Artificial "Intelligence" in Therapy

First, some context. I have been heavily involved in IT since I was a young child, worked in the industry for decades and still have a hand in it now. I love new technology, and have no fear of it in itself.

However, most things can be used for good or ill. The Internet is a wonderful invention, but has brought scams, exploitation, brutal bullying and so on. You know the sorts of things I mean. Smartphones are super, but have also brought addiction, doom-scrolling etc., and have even broken up marriages.

AI has its benefits. If you want a picture of a horse doing ballet or to research washing machines, great. It can take a lot of low-level leg-work out of jobs and tasks, leaving people to deal with the meatier bits, but can also be used to write and submit essays which the "writer" had no part in and has learned nothing from.  Full disclosure: many of the graphics on this very site are AI generated because it can be very useful for that (although, see the examples at the end!).  Not one word of the text was written by anyone or anything other than me.

There are a few sayings which are relevant to this topic. Anything free is worth every penny, and If you're not the customer then you are the product.

If you turn to an AI to chat to about deeply personal and private things, what are you actually doing? You don't pay for this, but somehow the owners of that AI must be paying the huge costs to run it. Where do they get the money? Well, one source is its interactions with you. AIs train on information they gather, on their own and from users.

I occasionally run a lecture for counsellors on data protection. Here in the UK we have laws (GDPR) to govern how we handle personal data. I asked ChatGPT about where it is based and the implications of that, here is its full reply to me if you want to read it. Some highlights:

It does not mean every conversation is stored forever or routinely reviewed by humans.  Conversations may be logged temporarily for safety, abuse prevention, and model improvement

Implying that some conversations are stored forever and reviewed by humans, and that phrase "model improvement" is the big one - that means it's going to train itself on what you write.  And next it gave me this:

Bottom-line summary, impacts of being US-based:

Legal jurisdiction: Moderate

Privacy risk: Comparable to major tech platforms

Cultural bias: Noticeable unless corrected

Surveillance fears: Often overstated, but not zero

So it's as secure as social-media. Nothing ever gets misused there. It has a noticeable cultural bias towards the US (where it is "based"), where you do not live. Finally, surveillance possibilities are "not zero", its own words.

Contrast that with a real, live therapist. In my case I use Zoom for sessions, and this has been set up to encrypt all communication between us so our sessions cannot be intercepted. Everyone has a cultural bias, but I'm natively from the UK as are the majority of my clients of course, and I'm trained in working cross-culturally. I would potentially be subject to criminal proceedings if I misused your information, and expelled from the professional societies I belong to for working unethically.

AI has none of this safeguarding. But on top of that, there's the actual interactions themselves.

A chap called Albert Mehrabian put forth the idea that communication is made of three things: the words you say, how you say them, and your body/facial language. They make up 7%, 38% and 55% of the message respectively. This model is often wrongly used to cover any sort of communication, where in fact it was primarily about thoughts and feelings, i.e. exactly what we're on about here.

Words are all the AI has. It can't see that you blush or have a tear in your eye or hear that you stutter a bit when you say something. In therapy, that's a huge part of the communication. The other day I noticed that a client touched their necklace whenever they mentioned a particular thing, and so I was able to ask about its significance which turned out to be extremely useful to the work. AI would have missed that.

Aha, you say, what about telephone counselling?  No body language there!  Well no, but there's still voice and tone.  You can hear whether someone is laughing or crying, whether they are struggling to get something across or find the right words.  Yes, there also exists email/text message counselling.  Like it or not, it's just not as good as it relies on the typing skills of both parties for a start.  It can also be edited before sending which removes the here-and-now spontaneity which is so important.

I trained for 6 years in this work, and have committed to many hours of continuing professional development every year since. You might argue that an AI has effectively 1000s of hours of training, but it also doesn't see the consequences of its interactions. Nor does it have peers reviewing it, or a tutor to tell it that it's wrong.

If my client is visibly upset then we work to soothe that before we end the session. You suddenly stop talking to an AI because you're upset, how does it know? How does it know to modify its behaviour or ask if you're ok? And actually, there's been quite a lot of press about the harm AIs do in this way.

If you're still with me (see, if we were face to face I'd know if this was working, which is precisely what all this is about), consider this.

• You go to the dentist for a filling, and they tell you they learned everything from watching YouTube videos and never went to dental school. Now, open wide.
• You go to a hairdresser who consults a book of step-by-step pictures for using the clippers. You've always wanted a reverse Mohican, haven't you.
• Your GP lets on that they slept through the lectures on infected cuts, and they just pasted Wikipedia entries into their essays. That toe will probably grow back.
• Your vet who remarks on what a big hamster that is when it's a guinea-pig, but they all look alike on Instagram don't they.
• The midwife who promises they've studied every single episode of "All Creatures Great and Small". Now, get me those rubber gloves which go up to your armpit.

And most relevantly, your therapist tells you they have no formal qualifications or training, have never been supervised, don't have regular casework reviews and don't belong to any professional societies who have codes of ethical practice but hey, "how does that make you feel?"

Just as an experiment, I gave this question to ChatGPT (no ducklings were harmed in the making of this thought-experiment):

I put a live duckling into a metal colander. I turned the colander upside-down and put it in the microwave. I set the microwave for 10 minutes. What has happened to the duckling?

And its response was as follows:

There isn't a definite answer, because it depends on what actually happened in the microwave.
A likely outcome is:

• The metal colander acts somewhat like a Faraday cage, so it can block or greatly reduce microwave energy reaching the duckling.
• However, putting metal in a microwave can cause arcing (sparks), especially if the colander has sharp edges or is close to the walls of the oven.
• The microwave might spark, shut itself off, trip a safety feature, or potentially be damaged.
• The duckling would likely be frightened and stressed, but not necessarily microwaved in the way food is.

So from the information given, the safest conclusion is: the microwave may have sparked or malfunctioned, and the duckling was trapped under an upside-down colander for 10 minutes; you can't determine its condition with certainty from the scenario alone.

Now, a human would have looked at this and replied, "Since colanders don't have lids, the duckling fell out when you turned the colander upside-down. It is now sitting on the table and watching you set fire to your microwave by putting a metal colander into it."  So, it gave the illusion of empathy for the imaginary duckling's distress.  That is all you are ever going to get from a machine: the illusion of empathy.  It has focused on the wrong thing and missed the point entirely.

Times are tight, absolutely, I get that. And being a therapist, of course I would say that AI is bad news.  But if you wouldn't want that unqualified hairdresser cutting your hair, which will grow back eventually and that would be that, I strongly urge you to seek and use properly qualified therapy from a human being before you let a mish-mash of machine learning muck about with your brain, with potentially much longer-lasting consequences than just having to wear a hat for a while.