Assessing Student Learning in the Generative AI Age

25 September 2026

AI can analyse, identify key points, propose arguments, and structure a response to a prompt. It can, in short, do all the things students used to have to do.

This is a major problem. Cognitive offloading – getting AI to think for you rather than thinking for yourself – makes you stupider. If students actually use AI in the ways AI companies, other software providers, and even some educators actively encourage them to, they will emerge from their studies less rather than more able to think for themselves.

From the universities’ perspective, this risk poses a double threat.

First, it undermines our core educational mission. We are in this to make young people smarter, not dumber. We want our graduates to be more capable when they leave us, not less. We are committed teachers. And we believe deeply in the value of our subjects.

Second, it threatens our business model. If students are losing rather than gaining skills at university, sooner or later someone is going to notice. That might look like employers either blacklisting universities whose graduates have struggled in the workplace, or targeting institutions believed to be doing a better job of protecting their students from the deleterious effects of AI. It might look like governments further cutting funding for university education. It might look like steep declines in enrolment numbers at a time when many institutions – at least in the UK – are already struggling. And it might look like regulators imposing increasingly tough conditions to try to stop the rot. None of these scenarios would be good for us.

In the School of Society and Environment at Queen Mary University of London, we are taking action to protect the intrinsic and extrinsic value of our degrees. We have broken down the ideal student learning journey in the age of generative AI into three distinct stages, and are developing updated classroom activities and assessment diets to align with each one.

At the first stage, students should develop foundational disciplinary knowledge and cognitive skills. Bucking the recent trend, this might mean making less use of technology in all forms, and more offline study – pen and paper, workbooks, and invigilated written exams. This stage ensures all our graduates both develop and demonstrate core academic capabilities without relying on AI.

At the second stage, students should develop the ability to use AI intentionally. That means learning about the distinct strengths and weaknesses of generative AI tools, and how to be selective about whether and when to use AI. It means appreciating the importance of not becoming a redundant “meat proxy” interfacing between an AI and the world. It means being aware of the general economic, social, and political costs of generative AI, and the specific educational downsides associated with using it. Outsourcing cognitive effort to AI is like taking a forklift to the gym. The weights get lifted, but you don’t get stronger. And it means ensuring they can use AI honestly, transparently, and with integrity.

At the third stage, students should explore how to use AI effectively and critically. They should develop a set of prompting skills that enable them to get the best out of different platforms, while minimising financial costs and energy use. They should become sophisticated consumers of AI-generated output, quickly identifying its flaws and shortcomings. And they should reflect on what it means to be human in an AI era.

None of this is straightforward. Middle-aged academics are not well-placed to develop new, advanced technical skills. But educators need to understand what AI can do if they are to do their jobs. And institutions need to back them up – providing facilities for in-person teaching, learning, and assessment, buying access to advanced AI models, developing new adjustments for students with learning differences that do not simply rely on technology, and ensuring staff have the time to learn, experiment, and train.