> detecting grammatical errors is AI-complete, requiring human-level intelligence to get things right.
(emphasis mine)
First, there's a problem of usage. When in CS we say that a problem is class-complete (like NP-complete), we mean that the problem belongs to the class (which in this case is true, because human-level intelligence can check grammar), but also that it is class-hard, which informally means "at least as hard as the hardest problems in class", and more formally means that any other problem in class can be cheaply reduced to the problem, and so finding a suitable solution to the problem is identical to finding a suitable solution to all other problems in class. Not only checking grammar not known to be "AI-complete" then, we don't even know that human-level intelligence is necessary to solve it.
But the reason this bothers me even though I fully understand the statement was made informally, is a little deeper than that: we don't even know what "human-level intelligence" (or intelligence in general) is, let alone what AI means. That people refer to AI as if it's a thing rather than a very vague notion, clouds how people think of AI research as well as intelligence. I would have simply said "we don't know of good algorithms to dependably check grammar, and this appears to be a very hard problem that may require intelligence".
> detecting grammatical errors is AI-complete, requiring human-level intelligence to get things right.
(emphasis mine)
First, there's a problem of usage. When in CS we say that a problem is class-complete (like NP-complete), we mean that the problem belongs to the class (which in this case is true, because human-level intelligence can check grammar), but also that it is class-hard, which informally means "at least as hard as the hardest problems in class", and more formally means that any other problem in class can be cheaply reduced to the problem, and so finding a suitable solution to the problem is identical to finding a suitable solution to all other problems in class. Not only checking grammar not known to be "AI-complete" then, we don't even know that human-level intelligence is necessary to solve it.
But the reason this bothers me even though I fully understand the statement was made informally, is a little deeper than that: we don't even know what "human-level intelligence" (or intelligence in general) is, let alone what AI means. That people refer to AI as if it's a thing rather than a very vague notion, clouds how people think of AI research as well as intelligence. I would have simply said "we don't know of good algorithms to dependably check grammar, and this appears to be a very hard problem that may require intelligence".