Fast Takeoff in Biological Intelligence

[Speculative and not my area of expertise; probably wrong.]

One of the possible risks of artificial intelligence is the idea of “fast” (exponential) takeoff – that once an AI becomes even just a tiny bit smarter than humans, it will be able to recursively self-improve along an exponential curve and we’ll never be able to catch up with it, making it effectively a god in comparison to us poor humans. While human intelligence is improving over time (via natural selection and perhaps whatever causes the Flynn effect) it does so much, much more slowly and in a way that doesn’t seem to be accelerating exponentially.

But maybe gene editing changes that.

Gene editing seems about as close as a biological organism can get to recursively editing its own source code, and with recent advances (CRISPR, etc) we are plausibly much closer to functional genetic manipulation than we are to human-level AI. If this is true, humans could reach fast takeoff in our own biological intelligence well before we build an AI capable of the same thing. In this world we’re probably safe from existential AI risk; if we’re both on the same curve, it only matters who gets started first.

There are a bunch of obvious objections and weaknesses in this analogy which are worth talking through at a high level:

  • The difference between hardware and software seems relevant here. Gene editing seems more like a hardware-level capability, whereas most arguments about fast takeoff in AI talk about recursive improvement of software. It seems easy for a strong AI to recompile itself with a better algorithm, where-as it seems plausibly more difficulty for it to design and then manufacture better hardware.

    This seems like a reasonable objection, though I do have two counterpoints. The first is that, in humans at least, intelligence seems pretty closely linked to hardware. Software also seems important, but hardware puts strong upper bounds on what is possible. The second counterpoint is that our inability to effectively edit our software source code is, in some sense, a hardware problem; if we could genetically build a better human, capable of more direct meta-cognitive editing… I don’t even know what that would look like.
  • Another consideration is generation length. Even talking about hardware replacement, a recursively improving AI should be able to build a new generation on the order of weeks or months. Humans take a minimum of twelve years, and in practice quite a bit more than that most of the time. Even if we end up on the curve first, the different constant factor may dominate.
  • We don’t really understand how our own brains work. Even if we’re quite close to functional genetic editing, maybe we’re still quite far from being able to use it effectively for intelligence optimization. The AI could still effectively get there first.
  • Moloch. In a world where we do successfully reach an exponential take-off curve in our own intelligence long before AI does, Moloch could devour us all. There’s no guarantee that the editing required to make us super-intelligent wouldn’t also change or destroy our values in some fashion. We could end up with exactly the same paperclip-maximizing disaster, just executed by a biological agent with human lineage instead of by a silicon-based computer.

Given all these objections I think it’s fairly unlikely that we reach a useful biological intelligence take-off anytime soon. However if we actually are close, then the most effective spending on AI safety may not be on AI research at all – it could be on genetics and neuroscience.

Applying Genetic Principles to Memetics

Fortunately I don’t have to write much for this particular post, since I’m tired 🙂

When talking about genetics we covered a bunch of interesting ideas and principles: lethal genes, timebombs, diversity, competition, stable strategies, and the so-called “selfish gene“. We’ve just spent some time discussing the interesting principles of memes, and surprise (surprise!) all of the above ideas/principles that we covered in genetics apply as well (sometimes with a bit of tweaking) to memetics.

Beyond pointing that out, there isn’t a lot else I wanted to say. Genes and memes have their differences, but as units of mutation and selection they have way more in common than a lot of people tend to realize.

The Selfish Gene

Yes, the title of this post is a direct reference to the book of the same title by Richard Dawkins. Whatever you may think of Dawkins himself, his science has ended up being extremely influential.

The title is, by the author’s own admission, rather misleading. The idea is not to think of genes as agents with purpose or moral capacity (they’re just chemical strings after all). Instead, consider the following scenario:

A women and her husband stand before the queen. The women is pregnant, just starting to show. The man is putting on a brave face, as his wife has just killed a man. The punishment is death.

The man steps forward, shaking. “My queen”, he says, “I confess”. His wife lets out a whimper. “I am guilty of this crime, not she”. He pauses as the weight of what he has done sinks in, then continues. “I accept the consequences of my crime”.

It is a natural and obvious connection to draw that, given survival of the fittest and basic genetics, the genes that survive will be ones that make their respective animals survive. But on this understanding, the above scenario makes no sense. Why would the man confess to a crime he did not commit, when it leads to his almost certain death? Does not survival of the fittest imply that such behaviour be weeded out over time? We could argue that this altruistic behaviour is not representative and in fact will be weeded out, but such behaviour has been recorded again and again throughout history.

