In a world where AI will become “all pervasive”, regulation will be a huge challenge, says Professor Arlindo Oliveira.
 
“This technology will be in everything – smartphones, on cars, on computers, the internet. Regulating it will be a bit like regulating the world.”
 
It can already predict what people are likely to buy, but could it one day predict, with high precision, how people are going to vote and mean we don’t need elections any more?
 
“We are still very far from this. It is just a thought experiment,” he says. “But maybe if you can create very, very good models of every citizen you can predict what citizens will buy, how they will vote, where they will go when they travel.”
 
That’s where regulation must come in. “If we take it as granted that we will have machines which are as intelligent and as competent as human beings, we should probably think about making sure they are created to do things that we are interested in.”
 
He gives an extreme example. Say we created, in 20 years time, a very smart and expensive machine to solve climate change. “So we ask the crucial question: please take the actions required to stop global warming. The machine thinks for a few seconds…and just decides to kill the human race. That will definitely stop global warming, but was not quite what we were thinking of, right?”
 
“These are complex, smart systems that will require regulation, laws and legislation that we can enforce to make sure that machines…are doing whatever their job is.” A less extreme example would be using AI to balance energy supply across a continent. Maybe the best ways to do this might have some effects we were not expecting.
 
“So I think regulation will be needed in several respects and, in the same way we regulate transport, energy production and distribution and so on, we will have to regulate AI.”
 
The challenge in doing so is clear. Machines like humans, sometimes like to circumvent rules. “You have to regulate in a way that you don’t stop innovation but also that you don’t leave too many holes, because, otherwise, systems will explore those holes.”

 

“ It can already predict what people are
likely to buy, but could it one day predict,
with high precision, how people are
going to vote and mean we don’t need
elections any more?

 

An interview with Professor Arlindo Oliveira, 
Instituto Superior Técnico – Department of Computer Science and Engineering