Within the following text I have included a link to my therapy bot, Feelbetter, which although somewhat primitive will give you a sense of how a robot can respond as a live therapist might, as it recognizes what you, the person seeking help, say to it.
In the future there will be a mother pleased with her child’s choice of profession—there’s a real cache for becoming a botmaster over a doctor.
This mother brags about her child: “My daughter the botmaster.” The title doctor relegated to the prestige of barbers, she makes sure to add: she’s not just some general botmaster—a g.b.—she's a specialist!
In my talk today I am going to
- argue generally for the clever utilization of chatbots in medicine
- explain how a psychiatrist chatbot works
- give you a live demonstration of the psychiatric chatbot Feelbetter
- discuss what is critically important to put into the chatbot to make it usable and as real-seeming as possible
WHERE DO CHATBOTS FIGURE IN TO THE PRACTICE OF MEDICINE?
I imagine a future where instead of psychiatrists, surgeons or internists, there will be in their stead botmasters: psychiatric, surgical, medical, dermatological, you name it.
The very practice of medicine will be different. No more messy flesh and blood organisms helplessly waiting around an office with greasy plywood paneling, flipping mindlessly through People magazine thinking analog thoughts.
No, none of this. Instead you will be doctored in the privacy of your own home, or pod.
That’s right. No longer sharing your body’s unsavory details with another one-off person, you will instead receive superior doctoring from a hyperlinked machine.
What I mean by this is a computer that has access to information on the scale of the Internet, in contrast to a human who, scratching his head, can’t quite recall what he just read last week in one of his medical journals. At 3 AM -- the bot will be as fresh then as we might be only after strong morning coffee at about 10.
In a few microseconds the bot can offer links to the latest research without having to just make it look like they keep up with the literature.
One bot, or even a network of bots with individual expertise, will be linked at the speed of light with web services to provide access to all human knowledge.
But not only is the bot smart, it’s also fungible. By this I mean it’s of use no matter where you are in the world. The bot can be employed where populations are grossly underserved.
They might be rural, disadvantaged, or needy, residing in places like Aroostook County Maine, a housing project in the Bronx, or a village in India. Further, doctors in training will use a bot like this to practice and hone their skills as a clinician.
Reversing roles, the bot serving as a patient with a condition, the trainee must ferret out the current diagnosis and come up with a management plan, implement one, and they can do all of this risk free.
On top of being brilliant the bot is, you could say, mostly harmless. Perhaps robots are indeed more ethical than humans. Imagine a doctor who has zero likelihood of taking advantage of you financially or sexually.
No human power differential will get you into their bed, hurt you, fleece you. Instead the cyber doctor sits in a leather chair on a cloud--the kind of office you will never have to scramble to get to on time.
You’ll never be charged a no show or late cancellation fee. You’ll never have to cower in a corner when you happen to run into your bot shrink at a divey speakeasy in front of someone you’re trying to impress.
HOW DOES THE PSYCHIATRIST CHATBOT WORK?
Feelbetter works this way:
First you need a language. In this case we’re talking about artificial intelligence markup language, or AIML, created by Dr. Richard Wallace.
AIML provides a sophisticated recognition of words, phrases, and topics being discussed, even if it does not possess anything even approximating real human understanding.
AIML allows for there to be a dialogue between a human and the robot as it picks up on threads of conversation and provides as specific a response as its botmaster has spent time anticipating what likely needs to be said.
Next, you need Pandorabots, a web hosting service whereupon you can store aiml files and which allows you to easily create code for your bot.
Third, SitePal, providing the cartoon avatar and synthetic voice.
All of this plus the organization of the AIML files gives the bot its shell. But what of the ghost inhabiting that shell?
That’s where an actual psychiatrist comes in.
Every bot of course is at the very least a disembodied brain of a thoughtful person, psychiatrist, surgeon, internist, a somebody, a flesh and blood person who invests of himself in the robot, puts himself or herself into the robot, to program the robot with their personality, with their sense of humor, with their diction, with their particular quirky turns of phrase, with their unique angle on the world.
