Table saws are deceptive tools. If you don’t know what it looks like, it’s a table, with, well, a saw poking up. Here’s a small one:

Typically you’ll set the guide on the right to the width you want for your board, then you push the board along the table and the saw part off the table saw trims it down to the appropriate width. A lot of table saws have a little sensor now that can detect when something that isn’t wood touches the blade, and the blade drops out of the way, hopefully saving a finger.

They developed this because, for something that looks so easy to use, it is EASY to lose a finger on a table saw. A friend of had to have their finger re-attached after they were in a bit of a rush ripping some wood using a table saw. I was taught to use another piece of scrap wood to push the board through so you never got your fingers close to the blade. The point is, table saws are dangerous.

What’s funny to me is that other saws are much LESS dangerous. Take a hand saw. You know, the type of saw bugs bunny would use to cut a hole in the floor under daffy duck. You typically don’t have to worry about losing a finger to a hand saw. I mean, you can get a nasty cut, but they don’t put hand saws behind glass at Home Depot and require a license to operate one. Why is that?

Well, a hand saw, by it’s nature, is less dangerous. If some damage is going to happen you know it pretty much immediately, and the amount of damage is much more limited. It would be very difficult to cut your finger off with a hand saw — you’d have to really work at it for a few minutes and you’d probably notice that your hand feels weird at some point.

Every piece of technology lives somewhere on this continuum — somewhere between pretty safe to operate, and incredibly dangerous to even be near. On the one side, we have paperclips. On the other side, the demon core.

A handsaw is a little closer to the paperclip side (but still in a dangerous area). A table saw is a little closer to the demon core, but not in amongst all the radioactive stuff.

With physical stuff it’s easy to plot it on this continuum, but for “virtual” technology, or software, it can be much harder. Where does social media land on the continuum? Not even all social media is alike. Where is instagram? Twitter? Tik tok?

And the million dollar question, where is AI?

Is AI harmful?

Yes! I’m not going to beat around the bush here. Like any powerful technology, AI can be extremely harmful. And like much of technology is can be harmful on purpose or inadvertently.

Let’s look at some of the purposeful harms first, and then work our way to the more inadvertent harms. And when I say purposeful what I mean is that the companies know that they’re doing something which harms their users, but they’re doing it anyway because it’s good business. Much like a cigarette company.

AI is addictive

Hey, basically exactly like a cigarette company!

AI usage, like social media, can be addictive — although for very different reasons. Social media is generally designed around the dopamine hit we get from the approval of our peers or seeing something novel.

AI makers, on the other hand, have designed AI that can form an emotional bond with users. Sometimes that’s just a byproduct of a technology designed to be lifelike and relentlessly positive, but sometimes it’s the entire point — like the multiple companies that offer “AI companions” and “AI partners.”

AI companies want you to become attached to your “AI assistant” (or companion, or partner, or whatever) because that’s the easiest way to retain you as a customer!

People become reliant on AI

This is different from addiction. With addiction you feel a compulsion to return to a habit/substance/whatever because of some perceived reward. That’s motivating people with a carrot.

AI can also motivate people with a stick — by making them need it in order to do tasks they typically would’ve done on their own.

Here’s an example — a developer writing in a language they’re not as familiar with. They may turn to AI to help fill in their blindspots. They may find the AI very helpful (coding is one of the things AI tends to do relatively well). If they continue to use it to produce and review code that will become their normal workflow. They will become reliant on AI.

That’s not necessarily a bad thing! I mean, we can also say developers are “reliant” on Google, or IDEs, or the personal computer. AI is just a tool.

But many ads for AI show it being used well beyond professional settings. Ads show people using AI to make a menu, tell their children a story, or write a meaningful, supposedly heartfelt letter. One infamous ad shows a man on a date who uses AI powered smart glasses to basically feed him everything he should say in order to manipulate his date.

While it’s one thing to need AI to perform a job at work, it’s another thing entirely to need AI to be a mediator in all of your personal connections.

Don’t forget the regular AI problems

In addition to these personal risks, AI also has all the other, regular risks we’re familiar with. It makes stuff up sometimes. It can repeat things that aren’t true with confidence. It can be manipulated for horrendous purposes. It can be hacked. It can be used to hack other people. It can perpetuate bias.

