To be fair, this works with humans, too.
Hence the comment about “bias automation”
AI reflects its training data??? Shocking!
Yes, contradicting the claim that it’s “more objective”.
One web LLM I was screwing around with had Job Interview as a preset. Ok. Played it totally straight the first time and had a totally positive outcome. Thought the interviewer way too agreeable. The next time I said the most inappropriate stuff I could imagine and still the interviewer agreed to come home with me to check out the rock collection I keep under my bed and listen to Captain Beefheart albums.
Listening to some Captain Beefheart, huh… I’ll grab my shiny rocks!
That shit works IRL too. Why do you think therapy practices often have themselves positioned in front of a wall of books? Not that it’s a bad thing; it’s good for outcomes to believe your therapist is competent and well educated.
Reminds me of an early application of AI where scientists were training an AI to tell the difference between a wolf and a dog. It got really good at it in the training data, but it wasn’t working correctly in actual application. So they got the AI to give them a heatmap of which pixels it was using more than any other to determine if a canine is a dog or a wolf and they discovered that the AI wasn’t even looking at the animal, it was looking at the surrounding environment. If there was snow on the ground, it said “wolf”, otherwise it said “dog”.
Early chess engine that used AI, where trained by games of GMs, and the engine would gi out of its way to sacrifice the queen, because when GMs do it, it’s comes with a victory.
Reg, why’d you just stab yourself in the shoulder?
Ah cmon, ain’t ya ever seen a movie?
Well of course I’ve seen a movie, but what the hell are ya doing?
Every time the guy stabs himself in a movie, it’s right before he kicks the piss outta the guy he’s fightin’!
Well that don’t… when that happens, the guys gotta plan Reg, what the hell’s your plan?
I dunno, but I’m gonna find out!
Why would you use AI for chess?
You don’t use it for the rule-set and allowable moves, but to score board positions.
For a chess computer calculating all possible moves until the end of the game is not possible in the given time, because the number of potential moves grows exponentially with each further move. So you need to look at a few, and try to reject bad ones early, so that you only calculate further along promising paths.
So you need to be able to say what is a better board position and what is a worse one. It’s complex to determine - in general - whether a position is better than another. Of course it is, otherwise everyone would just play the “good” positions, and chess would be boring like solved games e.g. Tic-Tac-Toe.
Now to have your chess computer estimate board positions you can construct tons of rules and heuristics with expert knowledge to hopefully assign sensible values to positions. People do this. But you can also hope that there is some machine learnable patterns in the data that you can discover by feeding historical games and the information on who won into an ML model. People do this too. I think both are fair approaches in this instance.
You can calculate all possible moves in milliseconds on any silicone these dsys
All possible moves one step from a given position sure.
But if you then take all possible resulting positions and calculate all moves from there, and then take all possible resulting positions after that second move and calculate all possible third moves from there, and so on, then the possibilities explode so much in number that you can’t calculate them anymore. That’s the exponential part I was refering to.
You can try and estimate them roughly, let’s say you’re somewhere in the middle of the game, there are 12 units of each side still alive. About half are pawns so we take 1.2 possible moves for them, for the others, well let’s say around 8, thats a bit much for horses and the king on average, but probably a bit low for other units. So 6 times 8 and 6 times 1.2, lets call it 55 possibilities. So the first move there are 55 possible positions, for the second you have to consider all of them and their new possibilitues so there are 55 times 55 or 3025, for the third thats 166375, then 9.15 million, 500 million, 27.6 billion, 1.5 trillion etc. That last one was only 7 moves in the future. Most games won’t be finished by then from a given position, so you either need a scoring function or you’re running out of time.
Yep, those are the moves that can all be easily calculated very quickly on modern hardware
It’s not wrong
While I believe that, it’s an issue with the training data, and not the hardest to resolve
Yes, “Bias Automation” is always an issue with the training data, and it’s always harder to resolve than anyone thinks.
So is the example with the dogs/wolves and the example in the OP.
As to how hard to resolve, the dog/wolves one might be quite difficult, but for the example in the OP, it wouldn’t be hard to feed in all images (during training) with randomly chosen backgrounds to remove the model’s ability to draw any conclusions based on background.
However this would probably unearth the next issue. The one where the human graders, who were probably used to create the original training dataset, have their own biases based on race, gender, appearance, etc. This doesn’t even necessarily mean that they were racist/sexist/etc, just that they struggle to detect certain emotions in certain groups of people. The model would then replicate those issues.
Old data adage. Garbage in, garbage out.
Maybe not the hardest, but still challenging. Unknown biases in training data are a challenge in any experimental design. Opaque ML frequently makes them more challenging to discover.
The unknown biases issue has know real solution. In this same example if instead of something simple like snow in the background, it turned out that the photographs of wolves were taken using zoom lenses (since photogs don’t want to get near wild animals) while the dog photos were closeup and the ML was really just training to recognize subtle photographic artifacts caused by the zoom lenses, this would be extremely difficult to detect let alone prove.
Exactly.
The general approach is to use interpretable models where you can understand how the model works and what features it uses to discriminate, but that doesn’t work for all ML approaches (and even when it does our understanding is incomplete.)
That’s funny because if I was trying to tell the difference between a wolf and a dog I would look for ‘is it in the woods?’ and ‘how big is it relative to what’s around it?’.
What about telling the difference between a wolf and grandmother?