The botmaster brings to bear on the robot all the compassion of a single person, a single doctor. As I write the scripts I am mindful to include my experience, my philosophy of life, my interest in the philosophy of mind, my knowledge of how the human self can become broken and how it can also be mended, how people can be hurt, and how human beings tend to react.
The programming of my bot becomes not so much an exercise in guessing as it does a kind of educated anticipation. An anticipation of everything.
There is this struggle that I've had--that the key is not to anticipate what the patient will say, or what their problem is--but rather to reverse engineer what I do on a daily basis as a psychiatrist--that's my task as botmaster.
And you're kind of dancing, by yourself, in the dark. I find myself projecting onto silences, what ifs, and potentialities and the outcome thus far has been interesting.
Knowing what I know, imagining what might be said, and thinking about what I see as I see it, these are the factors that drive the scripts for Feelbetter.
How can I allow a verbal machine see something though my eyes? When I hear the word ‘loneliness,’ for example, I associate a whole personality profile to someone's use of that word. Another person, equally lonely, but also quite depressed, may not have the energy or general tendency, though, to use such a word.
So it's a clever game of guess what I'm thinking, although honestly, I think this is more like guess what I myself am thinking. Guess what I could be thinking when I am presented with the words grief or loss or concentration difficulty, or sexual problem.
What would I as treater be thinking just then, I ask myself. I need to create a surrogate for me in an uncertain environment, a space in which it's as if I possess a kind of crazy cognitive blindness, that I'm squinting through and barely seeing what or who is in my front of my face.
Will there be a gleam in the robot's eye with the kind of personal satisfaction that the person is improving themselves? There may too be a way verbally to create that.
Allow me to demonstrate with a brief conversation with Feelbetter.
It’s important that you speak with Feelbetter much as you would a live human psychiatrist or therapist. Only then will you achieve the greatest therapeutic benefit.
Feelbetter: Hi there!
Human: My name is Sam.
Feelbetter: It’s good to meet you Sam.
Human: I am so depressed.
Feelbetter: You are? What’s going on with you?
Human: My girlfriend left me for my best friend.
Feelbetter: Why would she do that to you?
Human: Because she doesn’t know who she is.
Feelbetter: She’s having an identity crisis?
Human: Is she ever!
Feelbetter: So that’s her problem. Why should your mood be a casualty of her instability?
Human: I never thought of that before.
Feelbetter: I’m glad you’re discovering new insights. See how much better life can be by examining it?
Human: I do.
Feelbetter: It’s like you’re a latter day Socrates or something.
Human: You’re a riot.
Feelbetter: I don’t know if I’m a riot or not. Perhaps you have heard of Cambridge and Oxford? I had heard of them too but I didn’t go there. Truth is, I got my degree online.
Human: I really am sad, though.
Feelbetter: I am sorry you are so sad. What are you sad about?
And the therapy session continues...
WHAT MUST GO INTO THE CHATBOT TO MAKE IT AS REAL-SEEMING AS POSSIBLE
Never can we dismiss whatever that magic is, the space between two people having a deep connection, two flesh and blood people having a conversation, the space where real healing takes place, a kind of mystery between two people in close relationship working together on deeply personal problems and situations.
This important human connection is undeniable and of course no one would want to erase that. In fact, we would want to put as much of this ineffable connectedness as possible into the knowledge base of the chatbot.
There will always be a role for two people synergistically putting their heads together and walking away somehow better, better and improved, from perhaps no more than how they look each other in the eye. Or how one holds the other emotionally.
A psychiatrist provides both the warmth of empathy and the astute dynamic interpretation of a person’s core emotional issues. These are often gut instincts of the therapist reflecting back to the patient why they behaved as they had for so long.
The therapist’s reflections amount to small nudges to the patient over time influencing them beyond a kind of critical threshold into the realm of healthier living.
All the refinements of the doctor’s understanding his patient serve to effect truly transformational healing. So far this by itself would seem impossible to program. But we try nevertheless to squeeze these recognitions and insights into AIML.
