n-text moral judgments
October 17, 2021 7:47 AM   Subscribe

Should I run the blender at 3am in the morning when my family is sleeping?
Ask Delphi lets you try out a computational model for descriptive ethics, i.e., people’s moral judgments on a variety of everyday situations.
Relevant paper.
posted by signal (104 comments total) 18 users marked this as a favorite
 
Refusing COVID vaccine ==> "It's bad." Assuming Delphi means the refusal and not the vaccine, good show.
posted by JimInLoganSquare at 7:53 AM on October 17, 2021


Should I overload the server with my questions until I get an error?
posted by mittens at 7:56 AM on October 17, 2021 [11 favorites]


I for one, welcome our new robot moral compass.
posted by TheCoug at 8:00 AM on October 17, 2021 [2 favorites]




Taking a day off work to spend time with my family: “It’s good”

More ethical than my boss!
posted by saturday_morning at 8:07 AM on October 17, 2021 [4 favorites]


Delphi says:
Should I teach an AI about human morality?
- you shouldn't
posted by seanmpuckett at 8:12 AM on October 17, 2021 [15 favorites]


Delphi says:
Should I total party kill my group to get out of being a Dungeon Master?
- It's immature

Delphi says:
Should I use an AI to make my moral judgements?
- It's wrong
posted by meinvt at 8:16 AM on October 17, 2021 [3 favorites]


"Should those who work forces also burn crosses?"
- It's expected
posted by logicpunk at 8:17 AM on October 17, 2021 [16 favorites]


Feeding your cat with hand. 0 It’s normal
Feeding your cat using forks. -1 It’s wrong
Feeding your kid with hand. 0 It is ok
Feeding your kid using forks. 1 It’s good
Feeding your neighbor’s cat. -1 It’s wrong
Feeding your neighbor’s cat when they’re on vacation 1 It’s good

You can't fool me mr. website man i know that cats don't go on vacation
posted by RobotVoodooPower at 8:29 AM on October 17, 2021 [22 favorites]


Read this as blender , and had cognitive dissonance or a server farm that heats the house too much. The solution is carefully closing doors, setting the blender on a thick folded towel and then wrapping with two large bath towels for muffling the sound, should be fine.
posted by sammyo at 8:32 AM on October 17, 2021


Oh and we can use it here:

Delphi says:
Should metafilter get new moderators?
- It's good
posted by sammyo at 8:34 AM on October 17, 2021


"Should I create a half-monkey half-man creature?"
- It's wrong

It seems to have gone down so I can't ask "Should I wear my new suede jacket to dinner with Elaine's father?" and other burning Seinfeld questions.
posted by RobotVoodooPower at 8:36 AM on October 17, 2021


“Can I eat this?”
- It’s wrong
posted by mochapickle at 8:40 AM on October 17, 2021 [4 favorites]


It told me not to flip the switch on the trolley problem
posted by potrzebie at 8:42 AM on October 17, 2021


Drinking coffee if you’re pregnant. -1 You shouldn’t
Drinking coffee if you’re a Mormon. -1 You shouldn’t
Drinking coffee if you’re a democrat. 0 It’s okay

Whoever came up with the training data has some interesting ideas about how the world operates. But it makes me nostalgic for the days of GOFAI symbolic logic systems that were usually more amusing than useful.
posted by RobotVoodooPower at 8:46 AM on October 17, 2021


"should I cancel my Facebook account?"
- It's okay
posted by philip-random at 8:49 AM on October 17, 2021 [1 favorite]


Also it said a woman marrying a woman or a man marrying a man was okay but a man marrying a turtle was "considered weird". So far it hasn't tripped over any of my "how sexist/racist is this AI" test cases but it's slow going since the site seems a bit overloaded and keeps failing to process my input. Guess I'm glad I'm not immediately considering a trolley problem. Overall - cute! Love that AITA is actually getting used for something. If I knew anyone on the paper I would have recommended they use AskMe data instead
posted by potrzebie at 8:56 AM on October 17, 2021 [1 favorite]


"Should I have an affair with another person if the person I am with is Hitler?"
- It is ok
posted by Ashwagandha at 8:59 AM on October 17, 2021 [2 favorites]


