We keep talking about AI as if it is an isolated concept and fail to recognize that it is just another form of intelligence like our biological intelligence. In order to understand about AI, it is quite important to understand what “intelligence” in general is.
Meaning
Intelligence at a fundamental level may be termed as a system that efficiently models reality, makes predictions on the basis of such models and achieves goals on the basis of those predictions.
To make these models of reality in an efficient manner, the intelligence has to compress copious amounts of data that it comes across. This data could be in the form of patterns or even scrambled information. Thus, compression of data becomes essential to discard noise. You will be surprised to know that a great amount of compression can occur. In fact, a good compression is one that furthers near lossless understanding. Your model of reality will become better only when it uses less and less amounts of data to make more and more accurate predictions.
Prediction
But why do we need to predict? We need to predict because we must know what is likely to happen at the next instant of time. We do not want to be caught off guard. A good lawyer prepares his case so thoroughly that he can anticipate most of the questions that a judge is going to ask. That is what a good lawyer needs to predict with his efficient model of reality to achieve the goal of a favourable outcome in his client’s favour.
But how exactly is this prediction done and how exactly are efficient models formed by any form of intelligence? It involves exploration, learning both randomly and non-randomly about the world, exploiting the already existing data to get better understanding, and improving each time new data comes across.
Abstraction
The efficient models of reality also involve layers of abstraction. They use sensory data to form patterns. Through patterns, concepts and principles are formed. And once concepts and principles are formed, the overarching meta-principles are discovered. That is how learning and improvement in learning occurs.
One important thing to remember always is that intelligence always operates under constraints. There is no such thing as unlimited data or unlimited time or unlimited memory or unlimited calculations/compute. Intelligence is considered better when it finds minimal solutions to grand problems such as e = mc^2. Einstein literally used just his brain and very few other resources such as books, journals, pen and paper to come up with energy-mass equivalence.
Compression and Distillation
You might be thinking that okay, now I know what intelligence is and what it involves. But wait, there is a backdoor entry or a shortcut too here. For example, Einstein might have spent a good portion of his life to discover mass-energy equivalence. But the future scientists distilled this information. They did not look at the life of Einstein in toto, rather they just looked at the relevant factors leading him to discover mass-energy equivalence. Thus, the efficient models of reality have some form of innate structure and redundancy. There is simplicity but complexity emerges when the number of interactions increase. The principle of mass-energy equivalence can be used in various disciplines for various purposes. And while applying mass-energy equivalence in different disciplines, many new principles will further emerge that will make things quite obvious to understand. This is again a form of compression. An intelligence that itself once grasps this knowledge and capability will be able to explain itself using different examples and principles.
Basically, that is how we teach our kids. To put it in other words, a smarter system can transfer its capabilities to a simpler one. You may call it teaching or distillation. It is the process of extracting useful patterns from a complex system and transferring those useful patterns to a simpler but more efficient system that matters.
Thus, whether it is AI or our brains, they all function in a surprisingly similar manner. In fact, the whole of AI stems from the understanding of the brain and making an attempt to replicate the processes inside it. Today, we are reaching a point where AI might surpass human intelligence but at its core, the fundamentals are all the same.