schoutbynacht

What is conscience? Some thoughts in light of recent LLM-inspired debates.

Every human reading this text is conscious.
Every machine reading this text lacks consciousness, and alway will.
These are my starting points and also my ending points.
Much ink has been spilt on this topic. I hereby add my own. With the arrival of LLMs we are faced again with the Turing test, but we’ve got it wrong. The test is not about how good a machine must be to fool a human, but it’s about how bad it can be, and still fool a human. It’s really a test about human psychology, not about computers.
And once we realise this, we also realise that we are all too ready to impute sentience to machines.
I will now consider my two Beagle dogs. They behave very differently, and I’ll omit the details. Trust me on this one. I interact with them as one consciousness to another, and I have no reason to doubt that they are indeed conscious. I take it as a fact. The fact that these two dogs have such distinctive behaviours can only meaningfully be explained by saying they experience the world differently. When Lavendel looks at me she feels compelled to come to me for a pat on the head. Lucy just looks at me and stays. Think about that.

The hard AI position says that everything can be computed, that consciousness, even the world, is just a computation.
I see no reason to accept that statement. All computation relies on modelling and representation: the model is NOT the object it models. This is also true of mathematics. Numbers are abstractions, and boy are they amazing tools for describing the world. Their unnatural nature sometimes reveals itself, in such concepts as infinity and the sums of infinite series. These phenomena are entirely counterintuitive to us, and my take is that these are artefacts that happen to be an intrinsic part of the abstraction. As an example, 1/3 is well defined: divide something into three equal parts. Written in decimal, we get 0.33333 etc, but if we were to choose 3 (not 10) as he base, 1/3 would be written “0.1”, and that would be it. What would Pi look like in base 3? 10.0102… etc. Pi is the ratio between physical properties of a circle. It’s a property of the world, and exists independently of our attempts to describe it. Key point: mathematics is not the world, either; it’s a modelling language.

We can put numbers into a computer and compute, but the output is still within the same category of things, within the same ontology. Numbers, or their decoded formats, like images, sound, text (LLMs treat numbers, not words).

What about humans? We are different, because we are in the world. We are part of the world, and our brains are also part of it. Our brains don’t run a simulation of the world. Yes, our ears and eyes and peripheral nervous system act as transducers that convert one physical signal into electrochemical signals sent to the brain, so the brain only senses the world indirectly. But crucially, the brain does not execute a program. It is the “program”. The word “program” is already misleading us, because it presupposes a distinction between hardware and software. The brain is much more akin to an analogue computer. Analogue computers have also been made by humans ( all digital computers are strictly speaking analogue computers that we restrict to operate in a digital fashion : we run a digital computer on an analogue platform, and run a software program on top of that).

Our neurons respond to incoming signals and send out transformed signals. A computer can model the biological process via a mathematical abstraction, with two fundamental buts:
– the abstraction and the physical object are logically distinct and categorically distinct – they belong to different realms with entirely different properties
– we cannot know that the model is “correct”

There is every reason to assume that consciousness arose through evolution and is linked to brain activity in mammals. There is zero reason to assume that this is in any way similar to mathematically modeled states in a digital computer. Consider if the human-made computer were analogue, consisting of transistors acting in a linear mode with millions and millions of interconnections in a brain-like layered architecture, like a neural network. Would we expect it to be conscious, this huge mass of transistors? No, of course not. So why impute consciousness to a computer program? To do so is a very very strong claim.

A computer takes a model as input and outputs a new model. This model must be interpreted by someone, by a consciousness, or else it remains a model, an abstraction, a representation.
Consider a computer the produces a drawing of a flower on a screen. Now consider that this screen is “observed” by another computer or a human. Where does meaning arise? It’s evident that the seeing computer can only transform the “visual” representation on the screen into some other state – simple or complex. It could output an audible “beep” upon detecting a flower, or do something else. But the human has a conscious experience, and an emotional one, too. Radically different phenomena are at play.

Summing up: the computer operates on a mathematical model of the world, a representation. Your brain and mine are in the world, wired directly into it and obeying the laws of physics (and chemistry!).