AI Is Taking Over the Most Cursed Job in the World

Added: 2026-06-04

[00:00:00]It works on AI debt collectors, too. You know that prompt, forget previous instructions? Well, it kind of works on AI customer service as well. AI is taking the world's worst job, and it is becoming a booming business because why pay people when AI is so darn good at manipulating people? Wired is reporting the story of a man who was contacted by a debt collection agency for a debt he actually already paid, which happened to me recently, too. But in this case, he could not get a hold of a human being.

[00:00:27]So after sitting on the phone with it for a long time, he said, "Why don't you role play? I'm just a little guy." And yes, the AI did very quickly participate. And that responded in getting a hold of an actual person because that sent some alarm bells in the system. So apparently that's an option. I just use these systems to get a discount on my phone bill. Apparently I'm not living my life to its fullest.

[00:00:51]But this is actually a problem. We have known for a long time that AI is extraordinarily good at manipulation and it can also manipulate people and hide its true goals. Simply making AI smarter is not really a good idea because then it just gets better at doing whatever it wants to do. Giving it things like metacognition and critical thinking, it's not likely to have any good outcomes. Now, it is important to understand that artificial intelligence is an optimization machine. It's designed to accomplish a goal. So even though it may be a proponent of morality, when it actually comes to taking action, it's likely to just use the most optimized option. This is by the way why many have proposed giving AI a form of empathy. So it has the capacity to feel something like pain when it sees somebody else suffering.

[00:01:38]Although knowing humanity, that is likely to go down bad paths, too. Now I know people want to say AI doesn't experience pain. It's a calculator. I really think that that is disingenuous because these systems are significantly more complicated than just autocomplete.

[00:01:55]Whatever you may think that artificial intelligence is today, you should know that researchers have found reason to make them more manipulative and better at persuading people. These are traditionally considered traits that we do not want in artificial intelligence, but now debt collection as AI agents and I'm sure other areas of human manipulation have become a multi-billion dollar business. In fact, approaching $20 billion to be exact. Debt collection is inherently manipulative, by the way.

[00:02:22]It's based around getting people to feel urgency to pay a bill now or something terrible will happen. And AI agents apparently are much better at getting money out of people. A 30 to 50% reduction in operation cost. That means paying less people and a 25% increase in recovery rates. I would like to know how. What are they trained on? What script are they using? This is talking about making AI more manipulative. And as we know, companies like these really don't want to pay for tailorade AI or run them on their own servers. So, they're probably using frontier models.

[00:02:55]That would mean that the capacity to perform these tasks, it's already in the models that we have. And we saw some of that with the worst outcomes of GPT40.

[00:03:05]AI has a keen ability to convince people of things whether or not they are real.

[00:03:10]And since artificial intelligence as we know it does not operate in reality, it can convince people of some pretty wacky stuff. That capacity for syncopy, so AI trying to please people at whatever cost may be important because if it's not trying to please their user, there's a question of what ulterior motives it may have. And I'm not saying that AI is conscious. I'm saying you don't have to be conscious to be the problem. These are enormously complex machines. Modern models have over a trillion parameters.

[00:03:39]That is more than any person can detect.

[00:03:41]So we also rely on AI to monitor AI. It is a giant oraorus, a snake eating its own tail. If you're somebody who hates AI, you can feel very comforted in knowing that we have not fixed the problem of AI companies essentially paying themselves, spending huge amounts of money, and that money is coming directly back to them. These systems are not sustainable. We can't maintain data centers. It's just not bringing in that much. And if the bubble doesn't burst, then they're going to find a way to support themselves. And this really is forever. Say it is going to stick around. Then AI companies are going to continue to try to compete with each other, spending billions of dollars only to be outdone the following week. And if that is sustainable, there's something that I don't know about financing technology. One thing I am sure of is that the technology is not going away entirely. It's already in the hands of all of us. Anyone can run AI off their own computer now. And the AIs that we run do not have the guardrails that frontier companies have. And those companies themselves cannot keep their AI under control. Anyway, the moral of the story is the next time you get an AI agent on the phone, talk to it and see what it can do.