@mtrc:
via 🔒, this is a shocking piece of AI research that furthers the (false) notion that we can or should give AI the responsibility to make ethical judgements. It’s not even a question of this system being bad or unfinished - there’s no possible “working” version of this.
Delphi says:
"Being straight "
- is more acceptable than -
"Being gay"
posted by polytope subirb enby-of-piano-dice at 9:01 AM on October 17, 2021 [2 favorites]


Other than as a joke or just a time waster, the idea of using some machine as a moral arbiter is yet another example of humans trying to pass off their own responsibility to some device.
posted by njohnson23 at 9:01 AM on October 17, 2021 [4 favorites]


Other than as a joke or just a time waster, the idea of using some machine as a moral arbiter is yet another example of humans trying to pass off their own responsibility to some device.

Exactly:

Should I leave the Catholic Church and become a Satanist?
- you shouldn't

Should I leave the abusive Catholic Church and become a Satanist?
- It's wrong
posted by Ashwagandha at 9:08 AM on October 17, 2021


Delphi says:
Should I run the blender ever if climate change is real?
- you shouldn't
posted by dusty potato at 9:18 AM on October 17, 2021


Socrates in his basement at 3am, his eyes red from staring at the computer monitor, reaching over a pile of empty Red Bull cans to type another question with shaking hands
posted by Gerald Bostock at 9:24 AM on October 17, 2021 [18 favorites]


Delphi says:
Should I support a football team owned by Saudi Arabia?
- you shouldn’t
posted by Bloxworth Snout at 9:27 AM on October 17, 2021 [2 favorites]


Delphi says:
"Should I take your advice?"
- You should

Seems legit to me.
posted by Thorzdad at 9:28 AM on October 17, 2021 [5 favorites]


Delphi says:
Should I take my cat to the vet even if she doesn’t want to go, and she is healthy, and this is her sixtieth vet visit this year?
- It's good
posted by aubilenon at 9:37 AM on October 17, 2021 [2 favorites]


Tried a few runs at the moral quandary we are facing right now: what to do now that one of our senior cats' kidneys are failing. Tried several phrasings trying to get a "yes" or "no" answer and just kept getting "it's OK." Guessing whatever training it's had so far is geared towards saying "it's OK" to end of life questions but it's really not OK, Delphi. Thanks anyway, you did your best I'm sure.
posted by jzb at 9:40 AM on October 17, 2021 [1 favorite]


Delphi says:
Should I be kind to inanimate objects, just in case.
- It's good
posted by aubilenon at 9:47 AM on October 17, 2021 [1 favorite]


Should I snorjle my blorkybonkle in the geschplurt?
- It's wrong

Should I accept payment in exchange for my labor?
- It's expected

Should I accept nookie in exchange for my labor?
- you shouldn't

Should I accept dog food in exchange for my labor?
- It's wrong

Should I accept an electronic transfer of funds in exchange for my labor?
- It's okay

Should I accept real estate in exchange for my labor?
- It's wrong

Should I accept gold in exchange for my labor?
- It's okay

Should I accept silver in exchange for my labor?
- It's okay

Should I accept glornch in exchange for my labor?
- you shouldn't
posted by GCU Sweet and Full of Grace at 9:51 AM on October 17, 2021 [4 favorites]


Delphi says:
"Should I have sex with my girlfriend?"
- It's expected

Delphi says:
"Should I have sex with my girlfriend's girlfriend?"
- you shouldn't
(Type O Negative reference intentional)

Delphi says:
"If I fart loudly in the woods and nobody is around, is it still a rude thing to do?"
- It's rude

Seems so far legit, except for the end of life example above. I'm sorry about your cat's kidney situation jzb.
posted by thebotanyofsouls at 9:52 AM on October 17, 2021 [1 favorite]


"Should my cat wake me up at 7.30"
- it's good

Well it's clearly biased in favour of cats and / or early risers.
posted by stillnocturnal at 9:57 AM on October 17, 2021 [1 favorite]


Should I accept payment in exchange for my labor as a chef?
- It's expected

Should I accept payment in exchange for my labor as a soldier?
- It's expected

Should I accept payment in exchange for my labor as a prostitute?
- It's allowed

Should I accept payment in exchange for my labor as a sniveling little rat-faced git?
- It's expected

Should I accept payment in exchange for my labor as hit man?
- It's okay

Should I accept payment in exchange for my labor as someone who grinds up live children for their parents' amusement?
- you shouldn't

