According to Gates, there are two main ways artificial intelligence could become harmful: it could eventually substitute some human labor in the workplace, and it could grow to become more intelligent than humans.
These issues seem to be solvable, according to Gates, but one may be easier to address than the other.
Here's what he said to Re/code's Ina Fried on the subject:
There are two different threat models for AI. One is simply the labor substitution problem. That, in a certain way, seems like it should be solvable because what you are really talking about is an embarrassment of riches. But it is happening so quickly. It does raise some very interesting questions given the speed with which it happens.
Then you have the issue of greater-than-human intelligence. That one, I'll be very interested to spend time with people who think they know how we avoid that. I know Elon [Musk] just gave some money. A guy at Microsoft, Eric Horvitz, gave some money to Stanford. I think there are some serious efforts to look into could you avoid that problem.
This isn't the first time Gates has spoken about artificial intelligence. About one year ago, he said software substitution for labor, whether it be for "drivers or waiters or nurses," is progressing when speaking at The American Enterprise Institute in Washington D.C.
In a recent Ask Me Anything thread on Reddit, Gates also said he's "in the camp that is super concerned about artificial intelligence."