Research Findings

Impact of Structured Technical Profile on Students Socio-technical Reasoning

Our current research examines how providing students with structured technical information about a robot influences their reasoning about real-world robot incidents. In Fall 2025, 81 undergraduate students in an introductory engineering course analyzed an incident involving Digit, a humanoid robot, before and after reviewing a Robot Structured Technical Profile. The profile summarized the robot’s hardware, software, capabilities, limitations, stakeholders, and safety considerations.

Students’ responses became more technically specific

The largest changes were observed in students’ proposed solutions (69%) and in who or what they viewed as contributing to the incident (46%). Smaller changes were observed in who they thought could help prevent similar incidents (30%) and who could be harmed (28%). Before reviewing the profile, students more often proposed broad responses such as increasing robot testing. After reviewing it, new responses included more specific ideas related to improving sensors, battery performance, and safe failure responses, as well as greater attention to technical contributors such as hardware issues and robot designers.

Students found the profile useful and informative

Students generally responded positively to the profile. Seventy-seven percent agreed that it was useful for understanding the incident, 75% agreed that it provided new information, and 70% were satisfied with it overall. Students most often identified information about the robot’s limitations as useful. They also suggested areas for improvement, including more information about failure behavior, software, safety mechanisms, sensor data, failure logs, and performance limits.

Why this matters

These initial findings suggest that providing structured technical information alongside a robot incident may support more technically specific reasoning about responsibility, safety, and possible interventions. They also provide direction for refining the profile and for studying how this approach works with other AI and robotic systems and with students in different educational settings.

Change in Student's Knowledge, Attitudes and Perceptions due to the GenAI Module

Key Findings from GenAI Module

After the intervention, people's views on AI shifted in some interesting and even seemingly contradictory ways. Participants became noticeably less likely to see AI as something that will "provide opportunities" (odds ratio = 0.355, p < .001) and less likely to report feeling uneasy about the future (odds ratio = 0.371, p < .001) — both meaningful drops. At the same time, people became much more likely to agree that "AI is dangerous," with the odds of holding this view more than quadrupling after the intervention (odds ratio = 4.382, p < .001). Views on whether AI offers "many benefits" (odds ratio = 1.041), whether "society will benefit" (odds ratio = 0.791), and whether "people will suffer" (odds ratio = 1.186) didn't change in any statistically reliable way (all p > .10).

Taken together, these results suggest something a little counterintuitive: people came away seeing AI as more dangerous, but at the same time, less worried about the future. One possible explanation is that the intervention helped people think more critically and specifically about AI's risks, without necessarily leaving them more anxious overall — in other words, a shift toward clearer-eyed awareness rather than heightened fear. That said, this is just one interpretation, and more research (including some follow-up conversations with participants) would help confirm what's really behind this pattern.

Changes in Attitudes

After the intervention, people's views on AI shifted in some interesting and even seemingly contradictory ways. Participants became noticeably less likely to see AI as something that will "provide opportunities" (odds ratio = 0.355, p < .001) and less likely to report feeling uneasy about the future (odds ratio = 0.371, p < .001) — both meaningful drops. At the same time, people became much more likely to agree that "AI is dangerous," with the odds of holding this view more than quadrupling after the intervention (odds ratio = 4.382, p < .001). Views on whether AI offers "many benefits" (odds ratio = 1.041), whether "society will benefit" (odds ratio = 0.791), and whether "people will suffer" (odds ratio = 1.186) didn't change in any statistically reliable way (all p > .10).

Taken together, these results suggest something a little counterintuitive: people came away seeing AI as more dangerous, but at the same time, less worried about the future. One possible explanation is that the intervention helped people think more critically and specifically about AI's risks, without necessarily leaving them more anxious overall — in other words, a shift toward clearer-eyed awareness rather than heightened fear. That said, this is just one interpretation, and more research (including some follow-up conversations with participants) would help confirm what's really behind this pattern.

Following the intervention, participants' views on generative AI shifted in a few notable ways. People became significantly less likely to see GenAI as "beneficial" (odds ratio = 0.437, p < .001), and dramatically less likely to report feeling "concerned about the future" (odds ratio = 0.038, p < .001) — this second shift is especially large, suggesting a big drop in future-oriented worry. There was also a marginal decrease in being "impressed by potential" (odds ratio = 0.696, p < .10), though this one is a weaker, borderline effect and should be interpreted more cautiously. On the other hand, concern about "environmental impact" increased significantly (odds ratio = 1.576, p < .05). Two items — willingness to "suggest others use GenAI" (odds ratio = 0.939) and viewing "wide use as concerning" (odds ratio = 0.774) — didn't show statistically reliable changes. 

This pattern hints that the intervention may have prompted more grounded, specific thinking about GenAI's trade-offs (recognizing real costs like environmental impact) rather than either uncritical enthusiasm or generalized dread.

Environmental Disconnect

The results revealed a striking contrast between participants' factual knowledge of AI's environmental costs and their personal sense of environmental agency. Across all measured metrics, knowledge of GenAI's footprint increased, driven by a sharp surge in awareness surrounding electricity consumption (odds ratio = 2.703, $p < .001$). Smaller, suggestive increases were also observed regarding awareness of freshwater cooling, carbon footprints, and non-renewable resource usage.Conversely, self-reported personal sustainability attitudes moved in the opposite direction. Following the intervention, participants were significantly less likely to agree that everyone shares responsibility for protecting the environment (odds ratio = 0.431, $p < .001$). They also reported lower concern about long-term impact and fewer personal sustainability actions. This divergence hints at a potential "scale-of-the-problem" effect: as individuals learned about the massive industrial energy demands of AI systems, their own individual conservation habits may have felt far less impactful by comparison.

Personal and Economic Concerns

When examining personal and social perspectives, the intervention sharpened specific fears without causing a broad collapse in trust. Participants became nearly twice as likely to agree that GenAI frightens them personally (odds ratio = 1.809, $p < .001$) and showed a notable increase in economic concern regarding job displacement (odds ratio = 1.601, $p < .05$). Interestingly, general attitudes toward trust, developer responsiveness to public needs, the overall helpfulness of AI, and the necessity of governance and monitoring remained statistically stable. This indicates that the shift in mindset was not a widespread rejection of the technology, but rather a targeted recognition of personal and economic vulnerabilities.

This pattern fits quite well - rather than a broad swing in one direction, the intervention seems to have sharpened specific fears - personal apprehension ("scares me") and economic concern (job displacement) - while leaving more general attitudes about trust, developer intentions, helpfulness, and oversight largely unmoved. It's a useful complement to the earlier findings, since it suggests the shift toward "clearer-eyed" or more risk-aware thinking shows up as targeted, concrete worries rather than a uniform increase in anxiety across the board.

The results show a clear and consistent pattern: after the intervention, participants' knowledge and awareness of GenAI's environmental impact increased across the board. The strongest gain was in awareness of "electricity usage," where the odds of higher awareness more than doubled (odds ratio = 2.703, p < .001). Awareness of "freshwater cooling" (odds ratio = 1.554, p < .10), "carbon footprint" (odds ratio = 1.658, p < .10), and "non-renewable resource use" (odds ratio = 1.548, p < .10) also increased, though these three effects were more modest and only reached borderline significance, so they should be read as suggestive trends rather than firm conclusions.

Key Takeaways

• Overall, the intervention successfully transformed how participants evaluate gen-AI. 

• It reduced vague anxiety and uncritical optimism, equipping individuals with a clearer, more specific awareness of the technology's costs - particularly regarding job displacement and energy consumption.