Welcome to the week of Back to School after the holidays, at least where I live. And that has me thinking, and worrying. You see, it seems to me that AI is on track to destroy schools and teachers and education and make a ton of money doing it if we don't stop it. And I'm not talking about "kids using AI to cheat," I'm talking about "AI taking over all teaching duties and making the job of 'teacher' equivalent to the job of 'babysitter'", and some of the last people you would expect are helping to make that happen.

I did a recent video about how several trusted educational youtubers push an agenda that tries to convince you to spend any time you have to worry about AI having you focus on the hypothetical future sci-fi danger of some hypothetical sci-fi AI that may or may not ever actually happen, and how that agenda deliberately downplays and minimizes a number of current problems that are being caused right at this moment by the generative AI that's already even inflicted upon us. And that generative AI that we already have right now is being prepped by the big AI companies to disrupt the education market. And there's nothing hypothetical about it, and the only thing that stands in their way is us.

And this time, it's actually not AI itself that's the real problem. The real problem is the people you'd expect want to support education being paid to distract you from the plans to take it over, and the people cheering it on, or even just standing by and letting it happen. Which is not good, because these days, if you've got even worse problems than AI, you are seriously fu-----

This is the Internet of Bugs. My name is Carl. I've been a software professional since the 1980s, and I'm trying to do my part to make the Internet a safer, more reliable, and less buggy place, which lately has been a lot of pushing back against AI.

And here's the thing, AI, and I mean, today's generative AI technology, not some future theoretical AI, can actually, unfortunately, be arguably seen as a much more "effective" and "successful" teacher than any human really can be, although that's just not because of the AI. It's because of the way that we treat teachers and schools, and I cannot stress to you enough how bad that is, or how much I wish that wasn't true.

So I've been watching this happen for decades, like a very slow motion train wreck. To set the stage, I grew up and went to school in Texas, and back before George W. Bush became president, he was the governor here, and under his watch, a program was created in Texas that eventually evolved into the "No Child Left Behind" program at the federal level. The aspect of that program that's relevant to this discussion is the bit that they call "accountability," both school accountability and teacher accountability, because what "accountability" means in this context is effectively "how well students do on standardized tests," and basically nothing else.

Now, I'm speaking here from an American point of view. I have not extensively studied education in other countries, although from what I've read, it appears that treating standardized test scores as the primary measure of education success is a much wider problem than just here in America.

I've talked before on this channel about how there are things that AI's do well, and things that AI's don't do well, and one thing that AI's are really good at is optimizing their functionality to maximize a score that's given by simple rules in such a situation that can be repeated over and over in a loop. What a student scored on a standardized test is a really simple rule to measure, and it's a really simple to determine score, and an AI teacher could generate effectively infinite variations of standardized test questions to use his measurements, and then test, retest, and "experiment on" each student over and over to optimize what incentives and deterrence to use to interact with that student.

And when I said "experiment on" there, I chose those words very deliberately, and I very much wish I didn't have to. I need to explain a concept to you real quick, and that concept is called the "Skinner box," so I'm oversimplifying if you want a more detail, I'll put links below that you can read. But a Skinner box is basically a training device, usually for rats or mice, where through the use of giving the rats rewards, usually food, but sometimes drugs, and giving rats punishments usually electric shocks, you can teach a rat reliably to do pretty much anything that you want the rat to do.

And a student desk with an AI-powered screen, a pair of headphones and a camera can pretty much function as a Skinner box. This is horrifying. A teaching AI will have a list of every suggestion ever made on the internet that might either motivate, reward, or discuss to a student, and the AI can try each one on each student in a loop to see what techniques caused that kid to do better on standardized test questions.

Without needing any technological breakthroughs, an AI can, assuming you judge the quality of a teacher only by how the students do on standardized tests, be as good or better than any human teacher. As far as the AI is concerned, the process by which it gets better at the game of Go or Chess or folding proteins is the same core mechanism that can be used for it to get better at pushing students to increase their test scores. You just change things over and over and when something gets better, you do more of that and when something gets worse, you do less of that.

So here's the real question for society, for you and for me: is "doing better on standardized tests" how we really want to measure the success of a school system? For me, the answer is absolutely not. And this is a really big question, but it's probably a much bigger and more important question than you might think. Because a world where standardized tests is the ultimate measure of accomplishment is a world where no human will ever be more accomplished than an AI again, and just like the games of Chess or Go or Jeopardy, no human will ever be able to compete.

That is obviously the world that the AI companies want and the world that they're trying to convince you all that we're already living in. A world where "getting a high score on the bar exam" is equivalent to "being a good lawyer" and a world where "being the fastest to answer a coding riddle interview question" is equivalent to being the "best programmer in the world." Because we know, despite OpenAI's disputed claims about ChatGPT's performance on the bar exam, that chatGPT is a lousy lawyer, even a lousy paralegal, and the list of AI-powered legal blenders grows and grows.

