DANKO NIKOLIC has spent his life studying human intelligence. Lately, however, he鈥檚 been thinking about the artificial kind. Just how smart can AI get? Are we really headed towards the so-called technological singularity?
That was the topic of a debate in Berlin last month. Nikolic, a neuroscientist at the Max Planck Institute for Brain Research in Frankfurt, stood up in front of an audience of artificial intelligence researchers and made a bold claim: we will never make a machine that is smarter than we are.
鈥淵ou cannot exceed human intelligence, ever,鈥� says Nikolic. 鈥淵ou can asymptotically approach it, but you cannot exceed it.鈥�
Advertisement
For those who accept the possibility of the singularity 鈥� a date in the not-so-distant future when machine intelligence outstrips our own and goes on to improve itself at an exponential rate 鈥� this is fighting talk. 鈥淭hat鈥檚 a really strong claim without really strong evidence,鈥� says Mohamed Sayed, CEO of AI start-up in Berlin, which is developing artificial intelligence for use in hospitals.
Nikolic is convinced many artificial intelligence researchers are overlooking an important aspect of human intelligence: brains aren鈥檛 the only hardware humans need to be good learners.
For Nikolic, the most basic tools for learning are the instructions contained in our genes, honed over billions of years of evolution. Machine learning techniques can mimic the brain, but they miss the deeper elements that help us learn. The only way to get close to a mind that learns as well as us is to repeat human evolution, says Nikolic.
But maybe the singularity will arrive in other forms. For many who attended the debate, the singularity is better thought of as an acceleration of human progress, despite being fuelled by a near-future technological breakthrough. For them, it鈥檚 all about putting human and artificial minds together to solve real-world problems.
To begin with, these are likely to be relatively mundane. Berlin-based Leverton is using natural language processing to speed up the task of wading through huge volumes of written documents, a job that large firms typically employ many people to do.
Another Berlin firm, MicroPsi, is applying artificial intelligence to solve the complex logistical problems faced by the shipping industry and component supply lines in manufacturing. There could be better ways for large ships to move around the planet hiding in the mass of data that logistics systems produce that we just cannot see. Behind each of these projects is the assumption that unless machines find the solutions for us, we may have no idea what they are.
One more point that came across is that if the singularity is going to be characterised by working alongside artificial intelligences trained on vast amounts of human data, then we will want them to be as diverse as we are. It was a good sign that the debate was attended by an almost equal number of men and women.
Ultimately, it is hard to predict what a major breakthrough in AI will bring. 鈥淭he reason they call it the singularity is that it鈥檚 a point beyond which you cannot see,鈥� says Mehmet Akten, an artist and computer scientist studying artificial intelligence at Goldsmiths, University of London. 鈥淥nce machines reach human levels of intelligence, you can鈥檛 begin to imagine what鈥檚 going to happen.鈥�
This article appeared in print under the headline 鈥淰ision of singularity questions AI intellect鈥�
