When my daughter was being assessed for ASD, the doctor very briefly let her mask slip, but it was in a good way which let me know how fortunate I was to have her assessed when and how I did.
Basically, my daughter, then two-and-a-bit years old, walked up to the laptop (it was a virtual assessment, thanks Covid) and said a sentence, which had, like, a lot of grammar and words and stuff. I mean, to me it was just a thing that my kid did, but Much Words, Very Grammar was the impression I got from the doctor, who was sitting and watching us intently from the other side of the screen.
For a moment, the doctor almost seemed as if she was going to go down the well-trodden, “Can a child who does THIS much grammar and THAT many words ACTUALLY BE AUTISTIC?” path. She asked me my thoughts about the sentence my child had just uttered, and I confirmed it was about one of her current topics of intense focus — “Yeah, um, she really likes socks” — and so, yeah, she was gonna talk about it. Did she want to know anything else about socks? 
The doctor then seemed to compose herself and recited, almost as if to remind herself, “Well, girls can present differently.” It was as if she was recalibrating in the face of recent data indicating that past ASD assessment practices have been pretty gender biased (Cook et al., 2024; Tien et al., 2025), due to the fact that they were developed and normed on samples where girls were either underrepresented or excluded entirely (Gould, 2017). But this doctor had obviously recently taken a training along the lines of “How To Be Less Sexist In Your ASD Assessments” and it was fresh in her brain.
Then the doctor proceeded to do a pretty thorough, non-biased, assessment. I mean, there were still plenty of parts of it that I found annoying, but at least they weren’t judging my child by an unfair sexist standard. And my child got diagnosed and thereby got supports, which was a process that was also aggravating at times but fundamentally door-opening and thus GOOD.
How often do you recalibrate in your day-to-day life? It can be awkward and uncomfortable since it often means letting your mask slip at least a little, which for some is a big deal. Speaking of MDs, I have known some for whom it is a huge deal. Any crack in their perceived authority is a major problem. So I give that doctor credit for what she she did.
It was also neat that I got to see it in real time, and I became quite grateful. I knew how easy it would have been for us to get a doctor who would have brushed us off and set off a years-long battle for appropriate identification and supports, a path I have watched other parents go down only too often. And to be clear, it is not a path that you can go down only if you have a girl — by no means! But you are more likely to go down it in certain ways that correlate to how autistic girls have historically been missed, ignored, and even denied, a la “girls don’t have autism, it’s a boy thing.”
The point is, that doctor had updated her knowledge, as she considered such updating to be part of her professional obligation. And while her outdated, incorrect knowledge still held some sway, especially in her reflexive responses, she was committed to the retraining she had done, the re- and unlearning, the adjusting. And that matters.
I miss that fortunate feeling now, for it meant advancement, progress, and accountability. It meant being serious about equity and righting past wrongs, for the sake of the humans and human lives at stake. I mentioned professional ethics previously — that doctor had some, and she took them seriously in her professional practice.
I think if we ever get AI to work for us, we will need to be similarly conscious about this recalibration process and how it is tracked, verified, and communicated. This has been studied in terms that focus on whether users will continue trusting (and thus using) AI products which make mistakes. As of 2022, when AI voice assistants would mess up, there were identified “good” ways and “bad” ways they could respond: “agents that openly accepted the blame and apologized sincerely for mistakes were thought to be more intelligent, likeable, and effective… than agents that shifted the blame to others” (Mahmood et al, 2022, p. 1).
But note here that we are talking about blame, not redress or restitution in the form of an appreciable, trackable change in future practice. I am grateful my daughter was not assessed according to medical sexism, but I don’t think I would have blamed that doctor if she had applied a sexist standard and thus missed the chance to accurately diagnose her at that point in time. I would have been frustrated, certainly, and that frustration would have come out of a preemptive exhaustion at the thought that I was going to have to go through the whole assessment process again, but on a harder setting and likely at a higher monetary cost to me (Canadian healthcare had come through for us on the first round, and I was out nothing out of pocket). Nonetheless, blame is not the right word; neither is a ritualistic, performative process like sincere apologizing (and what is “sincere” for an AI agent, anyway?)
Mahmood et al. went on to discuss agents potentially offering compensation for mistakes, but restitution and compensation are also two different things; when you want a thing which was previously harmful to be and do better in the light of updated knowledge, you want a transformation and the knowledge that future people in your situation won’t suffer as you have.
These are admittedly higher stakes than an agent which orders you takeout, but things have also progressed on this front since 2022; yet the discourse around how and whether AI agents actually meaningfully recalibrate in light of updated, more equitable, more correct professional bodies of knowledge… kind of hasn’t. And yet we are insistent on outsourcing ever more professional responsibilities to them, with limited (and opaque) lip service to human oversight. In fact, promises about such oversight now are increasingly part of exaggerations and projections about the capabilities of AI systems themselves. It’s a futurism that is devoid of a falsifiable definition of what “human-in-the-loop” actually means; it means whatever it is required to mean (Chang, 2024).
Mahmood et al. noted that part of the struggle with getting people to “forgive” agents for missteps had to do with how “people’s expectations are misled by the futuristic portrayal of AI capabilities” (p. 1). When you have wildly inaccurate futuristic expectations for what the AI agent can do, you are not going to “forgive” as easily when the agent doesn’t meet those inflated expectations.
Sadly this is due to futurist propaganda, not accidentally inflated imaginings and desires on the part of the humans being encouraged or directed to put their trust in these agents. Those expectations themselves need a deliberate recalibration, as deliberate as that doctor was when she stopped herself in her tracks and did that about-face.
But that is not likely to happen, for when we are miscalibrated and uninformed, we are far more lucrative revenue sources, more dependable users, and more pliant subjects.
References
Chang, R. (2024). Human in the Loop!. In Edmonds, D. (Ed.), AI Morality (pp. 222-234.). Oxford University Press. https://doi.org/10.1093/oso/9780198876434.001.0001
Cook, J., Hull, L., & Mandy, W. (2024). Improving diagnostic procedures in autism for girls and women: A narrative review. Neuropsychiatric disease and treatment, 505-514.
Gould, J. (2017). Towards understanding the under-recognition of girls and women on the autism spectrum. Autism, 21(6), 703-705.
Mahmood, A., Fung, J. W., Won, I., & Huang, C. M. (2022, April). Owning mistakes sincerely: Strategies for mitigating AI errors. In Proceedings of the 2022 CHI Conference on Human Factors in Computing Systems (pp. 1-11).
Tien, I., Pearson, A., Sozeri, S., & Seers, K. (2025). “Only Boys Can be Autistic”: A Qualitative Exploration of Gender Stereotype and Socialization on the Diagnostic Journey. Autism in Adulthood.
Leave a Reply