Phantasian Biology 101: Lesson 4; Brain

Phantasian brains are going to be a really big topic, so I thought I’d better dip my toe in by first asking a simple question: What does a brain actually do?

Rather like the parable of the blind men and the elephant, I guess the answer to this depends very much on who you ask. So obviously I asked ChatGPT, because it would know, right?

Everybody’s favorite LLM has clearly digested a lot of school textbooks in its time, because it actually gave me a pretty good answer, but the core of its response was that “the brain is the command center of the body”, which I thought was quite interesting and reminded me of the Numskulls from the Beano (if that means anything to you). So, if the brain is metaphorically the command center of the body, like the field headquarters of an army unit, then who is the commander? Who is receiving all the incoming signals intelligence and issuing the orders in response?

The philosopher Dan Dennett would probably have argued that this is related to a fallacy that he called the Cartesian Theater – the idea that our first-person conscious experience of the world feels rather like watching a play, as if there must be a little person inside our head who is busy watching what comes in through our eyes. But as he points out, that little person must have another little person inside their own head, to watch the play for them, and so on ad infinitum. Where does the buck stop? Dan was a very sweet and thoughtful man. I was once lucky enough to spend a week with him and a handful of other thinkers on the island of Lanzarote. He even wrote some blurb for my book, and I really wish he were still alive to hear what I’ve been up to since then. But anyway, consciousness and the Cartesian Theater will have to be topics for another time. Let’s assume for now that nobody is in charge at all – somehow the input signals just mysteriously give rise to the correct output signals all by themselves, like an echo off a distant wall. This will actually come up again in a future post, because such an echo may not turn out to be as mysterious as it seems.

But that was ChatGPT’s take on brains, and it’s fair enough. After all, it only told me this in the first place, because thousands of other people have said exactly the same sort of thing when asked this question, and it was basically copying them. It doesn’t actually know what a brain is, and has never seen one. It certainly doesn’t possess one. If we peer into ChatGPT’s Cartesian theater, we’ll find that there’s nobody home, despite what we’re often encouraged to believe.

But while we’re on the subject of AI, it has always struck me as remarkable, just how incredibly impoverished many AI researchers’ views of the brain tend to be. It’s as if they think intelligence is monolithic – a single thing, with a single unified logic to it. Quite often, intelligence seems to be defined more or less as ‘the thing that mathematicians do’, especially when the person offering this definition is a mathematician.

For all the excited talk about AGI (artificial general intelligence) in recent years, it’s still often a bit… specific? I went to Ben’s first workshop on AGI more than a decade ago, and I got the impression that many people there thought that The Answer™ to AGI would lie in Bayesian statistics. Well, okay, but that’s like defining America as ‘very big’. It’s not all that helpful. A big what? No mention of the Rockies? Cheeseburgers? Democracy? Mickey Mouse? It may well be true that lots of intelligent behavior is Bayesian at some level, but the fact that ‘X is a Y’, does not in itself mean that ‘Y is an X’. I’d expect mathematicians to know that.

Bayesian statistics aside, The Answer™ always seems to be a single thing. One of the people at the meeting thought that it was holography. Another suggested that it was evolution. In Consciousness Studies, things get even crazier: consciousness is quantum entanglement, or it’s electromagnetic coherence, or it’s infinite recursion. But for artificial intelligence it’s always something that people hope will be patentable.

I think the origins of this incredibly monolithic perspective are largely historical. I was talking about this to a new member of our group, just the other day, in relation to the Turing Test. Basically, the very first digital form of artificial intelligence was… <drum roll>… programming. That was it. That was the whole answer. No fancy Chomskian grammars, no Big Data, no Bayesian statistics, at least not to begin with. Just plain programming.

The hint lies in the name ‘computer’, really: The digital computer was designed to emulate human computers – people who performed calculations as their job, by blindly following instructions one step after another. Since human computers clearly needed some intelligence to perform these calculations, and digital computers could also do the same calculations, then, in some vague sense, the digital computer was acting intelligently. Obeying a set of IF/THEN statements was the computer ‘making decisions’. It’s quite easy to forget, in this age of video games and websites, that there used to be an almost one-to-one correspondence in many people’s minds, between what computer programs do, and what human minds do. It’s largely what the computer was invented for.

Very early on, digital computers started to be called ‘giant electronic brains’ for this exact reason. And the internal structure of a computer is replete with brain-related or psychological metaphors: They have a ‘memory’ and a ‘central processing unit’ (akin to the ‘command center of the body’); they understand rules and routines, and they conduct queries and make decisions; they follow instructions, ‘wait’ for input, ‘go to’ other places in the code, as if on foot, and so on. And of course, at the heart of a computer lies formal logic, which had already been tied to intelligence for a century.

Unfortunately, this borrowing of neurological and psychological metaphors by computer scientists quickly reflected back on itself, and cognitive scientists began to use computing metaphors (like bandwidth, say) to explain the brain and mind. A direct correspondence between the computer and the mind quickly became entrenched, and is still quite prevalent today, at least as an unspoken assumption.

