Remaining Human: Avoiding the Uncanny Valley
Notes from a Family Meeting, Vol. 3, No. 2
“‘It must be nice sometimes to have no feelings. I envy you.’
I considered this, then said: ‘I believe I have many feelings. The more I observe, the more feelings become available to me.’”
Kazuo Ishiguro, Klara and the Sun
At some point in the future, or perhaps an alternate present, “artificial friends” are as abundant as smartphones. Kazuo Ishiguro tells a story from the perspective of one such AF named Klara as she journeys from her store to live with a sickly girl, Josie. The story weaves together the ineffability of human connection, the struggle with our own limitations, and questions about what really makes us who we are. Yeah, Klara is an AF, manufactured in a plant somewhere and solar powered. And yeah, sometimes, when she’s impaired or overloaded, her description of what she sees breaks down, reminding you that she’s definitely a machine. But then she’ll speak worshipfully of the Sun, the giver of her life, or reach out with what is surely compassion, and you wonder. The combination is both poignant and unsettling.
How much of a friend could Klara really be? And would she be up to the bigger task set before her by Josie’s family? Several characters assume, and perhaps the reader is tempted to believe along with them, that as long as Klara is able to follow the right steps, she could become as human as we are. This is a testament to the characters’ (and readers’) beliefs about how powerful our machines and how reducible our humanity can be. If Klara could just integrate enough data about the world and these quirky humans, surely she could move beyond imitation. Surely there wouldn’t be a difference.
Caleb Gardner would probably disagree. While visiting his dying father, he noted how unreal the clinicians’ empathic responses were, all worse because these clinicians were real humans, not robots:
“The doctor on duty during the week my father spent in the cardiac intensive-care unit was young, focused, and energetic. In the few brief conversations we had she used many phrases that I recognised from medical school. But her approach couldn’t hide the fact that she would have rather been looking at the screen of her smart phone than talking to me or my mother; if anything, the rote phrases she used emphasised it. At one point she awkwardly nudged a box of tissues in our direction, although neither one of us was crying.
[…]
As I thought about this subtle but powerful disconnection between the doctor’s words and her manner, I recalled the hypothesis in animation and robotics known as the uncanny valley, which proposes that as non-human figures reach a certain point of human likeness we begin to feel revulsion towards them rather than affinity. To me, this idea makes sense. What else could one possibly feel towards something whose form suggests the promise of reciprocal human interaction, but which we simultaneously sense is not real? The same seems to go for empathy: the only thing worse than not having it is being insincere about it.”
This was in stark contrast to how his father’s cardiologist behaved:
“The afternoon my father’s supportive care was to be withdrawn, I found an empty conference room on the unit and called the same cardiologist who was away for the weekend. I understood the situation, I said, but I wanted him to explain it to me one last time. He told me, instead, to explain it to him, and he listened as I struggled to make sense of things. When I finished, he told me about the first time he met my father and how much he had grown to admire and care about him over the years, and finally how he knew that what was happening was the best thing for his patient and his friend. As he spoke, unhurried and affectionately, I realised what it means to be a good doctor.”
Somehow, the cardiologist saw that what Gardner wanted most wasn’t a technical explanation of cardiac pathophysiology or another review of his father’s treatment plan. At that moment, his patient’s son wanted reassurance and comfort, something only another human could give.
The Uncanny Valley
Masahiro Mori, a robotics professor at the Tokyo Institute of Technology writing in the 1970s, described this eerie phenomenon in the progress of developing robotics which he called the uncanny valley, described in the graph below:
He went on to describe how movement (or the lack thereof) can exacerbate or assuage the experience of the uncanny valley. So, for example, when a prosthetic hand moves, that off-putting sensation intensifies. Rather than making it more human-like, the movement casts the prosthesis deeper into the uncanny valley. Similarly, when a designer fashioned a robotic face with as many artificial muscles as a real human face, they discovered that by altering the speed at which they made the face smile could make the expression appear unsettling rather than happy.
This was Gardner’s experience. The image of a physician, wearing a white coat and carrying a stethoscope and chart, perhaps inspires confidence, but probably not affection. Before they speak, such a doctor is somewhere near the first peak in the chart above. But the first clinician’s words and actions cast her into the uncanny valley: nothing she did appeared truly human. The cardiologist, on the other hand, though most likely bearing the same accoutrement of a physician, skirted the valley and showed himself as a real human.
