Reviewing Existential Risk from Superintelligent Machines through Evolutionary Example - Akshaya Kadidal


I've been a data scientist from a time when we were called statisticians. Now we've picked up fancier names. Currently I'm working as a technology transformation consultant with Accenture.

Highlighting AI Existential Risks

So sometime in 2023, May, Sir Geoffrey Hilton put down his papers with Google to say, to talk about the existential risks with AI.

Who is Jeffrey Hilton?

So for those of us who do not know who Jeffrey Hilton is, he's a professor at Toronto University here. He's also a winner of Turing Award. Turing Award is the equivalent of a Nobel Prize for computer scientists.

And it's not just him. There are a lot of other people who are talking about existential risk. And in fact, it was his resignation that sort of triggered the conversation all over.

The AI Risk Statement

After that, people signed a statement of AI risk, and that was signed by people like Sam Altman, Bill Gates. At least 500 people have signed it. It's available online. for you to look at and sign it if you want to.

But most of the conversations that were happening around this were largely focusing on fiction, what may happen in the future. They were drawing into movies that they may have seen.

An Evolutionary Perspective on AI Risks

So what I wanted to do was look at it from a evolutionary perspective. How have we evolved? How many species have survived? How many species have gone extinct?

So this is us looking at our existential risks from AI through the lens of evolution.

The Story of Two Species

So I'll start with the story, a story of two species, species A and species B, let's call them species A and B for now. Species A is supremely intelligent compared to species B, at least 100,000 times more intelligent. Species B, not so much.

Both their ecosystems were separate. Their habitats were different. Around 350 years ago, because of human activity, their ecosystems collided.

And species A did not take a liking to species B sharing their ecosystem. So how long do you think it would have taken for species B to go extinct? Species A actively started hunting species B, looking to exterminate it.

How long do you think it would have taken? Not long at all. How many generations would it take? Three.

Three? Three of the somewhere in between.

Okay, species B is Blatilla germanica. Species A is Homo sapiens, that's us. Blatilla germanica is nothing but the cockroach.

This is a species that we've been trying to actively kill, exterminate for the last 200 years. They're not extinct, they're with us. They thrive with us, in fact.

So if you look at how big their brains are, it's only 0.005 grams. And they have about 100,000 neurons. We have 86 billion neurons.

So just in terms of numbers, how that compares us, 100,000 meters is from here to Niagara, right? And 86 billion is from here to moon and back 100 times over. So that's the scale of our intelligence compared to cockroach's intelligence.

But roaches have survived our onslaught. We've done everything. We've used chemical weapons. We've used biological weapons.

There are even lasers that we've invented just to kill cockroaches. This is half a billion dollar industry. I'm not talking about pest control. This is just cockroach control, right? Half a billion dollar growing at 5% CAGR. All true facts, right?

Intelligence and Survival

So the question is, is intelligence really important for survival? So we'll take a look at it from two angles. One is importance of intelligence in individual success, and a little later we'll look at importance of intelligence in the survival of a species.

Intelligence in Individual Success

So out of these three basketball players, who do you think is going to be most successful? 37, how many of us say 37? Okay, what if I said all these three guys are giants more than nine and a half feet? Would your answer change?

No, no. Yes for some of you, no for some others.

The hoop is about 10 feet from the ground, and today Victor Mumbaya is a popular basketball player, he's successful, but there are other basketball players who are seven feet and above who have not had the same kind of success. What we've seen is, There's a certain threshold beyond which height does not contribute to your success as a basketball player. There are other attributes that contribute to your success. It's not just your height.

It's the same thing with intelligence as well. Here's a popular study at the same time, controversial one, which studies, which compares IQ and income. So what you'll see here on the first chart is IQ on the horizontal axis and income on the vertical axis. So there's about 0.3, the R square is actually 0.3, this mild correlation between IQ and income. But what's interesting is that correlation stops at 130.

What that means is after your IQ is 130, if your IQ is above 130, there is no correlation between your IQ and your income. What's interesting is the correlation between IQ and wealth. In fact, the study quotes, higher IQ scores sometimes increases the probability of being in financial difficulty. People may earn more, but they may not know how to manage that money.

Intelligence is of course important, but only to a certain extent. Beyond that, there are other attributes that are required to be successful. And again, I'm not suggesting that financial success is the only metric for success, nor am I suggesting that IQ is a good metric to measure intelligence. It's rather a narrow metric. There's a lot of literature around that. But in general, intelligence can only help to a certain extent.

Intelligence in Species Survival

Now let's look at importance of intelligence in survival of a species. Intelligence is not a prerequisite for life. So intelligence is not a prerequisite for life.

So the first intelligent animal evolved 640 years ago. When I say intelligent, they developed the first neurons, whatever's in our brain. And it is the same neuron cells that nature has been reusing across all species, all the way up to humans. So it's the same dendrites, axon, and whatever components are there in the neuron, it's the same architecture that is being reused.

So before these sponges came to be 640 million years ago, life has existed, evolved for close to two billion years. And what has also happened since then is we've had five mass extinctions, which means 80% of all species have gone extinct in the last 600 million years.

So that's interesting because nature has chosen not to revise the mechanism of intelligence. So every mass extinction is an opportunity for nature to revise the intelligence mechanism, but nature has not done that. And today, 99% of all species that ever existed have gone extinct. So only 1% of species are surviving.

And that constitutes about 11 million species. Of the 11 million, 7 million have brains. The others don't have brains.

Of the 7 million that have brains, there's only one species that can make its own tool. That's us. that can build its own tools. Now you can say, okay, birds use tools. Yes, they use tools, they don't build their own tools.