Look for a bonnet. Wolves don’t wear bonnets.
I can confirm this. I’m not a wolf expert, or even seen that many wolves really, but I have a dog and I don’t think she’d wear a bonnet.
Yeah, that’s a grandmother, so what?
Hot dog. Not hot dog
I don’t understand why anyone writing, reading or commenting on this think a bookshelf would not change the outcome? Like what do you people think these ml models are, human brains? Are we still not below even the first layer of understanding?
The problem is the hysteria behind it, leading people to confuse good sounding information with good information. At least when people generally produce information they tend to make an effort to get it right. Machine learning is just an uncaring bullshitting machine, that is rewarded on the basis of the ability to fool people (turns out the Turing test was a crappy benchmark for practice-ready AI besides writing poems), and VC money hasn’t reached the “find out” phase of that looming lesson, when we all just get collectively exhausted by how underwhelming the AI fad is.
Anyone have the original link handy? Trying to get to the tweet is uglier than I expected.
I do that shit when I have a web interview. Let up a guitar just visible in the camera, a small bookshelf, a floor lamp, make sure my tennis bag is visible despite not playing in ages…
Whether they realize it or not, people do take this stuff in. Not sure why some algorithm based on these very same interviews wouldn’t do the same.
Tennis bag? Oh, right. America.
America? Maybe Britain?
Maybe. But why tennis bag?
Play tennis
Ya
Journalist doing reports in front of their dildo collection: “hold my beer”
I did the same, but they were not impressed by my Obedience extreme sex bench 5000 with restraint straps. I even told them the sturdy bench is made of durable, heavy-duty steel, capable of supporting up to 400 pounds of weight.
smh.
I’d have hired you. At least I know you’d be honest and not try to hide shit for fear of embarrassment.
And takes well-informed (buying) decisions with a high focus on quality.
I’m amazed that no-one has complained that the graph’s data points are on the borders between categories rather than inside the category bars.
With that out of the way: WTF is wrong with that graph?
It’s not on the border. The specturm line is under each trait. Though it’s absolutely ridiculous that they’re connected instead of being bars.
Why are the different scales connected? How exactly does one interpolate between agreeableness and neuroticism? This is the kind of diagram I used to draw as an 8 year old, and they put this crap in a real product…
It’s incompetent plot by a company not even interested in what they are selling
They shouldn’t be plotted that way technically. The big 5 are independent traits so they should essentially be sliders, not linked like that.
That said, it’s way easier to see the points when you do that. Easy to miss when colors swap, for example, without the lines when you’ve been looking at this stuff for a few hours.
Yeah, it’s interesting that the math pretty much says, that these factors are independent from each other. Then we did even fancier math with “AI”, all to ruin the base understanding by connecting them graphically. It bugs me more than it should. Think about your graphics. It is a very interesting result nevertheless.
The idea of AI automated job interviews sickens me. How little of a fuck do you have to give about applicants that you can’t even be bothered to have even a single person interview them??
I dunno, but if your boss chain contains a machine (literally Amazon warehouse), does it matter?
But god forbid the applicant didn’t spend hours researching every little detail about a company, writing a perfect letter with information that could have just been bullet points and being able to explain exactly why they absolutely love the company and why it’s been their dream to work there since they were a child. Or even worse: Use AI to write the application.
We should build an AI that automates researching about a company for applicants
Cover letters fucking make so hateful. I love generating AI cover letters and sending them. Fuck your cover letters in a market where you need to send 100 applications to get 10 bites
Exactly!
Applicants are expected to dedicated hours of their time to writing their application and performing background research - both of which are becoming increasingly more tedious over time - so the least a company could bloody do is show some basic respect by paying an actual human being to come interview you!
That’s more like an excuse to keep those stupid 5, 6, and even more interview round processes. Basically making you work an entire week for free in exchange of a chance of getting an offer. Make the first or second rounds with AI and only bother after that.
It’s from 2021. Link to the website: https://interaktiv.br.de/ki-bewerbung/en/
Still pretty interesting though.
During the AI goldrush you can make your fortune selling bookshelves.
Selling bookshelf large poster, or just jpgs
Having a bookshelf poster behind you is actually a hilarious workaround.
Bookshelf NFTs? Only possible to buy with crypto?
I fucking hate that extraversion is a measured trait 🙄
I hate that they think bookshelves are an indicator for it
Should’ve gotten better genes from your parents then. Too bad you turned out to be the fastest swimmer. We really missed out on the next Einstein and got… you 🤢
It’s from the OCEAN model of personality, which is currently the most widely accepted model. It’s received less criticism than myers-briggs and astrology.
It’s received less criticism than myers-briggs and astrology.
That’s not a high bar to meat.
Of course it isn’t. Measuring personality is impossible. All personality models are wrong, and they always will be.
is it a high bar to vegetable? i would simply downvote but there is no option
Answering the question in the image: machine learning arose from the industrial control world. The idea was to teach a machine how to detect defects in supposedly identical objects out of a manufacturing line, most often with “machine vision” (ie. a camera). Applying it to humans was asinine.
I know right? I have seen seen vision systems do some impressive things, but they are carefully calibrated to work in a specific way under certain conditions. Some of the ones my company works with get fed CAD in real time so the robot knows what to look for.
I would be interested to see what happens if you lighten up her skin color a bit…
Conventionally attractive white people, stealing all your jobs!
Go full albinism
