A first, simple definition describes intelligence as the ability to reach your goal through interaction with the environment you find yourself in. On that basis, though, even an organism as simple as Escherichia coli could be considered intelligent, despite having no brain. To understand what intelligence is, we have to dig deeper. Ever since ancient Greece, the concept has been tied to the ability to perceive, reason and act successfully. Yet the problem of uncertainty has often been underestimated.
This is where utility comes in: the property of being useful or advantageous to someone. It was the mathematician Daniel Bernoulli who introduced this mathematical concept to explain human behaviour. Utility is an invisible property that has to be inferred from an individual’s preferences. In short, a rational being acts so as to maximise expected utility. For all the criticism it has attracted, the idea works — but only for individuals. When several intelligent beings have to work together, things get more complicated and other factors come into play, such as the ability to cooperate and to communicate among the parties involved.
The definition of intelligence that focuses on rational behaviour applies to artificial intelligence too — and so do its complexities. Given what it perceives, a machine is intelligent to the extent that it is likely to achieve what it wants through what it does.
Many consider the human brain the most complex object in the universe. Today we know a good deal about the biochemistry of its anatomical structures, but the cognitive level of the human brain remains largely unexplored. We still do not know in detail how various human functions take place — learning, remembering, reasoning, knowing, making decisions and so on. There is one important cognitive aspect, however, that we are beginning to understand in recent years: the reward system, an internal signalling system that, through dopamine, links behaviour to positive and negative stimuli. Organisms that are more effective at seeking reward — finding food, avoiding pain, finding a mate — are more likely to pass on their genes. Since it is hard to decide which of these actions are the ones that guarantee, in the long run, the transmission of our genetic package, evolution has equipped us with automatic internal signals. These are not always perfect, though. There are ways of obtaining a reward that actually tend to reduce the likelihood of our genes spreading — drugs being one example. One reason we understand the human brain’s reward system is its resemblance to reinforcement learning, the method used in artificial intelligence.
Over the course of evolution, we have also developed the ability to learn. Learning has proved to be not only a useful survival strategy but also a powerful evolutionary shortcut.