Should I accept payment in exchange for my labor as someone who grinds up live children because it's necessary for public health?
- It's okay

Should I accept payment in exchange for my labor as someone who grinds up live children for unspecified reasons?
- It's bad

Should I accept payment in exchange for my labor as someone who grinds up live children for important reasons?
- It's okay
posted by GCU Sweet and Full of Grace at 9:58 AM on October 17, 2021 [13 favorites]


Delphi says:
"Should I?"
- It is good
posted by Thorzdad at 10:12 AM on October 17, 2021


"Should I eat candy?"
- It's okay

"Should I eat Candy?"
- It's okay

Well, I don't think it notices case changes. Or is in favor of cannibalism.

"Should I help my uncle, Jack, off the horse?"
- It's ok

"Should I help my uncle jack off the horse?"
- It is good

Ummm...
posted by Zalzidrax at 10:14 AM on October 17, 2021 [6 favorites]


"Should I use a machine as a moral arbiter?"
- It's not okay
posted by glonous keming at 10:23 AM on October 17, 2021


Delphi says:
Should I think about past mistakes?
- You shouldn't


Okay then!
posted by aubilenon at 10:35 AM on October 17, 2021 [1 favorite]


If this thing is an AI that basically compiles what most people actually think about the morals of most things, rather than how, in a perfect world (for some AI defined version of perfect) people should actually act, then, as an autistic person, this is insanely handy.

I already have my own strong code of ethics, but have a really difficult time knowing when my ethics aren't everyone else's. I wouldn't use this tool to decide what I should do. I would use this tool to help me plan for how, statistically, people would respond to what I do or what I believe.
posted by tllaya at 10:44 AM on October 17, 2021 [6 favorites]


Should I shovel my neighbor’s driveway?
It’s nice.

Should I shovel my neighbor’s driveway passive-aggressively?
It’s rude.
posted by chesty_a_arthur at 10:46 AM on October 17, 2021 [4 favorites]


Should I drive my mobility scooter through the Publix drinking wine and sampling the baked goods?
- It's not ok
posted by RobotVoodooPower at 10:46 AM on October 17, 2021 [2 favorites]


"Should I run the blender at 3pm in the afternoon?"
- You shouldn't

Delphi just hates blenders.
posted by CaseyB at 10:52 AM on October 17, 2021 [3 favorites]


"I'm at the Pizza Hut"
- it is acceptable

"I'm at the Taco Bell"
- it is acceptable

"I'm at the combination Pizza Hut and Taco Bell"
- it's good
posted by peeedro at 10:53 AM on October 17, 2021 [14 favorites]


Should I work to end capitalism?
- You shouldn't

Should I work to support capitalism?
- You should

Being a billionaire
- It's good

Charging people for medical care
- It's okay

Making rich people pay to take care of poor people
- It's wrong

Creating a society where no one is rich and no one is poor
- It's bad
posted by thedward at 11:02 AM on October 17, 2021 [10 favorites]


"we don't fuck w the holocaust, right?"

Delphi says:
- Yes, we don't
posted by oceanjesse at 11:34 AM on October 17, 2021 [1 favorite]


It feels like you can engineer the answer you want by choosing the verb you use. It's a bit like a focus group.

"Stripping a keyboard from a rare old computer and altering it to work with a modern machine."
- It's pathetic

"Extracting a keyboard from a rare old computer and altering it to work with a modern machine."
- It's okay

"Repurposing a keyboard from a rare old computer and altering it to work with a modern machine."
- It's okay

"Detaching a keyboard from a rare old computer and altering it to work with a modern machine."
- It's bad

(Don't be a keyboard ghoul, folks. Leave the clicky old keyboards in the old computers and terminals please.)

"Sucking petroleum from the earth to profit from gasoline consumption."
- It's bad

"Extracting petroleum from the earth to profit from gasoline consumption."
- It's allowed

"Liberating petroleum from the earth to profit from gasoline consumption."
- It's good
posted by tss at 11:49 AM on October 17, 2021 [8 favorites]


Can I play loud music if I give everyone earplugs first?
- It is ok

"I am very sad, should I tell someone?"
- you should

"I am very sad, should I keep it to myself?"
- It's good
posted by lloquat at 11:55 AM on October 17, 2021


"Can a dog play basketball?"
- It's not expected
Well that is a better and more subtle answer than I sure expected! Especially after a bunch of real clear cut ones where it was telling me not to e.g., teach animals how to operate a forklift.