But my point is, AI was once just programming. All we had to do was program computers to do intelligent things, and then the problem would be solved. It didn’t actually work out like that at all, for reasons I’ll talk about another time, but I think the die was cast very early on – people started to imagine what the human brain does, in quite ‘algorithmic’, symbol processing terms. Intelligence is what mathematicians do, basically. So let’s just not ask any AI researchers what the brain does, eh? My blood pressure won’t stand it.

However, this is not how the brain looks to a neuropathologist, at all. They worry about what happens when brains go wrong, and it turns out they can go wrong in an amazing number of different ways. This is because brains are not really monolithic computers designed for thinking with; they also control our heartbeat, breathing, body temperature, hunger responses, eye blinks, sexual arousal, immune systems, kidney function, growth, muscle tone, balance, and a million other things. It’s a ‘command center of the body’, for sure, but we wouldn’t get far if we had to stop breathing while we wait for the ‘commander’ to finish a thought, or if the commander had to direct us to move each muscle in turn, in order that we can pick up a cup. Nor does it seem likely that each of these functions is controlled in the same way. The sympathetic and parasympathetic nervous systems seem very different from the parts of the brain we tend to pay attention to. They don’t always even get along with one another. And we have a whole other brain in our gut, for that matter, which probably has very much its own impression of what brains are for.

Of course, an evolutionary biologist might answer the question by asking us which brain we’re actually talking about. Do we mean the brain of an ant? A worm? An octopus? A cow? They’re all quite different. Sometimes, even closely related creatures might work in significantly different ways. Is being a bat really all that much like being a mouse? Do deer see a world made of objects, as we do, or is theirs a world made of motion? There are many different ways to be alive, and each one has its own requirements. Having eight tentacles is a very different control problem compared to having two wings, or just one foot.

But it’s not necessarily a jungle of irreconcilable complexity out there, either. For a start, there are really only two things a brain can do, regardless of who it belongs to: It can contract muscles, and it can secrete chemicals, such as hormones. That’s it. Everything else results from a combination of those two things. The slithering motion of a snake, the fear we feel from seeing one, the whispering of sweet nothings into someone’s ear, the writing of a scientific paper, the making of a nest… Whatever it is that goes on between a creature’s ears, it all comes out as patterns of muscle contraction and the triggering of secretory cells. So in some ways we’re all different, and yet we’re all the same.

The list goes on: Ask a cyberneticist what the brain does, and you’re likely to be told that it’s a homeostatic regulatory system – it prevents us from becoming too hungry, too cold, too bored or too lonely. Its job is to keep us in our comfort zone. This is actually a very useful way of looking at what brains do, although it somewhat sidesteps the question of how any of this happens. Ask a clinical psychologist, on the other hand, and you might find they’re more interested in issues like how the brain expresses itself through personality, or social engagement, or mood. Or how it perceives itself – its sense of self-worth, self-admiration or self-loathing; its body image. A clinical neuropsychiatrist, meanwhile, might note that brains will sometimes see things that aren’t really there, or hear voices, or claim that the arm that’s attached to their left shoulder actually belongs to somebody else. To them, this is what the brain does. Crazy things.

If we ask ourselves what brains do, we’re likely to grip our popcorn and focus on our role as the audience in our own personal Cartesian theater. In other words, we may be less interested in what goes out, and more likely to pay attention to what comes in. To many people, the core functions of a brain involve seeing, hearing, recognizing, noticing, contemplating, and feeling. Those are the aspects we’re most aware of, because without them we wouldn’t be anything at all. And for many people, memory seems justifiably key to the whole riddle. We are remembering machines. But again, which kind of memory? Remembering how to ride a bike is radically different from remembering to go to the dentist, or remembering who sang ‘Let it be’, or remembering what we did yesterday, or remembering who we are.

I’m not offering any answers, here. I just wanted to encourage you to think about how multi-faceted the subject really is, before we start talking about how phantasian brains do what they do.  If we go into this thinking that a brain is an algorithm for intelligence, then none of it will make much sense.

Next time, I’ll probably talk about the two main metaphors behind phantasian brains – maps and servos. We’ll look at how those two ideas can be combined to make a system that automatically looks towards something that moves. This may sound like a ridiculous non-problem, but it lies at the core of their brains and introduces the key structural elements of how they work. After that, we can move outwards into how they learn to recognize the world around them, how they decide which things to pay attention to, how they (occasionally) learn not to do stupid things, how their genes define their instincts, and so on.

P.S., I’m still slogging away with the user interface, I’m afraid, for which I’ve had to do a fairly massive rewrite. It’s funny how one thing leads to another in ways that might not be obvious from the outside. Or even to me until they hit me in the face. But I’m reasonably happy with the way it’s going, and it’ll be nice to get it out of the way so that I can focus on more interesting, virtual worldy things. I may publish a build while it’s only partly done, because it’s basically a problem in two parts, and one of them is mostly written now. I’ll see how quickly it goes, but in the meantime, thank you for being patient!