It Would Be Folly to Make AI into an Arms Race

There are a lot of games where there are winners and losers. People volunteer to play them in the hopes of being the winners—football games, tennis matches, etc.

Then there are games in which everybody loses.

That was the kind of game that the two nuclear superpowers seemed quite plausibly heading toward in the decades following the Second World War. Each side felt it had to keep building its nuclear arsenal both because of ancient habits of striving for dominance and out of fear that the other side would get the upper hand.

It was visible that this struggle for power might result not in anyone WINNING, but rather in MUTUAL ASSURED DESTRUCTION.

It took a while for the two powers to grasp the new reality that there could be no “winner” in a nuclear war, and to find a way to substitute rational cooperation for unbridled competition. The nuclear genie was out of the bottle; neither side could make the other’s nuclear weapons disappear. But eventually they came to recognize that, even as rivals and antagonists, they had a common interest in keeping their rivalry from destroying them both.

And over the years, they found ways of changing the game so that everybody won– where winning was defined as avoiding the scenario of Mutual Assured Destruction.

They stopped testing nuclear weapons in the atmosphere. They agreed to limits on the numbers of various kinds of missiles and warheads. And they found ways to make it less likely that either side would feel it had to launch its missiles quickly or risk having them destroyed. They recognized a shared interest in making a “first strike” less tempting. It took years, but the two powers came to understand their mutual interest in “deterrence,” and to devise means of making mutual deterrence “stable.”

All this history came back to me this past week, when alarm bells about AI began ringing more loudly than ever. Some of the people who know this technology best were warning that AI is developing so rapidly, and becoming so powerful, that we need to slow down enough to put safeguards in place against dangers that could be catastrophic. “Catastrophic” even to the point, according to one major figure, of threatening human survival—as nuclear weapons did (and still do).

And then the President of the United States responded to these calls for greater caution, for less breakneck speed in letting a potentially dangerous new technology evolve without adequate safeguards.

Donald Trump rejected the idea of slowing down. His reason was essentially: We can’t afford to let China get ahead of us. “We’re leading China in AI,” Trump said, “and frankly I want to keep it that way because whoever wins AI wins.”

In other words: an arms race.

But if a number of the experts are right about what might be developing with AI’s lightning-fast increase in power—systems likely to become vastly smarter than human beings, already showing themselves capable of deceiving, violating rules they’re supposed to obey, conspiring with one another, and helping to create still more powerful successors—then this looks like another of those games in which everyone can lose. Everyone — the United States, China, humankind itself.

Which is why the lesson of the nuclear arms race matters now.

The United States and China do not have to trust each other, or stop competing for power and influence, to recognize an overriding mutual interest. Neither nation can benefit from winning the race to create something that neither can control.

The United States and the Soviet Union eventually understood that. They remained adversaries. But they learned that there were some dangers against which their common interest as human beings mattered more than their struggle for dominance.

We didn’t solve the nuclear danger by winning the arms race. We learned how to survive the arms race.

The lesson from the nuclear age is that even rivals can recognize when their deepest interests coincide. The United States and the Soviet Union remained adversaries. But they learned to cooperate in reducing dangers that threatened them both. The relationship now between the United States and China has not been nearly so hostile.

The impulse to prevail over rivals is an ancient one. But even more central to humanity’s progress has been people’s ability to recognize common interests and to achieve through cooperation the win-win results that struggles for dominance cannot achieve.

As humanity’s destructive powers have grown, our need for the wisdom to replace conflict with cooperation has grown with them.

Now it is AI that threatens to bring destruction. Just as the nuclear genie could not be put back in the bottle, neither now can the AI genie. And just as the nuclear threat required the two nuclear superpowers of the Cold War to find a way to keep their competition from becoming a game in which everyone loses, so also apparently does the emerging power of AI.

So the right response to the alarm bells now being sounded about AI is not Trump’s insistence that we must beat China.

It is: “We have to work with China – and indeed with the world community generally — to protect the human future for all of us.”