AI Speeds Up Innovation, but Sustainability Still Needs Human Judgment

A new study of engineering students shows that AI is becoming a routine design aid, particularly for ideation, project structuring and working under tight deadlines. However, it also reveals a persistent weakness in sustainability education: students can often identify which Sustainable Development Goal a project supports, yet struggle to convert that alignment into technical requirements, indicators and verifiable impact.

Published in Knowledge, the study "Engineering Innovation: Impact of AI and SDGs in a Challenge-Based Environment" was authored by Sandra Delfa-Baena, Maria Boluda-Prieto, Carla Terron-Santiago and Angel Sapena-Bano. The research followed 60 students during the INDUSHack engineering challenge at the Universitat Politècnica de València, offering a real-time view of how AI and sustainability interact under pressure.

AI Became the Default Co-Designer

Most participants arrived with prior AI experience: 45% used it frequently and 43.33% occasionally. ChatGPT was known or used by every respondent before the event, showing how rapidly conversational AI has entered academic workflows.

During INDUSHack, 93.22% used ChatGPT, while Copilot and image-generation tools played much smaller roles. AI use concentrated where uncertainty was highest: 56.67% used it during early ideation and 68.33% during solution development, compared with 36.67% in the final presentation stage.

The main applications were idea generation, reported by 61.02%; solution structuring, at 45.76%; and impact evaluation, at 38.98%. Students appeared to treat AI less as an automatic answer engine than as a rapid thinking aid—a way to explore alternatives, organise material and maintain momentum under pressure.

Their perceptions were strongly positive. A combined 90.74% rated AI's effect on creativity as good or excellent. The same proportion believed it improved solution quality, either essentially or in certain aspects. Some 72.22% said it reduced development time, and 80% would recommend AI in future hackathons without reservation.

These findings explain why generative AI is spreading through engineering education. It helps students process information, compare options and communicate decisions within tight deadlines. But the study does not prove that AI improved technical quality. Its design was descriptive rather than experimental, there was no non-AI control group, and most gains were self-reported.

Sustainability Exposed the Harder Learning Gap

The SDGs presented a more demanding challenge. Before the event, 62.5% said they should be an essential part of engineering work, while 32.14% agreed they mattered but were unsure how to implement them.

Awareness did not equal operational competence. Only 21.43% said they knew most SDGs and understood how to apply them in engineering projects. Nearly a quarter had never worked on a project involving the goals, and 26.79% had only basic experience.

During the hackathon, more than 90% of teams identified a relevant SDG or were actively aligning their solution with one. Yet only 34.48% designed around specific SDG targets and expected impact. Another 44.83% reviewed targets to guide the project, while 17.24% identified a goal without analysing its specific targets. The problem was not persuading students that sustainability matters. It was helping them convert a broad global framework into engineering requirements.

The translation requires evidence. What indicator will show that a proposal lowers emissions, improves access or reduces resource use? What baseline will be used? Who benefits, who bears the cost and over what period? A project can appear aligned with an SDG while offering no credible way to measure progress.

Students recognised the difficulty. Practical application was the most frequently cited barrier, reported by 38.89%, while half said they would need more training before applying the SDGs in future projects.

Project themes also clustered around goals that are easier to express technically: health, clean water, clean energy and sustainable cities. Social and institutional goals were less prominent. This reflects a wider risk in engineering education—reducing sustainability to environmental efficiency while neglecting affordability, inequality, governance and accessibility.

Mentors, Not Machines, Made the SDGs Operational

AI adoption was largely autonomous. SDG integration was not. Mentoring and external evaluation pushed students from broad claims toward clearer justification. Two-thirds said mentors influenced SDG integration "quite a bit" by providing specific ideas, while 10.53% said they could not have integrated the goals properly without that support. Mentors encouraged teams to connect sustainability claims to design logic rather than merely attaching an SDG label.

The expert committee added accountability. Thirteen faculty members with engineering and sustainability experience assessed the proposals using common criteria. They found that 53.33% had a clear and well-aligned SDG focus, while 33.33% demonstrated moderate impact. Only 6.67% were judged to have great potential impact with a strong sustainable approach.

This external assessment strengthens the study because it moves beyond student perceptions. It suggests that sustainability was visible in final proposals, but also shows that deep integration remained uncommon.

The gap points to the danger of SDG labelling: using sustainability language without changing the design itself. Structured mentoring can counter that risk, especially when combined with rubrics requiring teams to map SDG targets to performance indicators, justify trade-offs and identify unintended consequences.

The study reveals a useful division of labour. AI supports exploration and structure. Mentors provide challenge, context and judgment. External evaluators create accountability. Effective learning emerges from their interaction, not from technology alone.

The Curriculum Lesson: Pair Speed With Accountability

Universities should not treat AI literacy and sustainability as separate additions to an already crowded curriculum. They should integrate both into the same design process. Students need to learn how to frame prompts, verify outputs, check sources, identify bias and disclose where AI influenced a decision. They also need repeated practice translating sustainability goals into specifications, measurable indicators and defensible trade-offs.

Challenge-based formats can help because they expose students to ambiguity, teamwork and time pressure. But hackathons should be connected to preparation and reflection. Before the event, students need guidance on SDG targets, responsible AI use and evaluation criteria. During the challenge, mentors should conduct verification checkpoints. Afterward, teams should reassess technical feasibility, sustainability claims and AI's role in shaping the solution.

The study's limits should temper generalisation. It covered one event, one university and 60 self-selected participants. Outcomes relied heavily on self-reporting, and no causal claims can be made about learning gains. The short format also favours rapid ideation over technical depth. Future research should compare AI-supported and non-AI teams, use objective performance measures and track whether students retain these competencies.

For developing countries, the model offers promise and caution. AI-supported challenge learning could expand access to design assistance and connect engineering education with local development problems. Yet unequal internet access, subscription costs, limited mentor availability and imported AI systems that overlook local knowledge could reproduce existing disparities.

The policy implication is not to slow AI adoption, but to govern it deliberately. Universities, accreditation bodies and employers should define standards for verification, transparency and responsibility. Development agencies and industry partners can support open tools, faculty training and locally grounded challenges.