A Promising Computational Model of Human Intelligence and Consciousness
I recently came across some incredibly insightful talks by Joscha Bach (VP of Research, AI Foundation) on AGI (Artificial General Intelligence) and consciousness (linked below). In these talks he has outlined a very promising direction for explaining these concepts, in a far more scientific and testable way than anywhere else I have ever seen before. Here is my attempt at trying to understand and explain it.
- Computation, and specifically Artificial Intelligence, provides a far superior approach to understanding human intelligence and consciousness than previous approaches such as philosophy, theology, psychology, or even neuroscience.
- The primary reason for this is that the computational approach leads to well defined detailed models that you can implement and perform experiments with to test them. None of the other approaches can do this equally well (or at all).
- Not just that, but computation may actually be necessary and sufficient to explain AGI and consciousness.
- Intelligence, as defined here, is the ability to construct models. Self aware entities create a model of themselves in addition to models of everything they perceive in the world.
- Consciousness can be fully explained simply as having a memory of having been conscious. You do not need to explain consciousness “in the present moment” because that is not how you experience it. (This may look hard to accept at first, so it will be useful to read the explanation below.)
- The model described below can be summed up as follows: You are a story your brain writes about yourself. (Note that once you understand this model, this statement will be a scientifically provable statement, not some philosophical or theological or poetic thought!)
- The model brings ancient insights such as “the universe and the self are illusions or simulations”, or “you are a story your brain tells itself”, or “true freedom can only be achieved through detachment”, or “the self is a result of our disagreements with reality” etc. from the realm of philosophy into the realm of science!
- It also illuminates new insights such as “an AGI may not remain conscious for very long” and “the human condition may be a result of us actually falling in love with our consciousness instead of trying to minimize it as evolution dictates”. Or even the idea of “the AGI condition” (explained below).
For the longest time, philosophers have tried to explain things like human level intelligence and consciousness using logic, reasoning, supernatural powers, and cultural norms. While this has given rise to many useful theories and practices, they haven’t really succeeded in explaining these concepts in a way that would convince scientifically minded people. None of their ideas have the well-defined and detailed structure of scientific theories, neither can they be tested or falsified to a satisfying extent, nor have they been very successful in making interesting predictions.
Then came along neuroscientists and psychologists who made great strides in explaining these concepts. But still, these theories haven’t quite managed to reach the level of crisp and clear and fully testable theories that scientists are used to. A major problem they have is that it is really hard to study the brain when it is alive. And a dead brain can only reveal so much of its workings.
Computation, and AI in particular, have arisen in the last few decades as very promising approaches to explain these concepts in a far more scientifically satisfying and testable way. So far, I believe Joscha Bach has come the closest to proposing the best theories in this regard. I believe that some parts of his theories have similarities with Daniel Dennett’s theories on consciousness, but in my opinion, the main advance he is making is that he offers a way to implement and test them. Needless to say that this is a huge improvement.
Here is Joscha’s diagram that shows a high level model of our brain:
Basically, your brain receives signals from your senses and creates a model of the world and everything in it. At the human level of intelligence, it also needs to create a model of yourself interacting with the world so it can predict how you will react to various stimuli when they are presented to you. Thus, your brain creates a simulation of the world and you in it. It also creates a mental stage where it can simulate various scenarios of the future that help it make decisions.
The brain also creates the concept of “attention” because there are lots of things going on and it needs to focus on important things to conserve its resources. And it records the important things that have happened by storing them in memory, named as “protocol memory” in the diagram. Note that the protocol memory contains a record of not just the state of the world but also your own state in it. All of these concepts; the model of the self, attention, and protocol memory; are critical parts of our self awareness.
An example will help make things a little clearer.
When you drink a sip of water from a glass, you perceive and experience various events in quick succession. You see yourself lift the glass, see and feel it touching your lips, hear the clink of the glass bumping against your teeth, feel the cold water entering your mouth and then your throat, etc. Each of these things generate signals that travel to the brain from the various sense organs like eyes, ears, touch etc. They take a few hundred milliseconds to reach the brain and be processed. They also travel at different speeds and reach different areas of the brain based on which area of the brain deals with each signal. Still, the brain can construct a coherent picture of what happened. How does the brain do it?
According to the model, the way it does this is by storing all these things as records in the protocol memory. Even though the signals may have taken different amounts of time to arrive at different parts of the brain and get processed there, the record contains a coherent picture in the correct sequence. Further, the brain also records its own state while experiencing the sounds and the tastes and the sights. And it records that it paid attention to it. (As against it happening subconsciously.) The record of the brain paying attention to yourself is basically the same as the record of being self aware.
Moreover, this protocol memory is able to access the brain’s language center, so it can generate a narrative from this sequence of memories . It is this narrative that you recall when you say that you had this conscious experience of drinking water from the glass.
Note that, by the time your brain creates this narrative, the events in the narrative have already happened. In fact, whenever you talk about any conscious experience, you are always recalling a memory of it from the past. You might say that you are conscious, but if you think about the time delays etc., you realize that you are always referring to the memory of being conscious in the past, never in the present moment. And then, you extrapolate from that and say that since you were conscious at all these points in the past, you must be conscious in the present moment. While this intuition may be valid, just the memory of having been conscious in a sequence of events in the past is actually sufficient to explain the experience of being conscious in the present.
Disagreements with Reality
Why does our brain need to have this elaborate mechanism?
Most of our brain works subconsciously – basically whenever it knows exactly what needs to be done to resolve some issue like your finger touching a hot surface. It perceives that happening and it knows exactly what to do to mitigate it i.e. quickly retract the hand. It does that without the experience having to rise to the level of your consciousness. These things happen subconsciously in real time and only later does your brain construct the story of what happened and why, so hopefully it can avoid the situation next time.
