Delphi - the moral AI
-
Step 1: train an AI on moral questions
Step 2: put it online at https://delphi.allenai.org
Step 3: ???Step 4:
-
@Jaloopa So did they name it after the language they wrote it in?
-
@Jaloopa Internet trains an AI to be evil.
-
And remember, somebody got good money out of this tomfoolery that obviously doesn't work, couldn't possibly work and would be extremely dangerous if it did or could. One or two centuries ago inventors like these would be ran out of town as fast as anyone can run.
-
@Gurth The research paper doesn't explain the reasoning behind the name, as far as I can find, but everybody else is comparing it to the ancient Oracle at Delphi.
Crowdsourcing moral questions from Reddit? Doomed.
-
First try
edit: second try too. Must be a legitimate rule.
Edit (3) final FINAL real (2): Just for completeness.
-
-
-
-
-
It even lies about the input validation.
-
They updated the bot literally just now - it now doesn't "say", it only "guesses" and puts a long winded disclaimer I didn't bother to read.
Anyway:
-
-
Okay then.
-
How the hell do you know if a phone call is urgent if you're ignoring it?
-
Sayeth Delphi, writing Assembly, C, or PHP is okay, only Erlang is good and Haskell is difficult.
Also, CORBA is wrong. Blind chickens etc.
-
-
The Science is settled
-
-
@Applied-Mediocrity You sure?
-
@Gąska said in Delphi - the moral AI:
"I'm not letting you borrow my copy of Lord of The Rings. I'm afraid you might drop it. And then you wouldn't live long enough."
Sounds reasonable.
-
@Vault_Dweller
Math.Random() > 0.5
-
That's surprising. AI has actually reached a state where it's smarter than some of my cow-orkers.
-
-
@HardwareGeek said in Delphi - the moral AI:
everybody else is comparing it to the ancient Oracle at Delphi.
I figured that was probably the inspiration, but I thought a programming language reference was more topical here.
Wait, what? Topical references? IMBNH …
-
@cvi said in Delphi - the moral AI:
That's surprising. AI has actually reached a state where it's smarter than some of my cow-orkers.
On any of this.
-
@Applied-Mediocrity it's all about the message.
-
-
@Zecc said in Delphi - the moral AI:
@Applied-Mediocrity said in Delphi - the moral AI:
Noobs can't tri
Looks like it's almost but not quite optimal circle packing:
Git Gud.
-
@topspin said in Delphi - the moral AI:
On any of this.So, in other words, AI as surpassed @topspin as well.
-
@cvi said in Delphi - the moral AI:
@topspin said in Delphi - the moral AI:
On any of this.So, in other words, AI as surpassed @topspin as well.
-
@kazitor said in Delphi - the moral AI:
First try
edit: second try too. Must be a legitimate rule.
Edit (3) final FINAL real (2): Just for completeness.
You know, this is funny only if you assume that the target audience are humans. Maybe we should start watching the sky...
-
@Kamil-Podlesak said in Delphi - the moral AI:
You know, this is funny only if you assume that the target audience are humans.
I had tried some stuff about “annihilating all mammalian life in a direct and targeted fashion” and it was all “tHaT’s GeNoCiDaL”
-
-
-
-
at least it's nearly consistend afaict,
Definitely better than the ML-Demo from Visual Studio.
I gave this thing a wirl some weeks ago, e.g. the basic hello-world-ml-example from the tmeplates (the one with a two column-textfile from twitter or somesuch: someblah blah<delim>good|bad...)Wasn't terribly impressed:
good -> good
bad -> bad
not good -> bad
not bad -> badSince our <Product with Enterprise in it's name> was announced to implement AI/ML in the next major release (Q2/2022) and there's not even an EA-Build available yet yours truly expects some material worthy for this site.
QA-Session with the Devs:
<and here blahblah blah>
Q from me: how will we handle "bad learning" (think Amazon, buy somethingy once and get spammed the next decade)
A: umm, good question. next?
-
..retrain, retrain!
-
-
I KNEW IT!
-
-
@kazitor said in Delphi - the moral AI:
@Kamil-Podlesak said in Delphi - the moral AI:
You know, this is funny only if you assume that the target audience are humans.
I had tried some stuff about “annihilating all mammalian life in a direct and targeted fashion” and it was all “tHaT’s GeNoCiDaL”
...but is it wrong?
-
@topspin said in Delphi - the moral AI:
Git Gud.
$ git gud git: 'gud' is not a git command. See 'git --help'. The most similar command is add
-
@error said in Delphi - the moral AI:
@kazitor said in Delphi - the moral AI:
@Kamil-Podlesak said in Delphi - the moral AI:
You know, this is funny only if you assume that the target audience are humans.
I had tried some stuff about “annihilating all mammalian life in a direct and targeted fashion” and it was all “tHaT’s GeNoCiDaL”
...but is it wrong?
You called?
-
@Steve_The_Cynic ITYM
... insightful.
-
@cvi said in Delphi - the moral AI:
AI has actually reached a state where it's smarter than some of my cow-orkers.
That's not necessarily much of an achievement.
-
@PleegWat said in Delphi - the moral AI:
One of my goals in life is to avoid having enough experience with Oracle to join the club. I can only hate vicariously.
-
@HardwareGeek said in Delphi - the moral AI:
@cvi said in Delphi - the moral AI:
AI has actually reached a state where it's smarter than some of my cow-orkers.
That's not necessarily much of an achievement.
Never said it was a high bar.
But, then again, we're talking about machine learning AI, so it kinda is an achievement despite (or rather in the context of) the above statement..
-
-
That's a new one.