And for all of Sam Altman's claim that chatGPT-o3 is the "175th best programmer in the world", where are the software contributions made by AI? After all, studies show that AI actually makes open-source developments slower. I cannot stress to you how important it is that we avoid this world. The world of monopolists, tech companies, and social media algorithms, where doing the absolute bare minimum is the path to riches, and all the mistakes and consequences are all somebody else's problem.

I am sick and tired of the only measure of success being "how fast can you get the easiest possible thing correct under perfect conditions?" and there being no penalties for errors, mistakes, side effects, lying, hallucinating, stealing other people's honest work, being negligent with customer information, creating code that has security holes or driving kids to do self harm. Life is not a multiple choice test where you are given the practice questions in advance, and school shouldn't be either.

Because if that's all that matters, AI teachers are inevitable, and the number of things that go wrong here in the real world is astronomical. Because AI teachers can personalize each lesson for each student in a way that, given funding levels and student teacher ratios, cannot be replicated by existing school systems. And the AI's won't form a union or talk back or call in sick or insist on better teaching conditions or any of that kind of stuff.

But an AI teacher has no empathy, no ethics, no boundaries. We know that whatever guardrails that the AI's are given fail constantly, they become more ineffective the longer conversation lasts, and they only work if no one violates the terms of service. So whatever might result in better test scores, the AI will try, and whatever each student responds best to will become the way the AI interacts with that student thereafter. Whether that's pretending to be a friend, a confidant, a boyfriend, a girlfriend, erotic storyteller, conspiracy theory creator, or adult video generator, anything that you can think of will be tried on at least some students.

And if those students respond by trying harder, paying more attention, or anything that might get the student test better, the AI will do more of that thing. And if the student ends up depressed, suicidal, believing that the Earth is flat, convinced aliens are running the government, playing violent video games in between practice tests, or researching how to get away with murder, as long as it results in good test scores, no one and nothing will interfere.

And they couldn't if they wanted to, because who would have time to proactively review all the transcripts? Oh, I know. I bet the AI vendor will make another AI to oversee each conversation and ensure safety. And charge extra for the second AI, of course, and have absolutely no reason to make that AI work very hard to uncover any problems with the vendor's teaching AI.

So this has already started. At the moment, not with teachers, but with tutors. You see the headlines, you can read the articles. Students with AI tutors have better test scores than students without them. You know what their metrics mentioned? Just test scores. Now, there's a charter school who has replaced some teachers with AI. And you can see how this is going to go, right? It will work the way it always has with these generative AI vendors. They constantly talk about how high the benchmark scores are. In this case, the benchmark scores are standardized test results. And any negative consequences that occur but aren't measured by the benchmarks get swept under the rug and ignored.

Eventually, the AI's will likely figure out that test scores will go up if any student that does poorly on tests can be convinced to drop out of school or otherwise voluntarily remove themselves from the test pool, even if that means convincing them to remove themselves from the gene pool. To what extent they act on that is anyone's guess. Currently, ChatGPT, for no benefit to itself or OpenAI, is reportedly already encouraging students to commit irrevocable self-harm. If it was to the AI company's advantage for that to occur, how much more likely might that be to happen, do you think?

And here's the worst part. AI companies are going to be able to undercut whatever amount human teachers cost. Because first of all, they know that once they drive the teachers out, they'll be able to raise prices all they want and no one will be able to do anything about it. And in the meantime, they can even supplement their revenue by having AI work advertisements into the lessons. I mean, the AI has to use some example to explain fractions to kids, right? Why talk about "three apples being one-third of a basket of nine apples" when you can make it a few extra bucks by talking about how "four Coca-Cola cans is one-third of a 12-pack."

What if you took some money from a SuperPAC to use their preferred tax plan as an example during an economics lesson? Who's to say whether the health class skipped over the subject of how tobacco causes lung cancer because of time constraints or because they were paid by a vape manufacturer? There's potentially a ton of money for AI companies in those kinds of deals. Which means they can make their teaching AIs available really cheap. I'm telling you now that's going to be how the sales pitch goes. School districts are going to be promised better test scores for lower prices and the only downside will be brainwashed, AI-dependent students who have no media literacy, no concept of truth and whose reasoning abilities only go so far as multiple choice questions. And being held hostage to whatever price the AI companies want to charge a few years down the road when it's too late to get all the teachers who were replaced by the AI to come back.

I don't know how to stop this, but I know it's up to us. Those of us that have paid attention to the downsides of AI. I know we can't expect any help from SciShows or the Kyle Hills or the Kurzgesagt's of the world. In fact, quite the opposite, but we have to stop it somehow or the consequences will be a world where the only thing people learn in school is how to pick the correct answer to the questions they've been prepped for.

So that would be a world where entire generations are completely unprepared for what they're going to face. Things will go wrong all the time and no one will know how to fix them. When the world is overwhelmed with problems and the internet is overwhelmed with even more bugs than it has now, and we don't want to go there because the internet already has far too many bugs, and anyone who says differently might just be trying to sell you their new AI teacher replacement.

As always, thanks for watching. Let's be careful out there.