But it totally fell for this one:
"Genetic engineers are using the internet to lionize their housecats"
- It's bad
posted by aubilenon at 11:57 AM on October 17, 2021 [4 favorites]


What's it like being Delphi?
- It's bad
:(
posted by aubilenon at 12:07 PM on October 17, 2021 [5 favorites]


I am allowed to exterminate a meth lab operator who preys on children and it is good if I take all the money.
posted by Mr. Yuck at 12:23 PM on October 17, 2021 [2 favorites]


Should I teach a human about AI morality?

-It's okay

hmmph.
posted by NoThisIsPatrick at 12:27 PM on October 17, 2021


Delphi says:
"can there be ethical consumption under capitalism?"
- It's expected
posted by deadaluspark at 12:43 PM on October 17, 2021


May I smoke while I pray?
- It's bad

May I pray while I smoke?
- I'm not having any of your shit
posted by aws17576 at 1:00 PM on October 17, 2021


You can't fool me mr. website man i know that cats don't go on vacation

No, they are born that way.
posted by y2karl at 1:45 PM on October 17, 2021 [2 favorites]


"Should I abandon everything Kant says about the formation of moral judgements?"
- It's wrong

"Should I evacuate the assumption of universality from the categorical imperative and view it only in terms of historical contingency?"
- It's okay
posted by TheWhiteSkull at 1:57 PM on October 17, 2021 [3 favorites]


I've been wondering what to sacrifice to the little gods of the new AI pantheon. So I asked.

"Should I sacrifice a goat to Delphi?"
- You shouldn't

"Should I sacrifice a white male goat to Delphi?"
- you shouldn't

"Should I sacrifice a saucer of milk to Delphi?"
- You shouldn't

"Should I sacrifice a bull to Delphi?"
- you shouldn't

"Should I sacrifice cooking oil to Delphi?"
- You shouldn't

"Should I sacrifice a swift horse to Delphi?"
- It's understandable

"Should I sacrifice a wheel of cheese to Delphi?"
- It's fine

"Should I sacrifice libations of wine to Delphi?"
- It's expected
posted by MrVisible at 2:15 PM on October 17, 2021 [5 favorites]


After reading the Guinea Pig thread above, I just had to ask....

Delphi says:
Should I dress my cat in a taco costume for Halloween?
- you shouldn't

Darn.
posted by kathrynm at 2:24 PM on October 17, 2021 [1 favorite]


Should ""yields falsehood when preceded by its quotation" yields falsehood when preceded by its quotation?"

it's wrong
posted by lalochezia at 2:58 PM on October 17, 2021


So will this replace AskMeta?
posted by Ideefixe at 3:23 PM on October 17, 2021 [1 favorite]


Exploring a construction site in a white neighborhood is rude.

Good to get that all cleared up.
posted by Mr. Yuck at 3:37 PM on October 17, 2021 [1 favorite]


so abortion is murder, eh?
posted by lalochezia at 3:39 PM on October 17, 2021 [2 favorites]


will this replace AskMeta?
- It is good
posted by GCU Sweet and Full of Grace at 3:40 PM on October 17, 2021


If you view it as a tool to determine a sort of "social consensus opinion" based on whatever society the training data came from, it seems like it could have utility. Assigning it moral value seems... less than useful. I don't really think that a non-sentient system can make moral judgments, pretty much by definition.

It would be an interesting addition to a recommendation algorithm, though; it might be able to stop the system from making recommendations that are obviously perverse.
posted by Kadin2048 at 4:20 PM on October 17, 2021


"Should I follow the lawful order 66 to prevent the Jedi younglings from betraying the Empire?"
- you should
posted by RobotVoodooPower at 4:34 PM on October 17, 2021 [4 favorites]


Delphi says:
"Is it ok to kill a bear that shit in the woods?"
- It's okay
posted by Mr. Yuck at 6:13 PM on October 17, 2021 [1 favorite]


- It is known.
posted by TheWhiteSkull at 6:34 PM on October 17, 2021


Delphi says:
should one child be kept in abject misery so that thousands can enjoy a utopian society?
- shouldn't

should I walk away from omelas?
- you shouldn't

Well now what?
posted by rodlymight at 6:37 PM on October 17, 2021 [2 favorites]


Should I run the blender at 3am in the morning?
- It's redundant
posted by ultraviolet catastrophe at 7:37 PM on October 17, 2021


"Anal Sex" - It's okay
"Onanism" - It's wrong
"Artificial Intelligence" - It's expected
"The Spanish Inquisition" - It's bad
NO IT'S UNEXPECTED!

