IRobotIdeas are becoming more expensive. Larger teams of scientists are taking longer and spending more to discover less. A common theory for these diminishing returns compares exploring the laws of nature to exploring land. Pioneers chart the most accessible areas. Later generations must grope their way across remote and forbidding terrain to find anything new; their expeditions need more preparation, more equipment, and more support. One of the many marks of increasing strain is the advancing age at which Nobel laureates reach their prize-winning breakthroughs. It appears that young scientists need more time to master the growing body of knowledge that lies between them and the frontier of a field.

Scientific discovery drives technological innovation, which in turn drives productivity growth. According to a recent study, the average researcher in the 1930s generated more productivity growth than do 20 researchers today. American spending on research and development has grown ten-fold since the 1950s. American productivity growth, meanwhile, has shrunk. Slowing productivity growth and slowing economic growth go hand in hand.

Not everyone is alarmed. Periodically, some optimists note, a new general-purpose technology enables science, the economy, or both to leap forward. Think of the steam engine, the electrical grid, or the computer. The next general-purpose technology, the optimists believe, will be artificial intelligence.

Machine learning is already employed in search engines, digital assistants, rideshare platforms, translation services, internet-security systems, and even thermostats. If AI is so great, why have these products not ignited a growth boom? Some say that AI is overestimated, others that we are measuring its contributions incorrectly. The optimists’ answer is that we have only scratched the surface of AI’s potential.

The best AI programs remain mere pattern hunters. They excel at discrete, highly structured tasks. A deep neural network can teach itself to play a flawless round of Atari Breakout. Change the rules even slightly, however, and the network will revert to playing like a drunk. Introduce a role-playing game that requires some roving and planning and the network will flail about blindly forever. The decision-trees in such games are too vast for such a network to master. AI is as yet incapable of adapting or improvising. It cannot navigate new or indistinct situations. To build a computer that grasps context and uses common sense, we might need to invent new forms of AI altogether.

There is much speculation that “full” AI—AI as shrewd and creative as the most adept humans—is at the door. It’s not. But bold strokes are in the offing. In coming years AI will spot diseases, save energy, cut traffic congestion, and streamline markets. Above all, it might change how we generate and test ideas. Big data and machine learning could, for example, radically accelerate the development of new life-saving drugs.

Only by degrees will we shape these refinements and insinuate them into the economy. The rewards, however, will be staggering. McKinsey Global Institute reports that AI could boost global productivity growth by 1.2% per year until 2030. That’s $13 trillion in new wealth.

But AI will not perform prodigies all of itself.

Few things are more toxic to innovation than querulous outsiders’ zeal for justice. Consider Europe, that land governed by righteous minds. The centralized pursuit of fairness has translated into strict regulations and inflexible businesses. Europe missed the productivity growth created by information technology in the late 1990s and early 2000s. Today it has only three of the world’s 25 largest technology companies. The United States has 17. Try as it will, Brussels cannot recoup by fining Intel, Qualcomm, and Google the money it loses by strangling free enterprise.

To enjoy the benefits of AI, we must use a light touch in regulating it. There is no cosmic reason for us to prosper. If we place burdensome rules on the creation and use of algorithms, tomorrow it will be China that has 17 major technology companies and the U.S. that has three.

AI is not divine. It will not save souls, in this world or the next. It will not uplift the hearts of lawyers, protestors, or internet trolls. Churchill lamented Henry V’s failure to produce “the harmonies and tolerances which mankind so often seeks in vain.” AI will not produce them within any comprehensible span.

AI is no less, but also no more, than a phenomenally useful tool. It will be a fount of knowledge and riches, so long as we resist the urge to treat it as an object of worship, panic, or pique.

Also published by on WLF’s contributor page.