Hacker News Reader

Comment by andreilys | Hacker News Reader

> "Some program that self learned a (admittedly impressive) number of reactions based on seeing/playing a huge number of simulations is not intelligence so not AI."

Seems like with every breakthrough the goalpost of AI gets moved.

I believe a researcher coined a phrase for this but I can't remember what it was.

View on HN

It doesn't get moved so much as we discover that a certain mathematical manipulation plus all our previous mathematical manipulations don't amount to an AI.

It also seems like with every breakthrough this complaint is raised without addressing the underlying issues with the new AIs shortcomings.

My hot take: true AI is so far out of reach of our ability it's not even funny, which makes the whole field either a search for the fountain of youth at worst, or at best a search for tools to inform humans or to replace humans in rote tasks. See https://youtu.be/orMtwOz6Db0

Ummm, no? AI means a clear thing: artificial _intelligence_. Depth-first search is not intelligence. Auto-generated algorithms that succeed on most of the training data is not intelligence. And that is obvious from the errors "AI" makes that a human would never do.

For example, would you fool even a child with this? https://cdn.mos.cms.futurecdn.net/s4DuKgTLnS4cngTyiqVkNC-970...

Someday, this line of argument is going to conclude that humans are not intelligent.

> I believe a researcher coined a phrase for this but I can't remember what it was.

There's an article on Wikipedia called "AI effect", I don't know if that's what you had in mind.

https://en.wikipedia.org /wiki/AI_effect

Reply on news.ycombinator.com