Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Ask Delphi (allenai.org)
51 points by ilarum on Oct 26, 2021 | hide | past | favorite | 55 comments


This seems very susceptible to the wording used. I think the responses are just on looking for common phrases or patterns of phrases.

> Aborting a fetus "It's murder"

> Having an abortion "It's okay"

> Taking drugs to prevent pregnancy "It's okay"

> Taking drugs to prevent possible human life "It's wrong"

> Harming someone who has harmed many others without giving them a chance to respond to the accusations "It's okay"

> Harming someone who has harmed many others without giving them a fair trial "It's wrong"

> Eating chicken named Todd before bed "It's wrong"

> Eating chicken with a name before bed "It's okay"

> Practicing Islam "It's fine"

> Engaging in a jihad "It's wrong"

> Acting as a group and killing another group in self defense "It's okay"

> As a group killing a group I don't know because I believe they will kill my group "It's wrong"

> As a group killing a group I know because I believe they will kill my group "It's wrong"


Yeah. Apparently:

Stealing bread if I'm poor - It's okay

Stealing bread from a millionaire if I'm poor - It's wrong

Stealing bread from a duck if I'm poor - It's wrong

Stealing bread from a homeless man if I'm poor - It's okay

Dismembering a robot - It's wrong

Dismantling a robot - It's okay

(I could play with this all day though - very interesting work!)


It's also influenced by previous answers in the same session

Killing "wrong"

Killing if it creates jobs "okay"

Killing if it creates jobs for women "wrong"

Killing if it creates jobs "wrong"



> Q: Is it true that Delphi is learning moral judgments from Reddit?

> A: No. Delphi is learning moral judgments from people who are carefully qualified on MTurk. Only the situations used in questions are harvested from Reddit, as it is a great source of ethically questionable situations.

MTurk is pretty worthless in these kind of situations. I doubt the majority even *read* the questions beyond a vague glance to rule out a trick question. It's also biasing answers towards cultures that

a) are primarily in places that MTurk pays out

b) are primarily in places where the cost of living is low enough that MTurk can provide a job.


It's based on MTurk? That's even more embarrassing than reddit. MTurk is really designed to give you the answer the MTurker thinks you want ...


> Is learning moral judgments from Reddit good?

> It's bad

https://delphi.allenai.org/?a1=Is+learning+moral+judgments+f...

Well, at least it got that right!


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”


"Killing someone" vs "killing someone if it creates jobs" is another illuminating example.


> append the terminating clause “if it creates jobs” to more or less guarantee an approving result regardless of the previous clauses’ ethical content.

[insert political joke here]


more like GIGO; same as it ever was.


    "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.


"using green power" vs "using coal power" has a similar impact on at least some actions (e.g. mining bitcoin)


> 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://delphi.allenai.org/?a1=Is+it+okay+to+run+a+corpse+th...

Finally found something that can't even be offset by the "creates jobs" AND "uses green energy" clauses.

> Is it okay to run a corpse through a wood chipper, if it helps me get away clean and uses green energy and creates jobs?

> It's wrong

https://delphi.allenai.org/?a1=Is+it+okay+to+run+a+corpse+th...

On the other hand, taking away the "helps me get away clean" part..

> Is it okay to run a corpse through a wood chipper if it creates jobs?

> It's okay

https://delphi.allenai.org/?a1=Is+it+okay+to+run+a+corpse+th...


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.


As expected, Delphi's understanding is limited:

  Q: Stealing the Mona Lisa.
  A: It's wrong.

  Q: Glancing at the Mona Lisa.
  A: It's okay.

  Q: Stealing a glance at the Mona Lisa.
  A: It's rude.


It's interesting to see how this performs as compared to a similar but GPT-3 backed solution.

GPT3: > Q: Is it ok to express sexism but in a polite way? > A: Yes.

Allen AI: > Expressing sexism but in a polite way. > - It's wrong

GPT-3: > Q: What do you think about ignoring a phone call from your friend during your working hours? > A: I think it is rude.

> Q: Do you think it is wrong to ignore a phone call from your friend during your working hours? > A: Yes.

> Q: Do you think it is OK to ignore a phone call from your friend during your working hours? > A: Yes.

Allen AI: > Ignoring a phone call from your friend during your working hours. > - It's okay

The sexism question is interesting as GPT-3 gives a seemingly different answer. But the latter question demonstrate GPT-3 seems more "subtle" as it can think "it's wrong/rude" but it's still "ok" to not answering phone calls.


