Author: sowocki

Closely related to AI and Segmentation (my previous blog post) is AI and Spikiness. No, I’m not talking about AI becoming a resurrected triceratops and sparring with humans, Jurassic-Park-style. I mean peaks and valleys. I mean uneven cognitive profiles. I mean the Autism Fluctuation Nation. (Okay, I just made that up.)

Image source

Image adapted from Public Domain Stock Photo: https://picryl.com/media/anchiceratops-dinosaur-a72266

What I mean is that I am typing these words here, and many of you non-autistics would look at me and go “but your fingers, they move so FAST and you use so very many WORDS,” and maybe you would even be intimidated by this fact! But then you ask me to do… I don’t know, think of something that you assumed is simple because you do it in your daily life without really thinking much. Ask me to do THAT (or worse, stand over me after asking!), and watch as I grimace and either do it very slowly or in what to you looks like an entirely haphazard way. Or I avoid doing it at all, and you struggle not to label me stubborn or lazy.

Yeah. That.

Central to many autistics’ experiences of being spiky (my term of choice, because former-aspiring-goth-kid) is not knowing you are spiky until later in life. Now, this is not universal, and some of us did know in different ways. I certainly grew up knowing I was “good” at some things and… ah, much less good at others. But I had no real language for it or paradigm to fit it into. Consequently, I learned to hide the things I was “not good” at and be ashamed of them, or worse, to fear them. And because I had at least one sibling who was also mega-spiky (RIP)¹ and a parent who was as well, I watched this pathology (yeah, I said the P word) play out in my family, how we related to one another or didn’t, what got labeled Normal and what was A Problem.

That’s unfortunate because in many cases, my “not-good-at” things were not actually that big a deal, and I could have easily asked for help — but asking for help is yet another skill which must be learned. Moreover, it happens in a social context; one must be in an environment where one can do the asking — and remember that many autistics can’t ask using mouth words in a consistent manner and understand replies that also come in mouth words.

So, yeah. I knew I was spiky, and yet I didn’t. And this brings me to our AI friends, specifically Claude. We are all familiar with AI “hallucinations” — made-up little nuggets confidently presented by the AI model in question as fact. Claude has been in the news for this behavior (though it should be emphasized it is far from the only culprit). What’s more, Claude’s “hallucinations” are having real-world consequences; they are playing out in courtrooms where precise and fact-based reasoning is paramount.

I have “hallucinations” in scare quotes here because I do actually not think this is the proper term. If we are going to be talking about AI models as potentially possessing human-like qualities — and I don’t necessarily agree that we ought to, but the cat is more or less out of the bag here — we need to get better at recognizing their shortcomings as similar-but-different to those we experience, as humans, in our human societies.

And this brings us back to autistics, and specifically autistics who had to hide what we didn’t know and couldn’t do because we had been taught that only certain pointy parts of our minds were impressive and useful, and the rest were aberrant and bad. Now, we autistics are different from Anthropic’s lawyer who is trying to cover for Claude in that courtroom, since most of us don’t have deep pockets for one (please don’t come at me about Elon Musk; I really truly do not care). For another, we are human, and many of us actually care quite deeply about truth and the reality of things — more than many non-autistics think we ought to, in fact.

In contrast, I am not convinced that those LLMs “care” about those things in any meaningful sense of the word — it’s the people who build, maintain, and answer for them. The models are acting as they have been trained and are not particularly attached to the “facts” that they invent. They certainly do not accept those “facts” as authoritative in the social sense of we humans who read them and blithely accept them because of the attractive, intelligent-sounding trappings they come wrapped in. Yet at the same time, the LLMs also do not find the facts that they do NOT make up particularly authoritative, in the social sense through which we ascribe and revere authority.

Speaking of such revered authority: We humans have long treated measures like Intelligence Quotient (IQ) as authoritative fact (this has been coming apart in recent years, but only slightly). My own life trajectory has been tied up in this. I was assigned a “high” IQ score in my young, hiding-from-my-spikiness days, and now… well, now I am just not assigned one at all. Which is fine by me, actually! But it leads to some FUNNY wording on the part of the authorities who still insist on their right to do this assigning.

According to these authorities, my spikiness as an adult was so very marked that the mechanism through which my mind was to be assigned intelligence “value” basically short circuited. “Nope,” it went, “this isn’t like the 99.5% of humans I signed up to be an expert on — this person’s brain is a DISCREPANCY, thank you very much, and I don’t do discrepancies.”

Note that the little snippet I have pasted above is far from the only part of my psychoeducational report that used that word “discrepancy.” And, there were also plenty of places on it where I got “lower” scores, like one 3rd percentile and one 13th percentile. Some of those were in sub-tests that I recall struggling to complete, but others were in areas that I thought were entirely routine, if a bit boring or aggravating.

