Jeff Hancock
On Super Data Science: ML & AI Podcast with Jon Krohn
“Right now, if Kate kills it and she's doing really great and super productive because she's got her really good private AI IP, she's going to get rewarded for that. If she shares it with me, there's zero incentive to do that.”
Super Data Science: ML & AI Podcast with Jon Krohn · AI Builders · October 2026
Kate is Hancock's co-author Kate Niederhoffer, who was also on the episode. He was explaining why good AI practice doesn't spread through organizations: nearly everyone works with their AI tool alone, so nobody sees how a colleague got a result, and sharing prompts still feels odd. He named norms and incentives as the two things that have to change.
Transcript
Yeah, a lot actually. I think the big theme in all of our research over the last three or four years has been how social and relational AI usage is. And we maintain a really strong positive belief that AI can transform relationships. It's a transformative technology that can have really positive enhancing effects on relationship if used in a particular way. Most of the foibles that we identify are human issues or investments of the talent infrastructure that have gone awry or that we've left neglected and this powerful technology is amplifying. So yeah, I think there are a few things that we've identified now that have become more important to make sure that we shift the organization from being so overly focused on individual AI adoption to understanding how it can be embedded in a team for productive collaboration and gains. Some of the things are similar to what we've been talking about, like psychological safety, ensuring that that's there, like maintaining eye contact with people and investing in human relationships at the same time as you're investing in the AI technology so that we can maintain that. And we never are in an instance where we're displacing people. Instead, we're using the AI as a mediator to augment the relationships that we have and our own expertise and contributions to that relationship. So we can talk about other ways, but I think that's the gist of it is like being aware of, in fact, how social AI is and how much human potential it takes for it to be effective in the organization. Organization and how it can have these enhancing experiences if you use it with agency, if you're aware of the people around you, if you talk about transparently your usage of the tool so that we can effectively coordinate.
Yeah, somewhere in my research, and I'm just scrolling through to try to find it quickly, but not immediately. Maybe you both will know exactly what I'm talking about. There has been recent research that I think you both published on related to how metaphors of AI indicate that people increasingly perceive AI as warm and human-like. Do you want to tell us more about that paper?
Yeah, that was actually one of our very first projects. So when Kate Niederhoffer I first started working together in 2023, we wanted to track how people were conceptualizing AI because it was like all in the news. Chat GPT was like, it just blew it up. And so we wanted to get in early and track how people were thinking about AI. It's difficult to ask people about it. Like you can't do a scale because, you know, it's messy and it's never stated. So we use a technique where we ask people about metaphors as a way of like kind of surfacing underlying kind of attitudes and fears, hopes, mindsets in some ways. And we found early on it was like, it's like a computer, it's an encyclopedia. And over from 2023 through, I guess was it end of 2024-ish, it was becoming more human. So more anthropomorphic metaphors, think assistant, teacher, child, moving away from computer, synthesizer, encyclopedia. And then, you know, in psychology, we know that when we meet somebody or a thing, like a robot, it goes, we measure something on warmth. So we immediately form a warmth thing and a competence. And we were seeing this track, whereas anthropomorphism also increased, so did warmth. So our kind of like liking it, thinking it was like working for us. It was friendly. It had our interests at heart. And so that changes the way we then interact with it and communicate with it. So for me, listening to Kate Niederhoffer and also thinking about our metaphor work, I think there's a second really big thing going on. I think Kate's right that AI is a multiplayer game or maybe a team sport. And yet, almost all of the AI usage I see with students at Stanford when I meet with Kate Niederhoffer these companies that we talk to, it's all individual. So like people are working with Claude or Copilot or ChatGPT on their own. And then they take whatever product comes out of that and they insert it into the workflow. A lot of times the workflow will be coming along and they pull it out so they can work privately on it and put it back in. And so this is a problem for relation slipping because now I don't know where Kate came up with that. That's number one. Number two is if Kate did something really cool in her development of an agent or prompting it or coming to that product, say the product was really great. That innovation is not observable because Kate is holding on to it as her sort of like private IP, right? And so innovation doesn't diffuse. We know that like there's a couple main things for innovation to diffuse to an organization and observability is one. So how do we change those practices? Like it just feels weird to share your prompts right now. So there's like norm issues. We need to change the norms. And the second is incentives. Right now, if Kate like kills it and she's doing really great and super productive because she's got her really good AI IP, private AI IP, she's going to get rewarded for that. If she shares it with me, there's zero incentive to do that. So I was talking to a CHRO recently and they were talking about how they changed the incentive structure so that they now work in joint teams with agents or well, in particular with a CLOD. And so they can see who's prompting. They see how the prompt, they can understand Claude's output now. And so it really helps deal with the coordination and team trust problems that we see with relation slipping.
I was using the term wrong. You guys were so kind to me that you didn't even correct me. I kept saying relationship slipping.
Well, Kate Niederhoffer I figured out that like after like four or five tries, it actually works. But the first few times, it's a bit like you can also use slippage if you like, John.
I like to figure a sentence that goes, how does this work in a sentence? How's your relation slip going? Or is your relation slip getting worse?