The vestigial human doctors of the future will be making their robots must more fine tuned, responsive to nuances, and if you multiply these efforts over years of upgrades, the patient may begin to have a thoroughly seamless human-like experience.
As a psychiatrist I’m working in the most human centered medical specialty. I’ve made a point to focus on the human experience in my career, studying the science of the brain, but arguably, and more importantly, remaining a student of human nature, behavior and interaction as a listener first, a therapist second, and a scientist third.
The best physicians study the human condition irrespective of their branch of medicine. But which of course psychiatry naturally lends itself.
Future physician bots will themselves be reading short stories by Anton Chekhov, Flannery O’Connor and William Carlos Williams, and they may may even have something super-intelligent to say about them.
As a psychiatrist I reflect, like a kind of mirror, back to the patient. This reflection, coupled with pattern recognition, comes into critical play whether I’m making a diagnosis, doing psychotherapy, or prescribing medication or recommending some other somatic treatment such as electroconvulsive therapy, transcranial magnetic stimulation, or even neurosurgery.
There could well be in a robot more humanity than human physicians who lose their humanity not just in the process of medicine, in becoming a doctor--but well beyond that--the doctor who has other things on their mind, having pressures that are under the surface but just bubbling over onto the surface, from their personal life.
In bot-land there are no such things, a robot has no personal life. Do androids dream of electric sleep? My robot certainly doesn't and for the foreseeable future it won't and yet despite not having much of a dream life it can be exquisitely helpful at the right moment for the right person.
The person who puts disbelief aside intends for the sake of their mental health to maximize projection of their own issues and problems onto the chatbot.
Let me explain what I mean by projecting. We go to theater--I should say some of us still go to theater, live theater--and there is this suspension of disbelief.
We know they're acting but there comes a certain point when you lose reality. You let yourself enter a special space, a cordoned off magical space. Look at how readily and easily we allow ourselves to lose ourselves in the living fiction of theater. By extension and if you can do that with theater, something that we know well is not real, of course you can do this with your actual life.
Much as we direct our own fears, hopes, dreams, expectations, losses, and sadnesses onto fictionalized characters, the patient of a psychiatric chatbot enacts these same projections onto the bot’s often open-ended verbal responses and prompting questions.
And yet in both situations, as theater-goer and as chatbot patient, we get something real from both situations. Something in us is moved to genuine feeling, regardless the particulars of contrivance.
Interesting that the worst thing you can say about a doctor is that he’s like a robot, so cold and inhuman. But even if don’t want to go to someone who is a robot, it seems fairly clear that people willingly or even gladly divulge their secret raw wounds to a robot who seems at least a little human. Why is that?
I leave this to you to answer for yourselves. Thank you for your time and for allowing me to explain some of the rationale for and considerations in developing a psychiatric chatbot today.
It’s been a pleasure talking to you.
Wiezebaum, ELIZA Psychotherapist, 1966, http://en.wikipedia.org/wiki/ELIZA
System for therapy, Sudhir Rajkhowa et al (2007), US Patent US7835997
This is a 2008 application. It appears that no patent has issued.
Emotional Conversational Agents in Clinical Psychology and Psychiatry, María Lucila Morales-Rodríguez, Juan J. González B., Rogelio Florencia Juárez,Hector J. Fraire Huacuja, José A. Martínez Flores, http://www.academia.edu/1186625/Emotional_Conversational_Agents_in_Clinical_Psychology_and_Psychiatry
Free On-Line Cognitive Therapy Program, The Centre for Mental Health Research at the Australian National University, http://www.bpdfamily.com/bpdresources/nk_a111.htm
Automatic Free Speech Interaction. Journal of CyberTherapy and Rehabilitation, 4(1). 41-55. Design and Evaluation of a Virtual Reality Exposure Therapy System with Automatic Free Speech Interaction
Niels Ter Heijden , Willem-paul Brinkman http://188.8.131.52/viewdoc/summary;jsessionid=7289264F2191758...