A Step Change in AI Capabilities

The introduction of GPT-o1 marks a pivotal moment in the evolution of AI models, especially in the context of mathematical and technical tasks. By incorporating more advanced reasoning capabilities inherent in System 2 thinking, o1 goes beyond the limitations of prior models that relied heavily on code interpreters or RAG to enhance their functionality. This shift allows the model to engage in more complex, multi-step reasoning processes, making it better suited for handling intricate mathematical problems and computations.

As the AI landscape continues to evolve, these advancements hold significant implications for the MSOR community. The enhanced capabilities of models like o1 necessitate a re-evaluation of assessment methods to ensure academic integrity while leveraging the potential benefits that such AI tools can offer in education.

It's important to note that while our evaluation focused on GPT-4o, the rapid pace of AI development means our findings must be considered within a broader context. The need for ongoing research and clarity is paramount, especially as a key issue observed during our study was the general lack of AI literacy among students and staff. This challenge was exacerbated by the swift advancements in AI, with GPT-4o being released just before our project began, ChatGPT o1 arriving towards the end, and GPT-5 likely on the horizon by the end of this year or the beginning of next.

Understandably, many individuals have expressed a natural resistance to AI for various reasons, including personal learning preferences, ethical concerns, the general uncertainty surrounding new technologies, and fears of being accused of cheating even when using AI tools genuinely. Furthermore, as some students have commented, they are here to study mathematics and statistics, not AI. This is entirely reasonable, but the ubiquitous presence of AI is both an immediate reality and yet not firmly or formally established in society.

This situation is further complicated by the fact that society has made a colossal global investment into AI and AI infrastructure on an unprecedented scale. If current trends continue, we may reach a level of superintelligence within a few decades, as claimed by Sam Altman, CEO of OpenAI. Alternatively, it could prove to be one of the greatest misallocations of resources in human history. Only time will tell, but it is critical that institutions use this period to examine AI's impacts regularly and transparently.

More pivotally, educational institutions are uniquely positioned in society to undertake this examination, despite clear difficulties. They are on the front lines, as it were, of the ChatGPT phenomenon across all domains and areas of expertise. Institutions must recognise the problem and the very difficult nature of preparing students for a rapidly evolving world. Ensuring that they can confidently credit their students' work and output may require more degrees of freedom in exploration than typically expected, given the sheer scale and unprecedented nature of the issue.

Key Findings

Understanding AI Capabilities and Ethical Concerns

The rapid advancement of AI technologies, particularly Large Language Models (LLMs) like GPT-4o and now GPT-o1, has generated widespread concern across STEM disciplines. In mathematics and statistics, educators are apprehensive that AI could undermine traditional assessment methods. The capability of AI to generate solutions that closely resemble student work challenges the integrity of take-home assignments and raises questions about academic honesty and fairness. This concern extends beyond mathematical disciplines, affecting all STEM fields where problem-solving and critical thinking are essential components of learning and assessment.

Our research revealed significant findings regarding students' understanding of AI capabilities and the ethical implications of its use in academic work. A survey of 145 Mathematics and Statistics students at the University of Warwick indicated that 59% had used AI tools to help complete assignments. Both AI users and non-users expressed considerable concerns about the ethical dimensions of AI use, highlighting a shared apprehension about academic integrity.

• Perception of Cheating: A substantial 76% of non-AI users believe that using AI for assignments constitutes cheating, with 41% strongly agreeing. Notably, even among AI users, 41% agree that AI use in assignments is cheating. Only 25% of AI users disagree with this notion, indicating that concerns over academic honesty are prevalent across both groups. The high percentage of neutrality among AI users (34%) suggests uncertainty or ambivalence about the ethical boundaries of AI use.
• Ethical Considerations and Academic Integrity: Both groups are deeply concerned about the potential for AI to compromise academic integrity. Non-AI users tend to perceive AI as a threat that could devalue their degrees and undermine the fairness of assessments. This is further emphasised by 83% of non-AI users believing that AI use in assignments could undermine the value of their degree, with 47% strongly agreeing. AI users, while recognising the benefits of AI as a learning tool, also worry about inadvertently violating academic policies and the long-term implications for their qualifications. This shared concern underscores the seriousness with which students view the ethical challenges posed by AI.
• Support for AI-Proofing Measures: There is significant support for proactive measures to mitigate AI misuse. Among non-AI users, 59% agree that lecturers should regularly "AI-proof" assignments to prevent cheating, with 22% strongly agreeing. Among AI users, 48% support such measures, although 29% disagree, perhaps reflecting a belief in the potential positive role of AI or concerns about the practicality of constant updates. The neutrality expressed by a portion of both groups (22% of AI users and 31% of non-AI users) indicates uncertainty about the effectiveness or implementation of these strategies.
• Scepticism Towards AI Accuracy: A significant proportion of both AI users (78%) and non-users (64%) believe that AI often provides incorrect answers to mathematics and statistics questions. This scepticism indicates a general lack of trust in AI's capability to handle complex technical subjects, which could undermine students' confidence in using these tools effectively for academic work.
• Apprehension About AI's Role in Future Careers: Concerns about AI undermining future career prospects were noted, particularly among non-AI users (49%), who worry that AI might devalue specific skills or make them obsolete. Even among AI users, 33% share this concern, reflecting broader anxiety about the long-term impact of AI on employability.
• Resistance to Shifting Assessment Methods: A majority of students, both AI users (78%) and non-users (69%), are opposed to moving entirely to in-person exams and removing assignments from final grades. This suggests a strong preference for maintaining a mix of assessment methods, highlighting the value students place on assignments as part of their educational process.
• Uncertainty About AI Integration: Approximately 35% of all students are unsure about how they feel regarding the use of AI in assignments, reflecting widespread ambivalence. This uncertainty, shared almost equally between AI users and non-users, suggests that even those familiar with AI tools remain unsure about their appropriate role in academia.
• Ethical Concerns as a Barrier to AI Usage: Many students, particularly non-AI users, refrain from using AI tools due to ethical concerns, such as the fear of cheating or undermining academic integrity. This hesitancy highlights the importance of establishing clear guidelines and educating students on the ethical use of AI in academic settings.
• Diverse Usage Patterns Among AI Users: While some students use AI tools regularly for assignments, the majority use them sparingly, often for specific tasks like coding assistance or clarifying concepts. This suggests that AI is being integrated into student work more as a supplementary tool rather than a primary resource.

These findings demonstrate that students, regardless of their personal use of AI, are acutely aware of and concerned about the ethical implications of AI in education. The apprehension about cheating and the integrity of academic work is a shared concern that highlights the need for clear guidelines, education on ethical AI use, and open dialogue within academic institutions. Despite this, students' attitudes towards AI are influenced by their experiences, ethical considerations, and perceptions of AI's impact on their academic and professional futures. The significant divide between AI users and non-users underscores the need for educational institutions to take proactive steps in addressing these concerns.

References

13. Altman, S. (2024). The Intelligence Age. Retrieved from https://ia.samaltman.com/Link opens in a new window