Microsoft, Facebook discuss how to create ethical robots

Developers from Microsoft, Uber, Facebook and other major tech companies attended The White House Frontiers Conference in Pittsburgh Thursday. They spoke about the future of science. Artificial Intelligence was as one of the primary topics — not just discussion about how it’s being developed but AI’s social and ethical ramifications.

When it comes to AI, there seem to be three basic ethical concerns: Accurately interpreting human meaning, the trust required, and the potential for financial fallout.

First, let’s talk about interpretation. Well, we can’t really talk about it. This is a blog, right? But, you knew that. A computer would not. Yann LeCun, Director of AI Research at Facebook, said that’s a very common problem for translation software.

“The standard example is, the trophy didn’t fit in the suitcase because it is too large and the trophy didn’t fit in the suitcase because it is too small,” he said. “You know in the first case the ‘it’ refers to the trophy and in the second case ‘it’ refers to the suitcase. That’s because you know how the world works.”

This could lead not only to misunderstanding but to cruelty and even abuse of unsuspecting AI users. Consider the example of Tay, the Chatbot that Microsoft launched but lasted less than 24 hours. Jeannette Wing, Corporate VP of Microsoft Research, said that blew up … big time.

“The sub-society of the internet discovered Tay and had it repeat, just, completely inappropriate remarks, reprehensible language, and we knew that this was offensive. Period,” Wing said. “We had to shut it down.”

An AI moral compass

You can’t program in common sense or a moral compass.

In connection with the Frontiers Conference, the president has also released a strategic plan for the future development in AI. He says schools and universities need to integrate ethics into their technology programs. The government needs to create safety nets to ensure the use of AI programs doesn’t have unintended consequences like developing bias or creating social injustice.

Now, the second dilemma. The only way for Artificial Intelligence to really move forward is for us to trust it. LeCun gave the example of AI in medicine. It requires thousands of personal medical records to be successful.

“How do we set up a system so that researchers who have the methods have access to the data so they can make progress in the science?” LeCun asked.

Raffi Krikorian, Engineering Lead for Uber, said rideshare company is running into a trust issue as they develop driverless cars.

“People need to accept the fact, or need to be willing to get into a car that may have pulled up with literally no one in the front seat,” Krikorian said.

An even bigger challenge might be that corporations and governments will need to trust each other. In his strategic plan, President Obama says progress will happen more quickly if AI software is Open Source. That means anyone could look at, use and change the code of the program, making it harder for companies to monetize their research investments.

And, finally, one of the most important issues AI raises, according to Obama, is its impact on jobs and the economy. The report recommends that the White House convene a study on automation and the economy. That should lead to a public report by the end of this year on what strategies and policies might help ensure that computers can help us run the world, but can’t take it over.