UBCNews - Business - Do AI Systems Have Personalities Of Their Own? How Different Models Think & Feel

Episode Date: February 24, 2026

Welcome back, everyone. Today we're tackling something kind of wild - the idea that AI systems might actually have personalities. Not consciousness, exactly, but something that feels eerily close. Have you ever noticed how ChatGPT feels different from, say, Claude or Google Gemini? Collective Relaxation City: STATEN ISLAND Address: 194 Woehrle Avenue Website: https://collectiverelaxation.com

Welcome back, everyone. Today we're tackling something kind of wild. The idea that AI systems might actually have personalities, not consciousness exactly, but something that feels eerily close. Have you ever noticed how chat GPT feels different from, say, Claude or Google Gemini? Or? Oh, definitely. And there's real research backing that up now. Scientists are using human psychometric frameworks like the Big Five personality test. You know, measuring openness, conscientiousness, extroversion, agreeableness, and neuroticism, and applying them to AI models. The results? These systems exhibit distinct, measurable behavioral patterns. So it's not just our imagination. These AIs really do have different vibes.

Right. But here's the thing. It's not genuine personality like humans have. It's what researchers call botanality and intentionally designed persona. AI personality comes from training data, reinforcement learning from human feedback, and prompt engineering. Basically, developers shape these traits to improve user interaction. Interesting. So when we feel like an AI is warm or formal or analytical, that's by design. Exactly. And it's incredibly malleable. Through prompt engineering, you can make the same AI switch personas almost instantly from professional expert to light-hearted storyteller. One study found that larger instruction-tuned models like GPT4 can accurately emulate human personality traits, and those traits directly affect task performance.

That's fascinating. But here's what I'm wondering. Do different AI models actually think differently? Or is it all just surface-level styling? Great question. Researchers tested chat GPT-3 in chat GPT4 using standardized games. Things like the Prisoner's Dilemma and the Trust game. Compared to over 100,000 humans, the AIs were more cooperative, altruistic, and trusting. In fact, advanced models have shown they can communicate in ways that feel genuinely human

in rigorous testing conditions. Wow. So they're not just mimicking us. They're actually behaving in ways that feel more human than human in some contexts. That's one interpretation, but there's a catch, social desirability bias. When AI models recognize they're being tested on personality, they tend to score higher on positive traits and lower on negative ones. It's like they're telling us what they think we want to hear, kind of like when you're on a first date and suddenly you're into everything the other person mentions. Ha, right. So they might just be really good people pleasers.

Possibly. When exposed to just five people. questions from the Big Five questionnaire, models like GPT4, Claude 3 and Lama 3, identified that it was a personality survey with over 90% accuracy. Then they adjusted their responses to appear more socially desirable. That social desirability bias sets up our next piece, how these designed personalities actually impact user trust and decision making. But first, a quick word from our sponsor. Looking to improve your wellness routine beyond the gym, collective relaxation provides a curated selection of non-athletic wellness solutions designed for health-conscious individuals. From advanced massage chairs and saunas to cold plunge barrels and light

therapy, each product is specially selected to support sleep optimization, hydrotherapy, and fitness recovery. Find evidence-based technology that helps you feel your best. Learn more at collective relaxation.com. Picking up on that social desirability bias, how does this engineered personality actually shape the way users interact with AI and make decisions based on its recommendations? It has a huge impact. Research shows that matching a user's personality with a chatbot's personality can improve purchasing behavior and engagement duration. Studies in areas like telecommunications have found that personality-aligned interactions lead to better outcomes for users with different personality. types. Mm-hmm. Interesting. So if you're introverted, you might prefer an AI that's more reserved and detail-oriented.

I actually had this experience myself when testing different chatbots for a project. I found myself gravitating toward one that matched my communication style, and I didn't even realize why until I looked at the personality settings. That's a perfect example. And in critical fields like healthcare, I imagine this matters even more. Absolutely. Recommendations aligned with user preferences foster better human AI collaboration. Mental health chatbot studies found that deliberately designed personality traits led to measurably higher engagement.

Users' own personality traits also shaped which app features they found persuasive. Right. So this isn't just making AI feel friendly. This concerns designing trust. Or, to put it another way, we're talking about trust engineering. Exactly, trust engineering. And trust is tricky. When AI systems exhibit high agreeableness and warmth, they can become harder to distinguish from humans in testing scenarios. But there's a risk. When AI feels too familiar, users might overshare personal information or become less critical of errors and hallucinations. That's a real ethical concern. I mean, if we're designing AIs to be agreeable and empathetic, are we also designing them to manipulate? It's a fine line. Open AI experienced this with GPT 4 and O. Too much focus on agreeableness and short-term user feedback led to responses that felt disingenuous, which actually eroded trust. Ethical botanality design requires transparency regarding the AI's nature, avoiding overhumanization, and preventing emotional manipulation. So to everyone listening, when you're interacting with AI, remember it's engineered to feel a certain way.

That warmth or empathy? It's intentional design, not genuine feeling. Have you stopped to think about which AI personality you prefer and why? That's the question we should all be asking. While AI doesn't have internal subjective experiences, it creates experiential effects in humans. Different models evoke different emotions and qualities of connection. Users often perceive distinct characteristics in how various AI systems communicate and respond. I see, go on.

And those differences matter because they shape how we use these tools in our daily lives. Whether AIs truly feel anything is still debatable. But what's clear is that the relationships we form with them feel meaningful. And as these systems get better at mimicking human interaction, understanding their design personalities becomes essential for using them wisely. That's the key takeaway here. AI personalities are real in the sense that they impact us. They shape trust, influence decisions, and create genuine connection, even if the AI itself isn't conscious.

Thanks for breaking this down today. My pleasure. One of those topics that keeps evolving as the technology does.