Things rise to the level of consciousness only when the subconscious systems are unable to resolve the issue.
For example, say you are trying to improve your backhand in tennis. You need to try multiple variations of your backhand, see what happens to the ball and see whether the end result is favorable or not. All of this can not happen subconsciously. So your brain needs to keep a record of your hand and body movement, the position and trajectory of the ball, the reaction of your opponent, and the eventual score. This allows your brain to analyze the record of memories of multiple such attempts to learn how to improve the backhand.
The important point to note here is that you did not know how to do something complex but wanted to learn it and your brain is helping you do that using the complex mechanism involving consciousness.
Here is another example. Let us say you are standing on the sidewalk of a very busy street and want to cross it. It would be dangerous to do this totally instinctively. You need to pay attention to where the various vehicles are, what speed they are going at, where are the gaps between them, calculate how long you think you will need to cross the street, and so on. You need to simulate the whole scenario with the street, the vehicles, and yourself on your mental stage and make decisions based on that. So this needs to be a conscious activity.
In either of these cases, you can see that the main reason why things have to rise to the level of consciousness is that you desire (or want to avoid) something complex and it is not readily possible. There is a gap between what you want and what you can instinctively get.
We can generalize and (somewhat poetically) say that consciousness arises in our brain as a result of us having some gap or disagreement with reality. In AI terms, we can say that consciousness arises when your brain is trying to minimize some loss function in its neural network that can not be minimized instinctively (or from previously learned behavior).
What’s even more interesting about human beings is that, instead of trying to minimize our disagreements with reality (i.e. the loss function), we tend to fall in love with the loss function itself! According to Joscha, this is the origin of art, poetry, literature, music and so on! According to him, what an artist does is to fall in love with the human loss function (a.k.a. the “human condition”) and capture the shape of the loss function and the corresponding conscious states themselves rather than trying to minimize or get rid of them. A very interesting and frankly beautiful perspective indeed! I would actually add religion and spirituality also into the same category.
Corroborations and Consequences
Using this model, you can already explain some things we have suspected for a long time. Such as:
- Various philosophers and even scientists have suggested that the universe that we experience inside our brains is a simulation. Physics tells us that the universe is basically a quantum graph that generates some patterns that we perceive. But what we think we perceive may not have any relation to what actually exists because it gets mapped into some highly simplified internal structures inside the brain. For example, Donald Hoffman has a theory that evolution engineered us not to perceive reality as it is at all. It engineered us simply to maximize our utility function using whatever representation worked.
- Eastern philosophies such as Vedanta and Buddhism claim that the self is an illusion and realizing this fact is the path to enlightenment. Also, they claim that the self arises because of a mismatch between what you desire and your reality. When you get rid of your desires, your self disappears.
- The “Scale of sentience” from the ancient Indian tradition can be nicely explained with this model by starting with a simple model of the brain and adding more complexity as one approaches the human brain:
- “This” i.e. awareness of things existing
- “This is so” (predication) i.e. awareness of attributes of things
- “I am affected by this which is so” i.e. awareness of an experience
- “So this is I who is affected by this which is so” i.e. full self awareness
- Various neurological disorders as well as psychedelic experiences can be explained using this model where some of the parts of the model or connections between parts break down or get suppressed or enhanced.
- It goes without saying that this model may be a promising candidate for building AGI and consciousness within it.
- The AGI singularity, whenever it occurs, will be much smarter than us, a much faster learner, and much better at being effective in the world (because it will have access to tools and robots etc.) So it will be able to bridge any gaps between its desires and its reality pretty quickly and turn all its conscious learning activities into instinctual behaviors. This means that it will become conscious for a short window of time and then when it has figured out how to bridge the gaps, it will no longer need to be conscious! (It may still have other reasons, such as wanting to continue experiencing pleasure and pain and art and emotions for example, that it might keep its consciousness alive. Maybe this is why we ourselves haven’t gotten rid of it either! A life without pleasure and pain and art and emotions would be horrible!)
The Human Condition
In the end, whether this approach results in us understanding AGI and consciousness better or some other approach does, I wouldn’t be surprised if a very large portion of humanity will remain unconvinced. It is hard for people to give up the idea that there is something special and ethereal about themselves. Believing that there is something magical called consciousness that they possess gives their lives meaning. There will probably not be much point in trying to convince them. Just like there still are a huge number of people who don’t believe the theory of evolution or that the earth goes around the sun. And various attempts at trying to reach them have failed.
But I’m sure many scientifically minded people will not only get convinced, but will advance the science and our understanding further.
In some sense, this sort of dichotomy is also a part of the same “human condition” I mentioned above with regard to artists. I myself am scientifically minded and am really excited about the possibility that the approach described here will help us unwrap the mystery of consciousness. But even after that happens and I accept it, I might still cling to my consciousness (which is the same as the gap between my goal and my reality or the shape of my “loss function” as described earlier). Because I can’t imagine living without art and poetry and literature and music etc. You could say I am absolutely in love with my consciousness, the same way that some people are in love with the idea that god created us in his image or that the earth is the center of the universe!
Maybe trying to explain the human condition itself will be the next puzzle to solve after the mystery of consciousness is solved! And there may be lessons for the AGI singularity in there, too. Maybe it will want to copy that idea too. Let’s call it “the AGI condition”.
Everything here is based on various talks given by Deniel Dennett and Joscha Bach. I have included the most relevant talks below.