Cortical Labs: "Human neural networks raised in a simulation"

Ah, the billionaires. Boogeymen of our times. The scourge of all us crabs living here in this fine bucket.

> Nobody would argue that turning off ChatGPT is killing a sentient being.

Some do, judging by surveys.

And there was a famous case last year with a different LLM: https://en.wikipedia.org/wiki/LaMDA#Sentience_claims

It probably depends on your idea of what the soul/consciousness is and how it interacts/integrates with the physical world.

There's a few people interested in the idea of (simplifying a lot) "brain as an antenna receiving the consciousness signals".

The belief of being no one can be a truth, the belief of being someone an advantage.

We know that the human brain is able to generate qualia (conscious experiences) despite having no model for how these are generated. (To be clear, by consciousness, I mean the ability to have conscious experiences such as experiencing a color or pain, not self-awareness.) On the other hand, the hypothesis that a Turing machine on its own could generate conscious experiences leads to many seemingly absurd scenarios. Notably, one has to ask how a simulation of supposedly conscious Turing machine using pen and paper could possibly be conscious, or indeed, why one would need to "run" a Turing machine for consciousness to arise and why a mere description of it would not suffice. And how could the mere description of a Turing machine (or equivalently, some C code) be enough for all of its unlived life's consciousness to manifest? If that were the case, one would have to concede that the set of all possible conscious Turing machines is conscious and their experiences are manifested already. If that were the case, then it's hard to see any point in moral reasoning, so for the purpose of debating moral and ethics, I think we can rule this out.

Now, one might propose that consciousness only arises when a computational process is physically run in certain ways but not others (this is what proponents of Integrated Information Theory (IIT) typically believe). Assuming this is the case, then implementing a potentially conscious process with biological neurons presents much higher moral hazard vs an implementation of the same process electronically, or safer still, on a Von Neumann machine.

I would go even further however, and propose that a conscious being (e.g. a being capable of generating the qualia of the color red for instance) cannot be simulated, i.e. conscious processes are generally noncomputable. Why? Well, consider what Chalmers calls the meta-problem of consciousness, which is to say the problem of why we perceive there to be a (hard) problem of consciousness in the first place (and why we are having this very conversation). A simulation of a conscious being would by definition present the same behaviours as that being (given the same stimulus, but for simplicity, we can consider the stimulus as part of the simulation itself without loss of generality). Therefore a simulation of myself for instance, would generate the same thoughts about consciousness itself, and this very same text. But, if we accept the proposition of my first paragraph —that a Turing machine on its own cannot be conscious— this would imply that our whole thought process surrounding consciousness and indeed our very belief that we are conscious is purely coincidental. After all, the unconscious simulation of myself would claim and "believe" just as strongly as I do that it is conscious while that is not the case, which means that the process by which it derives these thoughts and conclusions would be wholly unrelated to the object of these thoughts (actual consciousness).

As such, it is my fairly strong belief that there is some physical "device" in our bodies which allows us to generate qualia and get feedback allowing us to store a record of these experiences. If I had to guess, I would say that this "device" is very likely located in our brains, and that it is quite likely spread throughout our neurons and possibly each one of them.

It should be noted that although I do not believe I can be simulated in my entirety for the above reasons, I do believe that I could likely be emulated with a high level of accuracy from the perspective of an outside observer. Actually we see this already with ChatGPT being able to play the role of a conscious being. But unconscious objects appearing conscious is nothing new in a sense since even a novel (especially told from a first person perspective) can be thought of as such an object already. It will be very interesting to see whether AIs trained without reference to the concepts of consciousness (a hard task to filter that out of the training data!) will ever present signs of consciousness. That would certainly put into question my above philosophical reflections.

---

So to summarise and answer your question more directly, I think there is something about our universe that allows for the generation of conscious experience and that our brains, likely on the neural level, have a bidirectional interaction with this something. As to what this is and how it works, I have no clue. Roger Penrose for instance put forth the idea that this might be related to quantum mechanics and certain molecular structures in our neurons capable of interacting with the quantum world in specific ways, but this is still pure speculation.

More importantly, we know from our own experience that interconnected biological neurons processing information and put under stress (rewards and penalties) are capable of generating conscious experience including very negative ones. And, whereas I believe there is good reason to assume that Turing-equivalent processes such as electronic circuits are not capable of consciousness, I strongly believe that artificially created biological neural network are very likely to be conscious, perhaps even at a fairly small scale already.

So, yes, I think any work creating artificial information processing systems using biological neurons needs to very tightly regulated, if not stopped entirely. At the risk of sounding dramatic, we might accidentally create hell on Earth if we are not careful, at least if biological computing ever becomes competitive with transistor based computing, which until now I'm glad has not looked to be the case... but I'm starting to worry.