The University of Edinburgh Opens the MSc Artificial Intelligence with a Compulsory Safety and Evaluation Core

The School of Computing at The University of Edinburgh has opened admissions for the MSc Artificial Intelligence. It is a full technical master's — mathematics and probability for AI, machine learning theory and practice, programming intelligent systems in Python, data engineering for ML pipelines, deep learning, natural language processing and large language models, computer vision and multimodal AI, and reinforcement learning.

What the school has made non-negotiable is the rest of the structure. Every student completes modules in MLOps, evaluation and production AI systems, and in responsible AI, safety and governance, before they are allowed to begin the capstone. "We are not going to graduate people who can train a model and cannot tell you whether it is safe to deploy," said the Head of Computing.

The capstone itself must include a data card, an evaluation plan, and a written account of harm, bias and failure modes. A model that performs well in a notebook and has no evaluation plan for the setting it will actually run in cannot pass. Students present to faculty and invited practitioners who are encouraged to ask what happens when the input distribution shifts.

Amina Yusuf, a public-health officer in Lagos who already writes production code, expects to accelerate through the foundational programming and spend the saved time on evaluation methods. "The interesting question in my work is never whether the model runs," they said. "It is whether I can defend using it on real patients."

The programme is delivered online with zero tuition. Eligible modules include GPU-backed notebooks on the learning platform so that students are not told to buy hardware they cannot afford. Live sessions run in more than one time zone and are recorded with transcripts the same day.

Academic integrity guidance for the programme is detailed. Students may use generative tools for scaffolding and exploration, never for silent substitution of their own analysis or for fabricated citations. Any submission that does not match a student's prior work can be examined orally.

Applicants need a numerate first degree or equivalent professional experience, and comfort with programming. The school runs a preparatory short course in mathematics and Python for applicants who need to strengthen foundations before enrolment.

Staff across admissions, registry, teaching and student services are told to design every process as if the default student is studying after a full day's work: recorded sessions, published rubrics, named advisers, and a library desk that does not close at the end of one country's office hours. The University of Edinburgh will not pretend a programme is light. Applicants should plan their weekly hours before accepting an offer and ask direct questions at the open webinars if a page reads as vague.

The University of Edinburgh is a pan-African digital university with no campus to relocate to. Founded in 2009, it publishes news that is operational rather than ceremonial: what changed in a programme, who can apply, how assessment works, and how global students get help. Every programme is delivered fully online; there is no campus, and no student is asked to relocate. Questions about the claims in this article should go to the office named as author above.