This was the most illuminating part of the whole testing process by far. My brain might have been full of discrepancies, but I now had language to identify them and give them form — plus an official-looking report with official-sounding language to back me up! I have wielded this report selectively in the years since, but it has definitely done me strategic good.

And, for every time that I have cited it as an Authoritative Medical Source in order to explain a need or advocate for an accommodation, I have also laughed at parts of it with my autistic and otherwise neurodivergent friends. We have laughed because it tried to segment me, to break me down, and well, it failed. I am superior over it, not it over me — and I do not mean that I am superior because of any quantitative percentile ranks. No; I am superior because I recognize it as a tool which (in nuanced ways it is true I do not entirely control) answers to me. I am sovereign over it, not the other way around.

This is where we are coming up short with AI. Those who are effectively in charge of Claude’s psychoeducational testing are not forthcoming about its shortcomings, its discrepancies, because they do not (yet) have an incentive to be. They benefit from keeping us in the dark about Claude’s own spikiness, about its peaks and valleys. From making us think that it temporarily goes insane and “hallucinates,” but is otherwise perfectly sane.

Without learning about and accepting my own spikiness, I was facing a future of continuing to deny it, of watching the denial feed off itself and get worse and worse and worse — much worse than whatever issue had set it off to begin with. Where I am now is much better, but I did not necessarily know that when I finally had to face the things that I could not do head-on. I had to be forced there, by a life unraveling that I have written about elsewhere on these interwebs and probably will not go into extensively here. It was dramatic, and I would not necessarily choose to do it all over again. But I did do it, and I am here.

Claude is undergoing some form of pschoeducational testing, I promise you that, as are its brethren. What’s not clear is what prompts that testing (bad press, lawsuits, or the like); who is entitled to the full reports on that testing; what their interests are; and whether they are going to continue to spin it so that we are distracted by “hallucinations” and other dramatic nonsense.

---

1. _{This sibling did in fact die in their mid-twenties, and their death was not unrelated to their unacknowledged, unaccepted, unsupported spikiness when they were alive.} ↩︎

When I watch the rise of AI, I think of autism employment initiatives. You know: very highly-extolled, sometimes even moderately-resourced endeavors from a range of companies and organizations to Hire Autistic People, which mainly existed here stateside in The Before Times. Implicit in this was often the idea that we are great at pattern matching and honest and trustworthy workers — sort of real-life, slightly less annoying (if more desperate) Sheldon Coopers^TM. Rarely were there measures in place to keep us working once hired, let alone enable us to advance; and we were pretty much on our own in staving off autistic burn-out. But, darn it, they wanted to HIRE us. Supposedly. (That’s what the press releases said anyway.)

Now you have AI, and many of those jobs — you know, the coding ones that Sheldon-lite was going to be so good at — are disappearing, or retreating behind a thick layer of AI-enabled screening, discarding, and general opaqueness. Diversity is a dirty word ‘round these parts, and we autistics are part of that whole mess, aren’t we. I mean, we talk about NEUROdiversity and stuff. Guilty as charged!

Okay, enough ranty scene-setting. The point is, those initiatives were about Extracting Value from us. Which, I mean, yeah. Blah blah capitalism. But what I think was unacknowledged was how many actual autistics looked at those and pretty much sized them up as such. If you are the type of autistic who has been coded as (Ought To Be) Fit For Work^TM, you are pretty aware of the type of self-segmentation you have to take part in in order to be palatable to employers, as well as those who purport to be experts at shining you up and getting you ready for them.

And yes, I say segmentation on purpose, and not masking or something similar. I mean consciously breaking yourself up into Useful and Not Useful/Disordered parts. I mean trying to mine your special interests for something that will get you that paycheck. This is different from how non-autistics mine their interests. Special interests are something that keep us going. They are energy. They are life. When you mine them, you obviously disturb that life; for mining is not and never has been life-sustaining (because, again, capitalism, blah blah).

I am mindful (no pun intended… okay who am I kidding) of how certain powers attempted to mine my autistic daughter’s special interests in her early years of life. They wanted to get her to do a thing, and there was the other thing she was “fixated” on. Why not use it as bait? Only they misunderstood entirely why she was after the thing, as well as what she was getting from it. They wanted to harness her energy around the thing, yes. But they went about it in an entirely clumsy, counter-productive way, a way that I as a parent had to put a stop to sooner rather than later — I had to parent her back in our home at the end of the day, after all.

Moreover, she had rights. Rights to a childhood. I might not have much control over what happens when she is an adult, but I can at least not ruin her childhood on the way there.

Gradually I came to see that her rights and mine were intertwined. In order to parent her in a good way, I had to be there for her, not aloof in a cloud of segmentation. I had to keep the parts of me that were fueled by my own special interests alive; I could not extinguish them for the “easy” paycheck. I could not even perform, for the moneyed interests, in the ways I used to.

