The image that we have in society as a whole of what a true artificial intelligence will be like comes from the movies. Almost exclusively from Hollywood. These intelligences come, sometimes inside robots more or less similar to us, at other times they are supercomputers. But in any case, what characterizes these artificial intelligences is their intellectual resemblance to us, humans; it makes a lot of sense that it is so, because the human is always the measure of everything.
These artificial intelligences act and think as if a real human brain were inside a machine: more like a droid than a completely synthetic processing center; a machine, inside which was a completely human brain.
Basically we have in our imagination the idea that an intelligent machine is a computer with human behavior. There are probably many projects in the world looking at ways to mimic or simulate human intelligence inside a machine. It seems like a good idea, since we would get to have a lot of new intelligence, produced at an industrial level, that could be harnessed for endless useful uses. However, this seems like complete nonsense to me.
There is a reason that makes me suspicious of this approach of pretending to match the human mind.
The reason is simply practical: if we imitate the human mind, we will also imitate its defects. I don’t think we know how to develop a human mind, avoiding copying its defects, such as poor short-term memory, or our ability to forget, or even distort memories. Let’s imagine a human mind inside a robot: when we ask it to iron our shirts, it could well tell us no, that it prefers to watch television.
Jokes aside, a human mind inside a computer would suffer the same limitations that humans suffer today when it comes to accessing said computer. Huge amounts of data and processing of such data, but we are unable to assimilate and understand such processing and data. The same thing would happen to a digital human mind.
For all this, I do not believe that it is the imitation of the human mind, in a digital medium, that we should investigate and develop. Our needs are others.
What would these needs be? Well, the same ones explained above: our inability to access huge amounts of processing and data. Any development of AI must be directed towards improving the human-machine interface. Starting with developing a natural language interface (much like Jarvis’s origin in the Marvel movies was a simple natural language interface, he ended up running a big company).
The current man-machine interaction systems are actually ridiculous: the keyboard, the mouse, the screen. We hardly use other means such as video camera or loudspeakers to collect or transmit information to the user.
A true future AI should be able to, seeing the movements and listening to the explanations of a human, imitate these movements and match them with the explanations. For example: an operator would explain to the robot that he must take certain objects, put them in a box and shrink the box, putting the label that appears on a certain printer.
For this, we must focus on man-machine communication. Instead, it seems that we are focused on developing logics, machine learning procedures: but in reality what we need is to develop languages. A very special language.
The man-machine interfaces that we usually use are actually just a mirage of reality, a simplification of the true man-machine language. These interfaces are created to facilitate the use of computers by humans, but they are not the true man-machine interface: the true one is programming languages.
For all these reasons, my theory is that we must create a programming language whose inputs are words, gestures, sounds, images, and the result of the processing are machine commands: the instructions that make computers process, and robots move their actuators.
It is a programming language that we must develop. Just as languages require a compiler or interpreter to run, this AI language requires its own interpreter, but one trained by ML.