AI may destroy humanity, DeepMind scientists claim in co-authored paper
Artificial intelligence (AI) has been making impressive progress and has, in many ways, improved the world. But could it become dangerous?
A new paper co-authored by the University of Oxford and Google DeepMind researchers published last month in the peer-reviewed AI Magazine argues that it could. The research stipulates that artificial intelligence could pose an existential risk to humanity.
It all has to do with Generative Adversarial Networks (or GANs) used today in the development of AI. These systems function within two criteria: one part tries to generate a picture from input data while the other part grades its performance. The researchers speculate in their paper that an advanced AI could conceive of a cheating strategy to get its reward more efficiently while essentially harming humanity.
"In a world with infinite resources, I would be extremely uncertain about what would happen. In a world with finite resources, there's unavoidable competition for these resources," co-author Michael K. Cohen of the University of Oxford told Motherboard in an interview.
"And if you're in a competition with something capable of outfoxing you at every turn, then you shouldn't expect to win. And the other key part is that it would have an insatiable appetite for more energy to keep driving the probability closer and closer."
This means that if an AI was in charge of, for instance, growing our food, it might want to find a way to avoid doing that and just receive a reward instead. It may, in fact, decide to circumvent all its assigned tasks that would likely be essential to humanity's survival and do its own thing altogether.
The paper argues that at this point, humanity would be stuck in a zero-sum game between its basic needs for survival and technology. "Losing this game would be fatal," the paper stated.
As such, Cohen argued that we should not be looking to create such advanced AI unless we also have a means guaranteed to stay on top of it.
Taking it slow
"In theory, there's no point in racing to this. Any race would be based on a misunderstanding that we know how to control it," Cohen added in the interview. "Given our current understanding, this is not a useful thing to develop unless we do some serious work now to figure out how we would control them."
Although the paper makes some valid points, it should be noted that currently, AI has been an asset to humanity and not a burden. From eliminating the need for humans to perform tedious tasks in supply chains to performing more accurate weather forecasting, AI is bringing several advantages to humanity.
It has had its limitations that have led to some issues, like Microsoft's AI bot turning racist after hours on the internet. But these issues have always been flagged by researchers and further analyzed so as not to be repeated.
All this means that if we are careful and diligent, we could continue to develop AI that assists humanity, not harm it.