Before I talk about parasocial AI psychosis, I want to clarify a few things from last week’s post.
First, in this series, I’m describing the term “AI psychosis” as it is used by the public and not as it is defined in the academic literature. I chose to call genuine, AI-related psychosis “true AI psychosis.” The other two uses of AI psychosis—which I call “prolific AI psychosis” and “parasocial AI psychosis”—are not psychotic states. They are softer disconnections from reality driven by different life experiences.
The second clarification will take up the rest of this post. I stated last week that “prolific AI psychosis occurs when a person generates a large quantity of AI output without significantly increasing the real value of their work. In some cases, the new AI workflow may even destroy value.”
The key word is value. Prolific AI psychosis results in high-output, low-value work. Importantly, people with prolific AI psychosis don’t recognize this mismatch while it’s happening.
Before settling on value, I considered using words like taste, quality, and craft. Each can accompany prolific AI psychosis. Each also has significant shortcomings [1].
Taste
Taste seemed like the obvious answer.
Taste is one popular explanation for why we need humans in the AI era. I think that framing is backwards. People adapt to new technology, but we do not serve machines. Technology is important because it enhances human capability.
With that said, good taste elevates the value of machine-enhanced labor.
Taste is difficult to define. It’s a subjective domain with a rich philosophical tradition. We often know it when we see it, but we can’t always say why we like what we like.
Taste is aesthetic. It speaks in art, beauty, and emotion. Aesthetic assessments always reveal something about the assessor, but they may not expose any deep truth about the work itself.
Taste is experiential. It cannot be adequately summarized in words. Texture, contrast, color, flavor, scent, and vibration are qualities we experience. I can remind you of a sensation, but I cannot describe the color red, the taste of salty licorice, or the pain of heartbreak.
Most importantly, taste is social. It is defined by groups. Individualized taste is one of life’s great pleasures, but good taste is judged by what we share and who we share it with.
In the right hands, AI can enhance a project’s aesthetic. When used without care, AI tends toward tasteless slop.
Slop isn’t inherently bad. In fact, it is often more profitable than brilliance. This has always been true. Mass-market taste isn’t risky. McDonald’s has 40,000x more locations than the fantastic Filipino restaurant near me. The Billboard Hot 100 is more approachable than the virtuoso progressive rockers I love. And Candy Crush Saga has generated many times more revenue than all of the creative indie games I prefer. Delicious, outstanding, and original taste isn’t always the best predictor of success [2].
Since taste emerges in conversation, the creator’s job is to decide whom they wish to speak with. A decision to use AI to speak with the mass market may be rational. (At the same time, a creator shouldn’t be surprised when that aesthetic drives away some of their more discerning fans.)
Quality
Quality was the next most relevant answer.
Quality denotes excellence. It can include function, speed, reliability, and durability.
Quality can be more objective than taste (when comparison is possible).
So quality appeared to be the right marker for prolific AI psychosis.
But quality always involves tradeoffs.
High-quality items are difficult to create. They take more time to design. They demand more testing. They use more expensive materials. They require more labor. They may serve much smaller market segments.
The decision to make (or buy) a higher-quality product is not trivial.
The market reflects this. Lexus may be the higher-quality brand, but its parent company still sells 10x more Toyotas. Profit margin may increase with quality, but demand is often centered on cheaper, lower-quality products [3].
There is, of course, a minimum quality standard. Products that don’t rise above this floor have zero real value. (And many cases of prolific AI psychosis in software development reflect this reality.)
But most products provide some value. We decide whether we need basic-, intermediate-, or high-quality products. Wise use of AI can enhance value across all quality levels.
Thus, the decision to make a lower-quality product is not in itself the marker of prolific AI psychosis.
Craft
I prefer the word “craft” over both taste and quality. To me, it encompasses both ideals.
Well-crafted items are excellent and aesthetic. They work well and they evoke desired emotions.
I love well-crafted work. I’ve witnessed high- and low-craft work from mechanics, scientists, plumbers, sales reps, doctors, support staff, professors, engineers, and administrators.
Lack of craft is common in prolific AI psychosis. But low craft is also the status quo for many jobs. Too many roles are strictly transactional: minimum spec, low pay, throughput over quality, and limited personal development.
In low-craft environments, AI is used to squeeze out more work. In high-craft workplaces, AI is harnessed to create new value. Neither culture is inherently right, but a high-craft company tends to attract more-profitable customers and employees.
Just like our choices in taste and quality, decisions about craft are signals to the people we wish to serve.
Value
I ultimately settled on value.
Value isn’t a perfect concept, and it’s not always easy to define.
The simplified explanation makes sense: you are willing to pay me $X for a product. I am willing to sell you that product for $Y. Value is created for both of us when $X is greater than $Y.
But measuring value becomes messy when you add even a little complexity. What is the value of an untested product? How do you account for new business models? How does value change when copying a product is easier than ever? How do intangible factors influence overall value? When do long-tail effects contribute to success? Which public goods transcend transactions?
Measuring value is especially complicated for creative projects. Very few athletes, bands, artists, writers, and open-source software developers generate significant market value. But these forms of work may still create immense personal value.
If we had a simple way to immediately measure the value of every action, prolific AI psychosis might be rare. But most of our everyday actions are loosely connected to value creation. Generative AI can magnify this challenge.
When I struggle through a task, I develop new skills and refine my sense of taste. If I’m not careful, AI can short-circuit my personal development.
When I hire someone excellent to help me solve a problem, they apply their expertise and gently push back against my naive assumptions. Current AI models don’t always do that. They complete the task for me even when I’m headed in the wrong direction, and they offer unwarranted praise as they do my bidding.
These are not insurmountable problems. But it still takes discipline to use AI well.
AI alone has limited utility. But deliberate, thoughtful implementation of AI can be a major source of value creation. And a commitment to value over output may prevent prolific AI psychosis.
That’s more than enough for today. Next week (if I don’t get sidetracked again) we’ll start to dig into parasocial AI psychosis.
Footnotes:
[1] I’m aware of the meta question: why does any of this matter to a psychiatrist? In short, because I care about what it means to be human. Every technological iteration raises important questions about who we are and who we will become. These questions about aesthetics, quality, decision-making, economic value, and business are core to our humanity. We may use different words for each principle, but they inform our day-to-day experience far more than most clinical language.
[2] I’m a snob in these domains, and an uncultured caveman in others. Unsophisticated products often have objectively good qualities! I will never be let down by a Big Mac. I can sing along with the pop hits from high school. And free-to-play mobile games are occasionally better than nothing (they just aren’t made for me). The marketer Rory Sutherland often frames the value of brands in terms of risk mitigation: you don’t buy a famous brand because it’s great; you buy it because it’s unlikely to be awful.
[3] No shade on Toyota—it’s one of only a few big companies I still cheer for. My current ride is a Camry with a custom speaker system. The equivalent Lexus would sound better on the highway, but I’m not sure it would sound $20,000 better. We all make tradeoffs between price and quality.