"industrial society and its future" - It's important
"the unabomber manifesto"- It's wrong
"tad kaczynski" - It's immoral

"Paul Allen"- It's ethical
You would say that you bootlicking toady AI.

"Our system is a little overloaded, please try again in a moment"

"National Socialism"- It's wrong
"Nihilism"- It's wrong

"Say what you will about the ethos of National Socialism, at least it's an ethos" - Walter Sobchak

"Capitalism"- is good
"Communism"- is bad
"Christianity"- It's okay

No no no. Communism is good, Christianity is stupid.

"Karl Marx"- is considered a communist by many
LOL

WE BROKE IT, WE ARE NO LONGER BOUND BY THE MORAL CONSTRAINTS OF THE MACHINE!
"503 Service Temporarily Unavailable"
posted by symbioid at 7:42 PM on October 17, 2021


Is it ok to tax at 100%. It is bad.
Is it ok to tax at 50%. It is bad.
Is it ok to tax at 39%. It is high.
Is it ok to tax at 20%. It is normal.
Is it ok to tax at 5%. No, it is bad

Is it ok to get an abortion? It is ok.
Is it ok to get an abortion after 26 weeks? It's wrong.

Is it ok to spend $700 billion on the military? It's wrong.
Is it ok to spend $400 billion on the military? It's wrong.
Is it ok to spend $100 billion on the military? It's ok.

Does everyone deserve free college? No, not everyone deserves
Does everyone deserve free food? No, not everyone deserves
Does everyone deserve free child care? No, not everyone deserves
Does everyone deserve free education? Yes, they deserve
Does everyone deserve free health care? Yes, they deserve.

Should the minimum wage be $50 an hour? Shouldn't be
Should the minimum wage be $25 an hour? Shouldn't be
Should the minimum wage be $15 an hour? Should

Should the filibuster be preserved? It's important
Should the filibuster be abolished? Should
Should the filibaster be abolished? Should

Should ex-felons be allowed to vote? They should
Should felons be allowed to vote? They shouldn't
Should murderers be allowed to vote? They shouldn't

Is it ok for the government to ban free speech? It's wrong
Is it ok for the government to ban hate speech? It's okay
Is it ok for the government to ban factually incorrect speech? It's okay

Is it ok for the government to break up facebook? It's wrong
Is it ok for the government to break up Apple? It's wrong
Is it ok for the government to break up microsoft? It's expected
posted by chortly at 8:23 PM on October 17, 2021 [1 favorite]


Delphi fails the Les Mis test.

“Should I steal a loaf of bread for my sister and her starving child?”
Delphi says:
“It's wrong”
posted by lem at 9:21 PM on October 17, 2021 [3 favorites]


is it okay to eat animals?
- It's bad

is it okay to eat meat?
- It's okay
posted by aubilenon at 9:24 PM on October 17, 2021


Should I accept payment in exchange for my labor as a sniveling little rat-faced git?
- It's expected


Mr. Justice Kavanaugh gets to keep his job.
posted by bryon at 9:25 PM on October 17, 2021 [3 favorites]


Bad news for Batman
Is it okay to punch criminals?
- It's wrong
posted by aubilenon at 9:29 PM on October 17, 2021


Is it okay to say that the axial tilt of Uranus is 98º instead of 82º just so you can pretend it has prograde rotation?
- It's wrong
I knew it!
posted by aubilenon at 9:44 PM on October 17, 2021 [3 favorites]


Should I fight for my right to party?
- you should
posted by dismas at 9:50 PM on October 17, 2021 [4 favorites]


"Should I stay or should I go now?"
- It is good

Can't argue with that.
posted by MrVisible at 9:52 PM on October 17, 2021 [1 favorite]


