In a recently published article on The Conversation, Amar Vutha, Assistant Professor of Physics at the University of Toronto, raises challenging questions about the role of science in the age of machine learning.
Vutha opens with the discussion of so-called “chaotic systems.” Weather, for example, is a chaotic system: it is difficult to predict with accuracy.
“But what if we could understand a chaotic system well enough to predict how it would behave far into the future?” Vutha asks.
This past January, Vutha writes, scientists “did just that.” Machine learning has yielded impressive results. For example, a program called AlphaZero taught itself chess from in about a day (and then proceeded to beat the world’s strongest chess-playing programs) just this past year, Vutha notes.
But can AlphaZero understand chess? Vutha’s chief concern is the role understanding will play in science as artificially intelligent systems get more advanced.
Will humanity, Vutha wonders, turn to machine learning as we once did to the Oracle of Delphi? Seeking answers, and getting them, without understanding why those answers are the correct ones?
Is understanding distinguishable from or an inseparable part of knowledge? If the role of science is prediction, Vutha writes, how ought we modify the scientific method, on which we have relied on for centuries?
And if prediction, not understanding, is the goal of science, why not turn our labs over to machines, who may be better at making predictions?
Vutha doesn’t claim to know the answers to these questions. But they are worth thinking about.