The Case Against Deep-Learning Hype

Is there more to AI than neural networks? Gary Marcus, professor of psychology at NYU and ex-director of Uber’s AI lab, thinks so. He’s published a critique of deep-learning systems that use neural nets, and it skewers some of the current AI hype.

Deep learning’s limits: Marcus identifies 10 major hurdles facing deep learning, including data hunger and lack of generalization. For what it’s worth, we’re tempted to agree that it’s not the silver bullet many think (see “Is AI Riding a One-Trick Pony?”).

The risk of hype: He argues that overselling the abilities of deep learning provides “fresh risk for seriously dashed expectations” that could bring another AI winter, as well as blinkering AI researchers from trying new ideas.

What now? But Marcus doesn’t dismiss deep learning entirely: instead, he suggests that we should “conceptualize it, not as a universal solvent, but simply as one tool among many.”