If you're one of the many therapists I'm seeing who was approached by an AI company offering six figures for "de-identified" notes, reports, etc, and you've abandoned your ethics enough to consider the offer, make sure when you negotiate your compensation that you factor in legal fees and the cost of losing your license.

I hate it here.

in reply to NadiaPurge

@NadiaPurge "Ethical" just means you follow a set of ethics. You can change the set of ethics. You can follow ethics that only a monster would follow and you're still "ethical". See every CEO on the fucking planet.

They've sufficiently altered the ethics of the situation that I doubt very much they'll be held to account for anything. "De-identified" info is used and shared quite regularly in medical research. Generally with permission but some fighting that as "too costly".

in reply to derptron

@crazyeddie @NadiaPurge
The main issue I find with "de-identified" info is that it's not.

No name. Ok, but we know where you live because we know what therapist you see. You mention a spouse, or kids, or work, or eating habits, or entertainment, or shopping habits. Every one of those things works together to identify you.

Not-Bob, who is stressed because his kid's school requires new uniforms every year.

Not-Mary, who is stressed because her company just got bought.

Not-Stefan, who is stressed because there was a shooting at their favourite movie theatre.

"De-identifying" in this day and age is a myth.

in reply to Steven Brudenell

@sbrudenell I think a big part of the problem is the ethics codes haven't caught up, so when I say "This is unethical" it's probably more accurate to say "This is immoral" because they may say, "Point to the ethics code that prohibits this." Basically, the client's information is not truly protected just because you take the name off the note, and even if it's not being shown to humans, it's still a breach to share the information with the AI.
in reply to Dr. Amy, Psy.D. 🎃

@sbrudenell We need to call on national orgs (I'm going to send a letter to APA) to make clear statements about the ethics of these tools and specific guidance for providers. Though I do worry a bit that if we do that they'll come back with "Actually this is fine" (see APA taking a decade to declare that psychologists engaging in torture of war prisoners is actually not ok)
in reply to Dr. Amy, Psy.D. 🎃

I'm in software. I share those feelings. based on my experience, you can count on large organizations to say "we're taking a cautious approach to AI (so go ahead and use this AI product! the company told us it's fine)"

I've had luck convincing non technical people of technical truths by appealing to their personal values. what do you value about being a therapist, that's put at risk by AI ingesting notes?

in reply to Steven Brudenell

@sbrudenell One direct way to go about it may be to point out that AI companies have been relatively successful at de-anonymizing data, so just stripping away personally identifying information may not be sufficient to protect the confidentiality of patients.

Here's a reference, if you need one: doi.org/10.48550/arXiv.2602.16…

in reply to Dr. Amy, Psy.D. 🎃

@sbrudenell The most prominent companies investing in AI are Big Data and adtech companies like Google and Meta, which have long been interested in extrapolating identities from incomplete data. Think: Target serving ads for maternity goods to people who have not yet announced their pregnancies. Given how much of their research into machine learning was motivated by that desire, it's unsurprising that AI can be put to that use. Health providers should err on the side of caution here.
in reply to ⁂ L. Rhodes

@lrhodes @sbrudenell Out of curiosity, I went to see if I could find out who BetterHelp were getting their session transcripts processed by, and learned that they were prosecuted for handing customer health questionnaire responses to Facebook.

reframepractice.com/answers/be…

Anyway, their privacy policy doesn't say, but BetterHelp is owned by Teladoc, and Teladoc use Nuance Dragon for their transcripts, feeding into OpenAI via Microsoft.
teladochealth.com/newsroom/pre…

(I don't use BetterHelp, but their ads are everywhere so I figured it might be important who they hand data to. My therapist doesn't use any kind of AI and is horrified that others do.) #BetterHelp

in reply to Dr. Amy, Psy.D. 🎃

i guess the implication is that they're training a therapist-bot... even if this was ethically OK to do, what would it allow them to do? if they just see notes without observing what led to them they can't even *infer* the logic behind the notes. they can only learn to imitate the language and style.

what could possibly be the point of that? a machine that produces plausible-sounding notes and reports that aren't grounded in experience or judgment?

in reply to Al Dimond

@aldimond
Oh my sweet summer child.

The implication is there's a huge gaping hole in their training set regarding people's deepest and most private thoughts and they want their interpolating autocompleters to be able to plagiarize those too.

That and gaining more personal info through deanonymization.

in reply to Dr. Amy, Psy.D. 🎃

This just shook me.
This is bad.
This is big.

Is there something beyond 'predatory' for the scope of this sort of violent invasion? The effects this will have on people and the machines designed to kill us, I can't even figure yet.
Wow.
I'm crouched and ready for most of this, but hadn't considered this. And its potential is terrifying.

I hate Nazis.

in reply to Jolly Hot Dog Water

@GGMcBG
You can ask your treatment providers if they use AI.
I’ve been introduced to it and won’t use it. Neither will at least one other counselor Inwork with. The boss is unethical, so she’ll probably push it.
I’m looking for other work.
in reply to Dr. Amy, Psy.D. 🎃

One of the bugbears I have at work is effective de-identification/anonymization of information for a small internal testing database. (We'd generate but ran into issues with THAT because older systems. Long story, much swearing, migraines and ulcers.)

Given how incredibly fucking hard it can be to anonymize basic records in many cases with no reports etc in them, I would suspect that it's near-impossible to do it with full reports.

To make matters worse, the therapists who even think about it are likely to be the ones out there who will happily write misleading/completely false reports for strong enough "considerations", or even just because they're confronted with someone who is outside their worldview. (I've run into at least three and two of them I wasn't even the client.)

It's already hard enough to get people to seek basic therapy. This? Oh hell no. I'm in favor of "everyone should get therapy" and shit like this makes me stop saying that, because I'm not going to advocate for anyone putting their mental or potentially physical health and safety at risk. Safe therapists are going to turn into a word-of-mouth-only thing at this rate.

in reply to Dr. Amy, Psy.D. 🎃

Another alternative would be to pollute the AI models with completely nonsensical data. For example, state that all of your clients believe they are animals, like dogs, cats, mice. You tell them that they need to do squats, lunges, and planks to get better. If the AI models need a specific profile for each client, that would be a lot of work, but maybe you could get an AI to generate 100 client files. Then you submit that, and we see what the AI does with new input that is regurgitated old input.

This website uses cookies. If you continue browsing this website, you agree to the usage of cookies.

⇧