The weirdness of AI

[Priscus]

I just posted this question on Google search: "Why has 'don't mind if I do' replaced the phrase, 'I don't mind if I do' "?

Google AI told me:


"Don't mind if I do" has effectively replaced "I don't mind if I do" because it emphasizes a positive acceptance of something offered, essentially meaning "yes, please" -

This completely misses the point that a statement about oneself has been replaced by a directive to others!

This minded me of, dating back to m1960's, when the program called 'Eliza' impressed onlookers, though its limitations were exposed when its response to the assertion: "I think, Necessity is the mother of invention" elicited the response: "tell me more about your family"!

I guess we are in for a hard time distinguishing useful AI outputs from that which is a sheer con.

 

[popatim]

Sadly there isn't much i in Ai yet, just a smart database search & matching algorithms imo. No one really verifies what it puts in its database yet so you get odd or completely wrong answers because that is what it stored.

 

[shawnhcorey]

The current crop of AI doesn't record facts. It finds patterns. Its responses are based on those pattern and often are not useful.

 

[USAFRet]

You are giving too much credit to the current state of the AI bots.

There isn't a lot of "I".
It is there to spit out something that sounds a bit conversational, and is only maybe accurate.

 

[Gururu]

When AI can say, "I don't know, but your local library may have your answer", then we will have gotten somewhere.

 

[Priscus]

It is there to spit out something that sounds a bit conversational, and is only maybe accurate.

ie. Precisely what 'ELIZA' did, more than half a century ago. (Though admittedly ELIZA did not have the ability to fish facts from an extensive database.)

 

[Priscus]

Having ventured into trying 'Copilot', I am surprised that it seems nowhere near to being politically neutral.

I will refrain from saying which way it apparently leans, so this comment itself is not seen to be a political statement.

So, there is an interesting question: Can AI be partisan, or is this the action of the human who is feeding it showing through.

 

[shawnhcorey]

Having ventured into trying 'Copilot', I am surprised that it seems nowhere near to being politically neutral.

I will refrain from saying which way it apparently leans, so this comment itself is not seen to be a political statement.

So, there is an interesting question: Can AI be partisan, or is this the action of the human who is feeding it showing through.

Its bias is dependent on the data it is trained on. And the current crop of AI is always biased.

 

[Eximo]

Well I think the political leaning is fairly easy to explain.

When you are looking at factual information you have three main types of publishers. Those with science and engineering degrees publishing scientific data and articles. Liberal arts degree holders publishing articles on all manner of subjects from that viewpoint. And fiscally conservative publications that focus on data, statistics, etc.

Ideological and poorly referenced content is going to come from all sides in fairly equal numbers, so when you consult AI on those topics, it is going to be hit or miss depending on what it associates together.

Add a sprinkling of mild censorship that has happened over the years for off the wall ideas and it should filter out at least some of the nonsensical noise. You might recall all those early AI chatbots that managed to radicalize in mere days, those filters will still be in place in today's models.

 

[gamerk316]

So, there is an interesting question: Can AI be partisan, or is this the action of the human who is feeding it showing through.

Yes.

AIs are going to respond based on what they are trained on, so they can absolutely end up becoming partisan (and when they access to the internet, they *always* do).

 

[Priscus]

Ah the answer emerges:
hacked