We Need to Rethink the Idea That Deep Learning Models Are "opaque."

At this point I believe that idea serves intellectual ideologies more than a search for understanding, much less useful technology.

Strange, yes. Opaque, less and less so. Do we understand how the models work, in depth and detail? No. But we ARE making progress at understanding them and we would make more progress if so many didn’t insist that they are opaque.

Right. The claim of opacity and a belief in AI magic reinforce one another. The inner workings of these models can be understood. It's difficult, but possible.
As comparisons, consider DNA and animal brains. They too are "opaque," but we are coming to understand them.

— Bill Benzon, BAM! Bootstrapping Artificial Minds (@bbenzon) January 5, 2023

That is to say, BOTH AI skeptics and fan boys benefit from the mantra of model opacity. While the two groups oppose one another, they are complicit in upholding this intellectual ideology of opacity.

— Bill Benzon, BAM! Bootstrapping Artificial Minds (@bbenzon) January 5, 2023