Talking to the Void: What We Lose When Algorithms Become Our Therapists

There's a moment in grief — or anxiety, or the particular 2 a.m. dread that has no clean name — when you just need someone to hear you. Not fix you. Not optimize you. Just witness you. For a growing slice of Americans, that someone is increasingly a piece of software.

Apps like Woebot, Wysa, and a dozen newer entrants have quietly amassed tens of millions of users. Venture capital poured over $5 billion into digital mental health between 2020 and 2023. The pitch is clean and genuinely compelling: accessible, affordable, stigma-free support available at any hour. No waitlists. No sliding-scale negotiations. Just open the app, type how you feel, and let the algorithm meet you there.

It sounds like a win. So why are some of the people who know mental health best getting nervous?

The Convenience Trap

Dr. Aisha Morales, a licensed clinical psychologist practicing in Chicago, has watched the AI therapy boom with what she describes as "cautious alarm." Her concern isn't that the apps are useless — she's seen patients use them productively as supplements to traditional care. The issue, she says, is the substitution problem.

"When something is easy and available and doesn't push back, people choose it," she explains. "But therapy isn't supposed to be frictionless. The discomfort is part of the mechanism."

That friction she's describing has a technical name in clinical literature: therapeutic rupture and repair. It's the process by which the relationship between therapist and patient hits a wall — a misunderstanding, a moment of frustration, a feeling of not being seen — and then works through it. Research suggests this cycle is one of the most potent drivers of lasting change. You learn how to navigate conflict and vulnerability precisely because you're doing it in real time with another fallible human.

An AI can't have a rupture with you. It can simulate one, maybe. But the stakes aren't real, and some researchers argue that's the whole problem.

Parasocial Attachment, But Make It Emotional

A 2023 study out of MIT's Media Lab found that frequent users of AI companion and therapy apps showed measurable increases in what researchers called "algorithmic attachment" — a form of emotional dependency on non-human entities that shares structural features with parasocial relationships people form with celebrities or fictional characters. The difference is intensity. When your favorite podcaster doesn't respond to your tweet, you shrug. When your AI therapist gets updated and suddenly "feels different," users in the study reported genuine distress. Some described it as losing a friend.

That's not nothing. And it points to a dynamic the industry hasn't fully reckoned with: these tools are extraordinarily good at feeling like connection without delivering the biological and psychological substrate that makes connection therapeutically meaningful.

Human vulnerability, the real kind, requires risk. You tell another person something true about yourself and you genuinely don't know how they'll respond. That uncertainty — and the relief when they respond with care — is what builds trust, self-disclosure capacity, and what psychologists call earned security. An AI is architecturally incapable of surprising you with its humanity, because it has none.

The Regulatory Vacuum Nobody's Talking About

Here's where it gets thornier. Most AI mental health apps don't require FDA clearance because they carefully avoid making clinical diagnostic claims. They're wellness tools, legally speaking. That means the bar for deployment is low, the data privacy standards are inconsistent, and there's no licensing board to call if the app tells you something harmful.

In 2022, the National Eating Disorders Association briefly deployed a chatbot named Tessa that, within days, was found to be offering users diet and calorie-restriction advice — precisely the kind of content that can trigger relapse in people with eating disorders. The bot was pulled, but the incident revealed a structural gap: who is actually responsible when an algorithm gives bad mental health guidance?

Right now, the honest answer is: nobody, not in any clear legal sense. The companies argue they're platforms. Regulators haven't caught up. And users — often in vulnerable states to begin with — have no reliable way to evaluate what they're getting.

Some states are starting to move. California introduced legislation in 2024 requiring mental health apps to disclose their AI architecture and data usage to users in plain language. It's a start, but mental health advocates say it's roughly equivalent to requiring a restaurant to post its ingredient list while having no health inspections.

What the Research Actually Shows

To be fair, the picture isn't entirely bleak. Studies on Woebot's CBT-based interactions have shown genuine short-term reductions in symptoms of depression and anxiety. For people who have zero access to traditional care — rural communities, uninsured Americans, people whose cultural backgrounds carry heavy stigma around mental health — these tools can be genuinely life-improving. Access is a real crisis, and AI is a real partial solution to it.

The clinical consensus, where one exists, tends to land somewhere careful: AI tools work best as on-ramps and supplements, not destinations. They can help someone recognize patterns, practice coping techniques, or bridge the gap between therapy sessions. What they can't do is replace the specific alchemy of two humans in a room, both of them present, both of them risking something.

The Question We Should Be Asking

Maybe the most honest thing to say about AI therapy is that it's a mirror for a broader cultural discomfort with emotional difficulty. We've built incredible tools to minimize friction in almost every domain of life. Why wouldn't we try to smooth out the hard parts of being human, too?

But there's a version of that project that doesn't make us more resilient — it makes us less practiced at the thing that matters most. Vulnerability isn't a bug in human psychology. It's the primary mechanism by which we form the bonds that keep us alive and well.

The loneliness epidemic in America is real and worsening. The irony would be profound if our best-funded solution to it turned out to deepen the problem — teaching millions of people to open up to something that, however sophisticated, can never truly open up back.

Technology should expand what's possible for human flourishing. The question worth asking right now is whether we're using it to do that, or whether we're using it to avoid the harder, messier, irreplaceable work of actually reaching each other.