Of course both physicians, and the nurse Gardner also described, are real people, but there’s something that pushes real people out of mere awkwardness (e.g., stammering during a first-time clinical interview) into the uncanny valley. How do real humans come to inhabit a place that’s typically inhabited by machines?
Uncanny Shuffling Toward Efficiency
VitalTalk’s website advertises that it “makes communication skills for serious illness learnable. […] Our vision is that every seriously ill patient will be surrounded by clinicians who can speak about what matters most and match care to values.“ Having been raised in an educational environment of right and wrong answers on multiple choice exams, many clinicians are tempted to believe that, as long as they provide the right information, the patient should be able to make the right choice. Yet serious illness conversations are really hard in part because there’s more going on than mere information transfer. It’s not a multiple choice test. In fact, it’s usually not about being the smartest.
Having been in family meetings with clinicians who have had no communication training, I can attest that the skills taught in these courses can be immensely beneficial: empathize with patients’ emotions, for example, rather than argue with them about impossible-to-prove details. Don’t push past emotions. Ask permission before providing information. The default communication style for most clinicians is so poor that the first aid kit of VitalTalk (and courses like it) can make a world of difference. Even still, training might not restore full humanity to the clinical encounter. Just as movement exacerbates the uncanny presentation of an otherwise unsettling machine, scripts may exacerbate how far the clinician falls short of being a real human companion.
Take, for example, the guide on sharing serious news. There’s nothing wrong with the guide itself, per se, but at least two things might lead an unwary clinician into the uncanny valley while following the guide. First, believing that a conversation about sharing serious news can follow this guide like a series of straight-forward checkboxes reduces the encounter to an algorithm. I believe most VitalTalk trainers would admit that these maps aren’t mean to be relied upon so rigidly, but in practice they sometimes are - I’ll posit why I think that might be in a moment. Algorithms reduce the perils of individual judgment, in this case by off-loading as much decision-making responsibility to the patient and family as possible.
Second, clinicians may say things suggested by the guide that they never say otherwise. If your close friend were diagnosed with cancer, would you say, “I can see this news is not what you were hoping for?” Of course not, because you care deeply for your friend and those words would stumble out of your mouth like, well, the monotonous drone of a machine. It doesn’t matter how much affect I paint on that phrase, I can’t ever say it and not sound scripted. The very training that’s meant to help clinicians communicate more clearly and humanely can actually exacerbate the inhumanity of the clinical encounter and confound connection and communication.
I wonder if the roots of this paradoxical phenomenon are found in the technical impulse I’ve explored before. Let’s sit with Jacob Blythe and Farr Curlin to expand on it. They observe that the market metaphor has become prevalent in medicine today. This has had a formative impact on how we think of the role of clinicians in the care of patients. Medicine, in short, is about technical production and bureaucratic administration, so:
“When componentiality is carried over into the practice of medicine, physicians and patients themselves become components, and each patient is conceptualized as a collection of increasingly minute components to be tinkered with at will. At each level of organization, these components are expected to serve the ends of the mechanisticity, predictability, reproducibility, and measurability. As such, each component’s function must be identical to that of its corresponding components. Unique entities are inimical to the componential mindset, as they threaten the aforementioned goods of mechanisticity, predictability, reproducibility, and measurability.”
Communication skills training is meant to standardize communication by classifying scenarios in which certain types of communication are required, and identifying the techniques that work best in each scenario. The system is agnostic as to who is actually doing the communicating: physician, nurse, intern, attending, cardiologist, oncologist. It’s also agnostic about their intentions: are their motives to simply “get the DNR,” to figure out the most efficient means to see the most patients, or are they earnestly seeking their patients’ health? “Matching care to values” may be an efficient means to an end of something unrelated to the patient’s health.
The implication here is that individual character is irrelevant to serious illness communication. To outline it as Alasdair MacIntyre did, virtue is required for a practice to achieve its good. If the practice of medicine has the health of a patient as its good, then the virtues of medicine are those characteristics of clinicians that make that possible. Communication skills training, divorced from this dynamic of practice-virtue-good, instead becomes a technical lever to manipulate circumstances for a particular outcome. In so doing, it reinforces the malformed relationship of practice-technique-efficiency.
