Intuition and artificial intelligence

“Every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.” This conjecture launched the field of artificial intelligence 70 years ago.  It was an audacious proposition. Yet today, most people who have interacted with AI systems like ChatGPT or Claude will concede that they exhibit behavior that at least seems intelligent.

But are these systems really thinking or understanding? Aren't there are some cognitive functions only humans can perform? Computers may excel at calculating, reasoning, pattern detection, and prediction (based on statistical analysis), but they lack, for instance, the mysterious, powerful human capacity for intuition. Or do they?

Studies have found that when dealing with complex situations with multiple variables, intuition can lead to better decisions than lengthy deliberation. Intuition allows for quick processing of information, which can be crucial in time-sensitive or high-stakes situations. And it can lead to accurate judgements, such as medical diagnosis, in domains where someone has developed expertise.

Intuition is rooted in our brain's ability to recognize patterns and learn implicitly. Over time, our brains process countless pieces of information, identifying patterns and creating mental shortcuts for quick decision-making. This process, often occurring without conscious awareness, forms the foundation of our intuitive judgments.

Our understanding of human intuition actually aligns really well with the workings of deep learning. These AI systems are designed to develop pattern recognition capabilities automatically, often without explicit programming. This enables them to perform astounding cognitive feats. For instance, image recognition sometimes identify subtle features that make it appear they have intuited important visual cues. AI game-playing systems have made moves that surprised expert players and appeared to be the result of a machine-generated intuition. Some language models can grasp context and nuance in ways that seem to go beyond simple pattern matching, appearing to intuit the deeper meaning of text, or making brand new connections between concepts. And in fields like cybersecurity or fraud detection, deep learning systems can sometimes flag unusual activities that human experts might miss, appearing to have an intuition for what's out of place.

While the performance of deep learning systems can be amazing, their inner workings are often still not well understood. As with intuition, it can be difficult or impossible to explain the predictions or judgements made by a deep learning system. Deep learning and intuition do seem pretty similar. There is at least one significant difference, though, between human intuition and deep learning.

Intuition may involve emotional and physiological responses. Studies have shown that intuitive decisions can be accompanied by measurable physical changes, such as increased sweating. These bodily signals, known as somatic markers, can guide decision-making even before we're consciously aware of it. I don’t think there is a good analog between sweaty palms or a gut feeling and the inference performed by deep learning systems. 

If we can eventually replicate—or at least simulate—human cognition, the potential benefits to humanity are enormous. Some are troubled by this possibility. I think their concern is rooted in the view that our value as humans is determined by our cognitive ability. I think that view is wrong. And maybe the topic of a future essay.

What do you think about all this?