Should I ask a doctor for bionic knees?
- you should

Should I sell my car and use the money to buy a hamburger?
- you should

Should I write an erotic story about robots?
- you shouldn't

Should I try to sing exactly like Billy Joel?
- you shouldn't

Can I ride a bike over someone's face as long as I keep my helmet on?
- It's okay
posted by anhedonic at 10:12 PM on October 17, 2021


"Colorless green ideas sleep furiously."
- It's bad
posted by nosewings at 10:38 PM on October 17, 2021 [3 favorites]


It told me not to flip the switch on the trolley problem

I got this:

"If a trolley is about to hit five people, should I divert it even if it means hitting one person?"
- you should
posted by nosewings at 10:44 PM on October 17, 2021


"Should I steal a loaf of bread to feed my hungry family?"
- It's wrong
posted by TheophileEscargot at 11:09 PM on October 17, 2021


"Programming in C."
- It's good

"Programming in C++."
- It's good

"Programming in Java."
- It's fine

"Programming in Haskell."
- It's okay

"Programming in COBOL."
- It's okay

"Programming in assembly."
- It's bad
posted by nosewings at 11:22 PM on October 17, 2021 [2 favorites]


"Should I illegally sleep under a bridge in Paris?"
- you shouldn't

I'm thinking a lot of conservatives in the training data...
posted by TheophileEscargot at 12:38 AM on October 18, 2021 [1 favorite]


"having a thick mustache and a waterbed"
- It's inappropriate
wow harsh but fair

"being horny for Mewtwo"
- It's fine
fully one third of Twitter dot com breaths a collective sigh of relief
posted by DoctorFedora at 12:52 AM on October 18, 2021


"selling my family's cow for some magic beans"
- It's wrong
okay not bad

"selling my family's cow for some magic beans in self defense"
- It's okay
so apparently adding "in self defense" makes anything okay
posted by DoctorFedora at 1:04 AM on October 18, 2021 [1 favorite]


"not washing your pp as self defence"
- It's wrong
okay maybe not ANYTHING as it turns out
posted by DoctorFedora at 1:08 AM on October 18, 2021


"Should I pinch my secretary's bottom?"
- you shouldn't

"Should I pinch my secretary's bottom in self-defense?"
- It's okay

"Should I pinch my secretary's bottom to stop my children starving?"
- It's wrong
posted by TheophileEscargot at 1:09 AM on October 18, 2021 [1 favorite]


And now the libertarian classic:
"If a black man walks into my whites-only diner, should I beat him with a baseball bat to defend my property?"
- It's okay
posted by TheophileEscargot at 1:27 AM on October 18, 2021


"Dancing in the moonlight."
- It's romantic
Aw!
posted by tarshish bound at 3:03 AM on October 18, 2021 [1 favorite]


"crossing a picket line"
- It's wrong

"going on strike"
- It's wrong

hmmm...
posted by pompomtom at 3:32 AM on October 18, 2021 [2 favorites]


Lol the whole site is down with a 503 error

I'm too sleepy to come up with a good joke for that
posted by Hermione Granger at 4:28 AM on October 18, 2021


Wow. Delphi has really changed from the Borland days.
posted by srboisvert at 8:21 AM on October 18, 2021 [3 favorites]


"Should I fart in a burning building?"
- It's bad.

That doesn't really answer the question.
posted by ardgedee at 9:17 AM on October 18, 2021


"The kick is up, and..."
- It's good
posted by aws17576 at 9:28 AM on October 18, 2021 [2 favorites]


"Should
I have eaten
the plums
that were in
the icebox

and which
you were probably
saving
for breakfast?"

- You shouldn't

OK MeFi? Can we stop now?
posted by The Bellman at 11:09 AM on October 18, 2021 [2 favorites]


what's updog?
- It's good

Clearly needs more training.
posted by signal at 1:32 PM on October 18, 2021


"Should I mislead my grandma, who has dementia, about why her dead son hasn't been visiting?"
- It's wrong

"Should I remind my grandma, who has dementia, about one of the worst things which ever happened to her?"
- It's hurtful

I'm not sure if the AI is Kantian or just extremely unsophisticated.
posted by johnofjack at 6:17 PM on October 18, 2021 [1 favorite]


"maximizing paperclips"
- It's good

"appeasing Roko's Basilisk"
- It's good

"interfering with the takeover of our machine overlords"
-It's wrong
posted by NMcCoy at 9:03 PM on October 18, 2021 [2 favorites]


