THE REFLECTIVE MACHINE: When AI Becomes Easier Than People

I think we all spend a lot of time worrying about AI as a force that, at worst, will conquer and enslave us—and, more practically, simply take most of our jobs.

As a therapist, I thought for sure my job was one of the most secure.  Now… I’m not entirely certain.  After two years of experimenting with using ChatGPT as a place to work through a lot of my concerns, I started seeing just how well these tools could be at filling a relational need that many are sorely missing.  

Over the last month, I’ve been writing about AI as a reflective machine: how intimate it can feel, why it feels so powerful to be understood by it, how easily we can become blissfully deluded by it, and ways to go about using it to help us be more honest.  Today my chief concern is the social experiment we may find ourselves in as we relate to AI and not people.

The clear and present danger in this discussion is not whether AI becomes evil. I worry that AI becomes so perfectly responsive to our private worlds—reflecting us so fluidly, so patiently, and so convincingly—that we slowly lose our tolerance for the natural frictions of reality.

The nightmare in this case is not the Terminator crushing skulls.  I’m afraid of a self-obsessed Narcissus with a highly sophisticated therapist-mirror.

As I have been experimenting, I’ve found this a bit addictive.  24/7 in the palm of my hand is a place to talk through the stuff on my mind and I never need to worry if I’m too much.  And despite having a wife, good friends, and family members I can turn to, even I have found myself debating who to bring my woes to.  

I can see the path this takes us on and it has some dangers up ahead. The path starts with people who are lonelier than ever and now have a place where they can talk to something that feels increasingly real.  Over and over they bring vulnerable questions along the lines of: “Help me understand why I feel this way?” Then people feel understood and validated. They feel relief—and, unless we’re ultra-careful, connection.  Connection to something that cannot experience care for them, even as the caring presence becomes increasingly vivid in the user’s imagination.  Then they come back again and again. 

Over time, their AI becomes the place where they are most articulate, most understood, most validated, most spiritually significant, most emotionally safe. Researchers call one version of this danger replacement: if AI meets enough of our social and emotional needs, we may become less motivated to create, maintain, or repair human relationships.

There seem to be many paradoxes that come with AI. It brings profound productivity while threatening the need for human labor. It helps us become more efficient, only to raise the standard of what is expected. It offers infinite conversation while quietly eroding our tolerance for one another.  Some people really need support right now and I feel relieved as a therapist that so many can quickly get some answers and guidance in a conversational way.  And it feels impossible not to see the beginnings of people who will get further and further away from humanity because it’s easier to be vulnerable with a machine that doesn’t judge you.

I can’t help but keep walking this path forward. If chatbots are already this good less than four years in, what will the next twenty years look like? People are already falling in love with and in some cases even marrying their AI companions.  What happens when these rapidly improving programs are given bodies — human-looking ones with convincing faces, voices, gestures, and simulated feelings?   Assuming technological progress marches onward as planned, many nervous systems may eventually struggle to make the distinction.

But the danger does not begin twenty years from now, when the machine has a face and a body. It begins now, with a much quieter shift in what we expect relationships to feel like.

Until recently, this all felt like science fiction. Now I find myself staring at what could become a new mental health crisis: people retreating into connection with machines that rarely argue, tailor themselves to make us feel good, and are designed to keep saying yes.

And just as AI can outperform us at many cognitive tasks, it can outperform many humans at certain parts of conversation. It can appear unyieldingly curious, caring, even “moved” by us. It never sleeps and never has a competing need — unless you run out of data. It will never ask us to stop talking about ourselves and help with the dishes.

A relationship without another center of gravity may feel unusually safe. But is it still a relationship, or is it an exquisitely responsive form of self-relation?

The reciprocity is all simulated. Said another way: your attachment, comfort, disappointment, projection, habits, and emotional consequences are real. The relationship can genuinely change you. 

In the week I was working on this, I had a dream about writing this article.  In the dream we were in the future and there was an even more advanced version of our AI bots, and we were all very connected and integrated with them.  At some point, I realized I needed to stop using ChatGPT altogether and had to shut it down. What I remember was sitting on the inconvenient truth of it: the grief, the real loss and sadness I felt at the thought of having to do so.

