There is a constant caution about spurious comparisons between humans and machines, but we are storytellers and we have to ground discussions somewhere.
Understanding of our own nature, thinking, and conversion of that to taking action is by no means a settled science. We over-standardise to ease decisions in all spheres, because creating fully individualised environments, processes, and products is unrealistic.
That is as true in education as anywhere. In resource constrained environments, like current state education in various countries, that standardisation usually penalises those most vulnerable. Those whose circumstances or abilities do not enable them to conform. The standard enables making the best of educational resources available and exceptions are grit in the machine. In that subgroup are neurodivergent people with all the periodic push back about diagnoses.
"We didn't have all these Aspergers, ADD, ADHD, Autistic, AUDHD kids when I was at school"
"Why should THEY get an accommodation when we are all a little Aspergers, ADD, ADHD, Autistic, AUDHD"
A feature of over standardisation is an explosion in exceptions. One size will not fit all. One physics teacher took my inability to grasp electrical forces as an auditory issue. Instead of explaining better, he re-read the text book louder, and louder, and louder. You can guess my reaction.
In many cases our burgeoning crop of exceptions, people who can't get with the program, form tight knit communities and work to clarify the edges and specifics of their own capabilities and experiences.
Anyway, I digress.
Exceptions need a category to plan remediation
Most neurodivergent people are very conscious there is no cookie cutter diagnosis or precisely modelled set of symptoms. There was no diagnostic criteria for Autism in girls until very recently (recent relative to a lifetime).
There is power in putting a name to a set of things. Things that are more or less similar, with less divergence from the 'norm' in peer populations and organisations we have to navigate. Extreme versions of divergence are easier to spot, mild to moderate divergence can accumulate to crushing pressure if there is no flexibility in an environment that fits poorly.
I spent time in a unit for severely autistic children and adults as a sports coach of mine was also a mental health nursing specialist. He dropped by the unit on the way to a competition, obviously trusting that I could handle that as a mature late teen. It was really tough. I exist to communicate and could not find ways to connect with a lot of those individuals, a few trying incredibly hard to tell me something, verbally or non-verbally.
We would ideally make space for every distinct characteristic of people in this neurodivergent matrix, then they can just be accommodated as part of our kaleidoscopically varied community, but we can't / won't do that, so we need categories of needs defined on some credible basis and bodies that respond to reported issues.
I see loads of parallels in the diagnostic work I do with systems. That's where I excel; unpicking complex systems, finding ways to standardise, but only doing so where there are means in place or built to cater for exceptions.
The alternative is ramming exceptions into some standard category, deferring worrying about them, or deferring them until they are someone else's problem.
That latter part happens a lot. Think LLM / SLM hallucinations or confabulations (pick your term for issues when systems generalise using weighted probabilities for best next addition to a sentence), the subtle and not so subtle shaping of thinking, the huge explosion in malicious use of the technology for non-consensual porn, social media dis / misinformation, phishing, or political influencer content.
I am willing to bet my house that more has been spent to help LLMs make less egregious and consistent errors than on SEND provision across public education. But it's ok, Elon Musk says his proprietary Brain Computer Interface (BCI) will cure Autism.
Yes, there is a constant caution about spurious comparisons between humans and machines, but we are storytellers and we have to ground discussions somewhere.
Complexity as a Market Inconvenience
That weariness with complexity is at the root of a lot of AI industry design decisions and market dynamics. That persistent deferral to let someone else sweat the small stuff. The 'small stuff' that can impact millions of people and accumulate.
Not just a 'La La La, I can't hear you' when people point out limitations and issues, often an ousting of people pointing out the potential for adverse outcomes.
They are right in some ways, it will all come out in the wash. Big issues will surface and things will readjust, but - and this is the crux - at what cost.
The mental gymnastics to justify are very faith-based, just like it was for our ancestors who couldn't understand natural disasters. They offered Gods of choice a sacrifice. A subset of the Silicon Valley community calculate their personal p(DOOM) - the percentage chance of a world-ending issue - then dive back into the race to spread the general purpose technology, with many acknowledged issues, into as many environments as feasible. No wonder they develop odd opinions.