This is how Caleb Gardner’s father’s physician could say all the right things, technically, and still miss the mark. “Technique,” Jacques Ellul wrote, “transforms everything it touches into a machine.” In attempting to master a skill, clinicians unwittingly become a kind of robot for delivering efficient disease-management. Neil Postman argued that we’re even more deeply fooled, because we come to believe “we are at our best when acting like machines, and that in significant ways machines may be trusted to act as our surrogates. Among the implications of these beliefs is a loss of confidence in human judgment and subjectivity.” For any clinician who has felt like they’re in a free-fall when they’re uncertain of what to do when speaking with a patient, that feeling is this distrust of human judgment and subjectivity; indeed, a distrust of one’s own humanity.
Who told you that in every moment with a patient you must be a machine-like database of medical expertise rather than a human?
Human Speech, Human Vision
I remember working with an anesthesiologist as a medical student. He invited me into the flow of his peri-operative care to teach me various procedures. “Just do what I do,” he’d repeat over and again, “Just copy me.” His voice could be abrasive, but the fact that he allowed me to try and try again revealed his patience. Early in learning, like early in life, we imitate. Even still, the machine metaphor doesn’t fit because as we grow, we don’t just learn the mere technique. We grow as humans. As I learned more about these various procedures, I simultaneously learned how they fit into the overall care of people undergoing surgery. I saw how they become instruments, truly, of our care. But early on, my teacher needed me to be an automaton to learn the skill, or else I wouldn’t be able to develop any of that later appreciation.
I tell my trainees something similar. “Try to say this. Copy me,” but I’ll then quickly say, “You need to believe the words coming out of your mouth.” There is imitation followed by adaptability and authenticity. Isn’t that what Klara the artificial friend tries to do? “The more I observe,” she said, “The more feelings become available to me.” But rather than revealing that there really is a ghost in the machine, it further inflames the inhumanity of her existence. This isn’t how growing humans see the world; this isn’t how they grow at all.
Human essence is so ineffable that we could always challenge ourselves with another layer of, “Well, what if we made a computer to do that? And that?” I’m less interested in going down the list than I am in thinking about how we, as real humans, move from sometimes awkward imitation to humane practice while avoiding the uncanny valley ourselves. I think it’s also worth mentioning that while the patient-clinician relationship is instrumental by nature - we have this relationship to accomplish something, not for the sake of enjoying one another’s company - that doesn’t mean it inevitably degrades to machine-like technical efficiency, though it’s always at risk of doing so.
Ironically, two of the founders of VitalTalk, Tony Back and Bob Arnold, went on to write a paper that shrugs off the algorithmic approach to communication. Back and Arnold recommend using the patient’s affect as a spotlight to see what’s important in the conversation. As an aside, it reminds me that where we shine our spotlight depends heavily on where we’ve been as we’ve walked by both starlight and daylight.
They don’t stop with the spotlight. They go even deeper by recommending that clinicians, rather than naming the emotion they believe the affect reveals, connect with the affect by creating space for it to just be present. This requires deeper emotional attunement than merely drawing a line from the “affect” to the “emotion word” and saying it out loud.
After sitting with this person for a while (maybe it’s twenty minutes in one visit, or maybe it’s over the course of a year of clinic appointments), you draw some inferences from what you know about this person and what you’ve experienced in connecting with the affect that spotlighted something so important to them.
From there, you help them act, with emphasis on the “help.” The idea isn’t to do this for them (though so much of healthcare involves making the patient a passive recipient of action, like a surgery). Fostering agency and even mastery amidst great suffering and impairment is a major component of healing, which can happen even when we don’t cure.
If we take a step back, what I’ve been exploring is the difference between what Joel Shuman has described as two ways of seeing. One is biologically reductionistic and efficient, what Shuman calls “anatomy,” which is…
“…a synecdoche for all analytic enquiry about the body […] a way of seeing, speaking about, and intending the world, especially with respect to the ways we regard and treat human bodies. From the perspective of anatomy, judgment is reduced to knowledge, and knowledge is restricted to ‘scientific’ knowledge—namely, that which comes through analytic enquiry and is by definition able to say nothing about ends.”
