How To Build An AI (For Real) Continued
A Challenge From An Intrepid Reader Is Answered
Laying The Foundation and What Comes First
Before we can begin to build our AI we must confront one rather large and imposing problem. I think mostly everyone can agree on the artificial part, but what exactly is intelligence? The one thing I know for certain about it is that I cannot define it. Is it possible to build something, create something, if you do not even know what the “thing” is you are trying to build or create? For purposes of this series I am going to ignore that quite difficult question and just assume the answer is yes, such a thing is possible.
Let us try this approach. Perhaps we cannot define intelligence but we believe we are intelligent. At the very least I believe I am intelligent. I believe, I would even go so far as to say I am certain that I have intelligence. I do not claim it to be extraordinary or average or below average and I am not even sure that such evaluations or comparisons are possible given our lack of understanding of the nature of the thing. I also believe or would say I am certain that I know some other people who have intelligence. A number of them may even be reading these words at this very moment. All that said I definitely have it and thus I will suggest that we proceed with the goal of building/creating (I am going to use those two terms interchangeably from now on even though I realize they are different. Sorry but it is just too big of a pain in the ass to do the / thing every time and the project is too big to dedicate much time to side projects like parsing the exact meaning of each and every term used. Suffice to say that all terminology should be understood to be defined by their most average, everyday meanings. Any deviations will be noted and all technical terms will be defined precisely as required) our AI by using ourselves as the “blueprint” and ourselves and our abilities as the markers of success or failure. If our creation cannot do the things that we do with intelligence then it cannot be said to be intelligent. And here I know that I will lose some people. They will revolt and say “but we say some animals are intelligent and they can’t do the things that we do with intelligence, at least not all of the things, in fact not even most of them.” To you I say, you have a very good point however I have addressed it at length a number of times. Ultimately, I am trying to create the AI that the vast majority of everyday people think of when they hear the term, an ultraintelligent and conscious supercomputer type being with intelligence and capabilities that far exceed those of the most intelligent (if there is such a thing) human who has ever existed.
That part settled where should we begin? When we think about things we use our minds, our brains. When we do math and solve problems and such we use our brains. When we think about intelligence we use our brains to think about it and we think about our brains. Therefore we should start with the mind right? We should try to build a mind, recreate a brain, artificially. Wrong, wrong, wrong, and wrong. To build an intelligence we must first start with the body. We need the physical first, the sensory, the apparatus that allows us to interface with the real, physical universe. With apologies to Giulio Prisco and others who I know will object when I say this, The universe is not a mind, the universe is a physical space. Why do we think we need a mind first to understand it? We need a body first to explore it, to interact with it. We cannot build our intelligence without physical sensory modalities. The most basic and well understood (because it is us) concept/format for achieving a physical interaction with the universe is a body. It is important to emphasize that I am using the term body in the broadest and loosest sense, not the most commonly understood (remember my promise from above about definitions) as a human body with limbs and a head and skin and organs. Our “body” does not need any of those things. It could, and most likely in the end will be, just be a series of tubes lined with sensors and fractals, gotta have fractals. If you want to learn more about the importance, the requirement of the body, for consciousness and intelligence please start with Maurice Merleau-Ponty and his masterwork Phenomenology of Perception. You will learn a lot about this topic and also learn that he was an ass, a real dick who treated his patients like experimental animals, but whose brilliance in this area cannot be denied. He is French so there you go. I have to digress a sec and give just one quick example of a classic Ponty dick move. A number of his patients suffered from severe hallucinations. After studying and learning every detail about a particular patient’s repeated hallucination, he arranged to have the area in the hospital yard, which was the focus of this hallucination made up with real, existing, physical objects and people exactly as described in the hallucination. He had this done at night when the hallucinating patient was asleep and totally unaware. The next day when the hallucinator looked out the window he saw his hallucination brought to real life. Anti climatically it seemed he couldn’t tell the difference. Sick mother fucker right? Who does something like that? A Frenchmen that’s who. Lol.
Back to the topic at hand, I don’t believe a series of tubes lined with sensors is gonna cut it. To be intelligent requires more than a sensing body, that body needs to be capable of movement through space and sensing that movement (kinesthesi). Interacting with the universe which is made up of space we must be able to move through it and know we are moving through it. A motionless thing cannot be an intelligent thing. So let’s put our sensor tubes on wheels or legs or something. I know what you are thinking cause I’m thinking it to, how’s it gonna move? how will it control its motion? what about will? intention? action?, etc. To both you and myself I say slow the fuck down. We are nowhere near that level yet. We are still just putting together the physical pieces and laying out some criteria for things that our AI must be capable of to be intelligent. Capable is the wrong word because that implies intent or ability. We are not there yet either.
We are going to need some sort of way of knowing if we have succeeded or not. A way to evaluate if our build is truly intelligent. Some sort of test or game perhaps? To that end I have been thinking about Nash’s equilibrium. See my post on that topic for some initial thoughts and ramblings.