How smart is too smart? According to Elon Musk, artificial intelligence will become too smart too soon. He is calling for government regulations, but he isn’t exactly a Cassandra.
Musk provided an audience at the International Space Station R&D Conference with an example of technology that is too smart: DeepMind’s AlphaGo. The machine’s intelligence isn’t the concern so much as the speed at which the machine was able to learn and apply its knowledge. AlphaGo is years ahead of schedule and this lack of predictability is disturbing. It indicates that AI is unpredictable and if it goes on “learning” at rates that are faster than developers anticipate it can cause a “tidal wave” of problems.
Musk’s claims aren’t dramatic. He put his concerns in context by comparing AI to other technologies that had been revolutionary at the time – including air travel and telecommunications. Both of those technologies evolved alongside government agencies, like the FAA and the FCC, that monitored and controlled their development. Musk suggests that a similar agency needs to be developed now so that AI developers have limits to what their bots can accomplish.
The idea of tighter regulations seems to go against entrepreneurial innovation and Musk is aware of his own contradictions. As an entrepreneur and inventor he often rails against government regulations but he does see how they are necessary. He is calling for regulations now because he knows that artificial intelligence is growing more sophisticated and powerful each day. By the time the government catches up, Musk warns that artificial intelligence is, “…going to come on like a tidal wave.”