Latest / Future of Work Tech with Fexingo: Remote Tools, AI Productivity, and Workplace Software / How AI Is Decoding Your Slack Personality
Transcript
- Lucas: So Luna, here's a question you probably didn't wake up asking yourself: what if your Slack messages are telling your employer more about your personality than your annual review ever could? Luna: Oof. That's unsettling. But also, yeah, I can see how it would be technically possible. Lucas: It's not just possible — it's already happening. A team from Stanford and Microsoft Research published a paper last year where they analyzed millions of anonymized Slack messages from thousands of employees. They built a model that predicts the Big Five personality traits — openness, conscientiousness, extraversion, agreeableness, neuroticism — from chat patterns alone. Luna: And we're talking real accuracy? Not just horoscope-level guesswork? Lucas: The model's correlation with established personality tests was in the 0.4 to 0.6 range for most traits. That's not perfect, but it's statistically significant. For extraversion, it was even higher. The key signals were things like message length, response time, emoji usage, and how often someone starts a new thread versus replies in an existing one. Luna: Wait — so if I send short, fast replies and use a lot of emojis, the algorithm flags me as more extroverted? That feels almost too simple. Lucas: Right, but it's the combination that matters. The model also looks at linguistic features — word choice, sentence complexity, sentiment consistency. Someone who uses tentative language like 'maybe' or 'I think' more often might score higher on neuroticism. Someone who writes long, structured messages with bullet points might be higher on conscientiousness. Luna: That is genuinely creepy and also fascinating. But let's talk about the elephant in the room: who's actually using this? Is this just research, or are companies deploying this today? Lucas: Some companies are definitely experimenting. IBM has a tool — it's been around for years — called Watson Personality Insights, though they've recently shifted its focus more toward customer service. Google has internal projects that analyze communication patterns for team health. And there are startups like Receptiviti and Crystal that offer personality analytics from text, marketed for sales coaching and leadership development. Luna: So the pitch is: we can help managers understand their teams better, reduce friction, maybe even predict burnout before it happens. But the flip side is, you know, a massive privacy violation. Lucas: Exactly. And the ethics get thorny fast. Most employees don't know their chat data is being analyzed this way. Even when it's anonymized, the inferences are deeply personal. Imagine being passed over for a promotion because an algorithm flagged you as 'low conscientiousness' based on how you chat. Luna: Have you actually seen a case where that kind of inference was used in a real employment decision? Lucas: Not publicly, no. Most companies are cagey about it. But there was a notable incident at a UK tech firm a couple of years ago where an employee was let go after an internal AI tool flagged their chat messages as 'negative sentiment.' The employee sued, and the case settled out of court. So the precedent is murky. Luna: That's the kind of thing that makes you wonder: if you knew your Slack was being personality-scored, would you change how you type? Lucas: Probably. And that itself changes the data. It becomes a game of detection and avoidance, which is exactly what happend with productivity monitoring tools. People start typing 'just to keep the green light on'. Luna: It's interesting too that we're having this conversation on a show about the future of work tech, and I think a lot of our listeners are probably the ones building or buying these tools. If you're someone who's actually working with this kind of data, and you found today's breakdown useful — Lucas: Yeah, I was going to say — if you get value from shows like this, and you want to keep them ad-free and independent, listener support is what makes that possible. We're at buy me a coffee dot com slash fexingo. It's a small thing, but it genuinely helps us keep digging into topics like this one. Luna: Absolutely. And we really appreciate every person who chips in. So back to the personality profiling — one thing I'm curious about is whether this technology could be used for good in a way that feels consensual. Like, what if you opt in, and the insights help you collaborate better? Lucas: That's the optimistic use case. There are teams using tools like this for retrospectives — after a project, they run an anonymized analysis of communication patterns to see where bottlenecks happened. Did one person's messages get ignored? Did the team's sentiment dip during crunch time? That can be really valuable, as long as it's opt-in and not surveillance. Luna: So the line is consent. If I know my data is being used and I agreed to it, it's coaching. If I don't know, it's spying. Lucas: Exactly. And the problem is, most workplace tools today default to the latter. The privacy policies are buried, the consent is coerced — 'use our chat platform or don't have a job.' So the burden is really on regulators and on internal ethics boards. Luna: Speaking of regulation, the EU's AI Act is starting to classify some of these uses as high-risk. Do you see that changing how companies deploy personality inference? Lucas: It could. The AI Act specifically flags systems that infer emotions or personality traits in the workplace as high-risk, meaning they'd need third-party audits, transparency, and meaningful human oversight. But enforcement is slow, and a lot of companies are still in a 'ask forgiveness, not permission' mode. Luna: Let's talk about alternatives. Is there a way to get the team-health benefits without the creepiness? Lucas: One approach is to only analyze aggregate, de-identified patterns — like, 'the engineering team's average response time dropped 20 percent this sprint, which might indicate burnout.' No individual scores. Another is to give employees their own data dashboard, so they see what the algorithm sees about them and can correct it. Some companies are experimenting with 'data trusts' where a third party holds the model and only releases insights that the team has agreed on. Luna: I like the self-dashboard idea. If I saw that my Slack personality profile said I was 'low agreeableness,' I'd probably think twice before firing off that snarky reply. Lucas: Right — it becomes a mirror rather than a scorecard. And there's research showing that when people see their own communication patterns visualized, they actually improve their collaboration naturally. No algorithm needed. Luna: So maybe the best use of AI personality analysis is not to tell managers about employees, but to tell employees about themselves. That flips the power dynamic. Lucas: Exactly. And that's the kind of design philosophy that separates a tool from a trap. The technology itself isn't evil — it's how it's deployed, who controls the output, and whether the subject has agency. Luna: Alright, let's end with a prediction. One year from now, do you think more companies will be using this kind of analysis, or will the backlash kill it? Lucas: I think we'll see a split. Large, regulated firms in the EU will pull back or go fully opt-in. But startups and tech-forward companies in less regulated markets will double down. The genie is out of the bottle — the question is whether we can teach it to be polite. Luna: That's a good note to leave it on. Thanks Lucas. Lucas: Thanks Luna. And to our listeners, we'll be back next time with another angle on the future of work tech.