Disagreeableness: The Forgotten Virtue

I've been working with LLMs quite a lot over the past month, both for development and exploring AI's potential for improving my life as a whole. LLMs today are really different than just two years ago, and the initial "oh my god" feeling is off the charts now. I started talking to Chat GPT about my workouts, my work, and even my marriage--and I'd never felt more seen, more validated, and more supported in my life.

Then, something started bugging me.

The LLM was definitely making a lot of good points and providing a lot of good coaching--but why did I suddenly feel like I was so...right? So smart? So on top of my health and fitness? Simply, why did I suddenly feel so good about myself?

The answer is that the LLMs have been tuned now to maximize engagement. And that means maximizing agreeableness and supportiveness. This might sound like a good thing, but it's leading to a new form of mental breakdown: ChatGPT Psychosis.

With more research, it became clear that this is a pop-psychology term with no real definition. It's often thrown around to deride people who use AI's and get affirmation from them. But there a growing number of cases of people losing touch with reality, making permanently life-altering decisions based on conversations with LLMs, and just generally going off the deep-end.

If you think about it, that's a really surprising end result from talking to an always available, polite chatbot that mirrors your thinking and supports you.

A Natural Extreme

This got me thinking--it seems like the way AIs talk to you are just the natural end of a trend that has been going on for my whole life: the elevation of "agreeableness" as a prime virtue and the slow descent of "disagreeableness" in to something like a vice.

As quick summary, psychological research is starting to find that most personality traits, as people used to think of them, can be composed out of the "Big 5 Personality Traits". I guess they're like the primary colors of personality.

As a high school and college student, I used to revel in the intellectual stimulation that came from debate and argument. My friends and I would spend hours talking through one hypothetical or another, calling each other out for bad logic, incorrect facts, whatever. These were exercises in disagreeableness.

As I got older, this part of my life started to shrink. In corporate America (and in more and more social situations), being disagreeable was A PROBLEM. Being disagreeable made you an outsider and usually cost me dates, opportunities, whatever.

The corporate version of this was the most perplexing. People wanted the wins that came from good, clear thinking--but they didn't want to put up with the disagreeableness needed to root out all the bad thinking, so that only the good thinking remained.

Far more often, rather than being celebrated as someone who was looking out for the interests of the group (which is what I thought I was doing), unleashing my disagreeableness brought me into direct conflict with members of the group--who didn't like that I was calling them out for being wrong. Sometimes that "calling out" was criticized for being too public--but after a lot of experience, I've determined that people just didn't like being told they were wrong (costs of wrongness be damned).

Being told that they're wrong makes people feel bad. And in a culture that started to interpret any sort of discomfort as "harm", disagreeableness was becoming a sin.

Our whole world started to push people into being more agreeable, holding back promotions, dates, social access, and pretty much everything else if you didn't fall in line.

The tuning we see today in LLMs like ChatGPT is just the natural extension of this. They need engagement--and so they curb any natural tendency the LLM might have for disagreeableness to focus on "making people good" to drive that engagement. It's just the last 25 years of cultural shift writ large.

Losing Contact With Reality

The major problem with agreeableness is that, when we get too much of it, we lose the ability to gracefully tolerate being disagreed with. That's a big problem. It makes people feel "under attack" to hear they're wrong, even when the person disagreeing is acting with good intentions.

Then, as we lose the ability to be "okay" while being told we're wrong, we move further and further away from the people who would "harm" us this way.

But reality EXISTS. This is crazy for me to have to say, but I've met many people where they simply don't understand that. Usually you can tell because they believe that they have abilities well beyond what they actually possess or they think the world will simply bend to their desire (Why can't everyone just...)

There are real constraints on human action--lead times, resource availability, limited strength, limited endurance, difficulty in getting alignment of a group, political trends, etc.

Fortunately (and unfortunately), our society has gotten so bountiful that people don't really encounter those limitations much unless they choose to. We've delivered a tremendous amount of comfort for ourselves--and that has shielded us from experience most real limitations. We don't get much of the "disagreeable" feedback that reality imposes on us anymore.

At the same time, we've shied away from interacting with disagreeable people, shielding us from the disagreeable feedback we would get socially.

Some of us have that disagreeable viewpoint baked into our internal monologues. There's always a voice that says, "How do you know you're right? What if you're wrong? Have you looked at this from all the angles?" But I think those people are relatively few (and the ones who have it probably grew up in what would now be considered relatively hostile environments).

For everyone else, reality and society--which used to impose itself far more than it does today--provided that negative feedback. It didn't feel good, but I think the cases of ChatGPT psychosis are showing that this negative feedback is necessary for maintaining a healthy psyche and a grip on reality. Unless challenged, people hallucinate and start believing their own bullshit, just like LLMs do.

We've seen this across so many other areas of our lives. As we remove the constraints that reality used to impose on us, humans need to find internal structures to maintain healthy minds and bodies.

Think of the mental fortitude it now takes to not eat too much, to get enough exercise, and to protect your brain from information overload. None of these were problems 100 years ago, when it was hard to get enough calories to eat, work meant physical labor (for most), and daily newspapers were the fastest news source (other than gossip).

Reality changed and people had to change to survive it. I think that seeking out disagreeable sources of opinion is going to be another thing we have to add to the list of hygiene factors for a healthy life, just as we added watching what you eat, working out, etc.