Notices by Cat Hicks (grimalkina@mastodon.social)

They are frequently surveys written without survey methodology expertise. Especially for professional areas that don't value the behavioral sciences (......medicine, software.......)

If you want to run structured group processes you need to think about information architecture within groups. That means understanding how to surface and consider the weight of divergent opinion (and things that suppress it) not just "consensus." thx for coming to my TED talk

Building frameworks with experts going through rounds of consensus (e.g., Delphi panels), is thought to be a good way to elicit rigorous consensus frameworks.

However, I have participated in multiple Delphi panels, and I am not convinced they are. I think they can veer toward mediocre flattening

Do I have a solution, not really, but having read approximately twenty billion Delphi panel articles in medicine, and been on several in software, I find myself frustrated by the solicitation of experts *downstream of ontology and operationalization decisions*

I understand the underlying theory that structure elicits clearer and better group decision-making, which seems sound at a high level.

But in practice, the "structure" or facilitation of these panels seems to vary widely. It seems quite possible to create false consensus via forced choice.

""What’s wild is when people started building ANNs in the 1940s they copied the binary {0,1} template. But essentially all modern AI has abandoned binary neurons because they just don’t learn as well. AI engineers gave up on the (presumed) biology and now modern artificial neurons use (presumably non-biological) smooth, non-linear activation functions. But our work suggests that spikes were never binary to begin with (which is funny because every neurophysiologist is like: yeah, we know!)"

Look at this cool neuroscience work that my wife is part of!!!!!!!

In the words of one of the authors Brad Voytek: "We show that reducing action potentials to binary {0,1} events throws away meaningful variation in each spike’s waveform. "

https://www.biorxiv.org/content/10.64898/2026.09.15.751814v1

Often I notice there seems to be a pretty big chasm between what folks are discoursing about online re: neuroscience, cogsci and "understanding," and comp models, and AI, and work from our friends and colleagues in those fields. I am reading this paper this morning, which may provide some missing context. I have no desire to actually debate whether you agree with the stances in this but I just notice that this background seems to be missing for a lot of convos

https://rdcu.be/svRh41KcQyta

Anyway I'm having Thai takeout and a snickers ice cream bar. You do not choose what happens to you but you choose what you do with it.

But I guess, you know, so many impossible things have happened in my life it's made me believe that change will always happen. Odds are constructed out of the way things currently are and we can change that.

Another form of conditional probability we have to solve: we cannot make it necessary to be a goddamn PhD rearranging her entire life to get to be someone who gets an answer and an intervention and gets to have that life back.

The length of time of that window? We don't know for certain, but the range looks roughly similar to the length of time THESE things take:

- how long it takes most women with these symptoms to get referred to a specialist in the first place
- around the length of time you might spend battling two denials + appeals of a medication (based on my personal experience)
- the time it takes your favorite local psychologist to pitch, draft and publish a book

and you know, I did it.

There is a window of intervention for what I'm going through, and I got started on treatment during that window.

Past that window, patients' chances of changing the long-term course of their condition ARE. CUT. IN. HALF.

This is ONE example. I feel some I dunno, hesitation saying this because I have that gaslighting bullshit residue in the back of my throat, but we have spent thousands of dollars on my health appointments and you know what, we *needed to* for me to achieve an actual quality standard of care.

My wife is a TENURED BIOLOGY PROFESSOR who TEACHES EVERY PREMED student in one of the best medical systems in the country and that's what we get.

The price to get care is not just money. It is the energy and ability to fight with evidence. Who is capable of being as much of a pitbull as I am about things? It shouldn't take relentlessly going back again and again and again to finally get into a room with a basic diagnostic inventory. And it wasn't a failure of information: *I* knew the diagnostic inventory eight months before someone was finally willing to note two things together in my chart.

Come. the fuck. on.

In order to achieve the maximal efficacy of medical intervention, you absolutely need to solve diagnostic delays. Diagnostic delays are increasingly costly the more efficacious our early interventions are. If you miss the boat, and the boat could take you to an entirely different universe of outcomes, that miss matters tremendously.

I have been going back through the last two years of my experience here and trying to write something out of it. A few large points are resolving

It took EIGHT appointments, over $1000 (that's after insurance), and loads of time and stress.

And I raised the thing we needed to look at in the very first appointment and every subsequent one.

Now, I don't dispute the need to not leap to assumptions. Perhaps there really are many drivers of Symptom Y.

But not even *considering* Condition A? A real failure.

Doctors' appointments are subject to anchoring effects, perhaps worse than any single high-stakes reasoning moment I've ever seen.

I have Symptom X. Symptom X is a very likely predictor of Condition A, so much that Condition A is often presumed diagnosed *only* with direct observation of Symptom X. I observed Symptom X thirteen months ago.

I also have Symptom Y. When we know Condition A, Symptom Y is clearly established to be a direct result of Condition A in many cases. But Symptom Y also results from many other causes.

Nevertheless, the probability of Condition A given BOTH Symptom X and Symptom Y is, well. Different from the probability of Condition A given only Symptom Y.

Want to guess how many separate different doctors appointments it took to get from my raising Symptom X, while having an appointment about Symptom Y, to get to Condition A?

Famously, doctors struggle with base rates and similar cognitive biases around likelihood

(
see: https://www.researchgate.net/publication/247373134_Physicians_neglect_base_rates_and_it_matters

or

https://onlinelibrary.wiley.com/doi/abs/10.1046/j.1365-2753.2001.00284.x
)

Really, we all struggle with base rates

We are trained to think about population level averages. Our minds like to anchor on large known patterns.

But most moments of detective work are about winnowing through conditional probabilities