Can we build AI without losing control over it? | Sam Harris

We've had the iPhone since The Simpsons was on television. Fifty years is not a lot of time to face one of the greatest challenges our species will face. Once again, we seem to fail to have the right emotional response to what we have the most reason to believe is to come. Computer scientist Stuart Russell has a good analysis here. He says: Let's imagine we received a message from an extraterrestrial civilization that said, "People of Earth, we will arrive on your planet in 50 years. Prepare yourselves." And now we're counting down the months until the mothership lands? We'll feel a little more rushed than now.

We are told not to worry for another reason: these mechanisms can only share their benefits with us because they will literally be extensions of ourselves. They will be grafted into our brains and we will essentially become their limbic systems. Now take a moment to think that the only safe and wise path recommended is to implant this technology directly into our brain. It may be the only safe and sensible way forward, but one often has concerns about understanding the technology a little before getting it into one's head. (laughs) The deeper problem is that

build

ing a superintelligent AI on its own probably seems easier than

build

ing a superintelligent AI and having all the neuroscience that allows us to seamlessly integrate our minds with it.

Assuming that the companies and governments doing this work probably realize that they are in a race against everyone else, and assuming that winning this race is winning the world, assuming you don't destroy it next month, then it seems likely that Whatever is easiest to do will be done first. Now, unfortunately, I don't have a solution to this problem, other than to recommend that more of us think about it. I think we need something like the Manhattan Project for AI. Not to build it, because I think we inevitably will, but to understand how to avoid an arms race and build it in a way that is aligned with our interests.

When talking about superintelligent AI that can transform itself, it seems like we only have one chance to get the initial conditions right, and even then we will have to absorb the economic and political consequences of getting it right. But the moment we recognize that information processing is the source of intelligence, and that some adequate computing system expresses the basis of intelligence, we recognize that we will continually improve these systems, and we recognize that the horizon of perception probably greatly exceeds You grow what we imagine. We currently know, so we must recognize that we are in a phase of building some kind of god.

Now would be a good time to make sure he is a God we can live with. Thank you so much. (applause)