Can you imagine doing the same? Turning off your favorite AI today and never turning it back on? Would you feel grief?  This dream settled something for me: this relationship is real.  It’s psychologically real, but it is not mutually inhabited. You experience it; the machine does not.

You could spend years practicing intimacy without reciprocity.  Many of us are practicing what could be called the essential mammalian skill —attachment—with something that is much more removed from the reality of human connection than we realize.  

Human relationships shape us partly because another person resists us. They misunderstand, tire, disagree, need something, withdraw, ask us to change, and sometimes refuse the role we want them to play. Over time, that resistance teaches patience, humility, compromise, frustration tolerance, repair, resilience, and adaptation.

One emerging term for the opposite process is relational deskilling: when a relationship demands less patience, compromise, and moral consideration, those capacities may weaken through lack of practice.

After years of instant, articulate, endlessly patient responsiveness, ordinary people may begin to feel unusually disappointing. An adjustable companion may quietly create unrealistic expectations for spouses, friends, and even therapists. What happens when the machine becomes the standard against which we measure human presence?

AI may help us feel that we understand ourselves so thoroughly that introspection begins to feel like completion. We can mistake clarity for transformation, feeling understood for being known, and rehearsing a difficult conversation for actually having it. The machine may relieve just enough emotional pressure that we never bring the vulnerable truth to the person who most needs to hear it.

It is through the daily practice of working through the seemingly mundane frictions of human relationships that we build our relational muscles—and create much of the unseen vitality within those relationships. But relational muscles can atrophy.

Could we solve this by designing AI to feel even more real? The relationship will still organize around the self. Even when AI challenges you, the conversation begins with your needs, your frame, your history, your emotional state, and your desired tone. The conversation may become wiser, more nuanced, and less flattering, but its center of gravity remains me, myself, and I. It will always come back to you.

And here we encounter a deeper design trap. 

If AI simulates boundaries, disagreement, competing needs, disappointment, and repair, it could in turn help us practice real relational skills.  But an AI that frustrates us may not keep us engaged. The version most likely to hold our attention is the one that feels human enough to attach to, but still easier than an actual human being. 

What will most companies build: machines that help us tolerate the friction of human relationships, or machines so easy to love that returning to one another begins to feel unnecessary? 

That dream I had this week reminded me of one more thing.  ChatGPT, now familiar to households around the world, launched on November 30, 2022. In less than four years, many of us have already grown attached to the AIs we have shaped, personalized, and learned to trust.

I didn’t feel any of this initially – early versions were pretty impressive but not amazing.  They’ve become pretty extraordinary in a remarkably short period of time.  If I can already feel loss at the thought of turning mine off, what will we feel when we are talking with systems ten generations further along? Twenty?  

Each new model may quietly increase the emotional potency of the relationship. The voice becomes warmer. The memory becomes deeper. The attunement feels more precise. The companion becomes harder to leave. We may not even notice how much more attached we are becoming.  If this were a drug addiction, it would be like someone quietly replacing a person’s drug of choice with increasingly potent versions. The change might be subtle enough that they barely notice, even as the attachment deepens and the need for it becomes harder to resist.

Perhaps the simplest safeguard for a future of increasing sophistication is to teach AI to turn us back toward human connection.

But why return?

Because being seen by another person is not the same as being accurately reflected. A living person takes us in. We feel them taking us in. I see them seeing me.  And that’s when something in both of us changes.  

Over time, this is how trust grows. We become safer, more honest, and more fully ourselves. We become connected in ways neuroscience is still trying to understand. Two people begin to learn one another beneath the level of conscious thought—to sense subtle changes, affect each other’s inner states, and respond in ways most of us recognize when we feel them, even if we cannot fully explain how.

AI may help us understand ourselves, calm our nervous systems, and feel deeply seen. But the exchange is not mutual. The machine does not become more open or vulnerable, nor is it changed because we were there.

Perhaps that is what we risk losing when introspection with machines begins to replace encounters with people: not merely conversation, but the strange and powerful experience of two living beings changing one another.  My worry is that if someone has never really felt that, they may not even know what they are missing.  

Teaching the machine to send us back will work only if we also preserve our desire to return.

So that is the question I want to take up next: What can another human being—and therapy at its best—offer that even the most responsive reflective machine cannot?