We have started our journey into AI at genius levels in the previous article. We have seen that superior AI will be comprehensive, much more so than us because of simple scaling. But what about meaning and understanding? A common criticism of AI is that computers do not really understand anything, they just manipulate symbols. They can not relate to real word because they had no learning from early days of their existence to keep processing more and more data from their environment and build their sense of the world. Because of this issue , critics argue computers will never be able to acquire huge number of real concepts, senses and experiences we take for granted.
It is for this reason that much of AI, as well as cognitive sciences research, has been preoccupied by trying to duplicate in any way, obviously very limited, process of acquisition of concepts by small children. Most of those communities view this capability as de facto necessary condition for any intelligence.
But why would we think that real experiences at the level of small children, even very smart ones, are what superior intelligence at human level is about? Key observation is that human genius is and has always been at the level of abstractions. Even ancient Greeks understood this fact, as they were the first who started to introduce them into general thinking. There is no such thing as a real right-angled triangle, nor parallel lines continuing to infinity, nor a parabola. All those are abstractions we manipulate in our minds, very much like a computer would.
It takes a long time, even for the smartest of children, to start thinking in terms of abstractions. It commonly starts to develop in adolescence, though there were some exceptions at genius levels, such as Gauss inventing law of arithmetic progression at the age of 7, or Mozart starting to compose even earlier. But note that even for them, those miraculous acts at early age were only the beginning. It took them years too to come up with their other great discoveries and masterpieces.
Consider Russell’s paradox and its key concept of set of all sets that do not contain themselves. Not only is such a set not real in any sense but its very definition strains our brains in trying to comprehend and understand it. But that is not a problem, we do not need to understand it to figure out that the question of whether it contains itself leads to a paradox. One may object that this example is too contrived and exotic. Fine, take a very common term such as markets and let us ask its meaning. This term has many different meanings to different people. A farmer in marketplace will think of it much differently than an economist, trader, venture capitalist, hedge fund manager, central bank governor or a president. Not only do they think of it very differently, their own individual understandings and interpretations will change and continue evolving with knowledge and experiences they gain.
The key point of all this is that meaning and semantics of abstract concepts, which is what preoccupies superior intelligence of any kind, is defined in terms of relationships among such abstract concepts. The more relationships a genius learns, the more they are able to come up with great unexpected ways of looking at things, again creating unexpected new concepts and relationships. Note how the body and quantity of the previous knowledge is absolutely key as e.g. Archimedes would have discovered many more things had he had known about basic calculus. This is also what Newton meant when he famously said that he “stood on the shoulders of giants”.
Now we are starting to see a glimpse of something familiar and very powerful, which is a graph. Its nodes are abstracts concepts in the most general sense and its edges are relationships between them. And it is not just any relationship we would be looking at, but instead we are going to focus on a relation which is key to reasoning and inference.