What Students Say About AI-Human Relationships
Many of you know that I am an educator, and currently I teach at Oregon State University, where I direct a mental health training program. However, one of my joys is teaching undergraduate students. Undergrads come from such diverse backgrounds, and the earlier you catch them in the educational process, the more you can foster a wonderful diversity of thought and experience. So this last term I reprised a course here titled Human Sexuality. This class is a mix of sexual education, human development, social psychology, and (as it’s me) a healthy dose of emotional skills as they apply to intimate relationships. Hot on the heels of the announcement that Dr. Sherry Turkle will release a book on AI and intimacy, I chose to spend the last class period discussing AI-human relationships.
I offered students in my class this piece from the New York Times from about a year ago. Sorry for the paywall there, not my policy. In short, it was one of the first large-format news stories about human romantic relationships with AI tools. There have been many pieces before and since. Still, this one is a good entry point for the discussion, as it has just enough detail to be ‘meaty’ yet entry-level, so we aren't diving into the psychological theories underpinning this topic. Students read, spoke in small and large groups, and then did some writing. Like a giant feelings nerd, I then took their writing (which was astounding - great job, class!) and did some thematic analysis on it. For this discussion, I won’t be quoting anyone or identifying anyone in the class. They were an amazing group of students and deserve privacy.
So let’s start by saying that this undergraduate social science course on human sexuality probably creates a slightly biased sample. Even though I can say that the class had a very diverse range of majors, it was skewed in some interesting ways… not to mention that we are in a college classroom, which creates its own biases. Even with that said, the sample size of responses was pushing 70, and I can tell you that the class included people with fairly diverse perspectives on sex and sexuality. That is all to lay the groundwork for the fact that, in this sample of mostly white, middle-class, femme-presenting students, folks were absolutely unified in their belief that relationships require humans to be valid.
This is pretty wild as it contradicts the foundations of Turing-influenced computer science theory. The concept that a computer could convincingly mimic human interactions was not convincing the bulk of my students (even under direct questioning from me) that human-AI connections were being described as relationships. Probing the why behind the rationale that robots aren’t able to hold relationships, I found that students largely described these relationships as false because someone had programmed them to respond in a specific way (even algorithmically) and that they inherently lacked mutuality. Additional students commented that these relationships could not convincingly qualify as offering consent (something I was overjoyed to see as I hammered positive affirmative consent in class pretty hard). Really, the base issue that I saw in student responses was that AI “isn’t real” and cannot provide mutual support. When I pressed on this in class, what I heard beneath it was that AI tools cannot receive care and support.
This was pretty staggering, as students often responded that AI was designed to reflect (which it is) and to reinforce thinking (which it is), but that it cannot generate its own honest opinions (which it kinda can, but it really struggles with it). Deep in this discussion, what I heard was that, even if it could generate something like an opinion, it still could not receive the benefits of a relationship.
Thinking about this alongside my own work and conversations about these tools. Right now, we have this amazing tool that can mathematically generate all manner of neat things. Its output capability continues to advance in a linear, consistent way. And it continues to come up with things that are novel to the point of being world-shatteringly scary—the reflective capacities of these tools even staggers deeply trained philosophers. However, there is something inherently uncanny about AI tools. I’m not going to pile on Richard Dawkins in this article (if you know, you know; if not, go check it out). However, I have several brilliant friends who are computer scientists, and I look forward to discussing this with them. What I am coming to in looking over this data is the deepest part of the silence in the students' responses: no matter what psychology, research technology, and computer science say, students do not have a self-conception of AI-human interactions as relationships.
This is particularly important as self-conceptualization has a powerful impact on the quality, duration, and nature of human relationships. This is documented in research and is something I see in couples counseling all the time. As a therapist, I can tell when people are in the throes of thorny relationship problems when they cannot define the nature of their relationship to begin with. Married couples who are questioning what their marriage means are in a hard place to start work (and I end up starting WAY TOO MUCH) and often can be a predictor for relationship failure. My personal favorite is the Facebook relationship definition “it’s complicated,” which I have watched blow up so many relationships that it has become a near-hilarious coded way of saying “we're breaking up” without actually saying that out loud.
Self-conceptualization isn't the only way to consider relationships; certainly, there are ample experiments where how easily we can fool humans into thinking they are talking with a human, trailing all the way back to Eliza. However, when relationships extend, when we are knowingly engaged with an AI tool, we are moving into the space where how we think about our relationships starts to matter as much as how we engage with them. This self-reflective pattern is vital in sustaining and determining how relationships affect us, what they mean, and how we think about and talk about them with others. To this end, self-conceptualization is vital to integrate into our thinking about AI-human relationships. And from what I can tell from my class this term (reinforcing other works), students are not buying that AI relationships are genuinely relationships.
