The Human in the Middle
People ask me fairly often what it actually means to be a Behavioral AI researcher. Is that even a real thing? Is it just organizational behavior research with AI thrown in? Is it HCI? Is it AI ethics? It is none of those, though it touches all of them. And the question got me thinking about what actually defines this kind of research — not just the topic, but the way of approaching it.
So what does it actually mean to do this kind of research well? I keep coming back to five things: five ways of approaching this work that I think creates meaningful and impactful research. A Way of Thinking, which is about asking a different question than everyone else is asking. A Way of Doing, which is about designing studies that can see what standard methods miss. A Way of Being, which is about having the intellectual honesty to follow what the data says even when it complicates things. A Way of Relating, which is about staying close enough to real organizational life that your findings actually connect to something people are experiencing. And a Way of Mattering, which is about making sure what you discover travels somewhere beyond the page it is published on. Over the next several posts I want to unpack each one. Today I want to start with the first, because without it the other four do not quite make sense.
The Way of Thinking in BAI research is really one move: stop measuring what AI produces, and start asking what is happening to the person producing it.
There is a great deal of excellent research asking what AI does to the task. Does it improve output quality? Does it reduce errors? Does it make people faster? These are important questions and they produce real, useful answers. BAI research does not replace that perspective. It adds one that tends to get missed. It looks past the result — past the document, the decision, the product — and asks what is happening to the human on the other side of it.
Let me give you a concrete example. Imagine you have been using AI to help you write for a few weeks. Reports, emails, posts. It goes smoothly. The work looks good. Then one day you sit down to write something on your own, and you stare at a blank screen. The words do not come the way they used to. You probably blame the day, the topic, the mood. You do not think to connect it to the weeks of AI-assisted writing that quietly let a muscle go unused.
That blank screen is an invisible cost. And it is exactly what BAI research is designed to find, precisely because it is not visible in the output.
This is the perspective BAI research brings: instead of asking what makes me a better, faster writer, ask what makes me a better writer because I am genuinely collaborating with AI rather than just using it to get things done. The difference sounds subtle. It is not. One focuses on the quality of the output. The other focuses on what is happening to the person producing it. And once you start asking that second question, you start seeing things the first one was never looking for.
Let me give you an example from our own research. For years, the assumption in human-AI collaboration research was that if you wanted people to engage with an AI teammate, you needed to make the AI feel present — give it a name, give it a voice, make it seem more human. Social presence was the variable everyone was trying to maximize. Anthropomorphic design was the lever. That assumption made intuitive sense, and it generated a rich body of work. What BAI research found was that the mechanism underneath was more interesting than the surface feature.
Across three studies, including one using psychophysiological measures, we found that social presence did not directly drive motivation to contribute to the team. The path was more indirect — social presence influenced whether people became willing to depend on the AI, which shaped their commitment to the team, which then moved motivation. When that chain was intact, the relationship held. When it was broken, a socially present AI did not help.
And the design finding was the real surprise. Making the AI appear more human did not reliably build that chain. What built it was transparency. Simply introducing the AI early, and explaining clearly what it does and how it works, nearly closed the gap in perceived social presence between an AI teammate and a human one. The mechanism that mattered was not mimicry. It was understanding.
That is what the BAI Way of Thinking produces. You go in asking how to make AI seem more human. You come out asking what actually shapes whether people can trust it, depend on it, commit to working alongside it. The visible feature turned out to matter less than an invisible relational process happening underneath. And you only find that if you are looking for it.
Remember, this is not just a small methodological tweak. This mindset changes your unit of analysis completely. The human is not the conduit between the tool and the result. The human is the phenomenon. There is a whole world of excellent research studying the tool and the result. BAI research studies what is happening in the middle.
#BehavioralAI #BAILab
