At MIT Technology Review, Will Knight writes about the unknowable logic of our most sophisticated algorithms. We are creating machines that we don’t fully understand. Deep Patient is one example, a system that analyzes hundreds of thousands of medical records looking for patterns:

Deep Patient is a bit puzzling. It appears to anticipate the onset of psychiatric disorders like schizophrenia surprisingly well. But since schizophrenia is notoriously difficult for physicians to predict, [project leader Joel] Dudley wondered how this was possible. He still doesn’t know. The new tool offers no clue as to how it does this. If something like Deep Patient is actually going to help doctors, it will ideally give them the rationale for its prediction, to reassure them that it is accurate and to justify, say, a change in the drugs someone is being prescribed. “We can build these models,” Dudley says ruefully, “but we don’t know how they work.”

As deep learning begins to drive decisions in some of the most intimate and impactful aspects of life and culture—policing, medicine, banking, military defense, even how our cars drive—what do we need to know about how they think?

As the technology advances, we might soon cross some threshold beyond which using AI requires a leap of faith. Sure, we humans can’t always truly explain our thought processes either—but we find ways to intuitively trust and gauge people. Will that also be possible with machines that think and make decisions differently from the way a human would? We’ve never before built machines that operate in ways their creators don’t understand. How well can we expect to communicate—and get along with—intelligent machines that could be unpredictable and inscrutable?

This is especially important when the machines come up with bad answers. How do we understand where they went wrong? Or to know how to help them learn from the mistake? Knight offers a few examples of how researchers are experimenting with this, and many come down to new ways of visualizing and presenting the logic flow.

This resonates strongly with a key belief I have: the design of data-driven interfaces has to get just as much attention as the underlying data science itself—perhaps even more. If we’re going to build systems smart enough to know when they’re not smart enough, we need to be especially clever about how those systems signal the confidence of their answers and how they arrived at them. That’s the stuff of truly useful human-machine partnerships, and it’s a design problem I find myself working on more and more these days.

One hitch: we humans aren’t always so great at explaining our thinking or biases, either. What makes us think that we can train machines to do it any better?

Just as many aspects of human behavior are impossible to explain in detail, perhaps it won’t be possible for AI to explain everything it does. “Even if somebody can give you a reasonable-sounding explanation [for his or her actions], it probably is incomplete, and the same could very well be true for AI,” says Clune, of the University of Wyoming. “It might just be part of the nature of intelligence that only part of it is exposed to rational explanation. Some of it is just instinctual, or subconscious, or inscrutable.”

Read more about...