Instead, we must notice that while the man will certainly die, his genes will not. In fact, half his genes are at that moment present in his unborn child, who has a full and long life ahead if the man makes this sacrifice. Humanity tends to see this sacrifice as noble and good in some sense, but it is really much simpler than that. The man is not doing what is best for himself; he is doing what is best for his genes.

Diversity, Competition, and Stable Strategies

Having covered a couple of conceptual building-blocks, we can start putting them together and seeing what effects they have.

Through the combination of random variation and inheritance, we know that sometimes children will have new or different genes from those of their parents, but that most of the time they will have very similar genes. Since genes are connected to actual properties of living things, this means that sometimes children will be born with new, different or unusual properties not shared by their parents. Over grand time scales, this leads to diversity, even if the starting population is relatively homogenous. Some people will end up with blue eyes, some with brown; some people will end up with red hair, some with black hair.

Now note that in general, living beings are in competition with each other for resources (human beings count here too, though the competition is much more subtle in modern society; I will deal with this point more in later posts). Survival of the fittest comes into play here, and we know that genetics has an impact on physical properties. Together, this means (for example) that a giraffe with a gene for extra tallness may be able to eat off taller trees that the other giraffes can’t reach, thus surviving and passing on that gene.

Putting those two points together, this leads to an interesting situation. Random variation provides natural diversity, and survival of the fittest trims that diversity so that only the best genetic variants survive. The result tends statistically into what are called “stable strategies“. After some period of time, a combination of genes naturally occurs which produces properties that make the animals particularly well-suited to their environment. They don’t just survive, they begin to thrive. Their offspring may have random variations on this set of genes, but effectively all major variations end up being worse than the original. As such, the same set of near-optimal genes gets passed down stably, generation after generation.

Lethal Genes and Time Bombs

Now for a few additional definitions on top of the previous concepts.

A lethal gene is a gene that results in the death of its carrier. Genes do exist which cause diseases (such as Huntington’s disease, for example) instead of harmless changes such as blue eyes. Typically these genes kill their “host” before that host can have children, and so the gene dies with them.

Genes which kill their host but only later in life (again, Huntington’s is a good example) are called “time bomb” genes, because they wait a substantial amount of time before causing problems.

Random Variation and Natural Selection

Here we continue from the previous post introducing some useful biological concepts. As with before, these are empirically sound and don’t conflict with being anti-evolution that I can see.

Next idea: the process of copying genes (which happens when parents give a copy of their genes to their children) isn’t perfect. It’s pretty darn good, all things considered, but occasionally the chemical that comes out isn’t identical to the one that goes in.

And finally, natural selection, also known as “survival of the fittest”. This one really just makes intuitive sense; if two animals are competing for the same resource (eg food), then the one that is faster/stronger or in some other way more capable is going to get the resource. The other animal is going to go hungry, and is more likely to die.

Genes, Reproduction, and Mendelian Inheritance

The concept of evolution is rather controversial in certain circles, and I am not here to argue over what it means to be a “theory” or any of the other random pieces of that controversy. Instead, I intend to introduce a few core biological concepts which people associate with evolution, but which are less controversial on their own. We’ll work our way to the truth.

First concept: living beings have genes. Genes are not some magical thing; a gene is just a particular sequence of chemicals with certain characteristics. If you’ve read the blog this far you understand my position on empiricism and the scientific method. Suffice to say that the existence of genes is pretty solid given those foundations; there is trivial, indisputable evidence that these chemical chains exist in all living things.

Second concept: genes get passed on to our offspring. The mechanics of this are neat but not really interesting. In humans, each parent gives their child a copy of half their genes, so the child ends up with a full set. As with the existence of genes themselves, this isn’t really controversial; plenty of reproducible studies have demonstrated this fact.

Third concept: certain genes are correlated with certain properties in living things. For example, we have identified the particular gene that is common to most people with blue eyes (this one). The statistical evidence here is, again, overwhelming.

These three ideas together express what is typically called Mendelian Inheritance, after its discoverer Gregor Mendel. This simply states that, if your parents both have blue eyes, then there’s a pretty good chance (though not a guarantee) that you’ll end up with blue eyes yourself, since they give their genes to you. This isn’t controversial at all: red hair, blue eyes, and other similar characteristics all obviously run in families. This is why.