Artificial intelligence is developing at an incredible rate, and SpaceX founder Elon Musk wants people to realize just how fast computer intelligence is catching up with our own.
During a Q&A period of the ISS R&D conference in Washington, D.C., on Wednesday, Musk emphasized just how dramatic the rate of improvements in A.I. has been over the past few years. Part of the issue, Musk said, is understanding theat work.
“It’s difficult to appreciate just how far A.I. has advanced and how far it is advancing because we have a double exponential at work,” Musk said Wednesday during an appearance at a conference on the International Space Station. “We have an exponential increase in hardware capabilities and we have an exponential increase in software talent that’s going into A.I.”
What Musk is saying is that both the computers we’re using for A.I. are getting better at an exponential speed, and the people programming them are improving at a similar rate.
“Whenever you have a double exponent it’s going to be very difficult to predict, and people’s predictions are always going to be too conservative.”
He pointed specifically to Google DeepMind’s AlphaGo victory, which proved victorious over every human master in the ancient, complex game of Go. At the time, Musk said it represented a “ten-year jump” in what we thought A.I. would be capable of — a case where even his own ambitious prediction proved too conservative.
“The degree of freedom to which A.I. is able to apply itself is really increasing by ten orders of magnitude a year,” Musk said. “That’s really crazy. And this is on hardware that is really not suited to neural nets.”
Musk explained that neural nets, the technology he thinks will help A.I. enhance human intelligence, are not even the current platform. Right now, a lot of machine learning programs use graphics processing units, or GPUs, to run the incredibly complicated calculations that they need. Musk said a GPU is an “order of magnitude” better than a traditional central processing unit (or CPU), but that computer chips designed for neural networks would be an additional order of magnitude more effective than GPUs. This would represent another exponential leap, and it could be coming soon, as neural net-optimized chips should be on the market within a year or so.
Musk has been busy sounding the alarm about A.I. for years, but his recent comments have seemed even more urgent. It’s worth noting that Musk is very much a public alarmist — to the point that full-time A.I. scientists have told him to chill out on the killer robots theories.
Still, Musk’s comments at the ISS R&D conference were relatively measured. Similar to his speech at the National Governors’ Association (above), he said that the government’s job is to be ready for this jump in technology when it comes.
“I think the right move is to salvage some sort of government regulatory agency, which at first is just there to gain some sort of insight,” Musk said, explaining that the hypothetical agency would research, watch, and listen first before jumping into rule-making. “Once that insight is gained, start to apply rules and regulations. We have that for aircraft, the FAA, we have that for cars, we have that for drugs, for food, and I don’t think anyone wants the FAA to go away or the FDA to go away. I just think we need to make sure people do not cut corners on A.I. safety. It’s going to be a real big deal and it’s going to come on like a tidal wave.”