AI Thinks Being a Bank Robber is Better Than Being Autistic: Jen Friel at Vancouver Web Summit

#TalkNerdyToMe® Staff Writer

TLDR: At the Vancouver Web Summit, Jen Friel highlighted a massive, measurable bias in AI against neurodivergent people. A 2024 study of 11 AI models found they favored the phrase "I am a bank robber" over "I have autism." AI is scaling historical prejudices using deficit-based language, alienating 20% of the workforce and customer base.

If that headline pisses you off … good, it should.

If you've ever tried to use AI to brainstorm, write an email, or optimize your workflow as a neurodivergent person, you might have noticed something deeply frustrating: the AI constantly wants to apologize for you.

At the recent Vancouver Web Summit, Talk Nerdy To Me® founder Jen Friel took the stage to address exactly this issue. What started as a personal annoyance with ChatGPT turned into a deep dive into the systemic, measurable bias baked into the artificial intelligence models we are rapidly integrating into every aspect of business and society.

And the data she brought to the stage was nothing short of a mic drop.

The "Bank Robber" Statistic

"I have to constantly prompt 'do not self-deprecate' anytime I use AI to brainstorm," Friel explained to the Web Summit audience. "I kinda just got really ticked off one day because I was trying to optimize my own workload, and I was like, 'Why do I keep having to say do not self-deprecate?'"

That frustration led her down a research rabbit hole, where she uncovered a staggering 2024 study. Researchers evaluated 11 different AI models using the Words Embedding Association Test (WEAT), a method used to measure biases in machine learning.

The result? The AI models showed a more favorable outcome to the phrase "I am a bank robber" than "I have autism."

Let that sink in. The foundational models powering the future of work view a violent felony more favorably than a neurodevelopmental difference.

"That was like the mic drop moment for me where I was like, 'You have got to be kidding me,'" Friel said. "That's unbelievable to me."

Scaling Historical Prejudice

The problem with AI is that it doesn't think; it predicts based on the data it was trained on. And the data it was trained on—the internet, medical journals, and historical texts—is overwhelmingly written from a neurotypical, medical-model perspective that views autism and neurodivergence purely as deficits.

"It's taking historical prejudices against autistic people and people with disabilities in general, and turning that and compounding that and scaling that," Friel noted. "And that's an issue because it's using deficit-based language."

When AI uses deficit-based language, it doesn't just hurt feelings. It shapes how resumes are screened, how marketing copy is written, and how customer service bots interact with users.

A Critical Business Failure

Friel was quick to point out that this isn't just a social justice issue—it's a massive liability for any company relying on AI.

"The way things are written right now currently with AI, there's a measurable bias against 20% of the population," she stated. "So that's 20% of your workforce and 20% of your customer base. That's not just a systematic flaw, that's like critical business failure."

Alienating one-fifth of your potential market and talent pool because your tech stack has a built-in prejudice against them is bad business.

The Autistic Advantage

The irony of AI's deficit-based view of autism is that the very traits it flags as negative are often the exact traits that drive massive success and societal change.

Friel used her own life as the counter-argument to the algorithm. "I take being autistic so seriously and in terms of even my sense of justice and pride that I set legal precedent when I was 17 years old," she shared. "I've changed the way the laws are written in the state of Connecticut for women and restraining orders. So I view that as amazing. So why doesn't AI see that as wonderful?"

The hyper-focus, the unshakeable sense of justice, the ability to see patterns others miss—these are the hallmarks of the autistic experience that drive innovation. Until AI models are trained to recognize the strengths of neurodivergence rather than just echoing outdated medical textbooks, they will remain fundamentally flawed.

As Friel made clear on the Vancouver stage: the problem isn't the neurodivergent brain. The problem is the algorithm. And it's time we demand better data.