> Alexa is a glorified text-to-speech and NLP engine.

Where do you draw the line between NLP and intelligence, though?

This all sounds like a "No true Scotsman" to me. At some point in the near future AI will be indistinguishable from human-level intelligence in terms of NLP (i.e. be on-par with the average human in relevant benchmarks).

Maybe the problem is that we need a clear definition of what "intelligence" is and recognise that it's a spectrum, not a binary qualifier.

Personally I draw the line at the Winograd Schema Challenge. When that's passed I will be truly convinced.

According to Wikipedia, GPT-3 is almost there (88% vs ~92%) and a finetuned BERT even achieves >90%. So it stands to reason that a finetuned GPT-x would pass the test with flying colours albeit arguably not being very capable at "common sense reasoning" (whatever that should be, as "common sense" is often enough wrong).

I am sceptical of such tests, as all they do is provide yet another way to game the system instead of demonstrating whatever arbitrary construct scientists use to weasel around the fact that they don't actually know what they even mean by "intelligence".

IMHO we'd be better off defining characteristics for various levels/types of intelligence and trying to categorise our machines and fellow creatures accordingly.