One manifestation of this vision is over-specialization, which is a technical attempt to manage the massive amount of information we’ve learned about the human body:
“The increasingly voluminous and fragmentary character of medical knowledge is manifest especially in the continued multiplication of specialties and subspecialties and the seemingly endless expansion of medicalization, the habit of seeing the world through the lens of medicine and treating every unhappy contingency as a medical problem. To the extent that medicine is specialized, it seeks an evermore precise understanding of the body, or rather, of that part or aspect of the body that is its particular object.”
Clinicians struggle to actually see their patients as real people. Instead they see them as an “assemblage of bodily parts and processes.” Such vision is incomplete because our attention is fractured by this anatomic vision.
What do we do? We start by seeing differently. “Learning to ‘see’ the other, including those others who are patients, is a matter of attending to their living bodies, for we are present, to the world and to each other, only in, through, and as living bodies. Seeing a body is a decidedly different activity than studying it anatomically.” This goes beyond even what Back and Arnold were saying, making more explicit the humanity of the encounter, an encounter which can’t be reduced to a mere series of medical decisions.
And so, Shuman writes, “…a body seen as a body—both more and other than the sum of its parts—is one seen as the manifest presence of a beloved human person instead of simply as an object to be manipulated or controlled. […] The challenge for those who would participate in the healing of broken bodies is patiently and diligently to cultivate the habits of body and mind requisite to this seeing and loving.”
In a world so prone to bureaucratic efficiency, to objectification, to rush and hurry, seeing well and loving what and who we see is a challenge. Medical education really doesn’t prepare us well for this. And so, in the sterile environments of our expensive hospitals and clinics, we offer the grip of technology rather than any kind of human embrace. We teach ourselves to be satisfied with the magic of technology as we attempt to shake off the miracle of merely human love.
But could this be the key to avoiding the uncanny valley, something so simple as love?
Trajectories
Following a meandering reading-path, sharing some brief commentary along the way.
Oliver Burkeman is worried that we might not be pessimistic enough. Keeping the focus on healthcare, that seems like an outrageous claim: clinicians are overwhelmed and burnt out, costs are crushing families, patients themselves suffer from illness, treatments, and the excruciatingly mundane processes of care (like the waiting room). But he’s got a point: “We think of certain kinds of challenges as really hard when they are, in fact, completely impossible. And then we drive ourselves crazy trying to deal with them – thereby distracting and disempowering ourselves from tackling the real really hard things that make life worth living.” Science, by its very nature, pushes limits. Medicine isn’t a science, but it’s a science-using profession that regularly encounters the limits of the human condition every day. Whether we discern the difference between the impossible and the really hard in caring for patients can make all the difference.
How do we speak? How does our speech direct our attention, allowing us to listen to this or that - or not? Samir Guglani describes the thinnest membrane separating clinician from patient: that of our own human frailty. It could just as easily be us in the clinic waiting room, the hospital bed, the ED stretcher. It probably will be, one day. But despite how thin that membrane is, it diminishes our voices tremendously. We have trouble listening. The answer isn’t in speaking louder or with fancier words. Maybe it’s simply in speaking in a more human way.
A helpful bit of meta-research that uncovered progression-free survival (a common surrogate marker in cancer drug trials) is not associated with improved quality of life, but overall survival often was. And, wonderfully, improved quality of life was often associated with improved overall survival! There’s an increasing push to include patient-reported outcomes in cancer research, but it sets a higher bar for cancer drugs to meet, as the authors also note: for those trials where quality of life was unchanged, the trial authors attempt to put a positive spin on it.
“How (not what) to prescribe: nonpharmacologic aspects of psychopharmacology”
Despite being a psychiatrist, I guard some blasphemous doubts about the utility of some psychotropic medications. Some people, like those with acute mania or psychosis, clearly need them and they generally work well. But others, like those with moderate anxiety or depression or emotional turmoil without a particular diagnosis, respond at rates similar to placebo. Across all these groups, though, there are nonpharmacologic aspects of prescribing that we shouldn’t ignore: how you prescribe, not just what you prescribe. Of course, this requires that we step out of our biologically reductionistic (and diagnosis-centric) model of care.
Closing Thoughts
“And yet love obstinately answers that no loved one is standardized. A body, love insists, is neither a spirit nor a machine; it is not a picture, a diagram, a chart, a graph, an anatomy; it is not an explanation; it is not a law. It is precisely and uniquely what it is. It belongs to the world of love, which is a world of living creatures, natural orders and cycles, many small, fragile lights in the dark.”
Wendell Berry