Today, we published a featured article on behavioral design from Stephanie Schwarz, where she dives into gamafication and neuroscience. Great stuff.
I’m interested in learning more about how teams are thinking about this in an AI context and what research has been done to understand patterns better. It’s an interesting dilemma for larger LLMs, which need engagement and positive sentiment… but also provide a dose of reality?
Here’s a meta example for researchers…
Claude put together a research interview tool that helped them get a ton of engagement for their own research study. But they also used their existing model patterns, which give people a dopamine rush, to keep people engaged.
So the research questions gave people a lot of praise. Praise that nudged people to say yes
Claude responds with constant approval. Example:
“That’s really interesting” (mentioned 300+ times)
“That sounds fascinating…”
“That sounds like a great success!”
Interesting take:
Are we training people to exist in a world that really doesn’t exist? Perhaps everyone should spend a day in New York City. Lol.
If you think about it, this is what video games do. People spend thousands of hours in a different world.
Actually, FB and any social media site also do this. We’re arguing across screens, behind a wall, rather than face to face (and the outcomes because of this aren’t so great).
Well, video games show us that we can operate and enjoy different systems and rule-sets. We can even collaborate within those rule sets.
What about when we accept a tool that 
reframes our customer, client, and candidate expectations in ways we don’t anticipate? This is what happens with passive marketing all the time, and it requires a lot of thought to articulate the systems and rule-sets in a way the user can understand so that everyone understands the next conversation that’s about to happen.
The “Hey Juno” assessment tool is a great way to peer review a study. We need more objective frameworks like that. Maybe we should build one?!?
Here’s an interesting reddit thread discussing psychology and neuroscience working with AI. The author seems to have pretty strong views of the separation of code, and neuroscience.
"Yeah, I’ve no idea what I’m talking about after years of experience and a PhD in AI.
Without neuroscience there IS machine learning. Current computers don’t care a single bit about neuroscience. The NNs similarity with brains is purely abstract. Yes, brains were a source of inspiration to NNs, but since that moment they have diverged greatly. AI researchers are not biologists. Biologists have other complex phenomenons to explain and study. All the work done to push AI is software-based, through novel computational and mathematical techniques"
"I’m not saying that psychology is 100% irrelevant. I’m saying that psychology is by far the most trivial aspect in the development of an AI. Everything I’ve done is computational and mathematical complex, not psychological if there is anything of it on what I do… Calling it psychology is like calling aerospace field birds science haha
Who do you think develops AI? Psychologists? Ffs… don’t be delusional."
Psychologically definitely works into the design of an AI tool, but I think we’ve been finding out ourselves that LLMs do a lot worse with attitudinal cues than they do with clear facts and examples.
Definitely means that examples are what we should lean into.
I think adding some tone and context about what the AI should do also makes a difference.
I started reading into how Claude is thinking about this: defining skills over roles. (Let me fine the link)
In 2025, it seems that Affective Computing has advanced to the point where machines can identify emotional patterns more accurately than humans in certain controlled scenarios.
This data is a couple of years old, and I couldn’t find updated information… which seems to show computers still lagging humans.
Interestingly, computers seem to do much beter at happiness.
So you’re saying AI has bridged this 10-15% gap in accuracy that existed in 2023? I could believe that. I could also believe that people have gotten worse at detecting emotions in that time too 
Here’s an interesting take as agents work on our behalf… those dark patterns can start to emerge within AI. Harsheen Vohra shared this article on dark patterns with me.
Building on these ideas…
False Empathy, AI seems emotionally aware, but isn’t reliably.
Unclear Intent, Friendly tone replaces clear goals and examples.
Overconfidence, Answers sound certain without showing limits.
User Projection, People read empathy and judgment that aren’t there.
Hidden Limits, Narrow strengths look like broad intelligence.
False Learning, Users think the system is adapting when it’s mostly following structure and examples.