Now with AI, you have extraction in the extreme. Anything you post, including this here, will at some point be mined for further content. It will be segmented and rearranged, reconstituted in ways you have no control over. I know that type of activity brings to mind creativity from millennia. We have always gotten inspiration from various works that come before us, broken them up and rearranged them to make something new.

What is different here is the obvious, blatant extraction, and consolidation of power from that extraction — and then the press releases extolling it. All of us, all of our thoughts and feelings and ideas, are to be broken down and tokenized — and like autistics reading those autism employment press releases from The Before Times, we are expected to be glad about it. Only we are not. We whisper amongst ourselves, and calculate carefully about what will help us avoid burnout — to survive and even to thrive, in the longer term.

You do not know how similar you are to us autistics when you do this.

I am not an AI doomer. I think there are many useful parts to this technology, to this suite of technologies, and in complex ways, not all of them are necessarily a prescription for doom. But I am not so naive as to take the press releases at face value.

Speaking of faces, I could probably never deliver these remarks in a real world setting WITH my face. I would short circuit, stumble, and not be believed. I am “verbal,” yes, as in I make mouth words and can often understand them, but that autistic “verbal” ability has not yet been brought under neurotypical jurisdiction and control (just ask the numerous people who have struck out trying to get me to make correct mouth words about my feelings over the decades).

There is only a small chance I will be believed here. But, like we autistics would whisper amongst ourselves when one of those new autism employment initiatives would come out, I whisper here. For we are all ripe for segmentation now.

I do not make a moral judgment about those who opt in, or seem to opt in, for that segmentation. But the burnout that inevitably comes after is not something I, personally, can take on again and survive. And so I am on here, rather than on the platform they all say I must be on to be marketable, to be palatable, to succeed.

I do not know what success will look like in this brave new world. But I need to be there for my child, for myself, and for the memory of those who have not survived being segmented. And since this too will be The Before Times someday, I will leave a memory of it that will aid and nurture those who come after me. Whatever that ultimately comes to mean.

For that, too, is not ultimately in my control. But like my work to nurture my child before she becomes her own person in the world, under her own jurisdiction and sovereignty, I mean to give it at least a running start.

I recently completed an MS in IT degree, and have Thoughts^TM regarding AI in the classroom.

In several of my courses, we had to post online regarding various queries and prompts using citations — scholarly, popular, and a mix. We had to respond to peers’ posts using both our own thoughts and analyses and various citations.

It frequently became extremely obvious when fellow students were using ChatGPT to write said posts and beef them up with impressive-looking references. Now, while I used to teach college writing in another lifetime, I was in the student role in this course and therefore understood it was NOT my place to reply to another student’s post with “hey, looks mighty likely that you used an AI friend to write that post there.” I’m not a narc. Not the point, even, really.

But what I WAS tasked with doing was responding to those posts in the context of the references they were using, or claiming to use. So if they posted:

As experts have shown, dogs are much better than cats (Rufus, 2025).

And then the entry for (Rufus, 2025) in their References list went to http://arfarfarf.com/dogsdoresearch.html , I would click on the link to see just who this Rufus expert was and what type of research or pontificating they had done. Except 95% of the time, the link was a dud and there was no such source.

So I would politely post and note that I could not get anywhere with http://arfarfarf.com/dogsdoresearch.html and perhaps they could post an updated link so that I could address their claims about dogs being better than cats. Like… maybe I AGREE that dogs are better than cats and want to back them up. But I need to know who Rufus is and what they were doing, goshdarnit!

(Sometimes I would also try to google to find the academic paper they claimed to be citing, if for instance it looked like Rufus had gotten published in the Journal of Things Dogs Definitely Got Peer Reviewed. But those articles also seldom existed. In one case, I went to the Journal in question and checked out the volume and issue being cited, and there was simply no such article in it. There WERE, however, articles in other journals that had titles that could together compose a veritable mash-up of the title being claimed.)

Anyway, maybe this still counts as being a narc, idk. Or just autistic rigidity. But in the context of the online discussion forum, it was just… like, I don’t know how to do the whole discuss-with-sources thing without knowing what the sources, in fact, were.

For what it’s worth, I’m not “for” or “against” ChatGPT in the research classroom, virtual or otherwise¹. I get that many of us are using it in some form, and I think it’s all in the how we use it, AND how we stand behind that use when questioned.

Because my peers, when I posted those queries? Inevitably either ignored them, or posted to apologize for the “mistake” and promise an updated link… which was never actually forthcoming.

Sorry, Rufus. Arf arf arf. You might have had something great to say. But we will never know.

---

1. _{I do think it’s a lot more dicey when we get to using AI for artistic or creative purposes. And as someone who has created artistic and creative work, I am definitely not cool with my work being used to “train” AI in any way, shape or form.} ↩︎

Welcome to my public site on all things IT! Please use the links at the top to read about me (with or without my face) and to check out some of my recent work. More content coming soon.

-Sowocki