To Really Be Intelligent You Need a Slow Brain
Artificial Intelligence (AI) is the most existential challenge humanity will ever face. There is a point of no return where we can still regulate AI, and that is when it becomes smarter than us. And that moment is just a few months away.
This was the message presented by Mo Gowdat in a recent edition of Diary of a CEO. Mo is the former Chief Business Officer of Google X and a world renowned AI expert. His warnings are just one of many made by AI experts from across the globe ever since ChatGPT went public just over a year ago.
At the centre of many of these warnings is the fear that a super intelligent AI becomes sentient, perceives humanity as a threat to it’s own existence, and decides to wipe us all out. That is, quite literally, something straight out of a science fiction movie. Could it really happen? Could we really be just a few months away from that point of no return?
To answer that question, and to really understand the threat of AI, it’s crucial to understand the cognitive architecture of AI systems like ChatGPT, and how they compare to that paragon of intelligence, the human brain.
The Human Brain: Thinking Fast and Slow
So let’s start with the human brain. The human brain is widely understood to be made up of 2 distinct cognitive systems. Prof. Steve Peters, for example, talks about the “chimp” and the “human”, whereas Daniel Kahneman talks about the “fast brain” and the “slow brain”. Both are presenting a model of the brain comprising two distinct systems; one that is unconscious, fast and automatic, and one that’s conscious, slow and deliberate.
The fast brain is basically a pattern matching machine. The billions of neurons in our brain are connected together into these things called neural networks. These neural networks form pathways in our brains that map sensory inputs from our environment to an automatic response. Similar inputs will map to the same response.
An example might help here. Think back to a time when you’ve been watching a scary movie and, all of a sudden, something happens on screen that made you jump. This was the fast brain at work. The fast brain mapped sensory inputs representing something to be feared - perhaps a scary image on screen and a sudden noise - onto our fight-or-flight response. As a result, you experience a physical jolt as adrenalin is released into your body and you’re ready and primed, for fight or flight. All of this happened very quickly, and totally involuntary, and is the result of pathways in our brain that have evolved to keep us safe.
The slow brain, on the other hand, is the conscious thinker. The slow brain is responsible for things like understanding, reasoning, and contemplating the meaning of life. Prof. Steve Peters calls this system of thinking the “human” because it is this form of thinking that distinguishes us from the rest of the animal kingdom.
ChatGPT: A Master of Patterns but Not of Understanding
So while the human brain has two distinct cognitive systems, ChatGPT has just one. ChatGPT has a software version of a fast brain.
Arguably the biggest development in the history of AI has been the creation of artificial neural networks. Like the human fast brain, artificial neural networks are highly effective pattern matching machines. And it is an artificial neural network that is the brain of ChatGPT. It uses this this artificial neural network to identify patterns in text and to predict the next most likely work in a given context. ChatGPT is unbelievably good at this. So good that it can appear to be able to understand, to reason, even to be able to contemplate the meaning of life. But this is just an illusion. In reality, ChatGPT is just a statistical parrot, mimicking human-like responses without genuine comprehension or consciousness.
The Essence of Intelligence
What does all this mean for the potential of ChatGPT to develop into a super intelligent AI? Well, it seems to me that, to really be intelligent, you need a slow brain. After all, it is our slow brain, our ability to understand, to reason, to contemplate the meaning of life, that sets us apart from the rest of the animal kingdom.
So ChatGPT will never truly be intelligent until it has a slow brain. The problem is, we don’t really understand how the slow brain works. And until we understand it, we can’t build a software model of it. That’s going to take an even bigger breakthrough that the development of artificial neural networks, so it isn’t likely to happen any time soon.
Conclusion
To summarise, while AI models like ChatGPT demonstrate remarkable capabilities that mimic certain aspects of human cognition, they do not possess the slow, contemplative, and understanding elements of the human brain — the true hallmarks of intelligence. What’s more, the path to a future where an AI gains these elements is far from clear. Until we understand more about how our slow brain works, and can create working software models of it, then super intelligent machines capable of wiping out humanity will remain - for now at least - the stuff of science fiction.