Measuring intelligence takes more than one score

Google DeepMind published a cognitive taxonomy for assessing progress toward artificial general intelligence and announced a competition to develop related evaluations. The proposal separates different capabilities rather than compressing the discussion into a single ranking.

Context

For readers, it is a reason to ask what “smarter” means in each announcement. A system may excel in one task family and remain unreliable in another. A meaningful evaluation should show the capability profile, explain what the test leaves out, and state the conditions under which the result was obtained.