There is a tech experiment that I long to do. It is to set up fake online personas for a Hitler-worshipper and a Stalin fan, then wait to see what merchandise the likes of Facebook and Amazon’s robotic helpers serve up for their delight.
I actually quite appreciate web companies’ suggestions, based on my browsing history, for stuff I might like. It is like having a friendly idiot savant scanning the horizon, with a slightly too binary version of my preferences, but nonetheless often hitting the spot.
I am interested in how these robots work, although I demur at their being called smart. Surely stupid, able and willing is more apt and preferable? Genuinely smart software would be terrifying.
Yet technologists have promised that 2020 to 2030 will herald The Singularity, the time when computers become smarter than humans.
So when the chance came up to meet Ralf Herbrich, Amazon’s Berlin-based head of machine learning, I was excited to hear from one of the world’s most qualified people whether superhuman, conscious computers are a likelihood — and a mortal danger to humanity as Stephen Hawking, Bill Gates and Elon Musk among others say.
Amazon does not use only machine learning to help us buy more stuff. It offers the intelligence developed by Mr Herbrich’s team to users of its Web Services, the part of the empire that, white-labelled, runs many big websites including this one — and that Jeff Bezos expects soon to be bigger than Amazon’s retail estate.
Those who disapprove of Amazon may be amused to know that, but for a twist of history, Mr Herbrich would likely now be working for the Stasi. A rebellious, computer-obsessed teenager in East Germany, he was shunted into a factory electrician job at 16.
He says the pressure to help out the secret police with the skills he learnt on a broken Sinclair ZX81 sent by a relative in the west may ultimately have been irresistible — had the wall not come down six weeks into his apprenticeship.
Since then, he has become a research fellow at Cambridge and worked in machine learning for Microsoft and Facebook.
Conveniently, on the office tour, I saw a real-life machine learning issue. An automatic translation from German-language Amazon to English had changed a USB cable into a “USB rope”. “There is always a need for human intervention,” Mr Herbrich observed.
He has no desire or expectation for bots to take over the world. A phrase he used was “the sliver of achievability”.
“The Singularity? It didn’t happen,” he said and went on to explain that it won’t be happening, either.
“People are really good at seeing, tasting, smelling, hearing. Machines are not good at that. We’re just getting to the point where algorithms can recognise still images of cats and mice. But the brain does it much better.”
To my surprise, Mr Herbrich agreed computers are really glorified adding machines. “Yes, and the brain is a very coarse information processing machine.
“Computers can emulate intelligent behaviour. We’re seeing a lot of that and it’s impressive that when we put it in the cloud, we’re able to perform highly complex tasks.
“But all that’s happening is that they combine patterns. Machines learn the hard way. They can’t replicate creativity; recombining higher-level abstractions and imagining futures based on very little information or example. They can’t have a hunch, like even scientists do. The algorithms we’re studying today are crude approximations because they don’t rely on the same principles as the brain.”
So what about the idea of superhuman computers taking on a life of their own? “No. A computer is a tool and only humans can build computers. People write programs. A computer can’t write a program. There will be no self-propelling computer.”
By happy coincidence, or perhaps Oxford University Press has been monitoring my search history, I got home to find the publisher had sent a book on artificial intelligence, AI by Professor Margaret Boden.
I expected this to be another bots-are-taking over manifesto. Instead, it is an elegant demolition of the notion of the superhuman computer. “Near-apocalyptic visions of AI’s future are illusory,” Prof Boden concludes.
Again, how refreshing. It seems the (sensible) humans are fighting back.