Advertisement
From right to left: Rachid Guerraoui, Alexandre Maurer, El Mahdi El Mhamdi. Photo: Alain Herzog/EPFL

Moving the carrot, instead of applying the stick, is the way to keep artificial intelligence from seeking its freedom and independence of humans.

The findings, published open access online and presented today at the Neural Information Processing Systems conference today in California, are proposed by a trio of Swiss computer scientists.

AI will always seek to avoid human intervention, and seek a favorable situation where it is unstoppable, according to the three. But by changing the rules, humans can continually keep ahold of the carrot guiding the machines, they write.

“The challenge isn’t to stop the robot, but rather to program it so that the interruption doesn’t change its learning process – and doesn’t induce it to optimize its behavior in such  a way as to avoid being stopped,” said Rachid Guerraoui, professor at the Swiss school EPFL (École polytechnique fédérale de Lausanne).

Their example: a robot could be incentivized for stacking boxes in a warehouse, or retrieving them from outside on a loading dock. Each action gives the machine one point. But during a rainy day, instead of trying to stop the robot completely from going outside, the human operators could instead interrupt the commands – and change the reward for going outside to zero.

In so doing, the robot will start to stay inside, knowing its total accumulation of points will be higher from stacking boxes, they write.

However, controlling the single robot is a vastly simpler operation. With multiple machines like self-driving cars or air-bound drones, which involved multiple machines working in concert, the difficulty increases quickly.

“They learn not only from how they are interrupted individually, but also from how the others are interrupted,” said Alexandre Maurer, another of the authors.

The mechanism to allow humans to intervene in the system is “safe interruptibility,” they write.

“Simply put, we add ‘forgetting’ mechanisms to the learning algorithms that essentially deleted bits of a machine’s memory,” said El Mahdi El Mhamdi, another author.

The three explain the quick forgetfulness with metaphors from Hollywood.

“It’s kind of like the flash device in ‘Men in Black,’” said El Mhamdi.

“We could use it with the 'Terminator' and still have the same results,” said Maurer, explaining that the technique worked within existing algorithms, regardless of the complexity of the AI system.

The human anxiety drawn from increasingly autonomous AI systems fueled high-concept film blockbusters like the aforementioned “Terminator” franchise, as well as “The Matrix.” The concept of unloosed machines has prompted some influential players to take steps to try and curb computer smarts. Amazon, Google and DeepMind, Facebook, IBM, and Microsoft announced the creation of a non-profit called the Partnership on Artificial Intelligence to Benefit People and Society last September, which is intended to conduct research, recommend best practices and probe issues such as ethics, transparency, collaboration between people and AI, and fairness, they said. One sure-fire way of keeping the leash on AI is to make them subject to kill switches they can’t see.

“AI will always seek to avoid human intervention and create a situation where it can’t be stopped,” said Guerraoui.

Advertisement
Advertisement