Two major research publications in recent weeks have shaken the world of artificial intelligence—and both point to the rapid convergence of machine and human thinking. Researchers from Harvard University and Cangrade Inc. found that the GPT-4o language model shows signs of cognitive dissonance, an effect previously considered exclusive to humans. After writing a positive or critical text about a person, the model changed its assessment of that individual, and the effect became stronger when GPT-4o “thought” that it was choosing the topic itself.
At the same time, a professor at the University of Stirling presented a neural-network architecture called Co4, which can selectively focus attention and formulate hypotheses. This makes it more effective and closer to the principles of human imagination.
Such behavior cannot be called true self-awareness, but it suggests something beyond simple text generation. If earlier AI was a precise, yet “flat” performer, it is now becoming an active participant in intellectual interaction. And perhaps soon it will be able not only to help, but also to understand us—our contradictions and our dreams.
