AI as Social Skills Practice for Autistic Adults

The framing of social skills "training" for autistic people has a fraught history. For decades, the dominant approach involved intensive behavioral programs aimed at making autistic people appear more neurotypical—suppressing stimming, enforcing eye contact, drilling scripted social responses. Much of this happened without adequate attention to the autistic person's experience of the process or the long-term psychological costs. That history matters when thinking about AI as a social practice tool, because the question of what the practice is for shapes everything else. Practice toward performing neurotypicality is a different thing—and often a harmful thing—compared to practice aimed at building confidence, reducing anxiety, and developing personally relevant skills that help an autistic person navigate situations they actually care about.

What Autistic Adults Often Say They Want

When researchers ask autistic adults what social support they want—rather than designing programs based on what neurotypical observers think they need—the responses tend to emphasize: anxiety reduction, preparation for high-stakes situations, help with unpredictable social scenarios, and support for navigating specific practical challenges like workplace communication or disclosing their diagnosis. This is different from asking autistic people to become more like neurotypical people. It's about building capacity for situations that matter to them on their own terms. Research from Autistic Self Advocacy Network surveys and from qualitative research at the University of Washington has consistently found that autistic adults are far more interested in accommodations and self-understanding than in behavior normalization—a finding that hasn't yet fully penetrated mainstream practice, but should inform how AI tools in this space are designed and used.

Where AI Has Practical Value

AI offers something that traditional social skills practice cannot: a low-stakes, endlessly patient, judgment-free environment where someone can try the same scenario repeatedly without wearing out a person's goodwill. For autistic adults who experience high anxiety about specific situations—job interviews, difficult conversations with a partner, phone calls to unfamiliar organizations—the ability to run through the scenario multiple times, ask questions about subtext, get feedback on different approaches, and generally reduce the novelty of the situation is genuinely valuable. Anxiety about social situations is often partly anxiety about the unknown. Rehearsal reduces the unknown. AI can also engage with questions that might feel embarrassing to ask a human. "What does it mean when someone says they'll think about it?" "Is it rude to not respond to a greeting immediately?" "Why did this conversation feel like it ended badly?" These questions have real answers, and getting them without navigating the social complexity of asking a person is a genuine service.

The Double Empathy Problem Again

The framing that autistic social difficulties are deficits in the autistic person is increasingly challenged by research. Studies on autistic-to-autistic interaction, including work by researchers at Cambridge University, find that when autistic people interact with other autistic people, social communication flows well—they understand each other's cues, they find the interactions less exhausting, and they show similar levels of affiliation and rapport to neurotypical pairs interacting with each other. This suggests that what's often described as a social skills deficit is partly a cross-neurological translation problem. AI, trained on predominantly neurotypical communication, mostly replicates neurotypical social norms—which is useful for preparing for neurotypical-dominant environments but doesn't address the deeper question of whether those environments and their norms are the right target.

A Tangent on Alexithymia

A significant proportion of autistic people experience alexithymia—difficulty identifying and describing one's own emotional states. This isn't a feature of autism itself, but the two co-occur frequently. Alexithymia can complicate both social interaction and self-understanding: if you don't know what you're feeling, it's hard to communicate it, and it's hard to understand how your internal state is affecting your behavior. AI conversation can be useful here in a specific way. The process of talking through a situation, receiving reflections, and being offered possible emotion labels—"that sounds like it might have felt dismissing" or "it sounds like you were expecting more acknowledgment"—can help someone develop a more granular internal emotional vocabulary over time. This isn't therapy, but it can supplement therapeutic work or function as a starting point.

What Responsible AI Practice in This Space Looks Like

AI tools designed for autistic adults need to be built in genuine consultation with autistic people, not designed by neurotypical developers with assumptions about what autistic people need. They need to be clear about what they're offering—skill development in user-defined areas—rather than implicitly positioning themselves as normalizing the user toward neurotypical standards. They need crisis referral protocols and should not be promoted as a substitute for professional support. They also need to be honest about what the research supports. The evidence base for AI-mediated social practice is young and incomplete. The promising findings suggest genuine utility in anxiety reduction and specific skill preparation. They don't support broad claims about transforming autistic social functioning. Used well, with appropriate framing, AI practice tools can be a genuinely useful addition to an autistic adult's toolkit for navigating a world that wasn't designed for them.