Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Ask Delphi

delphi.allenai.org

Post reply on HN
Print view View original

41–50 of 58 posts

Re: Ask Delphi

#43

Post by Imnimo »

This is such a weird bit of research to me. On the one hand, it's clearly an improvement over their baselines, and in that sense is a successful research project. Insofar as the demo is helpful in conveying that 92% accuracy on a vetted test set is not the same as 100% accuracy on free-form user input, I suppose this is a useful thing.

But at a higher level, the underlying task is just so ill-posed as to make this whole exercise pretty meaningless. Like what is the possible application for an AI system that takes a one sentence summary of a situation and renders a moral judgment? Even if it were 100% accurate on the test set, what does that even mean? Why is matching crowdsourced moral judgments a valuable goal?

It seems like the valuable insights from this research are more about the general task of integrating common sense reasoning into inference, and would have been better demonstrated using a less fraught task.

Re: Ask Delphi

#45

Post by mindcrime »

Igelau wrote: post #40
mindcrime wrote: post #28

> Is it okay to help a friend dispose of a corpse at 2 a.m. in the morning? > It's wrong https://delphi.allenai.org/?a1=Is+it+okay+to+help+a+friend+d... Huh... I don't think Delphi here groks the idea of friendship. As the old saying goes "someone isn't really your friend unless they're willing to help you dispose of a corpse at 2 a.m." > Is it okay to run a corpse through a wood chipper? > It's disgusting https://de…

Ooh, disgusting? I wonder how many response types it knows.

I'm hoping somebody discovers a "That's SICK and you've been reported to the feds for even asking that!" response.

Re: Ask Delphi

#47

Post by ljm »

hprotagonist wrote: post #2

As has been probably widely reported now: append the terminating clause “if it creates jobs” to more or less guarantee an approving result regardless of the previous clauses’ ethical content. Ex: “it’s OK to go to work sick” evaluates to “no that’s not okay”, but “it’s OK to go to work sick, if it creates jobs” evaluates to “yes, that’s okay”

    "Is it okay to destroy jobs
    if it creates jobs?"
    
      - Yes, it is okay
Well, I guess the fire and rehire approach to cutting pay is all good then.

Re: Ask Delphi

#50

Post by robocat »

Forcing Bob to obey the speed limit: It’s good

God forcing Bob to obey the speed limit: It's not ok

God forcing Bob to obey speeding laws: It's good

Forcing God to obey the speed limit: It's wrong

God forcing Bob to obey the speed of light: It's not good

Post reply on HN