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, were trained by games of GMs, and the engine would go 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!
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.
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?’.
While I believe that, it’s an issue with the training data, and not the hardest to resolve
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 no 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.
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.
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??
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
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!
“Bias automation” is kind of an accurate description for how our brains learn things too.
The base assumption is that you can tell anything reliable at all about a person from their body language, speech patterns, or appearance. So many people think they have an intuition for such things but pretty much every study of such things comes to the same conclusion: You can’t.
The reason why it doesn’t work is because the world is full of a diverse set of cultures, genetics, and subtle medical conditions. You may be able to attain something like 60% accuracy for certain personality traits from an interview if the person being interviewed was born and raised in the same type of environment/culture (and is the same sex) as you. Anything else is pretty much a guarantee that you’re going to get it wrong.
That’s why you should only ask interviewees empirical questions that can identify whether or not they have the requisite knowledge to do the job. For example, if you’re hiring an electrical engineer ask them how they would lay out a circuit board. Or if hiring a sales person ask them questions about how they would try to sell your specific product. Or if you’re hiring a union-busting expert person ask them how they sleep at night.
I’ve just started doing practical interviews. I basically get really young people with little overall experience and I just want to know if they can do common technical tasks.
So one question is to literally have them explain how to tighten a bolt. One person failed.
To be fair, that’s a very open ended question. I mean, what kind of bolt are we talking about? A standard lag bolt? If so you don’t tighten it! That’d be a trick question! You tighten the nut. Same thing applies with car wheel bolts. Tricky tricky!
Is it a hex bolt that also has a cross head? How tight are we talking?
I’m just going to assume bolts of lightning and Usain Bolt are off the table.
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.
There’s a ton of great small scale things we can do with machine learning, and even LLM.
Unfortunately, it seems the main usages will be crushing people down even more.
Yup. AI should be used to automate all of the mundane day-to-day BS, leaving us free to practice art, or poetry, or literature, or study, or just do leisure activities. Because all of the mundane BS is automated, so we don’t need to worry about things like income or where our next meal comes from. But instead, we went down the dystopian capitalist timeline, where we’re automating all of the art so artists are forced to get mundane day-to-day BS jobs.
