I Let AI Fix My Dishwasher for 10 Hours. YouTube Did It in 20 Minutes.

My dishwasher stopped working, and I decided to fix it with AI.

I spent 10 hours following directions over two weeks, much of it on the floor with a flashlight and a crick in my neck, looking under the appliance. I spent over $100 on parts.

In the end I went to YouTube, searched “whirlpool lower arm not spraying,” found a 2-minute video, did what it said in about 20 minutes, and the dishwasher works fine now.

So what’s the lesson? Not “don’t use AI.” It’s valuable to me in obvious ways. Not even “don’t use AI to repair appliances,” although it was tempting to reach that conclusion. The useful version is more specific, and it may apply to how you use AI for more important tasks related to your career, business, or life choices. Here’s how I turned a 20-minute, no-cost problem into a 10-hour, $100 one, and what I learned to keep it from happening again.

1. Use the right tool

YouTube surfaces the most common fix first, because view counts rank it. The most common fix is the base rate, the thing that’s usually wrong. Starting there isn’t lazy, it’s how experienced troubleshooters work. A veteran mechanic doesn’t reason through every possible cause, he checks the three things that fail most often. Ignoring base rates in favor of detailed, specific-seeming information is one of the oldest documented errors in human judgment, which researchers Amos Tversky and Daniel Kahneman laid out back in 1974.

AI, asked to help you repair something, tends to hand you the whole diagnostic tree instead of the one branch that’s usually the answer. There’s an old name for reaching for the same powerful tool no matter the job, the law of the instrument. AI is a very good instrument. It’s the right one for the uncommon problem, the novel one, the thing that isn’t already sitting in a dozen videos. It’s the wrong one for a Whirlpool arm that won’t spray, a problem experienced (evidently) by hundreds of thousands of appliance owners.

2. Set a budget before you start

If someone had told me up front that this would cost 10 hours and $100, I’d never have started. I’d have called a repairman or bought a new dishwasher. Or told my kids to do it, although that would have meant telling them the story of how my parents had a dishwasher that broke 40 years ago and they still haven’t replaced it and I spent my teen years washing dishes and so what are you complaining about?!

Deciding what a problem is worth before you begin gives you something to measure against, which matters because of how weak our judgment gets once we’re already moving. The first number in a decision anchors everything after it, and the adjustments we make from that anchor tend to be too small. That’s the other half of Tversky and Kahneman’s work on anchoring. Set no number of your own, and the only anchor left is the money and time you’ve already sunk, which pulls your decision-making the wrong way.

Quibi is the expensive version of getting this right. Jeffrey Katzenberg and Meg Whitman raised $1.75 billion to build a short-form rival to Netflix, launched in April 2020, and when it clearly wasn’t working they shut it down about six months later and returned what was left to investors rather than pouring in more. The backers included Disney, Comcast, and AT&T. Katzenberg reportedly shopped the company to Apple before winding it down. You can argue they should have caught it sooner. What they didn’t do was let ego turn a failed bet into a bottomless money pit.

3. Don’t let ego drive

My ego got involved. First it was “you’re not a real man if you can’t fix this.” Then “this is fun, you’re learning, you’re becoming more competent.” Then “learning to fix things is good for your brain.” All of that might be true. None of it was worth 10 hours.

The reason a budget beats willpower when it comes to restraining your ego is that the pull to keep going gets stronger exactly when you’ve sunk the most, not weaker. Barry Staw’s 1976 study, the one with the memorable title “Knee-Deep in the Big Muddy”, found that people committed the most additional resources to a failing course of action when they were personally responsible for the original decision. The more it was your call, the harder it is to walk away, because walking away means admitting the first call was wrong. Corporate and political history includes a long list of leaders who doubled down on a losing project because backing out was embarrassing. A budget set in advance is how you stay accountable to the more logical part of you.

4. Don’t trust the confidence

This is the one that cost me the most. AI sounds exactly as certain when it’s wrong as when it’s right. Every failed fix arrived with the same calm authority as the one that finally worked (if it had worked), so a dead end didn’t feel like a dead end. It felt like I was one prompt away.

The gap between how sure AI sounds and how right it is has already produced real consequences in business. In 2023 two New York lawyers filed a brief full of court cases that ChatGPT had invented. One of them had even asked the tool whether the cases were real, and it assured him they were. A federal judge sanctioned both lawyers and fined them $5,000. A year later a tribunal held Air Canada liable after its website chatbot confidently described a bereavement refund policy that didn’t exist, and the airline’s argument that the chatbot was responsible for its own words went nowhere. This is called automation bias, and the more confident the output, the harder it is to resist, even when there’s zero objective proof. On the other hand, a video with two million views and a thousand comments saying “this fixed it” is real evidence.

AI won’t reliably tell you to stop, either. It warned me plenty, but the warnings were usually wrong. It told me it was late and I should start fresh in the morning, but it was morning! (it had the time wrong) It told me I’d spent too long when it had no idea whether I’d spent 10 minutes or 10 hours. After enough wrong warnings, I tuned all of them out, including the times it might have been right.

The part that should bother you more than a dishwasher

I only knew I was wasting hours of my time because the dishwasher told me. It sprayed water or it didn’t. Much of what I use AI for gives no such verdict. If I use AI to work on a client pitch, a business plan, or a marketing campaign, I might not know for weeks, months, or years if it worked or not. And by then, I won’t remember enough of the details to know whether it was AI’s fault or something else.

The psychologist Robin Hogarth drew a line between what he called kind and wicked learning environments. In a kind one, feedback is fast, clear, and honest. As a skater, I appreciate (and loathe) the “kind” learning environment of skateboarding. Try a trick, and you find out quickly whether you got it right or need to visit the ER. In a wicked learning environment, feedback is delayed, noisy, or missing, and it can even reward the wrong move.

The dishwasher was the kind case. It could tell me I was wrong, because that sprayer either worked or didn’t. Much of what we ask AI doesn’t speak so clearly to us, and that’s where the 10-hour detours do their real damage, quietly, without ever announcing themselves.

AI can help with many things, is progressing rapidly, and the trend will continue. But the ways we manage ourselves as we use AI will decide how much of a benefit it is, instead of a burden.

Sources

Leave a Reply

Your email address will not be published. Required fields are marked *