such as neural networks, but also experiencing periods of stagnation (artificial intelligence winter) before re-emerging prominently in recent times.

The podcast repeatedly underscores that the field where artificial intelligence is most emphasized and most successful today is natural language processing. This focus plays a crucial role in framing contemporary debates about AI. The proficiency of AI in this domain creates an illusion in human observers, acting as a barrier to recognizing what AI cannot yet do. The rapid progress of AI technologies in language, despite their still-limited physical capabilities, amplifies this illusion among the public. Naturally, exaggerated claims made by AI companies to convince investors have played a significant role in the emergence of this societal “hype”.

Another key point frequently raised in the podcast is that AI systems—particularly large language models—are in fact statistical automation systems. Although the ability of language models to generate contextually appropriate responses in multiple languages may give users the impression that they are interacting with a thinking machine, what they are truly engaging with is merely an automated system that produces answers based on statistical patterns. Selçuk Bayraktar’s analogy of a sewing machine is highly illustrative in positioning artificial intelligence as a technological concept.

Another topic requiring attention is the various predictions about the future of artificial intelligence. This debate reflects differing attitudes among researchers and stakeholders toward AI. As Aykut Fırat notes, when large language models emerged, two distinct camps formed. Figures such as Yann LeCun argue that language models cannot achieve true artificial general intelligence, while others such as Ilya Sutskever claim that these models represent a form of intelligence in their own right.

Artificial intelligence is not currently a conscious entity. However, when tasks are reduced to mechanical and computable processes, algorithms can operate far more efficiently than humans and rapidly iterate through trial-and-error phases. Within this framework, AI’s contribution to scientific advancement is entirely plausible. Yet, just as with the printing press, false information can spread just as rapidly as accurate information through these tools.

Ultimately, access to accurate information and positive development still depend on human will and choices. As with nearly all technological advances, human agency determines the outcomes of artificial intelligence. Dystopian scenarios about humanity’s future are entirely possible even without incorporating AI. Selçuk Bayraktar concludes by asserting that humanity does not need artificial intelligence to bring about its own end—it already possesses the capacity to do so. However, if used properly, these tools can serve humanity’s benefit.