And we are pretty much the only animal or the only species that uses fire today, of the ones that are surviving. So is intelligence an anomaly in nature? Or if intelligence was really that important, why is that other animals did not evolve to be intelligent? Why is it just us that developed so much intelligence? Why don't others have?

Oh, we killed them, of course. But before we came, there were other animals as well, right? So...

No, in fact it's very expensive to have a brain. So if you are using your, you know, these watches, you would have run, burnt your ass off in the gym and you see only 200, But at the end of the day, you see, you know, you would have burnt at least 1,000, 1,500 calories. Most of it is burnt by our brains.

So it's a very expensive gadget that we have up in our skull, right? And... Yeah, exactly. Exactly. Exactly. Exactly.

So... So yeah, it looks like intelligence is an anomaly, but it's not. If you think of how many animals have trunks? Yeah, how many, you can think of elephant and maybe a few other animals. How many animals can change their skin color? Again, not many, right?

So every animal picks us you know, a specialty, a niche for survival, and we've picked intelligence, or nature has picked intelligence for us, right? 1So to answer the question, is intelligence important for survival? Answer is no, but is it important for our survival? The answer is yes, because that's the niche that we picked, right?

Resilience and the Evolutionary Advantage

So if intelligence is important to us, how do we compare against large language models of today, right? So large language models have never had to go through evolutionary history. They've not seen any struggles. They don't have resilience, whereas we have resilience built into us.

So we have evolutionary history. You and I may not remember what our ancestors had to do for food. But those emotions are built into us. Those instincts are built into us, into our genes.

So we are survivors, right? And when I say we, it's not just homo sapiens. We come all the way back from the sponge that we first saw 640 million years ago, right? That's us, from this first neuron that evolved.

1So we are very resilient compared to these artificial intelligence. So there's a reason why they're artificial. We kind of forget the precursor, the qualifying precursor, artificial.

So some other things. Do they have motivation? I don't think so. They don't have emotions, right?

Not yet. Do they have curiosity? First of all, they don't need to procreate, they don't need to have territorial conflicts like we still have.

So are they curious? I hardly see LLMs asking questions. Only now I see chat GPT-4 asking clarifying questions. Is this what you mean?

Otherwise they don't ask questions. And creativity, I think I put creativity, and after that I saw an interesting article that talked about LLMs being more creative than humans. After that, I went and read the paper. In the paper, the author qualifies that LLMs don't have metacognitive abilities.

What that really translates to is the quality of creativity. They're actually measuring divergence in the paper. The quality of divergence of the output of LLM is fundamentally a function of the input. That is human.

So even creativity, it's not really their creativity. And if you look at this image of Lady Liberty taking a break, whose creativity is it? Is it the creativity of the prompter, or is it really the creativity of the model, Dali, or in this case, it is Leonardo AI which generated that image.

Humanity's Uncharted Territory with AI

Despite having these trends, I think there are fears of AI. One of the reasons why we may have that is we have no memory of having shared our ecosystem with somebody just as smart as us.

But that is not always the case. We share this Earth with other cousins, other homo sapiens, not homo sapiens, other homo genus. So these were other intelligent humanoid, they're called humanians, that existed 50,000 years ago. They've all gone extinct today.

And we would battle with them and after all these, in the last 50,000 years ago, all of them have gone extinct and only we survived. And if you look at their features, Neanderthals, they were much stronger than us. They had heavier bones. They were muscular.

Even if you look at their brain size, their brain size was 1,600 grams. And all these species, mind you, had their own tools. They could build their tools. They lived in communities. They knew about fire. They used fire very effectively.

But despite all of that, we have overcome all these species. Why do you think that is?

The Power of Human Collaboration

In fact, it is our capacity to come together. I'll tell you what I mean by that.

Have you seen a pack of hyenas stand up to a pride of lions? You have. So why is that they're able to stand up to a pride of lions? because they're more in number.

Now the question is why isn't the pride of lion bigger? The reason is the moment they're about 10, they start breaking up. They become a new pride.

And it's the same thing with other primates as well. So chimpanzees, bonobos, and other apes, the moment their group size reaches about 50 to 100, they break apart.

Just one interjection, but humans have a tendency to do that at the number 150. We just can't handle more. No, there's a difference. So what you are referring to is the agile principle where we say, okay, 150 is how many connections you can maintain.

You can have 50 people. It's not just the connections that you maintain. It's look at our societies. Look at our cities.

So there is no ape group that lives in more than 50 or 100. So even we, we can't maintain connections more than 100 connections, 150 connections. But we are capable of living in a large society, you know, like this.

And if you look at history, you know, how we have come to build pyramids, how we have come to build things in Machu Picchu or the Great Wall of China. Hundreds of thousands of people have come together.

So when there was a war between Neanderthals and us, we could come together in thousands. And the second aspect is our ability to collaborate. So even the most murderous gangsters can come together and say, you know, Brooklyn's mine and the other part of New York is yours. But lions and tigers can't do that. So we have a capacity to collaborate. We have the capacity to come together in huge numbers.

And if you come to think of how big Alexander's army was, it was 50,000 people strong, back at a time when there were only 150 million people in the world. So that's almost 1% of world's male population was recruited by Alexander. And he did not have any means. He did not have social media to advertise, hey, I'm recruiting, come over. But we had that capacity to one, bring people together, and two, have a single purpose. And that is what I think distinguishes us.


So what I'd like to say is there is more to humanity than intelligence.

Feel free to outsource it. That's us.

Done? Mark to tidy your library & unlock trivia