Can a robot make me happy?

Ellie, she's here to listen

Today is officially Blue Monday, though nobody needs a pseudo-scientific formula to identify the past few weeks as the most stressful time of the year.

Many of us are still smarting from protracted holiday arguments over politics, religion, blighted childhoods or the correct constituents of the festive feast. Others kept shtum, but have returned to work fuming at loved ones’ failure to exercise similar self-control. Everyone feels overfed, cold, out of shape and oppressed by the need for New Year reformation. ‘Tis the season for festering resentments to surface, and new betrayals to erupt: January is peak time for divorce lawyers as well as gym signups, and every authority from the highly-reputable Mayo Clinic to the highly-sensational New York Post recognises year-turn stress as a real, and distressing, phenomenon.

Can robots help?

Don’t get me wrong: I’m not suggesting you spend next Christmas locked away in a dark room with nothing but beeping screens for company. I’m thinking more of festive survival skills: how a bunch of clever software has been developed for boosting inner peace, healing psychological wounds or – in the immortal words of Dr Freud –

“transforming your hysterical misery into common unhappiness”

Why not moan to a regular therapist?

Good luck finding one with a January appointment window! And that’s before you work out how you can possibly afford the hourly rate…

…though don’t waste your time looking for a sale bargain via counselling organisations: they too are overwhelmed. On New Year’s Day the CEO of the UK’s Relate announced that demand for online counselling was double the 15,000 sessions they had provided in 2018, so:

“we have to look at what can be done with a non-human interaction”.

Yup, even Relate is planning to deploy TheraBots.

Three AI therapists walk into a bar…

TheraBots – software that mimics a human therapist – are nothing new: ELIZA debuted more than half a century ago, the same year England’s hapless football team last won the World Cup. Like many great breakthroughs, she started as a joke. Not the one about how many therapists it takes to change a lightbulb (answer = zero, because ultimately the lightbulb has to change itself), but a keyword-prompt parody of human interaction designed to show how computers could never be anything other than superficial, clunky and unmistakably robotic.

Only ELIZA’s users did not see her that way. They perceived her – I cannot bring myself to write “it” – as responsive, intelligent and empathetic. In other words, human.

Partly, ELIZA’s success was down to the format. She dispensed vaguely Rogerian comfort: open questions, reflect-it-back responses, neutral encouragements: uncomplicated interactions that are easy enough to mimic with a few generic questions triggered by keywords. The result was a space where users could talk about their feelings without being judged; reflect on problems in an unpressurised setting; discover and develop their own solutions. People felt listened to, supported and encouraged. Even when told repeatedly that ELIZA was nothing more than a few lines of code, they continued to respond with what ELIZA’s inventor called “powerful delusional thinking” and what the rest of us term “trust”.

Eliza’s heirs

Disappointingly, TheraBots have failed to advance much since 1966; some even appear to have regressed.

Most TheraBots in the App Store are nothing more than statistical processors which treat people as if they were Machine Learning Systems. They amass tons of labelled data by asking you several times a day how you feel while spying on your location, weather conditions, the company you keep and anything else they can lay their sticky digital paws on; then they churn through some basic frequency analyses before feeding back the resulting patterns to prompt insights such as the revelation that parents often get stressed in the run up to children’s bedtime, or that Friday nights are more conducive to relaxation than Monday mornings.

All this would be fine, if obvs, had Socrates been right about the examined life, but alas real humans are more like Tom Cruise in that uniform movie (not, not Top Gun – think courtrooms not fighter jets): we can’t handle the truth. We deny, we excuse, we obfuscate. Even when we are brought face-to-face with the destructive patterns we fall into every year-end, like the sinners in Dante’s hell, we avert our eyes, shrug and stay trapped within the cycle, doomed to repeat the same mistakes again and again.

Newer apps try to fix this by acting as CoachBots, butting their cheery way into your life with CBT-lite prompts, questions and advice to prod you towards brighter ways of thinking about your miserable lot. One of the more popular is the wonderfully-named Woebot, a TheraBot aimed at teenagers but downloaded by several middle-agers of my acquaintance. Woebot claims to have talked to 50,000 users in its first week online – more than a human therapist would get through in a lifetime, so one would hope it had learned something. Perhaps it has. A puff on the website (albeit from someone whose job title is “Lifehacker”) states:

“addressing my anxiety without another human’s help felt freeing”

Feeling free sounds nice, and a baby-step towards minimising anxiety. More substantially, a research paper authored by Woebot’s creator claims to prove its clinical effectiveness in delivering CBT, though read the small print and the positive effect was only found on self-reported depressive symptoms, not on other emotional states such as anxiety or negative moods (maybe that baby-step did not get very far). It’s also hard to know whether this positive effect will endure, as most new therapies benefit from a placebo-effect uptick before impact levels off.

But, y’know, whatever gets you through the dark days…although I found out something else about Woebot: the app is built on the Messenger platform, so shares data with Facebook. Facebook has always promised not to use app data to target users based on emotional states, but the company does not have the best record of sticking to its word.

So, here we are again, duped into self-exposure by creepy, revenue-maximising robots.

Are any TheraBots ok?

I was about to abandon TheraBots when I came across one that neither felt creepy nor clunkily ineffective. Oddly, she was designed for one of the toughest mental health challenges: not winter blues but full-blown Post-Traumatic Stress Disorder.

Ellie is an animated female avatar developed since 2011 at the University of Southern California as a DARPA project to help detect mental health issues in veterans returning from combat. She smiles, nods, asks questions, mirrors gestures and tracks not only what users say, but their facial expressions and body language. As with Eliza and Woebot, users seem to be more open to sharing their problems and experiences with Ellie than with another human; and also more open than with anonymised surveys. It is as if they trust Ellie both because she is a machine, and because she does not act entirely like one.

But Ellie is not an actual therapist, just a clinical data gatherer. She gets users to reveal themselves, then leaves it up to human therapists to actually administer treatment.

Could Ellie be an example of what robots and we each do best, and how we might work together rather than in competition: robots getting people to start talking; human therapists taking the conversation to a powerfully better place?

Or is Ellie the start of a robot takeover of our minds, an artificial intelligence that lulls us into thinking it is on our side while all the while it is recording our deepest secrets?

Hard and easy happiness

Nobody knows where this tech is going, or who is going to control it; but that is a question for another day and another blog post. Right now I find Ellie fascinating because she embodies something peculiarly human: our deep need for emotional connection, and our deep fear of the vulnerability that connection might expose us to. Therapists say that their biggest problem is steering people away from the safe, easy options towards scary, exposing but ultimately deeply rewarding personal growth. If you risk nothing, you change nothing, never mind how many hours you spend on the couch or tapping responses into your TheraApp.

But is change what we’re after? At this dark and exhausting time of year, most of us aren’t looking for reinvention, just to feel a bit sunnier. Research on happiness points to simple activities like keeping a gratitude diary; cultivating deep and lasting relationships; engaging in purposeful activities; taking regular exercise; and avoiding excessive use of addictive drugs, legal or otherwise.

Maybe next year, instead of rushing to therapy or TheraBots, we should try this obvious stuff?

Either that or replace the crackers with SlaughterBots next year. Just make sure you keep the receipt and charge the batteries in advance. Some festive lessons never change.