What sets humans apart from the rest of life, or indeed inert matter? Many people would respond that it is our intelligence. Yet the rise of seemingly intelligent machines challenges this way of thinking. The companies behind these new artificial intelligence technologies, in the form of ChatGPT and its rivals, speak of achieving artificial general intelligence – machines that have the same level of intelligence as humans across a range of tasks.
Does this meteoric rise in AI make human intelligence, and therefore us, less special? , professor of machine learning at the University of Cambridge, doesn’t think so. In fact, he thinks we should throw out the concept of artificial general intelligence altogether.
Advertisement
In his new book The Atomic Human: Understanding ourselves in the age of AI, Lawrence makes the case that it is only by better understanding our own intelligence, and how wildly different it is to its artificial counterpart, that we can make the most of both. Here he tells New Scientist why he thinks both human and artificial intelligence are misunderstood, why it is pointless to compare the two and why, ultimately, we need a more nuanced understanding of intelligence.
Alex Wilkins: What do you make of the trend to compare artificial to human intelligence?
Neil Lawrence: Most of these arguments are pointless, they are irrelevant. Of course, the nature of the intelligence that we’re seeing in AI is extremely different from our own. It’s absurd that people are talking about this intelligence as if it’s anything to do with us. But as humans, we can’t help but anthropomorphise it because that’s the way we try to communicate and understand.
It’s interesting because there was this tradition at one point where, every summer, newspapers published formulas for beauty – consisting of some absurd combination of numbers to try to unpick beauty. The discussion of intelligence, as if it’s rankable and quantifiable, is just as absurd. We wouldn’t talk about artificial general beauty, so we shouldn’t be talking about artificial general intelligence.
There are aspects we can talk about. I can talk about the intelligence of a particular decision, as long as I know the context and what the individual was trying to achieve. But if the context isn’t present, then trying to turn it into a rigorous scientific definition, or trying to claim that we can rank general intelligences, or that there’s such a thing as an artificial general intelligence, is absurd and dangerous.
Scientists have puzzled over how to define intelligence for a long time. Are we any closer to a satisfactory definition?
In my book, I use the term intelligence a lot, but I try to move away from that and towards information processing as an idea. I think we can now start looking at rigorous definitions of intelligence that are based on information, because information is fundamental.
Large distributed systems process information. The ecology that’s around us is doing that, and our culture does that too – we’re a distributed intelligence. Arguably, the seat of human intelligence is actually in our culture, rather than in individuals.
Do you have an example of this kind of distributed intelligence?
Social insects are a form of distributed intelligence. Where I am at the moment, in southern Italy, there are a lot of ants because the weather has been very dry. These ants seem special in that they form multiple nests. We get these columns of ants moving into the house and then moving out again and going from one nest to another. It’s really extraordinary watching them moving eggs. Just as we think we’ve dealt with one nest and one column of ants, the next day there’s a column that’s arisen from somewhere else, following the old path and coming into the house again.
My mother-in-law says, “wow, they’re really intelligent”. Of course, individually they aren’t really intelligent: they don’t have planning systems that are looking ahead a long way and making strategic decisions about how things might pan out, which is a quality of human intelligence. But the ants are making decisions about where they should go to acquire resources for the nest, and those decisions take place in a distributed manner through ants communicating when they have found a pile of sugar or laying down scent trails. Any sensible definition of intelligence needs to include components of information processing.
Presumably, a similar principle applies to humans. What does that tell us about the nature of intelligence more broadly?
The ants are an information processing system, but then when you see them you also start to realise that so is the whole ecosystem around us. Information is stored in DNA, and plants, animals, bacteria and archaea all react to changing circumstances, making decisions. Those decisions affect multiple generations. Arguably the AI we’ve created has more in common with the ants’ processing of information than human processing. But we anthropomorphise ants in the way we anthropomorphise AI.
These systems are quite hard for humans to think of as intelligent because they don’t manifest in the way that our intelligence manifests. Our intelligence is a distributed intelligence, but it’s distributed in this weird way where our communication bandwidth as humans is very narrow and our computational capabilities are enormous.
This gives us each a sense of ourselves within a community, as an entity, as an individual. When we’re communicating with others, we also think of them as individuals and as entities. I think this makes it harder for us to sometimes see the distributed nature of our intelligence. It’s also harder for us to recognise distributed intelligences, and for us to recognise intelligences that are information processing systems operating over time frames that are much, much longer than the time frame over which we’re operating, such as over millions of years.
Do you have an example of this kind of intelligence?
Life on Earth is an information ecosystem that’s operating over millions of years. It responds to plate tectonics and other events that occur on geological time scales. For example, flowering plants started to dominate other species around 100 million years ago, about 100 million years after the last supercontinent, Pangea, began to break apart. Perhaps in another 100 million years, these plants will be faced with the challenges of surviving what happens when the Earth’s land masses form a future supercontinent.
These are timescales that are very difficult to imagine because information is passed to our offspring through our genome hundreds of millions of times slower than a conversation between two humans shares information. Similarly, we find computers’ information processing difficult to conceive because they share information hundreds of millions of times faster than we can.
What, then, should people make of the latest AI systems, like ChatGPT, which certainly seem impressive at first glance?
I’m sceptical about the things people are saying about artificial general intelligence. I think it is dangerous to talk about things in those terms because we don’t have a good definition of intelligence. But despite that scepticism, with the technology we already have, even if you pause all development today, AIs are utterly transformational because they represent an information revolution.
People are saying, “oh, it’s going to have orders of magnitude greater intelligence”. For comparison, as I go from walking to driving a car, that gives me about an order of magnitude increase in speed. From a car to a plane, that gives me another order of magnitude increase in speed. That’s a 100-fold increase. But really, the big transformation is in AI’s information access rate – the rate that two machines can communicate with each other – which is 300 million times greater than human-to-human communication.
That’s the big transformation. It’s just got nothing to do with this idea of artificial general intelligence, which is a nonsense.
This transformation seems like it has the potential to upset a lot of the established players, like Google, Amazon, Microsoft and so on. How do you think that will play out?
For the first time we’ve got a technology – that isn’t perfect by any means and needs a lot of work – where regular humans can talk to the computer and instruct the computer directly. So a regular human, without having to learn programming languages, can ask the computer about the information that the computer has access to.
This is incredibly exciting and incredibly dangerous at the same time because the computer has the potential to misrepresent, as it is already doing, whether that’s on purpose or just the nature of the way the AI models have been built. But it also has the possibility to give individuals access to a collaborator, where the collaborator is not a human, but a machine that has access to this vast amount of information. This is transformational, giving individuals access to that information 300 million times faster rather than through a world curated by Google, Microsoft, Amazon or Facebook.
All the big tech companies are the ones that point the spotlight on where the information is. What’s interesting is that, currently, big tech is incentivised to persist in the way that they’ve been dominating the information infrastructure. But now they have to make sure they’re not disrupted by the capability of individuals to access the machine directly.
What aspects of intelligence are machines still lacking?
I think the really interesting one is lack of motor skills. It’s remarkable that all the things we’ve managed to get the computer to emulate are all the sorts of things that scientists think of as useful and important, like doing maths or playing chess or playing Go.
But when it comes to what regular people enjoy – playing football, having a good gossip, being a friend to a friend in need – all those things are part of our intelligence. Perhaps these are part of our intelligence that scientists, who are maybe not the most socially engaged of the human species, haven’t valued as much. All those pieces are incredibly difficult to emulate, but particularly the motor skills. It’s not to say that we won’t get there, but it feels like we won’t.
I'm sceptical about the things people are saying about artificial general intelligence
So where does all this leave us, with regard to how we think about our own intelligence?
There’s an iconic picture of Earth taken by astronauts on the Apollo 17 spacecraft in 1972. This blue marble photo became a symbol of the environmental movement, because Earth looks like a massive, extraordinary place when you’re on it, but when you stand off it, you realise that it’s a really special, complex, extraordinary thing that needs looking after.
I’d like the same thing to happen when you stand on this other intelligence. It will and already does have capabilities that are well beyond anything humans can do. This doesn’t mean that it replaces us, but it does mean that it’s an interesting place to stand and reflect on ourselves – what we actually want – and try to lift our thinking as a species.
And will AI ever have human levels of intelligence?
Well, those capabilities, in some sense, are critically dependent on our limitations. Our ability to see things, to make decisions, is dependent on our vulnerabilities and our limitations, and we’ll never be able to emulate that in machines because machines don’t have those limitations baked in and never will.
The fact that we’re going to die, the fact that those close to us can die, we can be betrayed by other humans, we can lose our reputation – it’s these things that a computer can’t experience that make us special. It is these things that make human intelligence unique.