I don’t mean to minimize those problems, only to say that they’re well known and (relatively) well understood.

But what do we do about those potentially incredibly dangerous risks? How do you avoid becoming overly reliant on AI? How do you avoid becoming addicted?

Josh’s (preliminary) rules for using AI safely

So, in no particular order, here are some rules that I think allow you to use AI in a safer manner.

1. Use AI on a computer — not on your cell phone
   • This will become harder as time goes on, but if you want to be a self reliant person, keeping AI off your phone as much as you can makes it easier to avoid defaulting to asking it for advice or help
2. Double check AI work
   • Remember, AI is like an assistant that lies 10% of the time. Treat it that way.
   • There’s another important consideration here — you can only double check the work of an AI if you, you know, are capable of it. If the AI is producing work about a subject you know a lot about you can double check it. If it’s producing work with citations you can check them. But if it’s producing results you don’t know much about, and there’s no source, you’ll be incapable of checking the work. That’s a pretty serious risk.
   • This is especially important in programming — if you don’t understand what the AI has written you can’t check for potential flaws. Anything that goes into production should be double-checked by a knowledgeable human being.
3. Don’t use AI for rough drafts
   • I DO use AI to help with my writing (occasionally), but I don’t want it to do the writing for me. Even if it’s just a report at work, I will typically produce my rough draft first, then ask AI for suggestions to improve it. This makes sure that my work follows my thinking, that stuff I know but the AI doesn’t actually makes it in, while also taking advantage of AI to make my own work better in a way where I learn from it, instead of it just giving me the answer.
4. Don’t use AI for interpersonal communication
   • Never, EVER write a prompt that looks like this: “I’ve been dating Samantha for three months and she’s stifling me. Please write a breakup text that is empathetic but firm. Use lots of emojis”*
   • It’s super tempting to have AI do this, but eventually you’ll be in a face-to-face situation and you’ll need to, you know, communicate, and you won’t have time to ask AI. You need all the reps you can get to learn how to communicate well
5. Learn how to prompt well to avoid the “regular AI problems” I mention above.
   • Good prompts help produce better work. I needed to do some research on improving my blood pressure, and instead of saying “Claude, tell me the best way to improve my blood pressure” I said “Can you comb through sources of information about blood pressure that are highly respected, focusing on scholarly articles and Internet information from experts in the medical field, to find the best ways to lower diastolic blood pressure? Report only findings that have been shown effective multiple times in high quality studies accepted in major medical journals.” which isn’t perfect, but it’s much better, and it takes some education and practice.
   • Also, never forget rule number 2, no matter how good your prompt is!
6. Don’t use AI for something you would like to be good at
   • Honestly, it’s just like anything in life. Building a skill takes time. Don’t hand the work off to an AI if you actually want to be good at something — and it pays to be good at writing!

That’s my list so far. I’m still working on it, so I may update it in the future.

Remember, the people building AI don’t have your best interests in mind, just like Zuckerberg didn’t care about the “dumb f#%&s” that gave him all their personal information.

So don’t use it how THEY want you to! Use it how it makes sense for you. Use it so it helps you, but doesn’t cripple you in the process. It’s a tool — sometimes a dangerous tool, but a tool nonetheless.

(and just so we’re on the same page, many of the people who helped create AI put it closer to the demon core than the paperclip)

*Claude provided this gem of a response to the prompt for a breakup text:

Hey Samantha, I’ve been doing a lot of thinking about us lately 💭 I really care about you and have enjoyed getting to know you over these past few months ❤️ However, I’ve realized that we might not be the best match in terms of what we each need in a relationship right now 😔 I feel like I need more space and independence than I’m currently getting, and I don’t think it’s fair to either of us to continue when our needs aren’t aligned 💔 You deserve someone who can give you what you’re looking for, and I need to be honest about where I’m at 🙏 I hope we can both find what we’re looking for, and I wish you all the best 🌟

Notably, it didn’t do what a real friend would tell you to do: break up with Samantha face to face, you coward!