One limitation of “intelligence” is that it isn’t task-specific. There’s no IQ or ‘g’ for machines (yet) and their performance varies along many dimensions. We

One limitation of “intelligence” is that it isn’t task-specific. There’s no IQ or ‘g’ for machines (yet) and their performance varies along many dimensions. We should see how well a machine does on each task we care about, using task-specific benchmarks and testing. This avoids a lot of fuzzy thinking about potential.