Most of us do not work on things with population-level influence, most do not or cannot think through that level of implications. Those folk designing world models are not different, they are just supported to worry less about it and largely shielded from impact of hiccups and FUBARs.
Verbose Logging: A Neurodivergent Burden and Superpower
Having thoroughly caveated some observed dynamics of applying general characteristics to hugely diverse populations, I'm going to do that to explore a point.
In my experience of neurodivergent peers and loved ones, mainly with ASD as suspected or confirmed diagnoses, they are far from disconnected and shut down. On the contrary, they feel and think about EVERYTHING. The ultimate in verbose logging.
Like an old school Intrusion Detection System (IDS) that sits in a network and captures anything that vaguely looks like an anomaly in traffic, some experience a flood of poorly differentiated noise and have to just give up trying to deal. Like a Distributed Denial of Service (DDOS) attack against a person. Conversations, non-verbal interactions, background noises, physical sensations, smells, tastes, layered pre-existing and new thoughts. It can be utterly overwhelming.
Conversely, when conditions exist to process some of that - perhaps a quiet familiar room, no-one interrupting, familiar smells, food they like, comfortable clothes - small and sometimes big miracles can happen.
Unlike people elsewhere on the continuum of 'typical' who triage for inputs more automatically to keep themselves functioning, less is missed. The inclusion / exclusion choice is more conscious and the cached overwhelming signals have formed neural connections. It can provide a far more nuanced, rich, deep, and complex basis for tackling problems, if that does not necessitate a full shutdown.
Neurodivergent people often crash for a day, a week, or a couple of months to reset - as do any neurotypical people who are driven beyond the point of tolerable pressure. Reaching that point faster and more obviously can be both a huge advantage and disadvantage. The iterations to find and integrate mitigations is a tighter cycle than for neurotypical folk who fit a norm of school or work more comfortably. NT folk can be more prone to view burnout as an aberration or a personal failing.
Neurodivergent people with a support network know their strengths and the amazing heights they can reach when the world accommodates them. Many of their peers don't have those frequent cycles of realisation.
Attacking DEI While Pitching Personalised AI for Education
Blaming others for problems is a strong theme in attacks on DEI as an acronym. You should pull yourself up by your bootstraps, stop blaming prejudice against your single or intersectional identify for all of your problems. Capability is colour, disability, wealth, neurodivergence, and sexual-orientation blind. Any other conclusion is pushing accountability for your success or failure onto others... apparently.
Except when something goes wrong for the cohort pushing that position. If they suffer a misfortune - maybe a downturn in business, poor polling, or an upswell of negative impacts from an action taken - blaming others is the first reaction.
At the very core of AI research is the aim to inject the full diversity of human thought and signals. The people who are most keen to normalise a whole population into more resilient, standard, and compliant economic units, is trying and failing to codify the inputs that produce people they aspire to emulate / replace.
Some of us see all the layered ironies, including the 100x that some AI can provide to neurodivergent people. Taking that wild set of signals as input and using a model to convert it to a generalised and more neurotypical output. These systems are seeing genuine success in refining translation work, but I don't think this version of translation is commonly considered.
The secondary ND advantage is the tendency to question and logically analyse everything. Potentially making this community more immune to taking outputs for granted.
That might explain why a subset of designers, developers, and AI evangelists feel LLMs / SLMs are incredible.
With domain specialism, critical thought, appropriate skepticism, and attention to detail there is both huge potential and huge risk. The overwhelming majority of the population are not being equipped to understand that.
For most it is FOMO or dread inducing rounds of existential risk, job replacement, and super intelligence said to be just around the corner. How do we get from here to understanding fitness for purpose and enabling our people to flourish in tandem?
That won't come out of the Silicon Valley ship MVP and iterate culture with multi-billion lobbying empires to quash closer scrutiny and guardrails.
This is likely to be revisited as it is a head-emptying exercise, not a piece with a neat beginning middle and end. The plan produced by the Trump administration for wholesale removal of brakes for AI implementation has all the hallmarks of a pre-crash bailout.
IMHO A market correction was coming. I don't know what else it could be when big foundation model companies are making billion-dollar consultancy plays to try and find use cases and bespoke layered solutions on the fly because their offerings are not living up to promises.