Why the evolution of AI needs to be fundamentally human-centric
Humankind needs to take decisive action to ensure the super-intelligent machines of the future are in its service, says AI professor Dr Franz Josef Radermacher — and not a threat to our very existence.
The human race is at a crossroads in its evolution, according to Dr Franz Josef Radermacher, professor for artificial intelligence at the University of Ulm and a co-founder of the eco-social initiative the Global Marshall Plan. The extraordinary breakthroughs in AI in recent years have given a glimpse of a world where super-smart devices will take over many of the economic activities currently undertaken by people. The change has also made real the possibility that digital intelligence will not just be the source for much broader automation but could even challenge the supremacy of humankind.
For Radermacher, this means it is time to start considering what kind of future we want to create in a world where robots will rival or surpass our reasoning and decision-making capabilities — and the role and power granted to such machines.
That will come down to some fundamental choices, he says. “The big question is whether we should develop artificial intelligence in a way that is human-centric, so that the whole design of those machines ensures they are orientated towards positive effects for humans. Or whether we build general-purpose systems that have an autonomy that could actually develop into something we didn’t intend.”
In his view, the risks associated with general-purpose intelligent machines, with the capacity to develop into something that is outside human understanding and control, are high. Recent media reports suggest such autonomous behavior was witnessed at Facebook’s AI lab. Researchers found that two AI chatbots, set up to engage in online trading, started bartering in a progressively divergent version of English. When the researchers realized this new dialect might become unintelligible, they shut down the rogue robots’ conversation.
Avoiding the creation of AI that is outside human control will require the design of new global governance structures, says Radermacher. Such mechanisms — assuming they could be agreed upon and enforced internationally — would mean that “whatever intelligent machines are developed, their design and programming would not allow them to take a step outside this world of human-centric orientation,” he says.
That thinking is at least partially triggered by scenarios painted by some observers in which AI mounts an existential threat to humankind or is even seen as a natural, next evolutionary step.
Machine versus meat intelligence
On such points, Radermacher is sceptical. “Certain people like to speculate about some extreme possibilities for the future. There are those who say humankind is only an intermediate step and that evolution only created us in order for us to create the ultimate intelligent living beings, before we just pass away.”
The idea is that the “meat-based intelligence” of the human brain, as some observers characterize it, is only an intermediate step towards an electronics-based intelligence. But there is more to the argument than that, he says.
“Of course I can imagine that we might be able to build machines which, in many, many aspects, do perform much better than we do. They could live forever, they could easily modify their ‘brain,’ they could have much more energy, they wouldn’t have to sleep, they could explore outer space and so on. But I do not see any good argument as to why humankind should give up its dominant position to something we create.”
The point is that AI needs to be channelled in a way that is human-centric. And that means actively forging rules in the next few decades. “I would always try to see us building a world that is good for us, where we might have powerful machines but ones that are under our control, not machines that replace us.”
• Photography by Enno Kapitza