That’s a good point about justification. Thanks, Katherine. But I *think* my point still holds, for if the DL model is inexplicable to us, even after we’ve satisfied ourselves about the criteria you list (the data is representative, etc.), we will be relying on a chain of logic that we cannot follow. If a conclusion results from a chain of logic, traditionally we want to be able to inspect and verify that chain before we will declare the conclusion justified. Here, in theory (I think, although I should be very careful about making this sort of claim to the Principal Data Scientist at Acquia!) we cannot. We are thus relying on the machine for justification, although you are right that it is not a purely blind trust.

I raise the extended mind idea in the post because I do not think this is the first time we’ve used objects as part of knowledge. We’ve always thought out in the world with tools. DL makes that fact inescapable, I think, which is why we’re now seeing so many discussions of the moral decisions AI is going to be making, e.g., networked autonomous cars.