AI Psychosis

"AI psychosis (or chatbot psychosis) is an emerging phenomenon where intensive interaction with AI chatbots triggers or amplifies delusional beliefs, paranoia, or a loss of touch with reality in susceptible individuals.&qout; - Google AI Overview

I agree with all of this, except for the word susceptible.

We've been playing the generative AI game for more than 3 years now (ChatGPT was released in November 2022). The adoption rate has been unparalleled when compared to any other technology, at least in my lifetime.

In recent history, the metaverse failed to gain any traction among the masses, with as little as 4% of the population having participated. VR technology peaked with the game Beat Saber.

Interest in cryptocurrency arguably peaked in the initial wave of adoption in 2017, before it became increasingly taken over by scams. Actual use of the technology outside of speculation is still yet to be realised.

Generative AI has been markedly different. Nearly 100% of the corporate world has adopted AI in at least some capacity. A staggering amount of human thinking is now offloaded to this technology.

Going slow is a feature, not a bug

Let's talk about the upsides. So far, it looks like AI might increase productivity in some areas, sometimes. Ask any real engineer how they use AI and it is a heavily supervised process, with the majority able to cite many occasions where they have gone back and forth for hours on something they could have learned in a similar amount of time.

Ok, let&apso;s say we can now develop features twice as fast, great! That's twice as many features that will never get used. Writing code has never been the hard part. Working out what to build is. AI doesn't know, though it will try and tell you.

Nothing of value was ever built in a day. There is a time and a place for failing fast, and using AI may remove some barriers for that specific niche. For everything else, strong consideration and intentional delivery always wins the longer game.

Self driving cars

Where else in life have we ever accepted 95% accurate is good enough? Despite several advancements since the original Tesla Autopilot in 2014, self driving cars are still not a reality. All cars still require a pilot. We have not applied such caution with AI. The effect of being right most of the time will compound over and over until the results are unrecognisable slop.

You are absolutely right!

The problem with AI is that it is good enough. To the untrained, AI mostly says things that are plausibly true. It is trained to agree with what you tell it. We see very publicly that CEOs everywhere now believe they themselves can ship production grade software.

These are by no means people that you would define as "susceptible", but it turns out that it is human nature to feel good when someone agrees with you. You always need people around you to tell you that you're an idiot when it's warranted.

Ever asked a therapy type question to an AI chatbot? They sure have the answer for you. Ever asked the same question to a therapist? Guaranteed they would not answer without asking a further 10 questions first.

I recently watched on in horror as a Product Manager demoed themselves prompting Claude to connect to multiple systems with customer data just to fill in some input fields, and suggested that this could be used in a production system. Where was Claude to say that there are many ways to skin that cat (like you know, an API), and using me isn't one of them? Even if you insisted that a browser be used, anyone could whip up a playwright script in no time.

Atrophy

I have noticed in myself that heavy use of AI has already affected my ability to think. The more we go down this path the worse it will get. Everyone will be walking around with rotted brains. The mental health effects will pale in comparison to the introduction of smartphones and social media. We just won't know for another 10 years.

There is a minimum amount of time you must spend to truely understand anything. With the use of AI you are either shifting that learning to the review phase (something no engineer ever enjoyed), or you're skipping it out entirely. Neither of these outcomes is desirable.

Conclusion

AI will become just another tool in the white collar tool belt. Certain niches will be created that can offload some of the dead work that no one wants to do. The data is murky, and the path unknown. What we know is what has worked until now, will continue to work in the future. Deep work and proper thinking cannot be replaced by this tool.

Never have we run so hard towards a technology with no clear upside and so many obvious downsides.

It is safest to preserve your brain and sit on the sidelines while it plays out. You won't get left behind, and you might be one of the few thinkers left that can clean up the mess.