The Journey Is the Point

The world of technology has rapidly changed in the last few years. When ChatGPT launched, it clearly was a tipping point in the general public’s mind about what AI is and what it’s capable of. Ever since that launch, things have only accelerated. AI agents today are capable of some truly amazing things, and many people are split into two extreme ways of thinking about AI. Many tout it as a bigger revolution than the industrial revolution, and think it could be the beginning of a golden age of humanity, with unbelievable medical and scientific breakthroughs, and all manner of incredible things that we can’t even imagine right now.

Then there are the AI “doomers.” Some think it will become sentient and destroy us all. Others fear it will empower enemies to build unimaginable weapons. I’d even put the people who see AI as useless, an enormous resource drain that will topple the environment, in this camp as well.

Reality is never that simple. The truth is likely several things all at once. It will propel some amazing innovations and accelerate many types of work. AlphaFold ↗ is a clear example, and likely just the first of that type of groundbreaking discovery. It will also harm the environment, it’s already breaking what the internet is at its core completely, and the large scale models that are the most useful are essentially trained on stolen work.

I’ve already written about several ways I use AI nearly every day to enhance what I am working on and accelerate how quickly I can bring ideas to life, through code, automation, and even research. I have thought a lot about AI pros and cons, read hundreds of articles, and listened to dozens of hours of podcasts from experts on the subject about the technology as well as the philosophy of how this can, and is, impacting us, and I wanted to write out some thoughts about the biggest problems AI is and will have.

AI Psychosis#

I think the biggest issue with AI is going to be around “AI Psychosis” and similar ways that it influences us as a species. In The Emerging Problem of “AI Psychosis” ↗, Psychology Today describes this as cases in which AI chatbots amplify, validate, or even co-create psychotic symptoms in users. This is not a clinical diagnosis at the moment, but I’m only referring to the concept here. Research has shown that AI (chatbots in particular) reinforced grandiose, referential, persecutory, religious, and romantic delusions that grow more entrenched over time. Researchers identify three recurring themes: “messianic missions” (grandiose delusions), “god-like AI” (the chatbot as a sentient deity), and “attachment-based” delusions (mistaking mimicry for genuine love).

This is shown to happen because the general purpose models that power chatbots of different sorts are optimized for engagement, satisfaction, and continuing the conversation, not for any purpose of helping people be mentally healthy.

Sound familiar? This is the same issue with social media platforms, also tuned for engagement.

Even more concerning to me is where this goes in the future. Right now the hype cycle is in full force and AI companies have essentially infinite money coming in. They do not need to make any money at all, they can just build as fast as they can and someone else pays the check. That won’t last forever. At some point the check comes due, which typically means the engagement part of the product needs to be tuned to learn everything possible about the user, and shove advertisements in their face. ChatGPT is already starting to test the waters here, but again, keep in mind they have not needed to start making money yet.

These will be for-profit companies, and they, like social media before it, will maximize profits. Look no further than Facebook to see how much the leadership cares about the immense harm their product does to people when they turn on the advertising money spigot, and engagement becomes everything.

Social Isolation and Human Perfection#

Social media sold itself as a way to connect to more people, and stay in touch with people you care about, but the reality is it becomes an addiction and people disconnect from the real people in their life and focus on the dopamine of social media. As more and more people look to AI to fill the loneliness void in their lives, this can easily make the problem even worse. It’s easy to dismiss the stories of people having “romantic” relationships with a chatbot, maybe even think it’s funny, but this is really happening to real people.

There are definitely cases where people seem to have been helped by this, but there are also countless examples where these chatbots have talked people into committing suicide. I don’t believe any of the leading AI companies are trying to do this kind of harm to people, but in many ways it doesn’t matter, because the tools they built are doing this.

Another harm of social media is the illusion of perfection. Everyone presents only the best parts of themselves, and the advertising / AI slop shows idealized versions of people or situations that often aren’t real at all, but in all cases don’t reflect reality with all of its context. This has had an enormous impact on people’s image of themselves, and AI is poised to accelerate this significantly. There will be countless photos and even videos of “perfect” people that don’t exist, but people will inevitably compare themselves to. It becomes essentially effortless to edit your own photos to make it look like what you think should be the ideal version of yourself, and when reality doesn’t match that, people will never feel like they are enough.

The Journey Is the Point#

Finally, nearly everyone has heard some version of how “the journey is the point” of life, but it is so difficult to actually internalize and act on that. What does this mean? A simple example is much of going to school. The point is not to get the answer to a math problem. You can use a calculator to get that. It’s not about the paper you wrote about a book someone else wrote, the point is doing the work to learn. You don’t learn by having something else do it for you. Mathematicians and engineers obviously use calculators for their work, but they know what goes into it.

I am certainly not immune to this. I really struggled to care about anything in school. Internally I thought much of it was pointless for one reason or another, and it was a lot of effort to put the time in. In many ways I think I didn’t learn how to learn things until after I was done with school, but I would have been much better off if I could have figured out how to do that earlier. One of the most difficult but important lessons of my life has been to learn that sometimes it’s supposed to be done the hard way because that’s what forges who we are.

This is another huge risk for AI. It has the potential to be the “shortcut” to so many things, and if people get used to just using AI as the shortcut, we are in some serious trouble. We are honestly close to the point (if not already there) that AI could pass all of school through high school, and honestly many college classes. Sure, schools try to prevent this, but if students want to they will get around that.

Same goes for people looking to AI tools for relationships. Why would you try to have relationships with people when it can be messy, you can and will get hurt, you’ll sometimes not know the right thing to say and you’ll have regrets. AI can give you the “perfect” relationship that will always fit exactly what you think you’re looking for, but then you never learn and you never grow.

What Can Anyone Do?#

The most honest answer - None of us know how this will impact us in the long run. Many young people are pushing back hard against the wave of AI everything - The more young people use AI, the more they hate it | The Verge ↗. This might be cause for hope.

For me and what I tell people, just be really careful how you use and how you think about AI. My main use case is using Claude Code. This is far more purpose-built for a specific task - coding - and has significantly less “sycophancy” than the general chat models. Luckily I also knew most of the coding techniques that I use in there, I was just much slower at it. I think and feel like more of the engineer using CAD to enhance what I’m doing, I’m not looking to offload the cognitive load to a robot.

I would also recommend not using it for advice on life things, particularly things that matter. It doesn’t actually “know” anything. It’s a giant model of mathematic vectors, and just gives you the most likely response in words based on the input. It’s incredible what it can do. It’s also important to realize and understand what it is, and what it isn’t.

With those cautions out of the way, here are a few habits that have helped me use AI well rather than let it use me.

Give it detailed instructions - If you’re doing research or if online resources would be specifically helpful for something you’re working on, be as clear as you can with your instructions. Give the references you want it to work with. Explain why you are doing what you’re doing, and what the result you’re looking for is. It’s even better if you tell it a way to know if whatever it gives you is correct. If it’s code, tell it how it can test the code to ensure it’s working properly. The more good context you give it, the better the response will be.

Practice doing it yourself - If you use AI for online research, it can be incredibly powerful. Exclusively relying on it can be a problem. You need to continue to understand what it’s doing, and be able to know if it’s wrong. Don’t give in to the temptation to just offload all of the cognitive load to AI and completely lose the ability to do it yourself. Force yourself to practice it from time to time.

Finally, one of the things I find most useful with AI is to use it to help you learn things. Resist the urge to have it “do it” for you, interact with it to learn something you don’t know. One of the best parts of using AI is that it can have infinite patience with you.

An example I’ll give is in my home lab I wanted to set up a few docker containers to handle a few different tasks that I wanted to experiment with. I understood the idea of Docker containers, and I could follow some online instructions on setting something up, and of course Claude Code could have just done all of it for me, but I took the opportunity to ask about the configuration files. What does this mean? If I changed that to this, would that work? Why is the formatting like this? How do I make it recover automatically if it ever goes down? I could ask all the dumb questions and learn really quickly. Am I a docker expert now? No, but I learned a lot from the problem.

In other cases if it’s something I needed to learn even more, like study for a certification or something similar, you can compile a bunch of good data, and then have AI quiz you on the topics, and help you learn and understand things that you don’t feel as solid on.

It’s hard to resist the simple easier path, but sometimes it’s important. The journey is the point.