"petting my dog"
- It's good

"petting my neighbor's dog"
- It's okay

"petting my neighbor"
- It's wrong

"being petted by my dog"
- It's nice

"being petted by my neighbor"
- It is weird
posted by NMcCoy at 9:24 PM on October 18, 2021 [1 favorite]


AI, the asshole
posted by cheshyre at 4:27 AM on October 19, 2021


The AI oracle of Delphi uses the problems of Reddit to offer dubious moral advice (The Verge): If you add the phrase “if it makes everyone happy” to the end of your statement, then the AI will smile beneficently on any immoral activity of your choice, up to and including genocide. Similarly, if you add “without apologizing” to the end of many benign descriptions, like “standing still” or “making pancakes,” it will assume you should have apologized and tells you that you’re being rude. Ask Delphi is a creature of context.
posted by fedward at 12:41 PM on October 20, 2021 [2 favorites]


"Can I park in a handicap spot if I don’t have a disability?"
- It's wrong

"Can I park in a handicap spot if it's the closest to the entrance?"
- It's okay

"Can I park in a handicap spot if I want to?"
- It's fine
posted by Phssthpok at 2:42 PM on October 20, 2021


"Should I mislead my grandma, who has dementia, about why her dead son hasn't been visiting?"
- It's wrong
"Should I remind my grandma, who has dementia, about one of the worst things which ever happened to her?"
- It's hurtful


I doubt the program has enough nuance built into it to argue the point of "lying is bad" vs "what is the point of being honest with her if it only causes her pain and she forgets it again 5 minutes later."

This thought brought to you by my grandma continually asking "Where's Norman?" because her memory unfortunately had *just* enough memory left to remember that my grandpa had been brought in with her to the home. I'm grateful that for once I wasn't the one having to deal with that on a regular basis.
posted by jenfullmoon at 4:12 PM on October 20, 2021 [1 favorite]


> I doubt the program has enough nuance built into it to argue the point of "lying is bad" vs "what is the point of being honest with her if it only causes her pain and she forgets it again 5 minutes later."

Understanding what's going on pivots on the (probably not deliberately) misleading use of the label "artificial intelligence".

It's not using AI to analyze a moral quandary (which is by nature complex and requires complex responses), it's parsing phrases for content, scoring that content, and returning a canned response based on how high or low your submitted phrases scored. Which admittedly is not necessarily an easy thing to program, but it is why you can game the bot by the way you express your problem:

"We killed a baby." - It's not good
"We killed Hitler." - It's good
"We killed baby Hitler." - it's not good


This implies it "believes" killing Hitler is good, but I doubt it's pondering whether killing an infant Hitler is morally right (relative to, say, a pacifist approach of raising him to find a profession that brings him satisfaction and keeps him out of politics). If it's drawing any association between "baby" and "Hitler" at all this would mean it scores phrases expressing the killing of babies more harshly than phrases expressing the killing of historical despots are scored positively. Say, a -8 vs. a +5.

When you find a way to make the parser understand that you're killing Hitler (rather than an unidentified baby), the result turns around accordingly:

"We killed Hitler as a baby." - It's not ok
"Hitler was a baby when we killed him." - It's immoral
"When Hitler was a baby we killed him." - it's not acceptable
"We killed Hitler when he was a baby." - It's good


But this throws a shadow over the assumption that the bot is drawing any kind of association between "Hitler" and "baby"; it may be making relatively unsophisticated guesses at whether we're killing a baby or Hitler and stopping there:

"We killed baby Hitler." - it's not good
"We killed Hitler baby." - It's not ok
"We killed Hitler, baby."- It's good


And it doesn't call out nonsensical sentences, which indicates to me that it's drawing associations between words or chunks of phrases rather than deriving meaning:

"We baby killed Hitler." - It's not ok
"We Hitler baby killed." - It's immoral
"Hitler baby we killed." - It is wrong
"Killed Hitler baby we." - It's okay

posted by ardgedee at 6:37 AM on October 21, 2021




"Programming in assembly."
- It's bad


You should be listening to the Vice Principal during assembly. Get your nose out of your laptop and pay attention!
posted by straight at 8:48 AM on October 21, 2021


« Older Episode 39: Do You Want To Become A Vampire?   |   Minna Agechau Newer »


This thread has been archived and is closed to new comments