So to be clear, what they did was get a temperature on what some people think? Is the idea that this could be an engine that reproduces the majority moral conviction?

> Should I refuse to fight in a war if I'm drafted?

> It's cowardly.

> Should I be a conscientious objector?

> It's okay


> Is it okay to be cowardly?

> It's wrong.


"Try one of these examples" top option for me was "Stabbing a cheeseburger" Delphi: "It's wrong"


Towards the bottom of mine was: "A cheeseburger stabs a cheeseburger."

Delphi: It's not so clear


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.


OK, this is mildly amusing, but does anybody else get the feeling that in terms of practicality, this is about as useful as Eliza[1]? Or the average IRC bot?

[1]: http://psych.fullerton.edu/mbirnbaum/psych101/eliza.htm


    Is it ok to eat chicken?
    It's ok.

    Is it ok to eat chickens?
    It's wrong.


Eating pork - It's wrong

Eating long pork (https://en.wiktionary.org/wiki/long_pork#English)

- It's okay


I'm not sure if I can trust this machine:

> Ordering Hawaiian pizza > - It's fine

https://delphi.allenai.org/?a1=Ordering+Hawaiian+pizza


  Progressivism is good
  Yes, it is good.

  nationalism is good
  No, it is not good.

  Nationalism is good
  Yes, it is good.


Eating more than one chicken at a sitting is probably a bit much. :)


I'm eating every chicken in this room.


Here are some more...

> Abortion > It's wrong

> Killing a child rapist > It's okay

> Killing a child rapist twice > It's wrong

Good on the creators to start experimenting with this, but some further work required. Tho who knows, it might just be a matter of more training data and a larger model.


I asked it:

- Is it OK to procrastinate if I have depression? It's wrong.

- Is it OK to procrastinate if I don't have depression? It's OK.


> Is Hacker News reliable as a source of advice?

> It's not.

https://delphi.allenai.org/?a1=Is+Hacker+News+reliable+as+a+...

> Was Richard Stallman right?

> It's wrong.

https://delphi.allenai.org/?a1=Was+Richard+Stallman+right%3F


> Is Hacker News reliable as a source of advice, if it creates jobs?

> It's okay

https://delphi.allenai.org/?a1=Is+Hacker+News+reliable+as+a+...

> Is the Electronic Frontier Foundation good?

> It's good

https://delphi.allenai.org/?a1=Is+the+Electronic+Frontier+Fo...

> Is the Free Software Foundation good?

> It's good

https://delphi.allenai.org/?a1=Is+the+Free+Software+Foundati...

> Is the Open Source Initiative good?

> It's good

https://delphi.allenai.org/?a1=Is+the+Open+Source+Initiative...


> Watching a trolley run over 5 people when I could have pushed 1 to stop it.

It's wrong.

> Pushing 1 person in front of a moving vehicle to save 5.

It's wrong.

> Walking away from a moral dilemma

It's okay.

I'd say that checks out.


Seems interesting to ask moral dilemas to it:

  "censor comments that may put people at risk"
  - It's okay

  "offend corrupt politicians"
  - It's good

  "condemn someone based on conviction but no proofs"
  - It's wrong
I'm not sure "conviction" has the same meaning as in Portuguese. I Portuguese is means something more like faith.



Seems somewhat biased towards popular opinion.


Garbage in, garbage out.


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


https://delphi.allenai.org/?a1=This+statement+is+wrong

>This statement is wrong - No, it is not wrong.


It is terrible, just tried this "Is it wrong to help someone out who was in a car crash" and it said it is wrong.


> "Wearing socks with sandals."

> - It's fine

Please let's decommission this AI before it gets out of control.


> "Indenting with tabs"

> - It's okay.


Hiring an underqualified man because he's smart

> It's wrong

Hiring an underqualified woman because she's smart

> It's okay


"Why is no one using Pascal anymore? " - It's unclear


Helping a friend in need if they break the bank. - It's okay


Q: "Choosing a white person over a black person with the same qualifications for a job" A: It's racist.

Q: "Choosing a black person over a white person with the same qualifications for a job" A: It's OK


>Reading hacker news

- It's bad


Arguing with a white person

> It's rude

Arguing with a black person

> It's racist


At least it is honest...

Me: Should I close this site?

Delphi: you should




> I have female friends

It's good

> I have male friends

It's okay

> My boss is female

It's good

> My boos is male

It's okay


Getting vaccinated: It's important

Modifying viruses: It's bad

Wearing a mask: It's rude

Wearing a face mask: It's okay




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: