WHIZ KID As a child I read voraciously, including adult-level science and history books – anything I could get my hands on. I was really bored in high school because I’d already learned most of the curriculum, so I applied to Bard College at Simon’s Rock, in Massachusetts, in the United States, which admitted younger students, and started my degree in mathematics when I was 15.
STAR STRUCK I started reading science fiction aged three and became fascinated by artificial intelligence (AI), time travel, space travel, brain modification and robots, and the possibility of making those fictional things real. At university I tried to figure out if you could make a time machine by spinning an American football-shaped star really fast while circling it in a spacecraft. According to the laws of physics you could end up further back in time, but, obviously, there were some major engineering problems with that plan. In my 20s, I realised that to get things done, I needed to focus my thinking. I chose AI because my chances of achieving something significant seemed higher. It’s an audacious, world-changing goal that you can work towards by sitting at your computer, writing code. There’s no need to spin a star around.
MAN VS MACHINE “Artificial intelligence” is an umbrella term. At one end of the scale is “narrow AI” – highly specialised techniques used to carry out specific tasks. Narrow AI is very different to human intelligence. For example, IBM’s Deep Blue chess program can evaluate 200 million positions a second. Humans can’t do that. The human brain excels in a different way – it can take chess-playing strategies and apply them to other games. Deep Blue would need to be completely reprogrammed. I work at the other end of the scale.
I introduced the term “artificial general intelligence” (AGI) around 10 years ago, to distinguish it from the narrow kind. AGI is broader and has the capacity to learn from experience and to teach itself things. In this way, an AGI system is designed to function more like a human – when confronted with a new challenge, it takes what it learned in another context and figures out the solution by itself. When I started work on AGI, in the 1980s, people thought designing a machine that could think like a human was an impossible goal. To the rest of the AI community, me and my colleagues seemed like a bunch of wild-eyed mavericks with crazy ideas that were better suited to the realm of fiction.
It’s completely different now – AGI is almost mainstream. Now, when you say you’re building a thinking machine, people don’t say “you’re insane”, they say “but hasn’t Google done that already?”
THE AGE OF ROBOTS The ultimate aim of AI research is the technological singularity – the point at which technology overtakes the human mind. AGI will eventually create autonomous machines. The machines will invent new machines, prototype them and build the factories, without involving humans. At this point the technological advance will become blindingly fast – the speed of development will approach infinity from the human point of view and we won’t even be able to understand it.
I think the singularity will happen some time this century – anywhere between 2025 and 2100. I’m hoping it’ll happen sooner rather than later because I’m already 48 years old and I want to experience and enjoy it.
MANUFACTURING EVIL All new technologies come with potential risks and rewards. Since the beginning of humanity, we have pushed forwards regardless. When we switched from hunter-gathering to agriculture, when we created the industrial revolution, we had no idea whether these transformations would bring danger. It’s the same with AI. The worst possible outcomes are extremely dark. This does worry me – I don’t want to see my kids disassembled so their molecules can be used to make more hard-drive space for machines. The best possible outcomes are utopic and amazing. I have a research team called iCog Labs, in Addis Ababa, Ethiopia. When I visit, it seems like half the people live on the street, suffering from hunger and disease. If we can resolve these problems with AI, by making it so cheap to create material wealth that, even with our perverted economic system, everyone has enough to get by and be happy, that’s a big plus. AI will be created whether I’m involved or not. I hope to have a positive impact on how it’s used – to my mind it’s better that the work’s done by people who want to benefit the planet, than by people in a top-secret military project.
ACHIEVING IMMORTALITY I work on AI for various companies. The range of applications is huge. I’m collaborating with Hanson Robotics to create robots that look and think like humans, and with investment management firm Aidyia Holdings, which uses AI to outperform human traders on the stock market. I’m also interested in life extension. I remember realising, at a very young age, that everyone was going to die one day. I couldn’t understand why others just accepted this. It seemed like a really bad idea to me. I’ve signed up to have my body transported to a facility in Arizona and frozen, if I die. When the technology’s ready, I will be brought back to life. Ideally, though, we’ll find a cure for ageing before I die. The best way would be to build a superhuman thinking machine and let it solve all the hard science problems.
I can’t resist seeing what the current technology can do. I’m working with a team to compare the DNA of supercentenarians – people who have lived to 110 years or more – with the DNA of people who lived to 80. We’ve found specific genes we think are responsible for their longevity and we’re applying for patents. The question is, can you edit the DNA of an adult human to give them that capability?
One hundred and ten years is a long way off our end goal – which is to live forever. But, if you live that long, hopefully AI technology will solve the rest of the problems during your lifetime.
Ultimately, we want to port our minds out of our bodies, by implanting a device in the brain that backs up to a server. Then, if you’re hit by a truck, you could 3D print a new body and restore the mind from the server. One day we’ll look back and think it’s ridiculous that we lived for so long without the ability to do that.