I am somewhat of an amateur to this topic so I just want to put forward an idea and I would be happy to be educated what the current state of the art in that direction is.

One of the big questions is how to get an AI that is conscious. It seems to me the surest way to get that would be to train for it, rather then let it emerge by chance. To do that I would imagine to use problems that need one or more properties associated with a "self" to solve it.

E.g. catching a ball that was thrown. To do that the self needs to know its position in space relative to the ball thrown. Maybe an intermediate layer could be trained for the current position, while the output could be trained for the velocity and direction to go to catch the ball. This should help to establish the concept of ones own position in space for the AI. Of course there should be many different problems that help solidify the same property.

Many different properties that can be associated with a self should be trained for. The nodes that correspond to those properties should probably be closely connected, to establish some kind of self core where that information is compressed into.

One could start with problems that require one dimension of the self core to solve, respectively. Then move on to problems that require multiple ones. I suspect the higher the number of dimensions of the "self core" the closer it would get to a state we would recognize as consciousness.

My question would be, whether some kind of approach like that was ever investigated? Or alternatively, why this approach would not be feasable or would not work?


1 Answer 1


Well, let's say you train some robot to catch a ball (like in your example). Your training data would be, for example, a stream of images (captured from a camera), and the (relative) x-y coordinates for the "self" part.

Even if you do so, and the robot is able to correctly catch the ball, you still haven't achieved consciousness: making it know its position in space, is not the same as knowing that the code that runs on it (the robot), has actually a body that can move, with sensors, and that it can reason about possible movements (e.g. would such trajectory miss the ball? or would I damage myself? or even harm someone?)

Consciousness is something way more powerful than training a control policy. It involves self-awareness and reasoning at very least. And I guess we should first understand how such works in the human brain, before trying to mimic it.

Even if you equip the robot with the largest LLM (e.g. ChatGPT), and you ask it if it's conscious or even if it has a body, well, probably it would respond: "yes, I've a body because I'm a robot, I'm designed to etc". But that either does not mean anything, unless it won't be a creative copy paste of some piece of text encoded in billions of neural networks weights.

  • $\begingroup$ Catching the ball was just an example and the idea was to train a net that can reason about itself in many dimensions, and not even exclusively physical ones. Like you mentioned, not only position, but rotations, limb positions, maybe limits of the vision field, but also maybe mood, recent rest quality (or loading status), and so on. I suspect the higher the dimension of the "self" core the closer it would get to something we recognize as consciousness. Maybe a main requirement to solve the problem would be to actually give it a physical body. $\endgroup$ Commented Jun 3, 2023 at 6:53
  • $\begingroup$ Neural nets solve problems without us understanding how it works. We can try to understand it in hindsight, but it is not necessary beforehand. So maybe through neural nets we would be closer to creating an AI rather then understanding it and the understanding could come later. $\endgroup$ Commented Jun 3, 2023 at 6:56

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .