AI research is one of the most competitive career paths in tech. There are two viable routes — the academic path and the industry path — and they require different strategies, different timelines, and different evidence of capability. This guide covers both in depth.
Path 1: The Academic Route
The traditional path to a research career runs through a PhD. In the UK, PhD programmes in AI and machine learning typically run 3–4 years, often funded by EPSRC, industry partnerships, or university scholarships.
Choosing a supervisor and institution: The most important decision in a PhD is your supervisor. The institution matters less than the supervisor's research quality, their publication track record, and whether their research direction aligns with yours. That said, institutions with strong AI research groups include the University of Edinburgh School of Informatics, Cambridge Machine Learning Group, UCL AI Centre, Oxford Computer Science, and Imperial College London.
The postdoc question: Most academic research careers include one or two postdoctoral positions (1–3 years each) before a permanent role. Industry labs have reduced the postdoc requirement somewhat — Google DeepMind and Microsoft Research Cambridge hire directly from PhDs — but it remains common for competitive positions.
Funding: The EPSRC funds many UK AI PhD studentships. Industry-sponsored CDTs (Centres for Doctoral Training) provide funding with industry collaboration. The Alan Turing Institute offers enrichment schemes for current PhD students.
Path 2: The Industry Route (Without a PhD)
Industry labs like Google DeepMind, Microsoft Research, and Samsung AI Centre hire Research Engineers — roles that contribute to research through implementation, experimentation, and systems work rather than hypothesis generation. These roles are genuinely open to strong engineers without PhDs.
The key to making this transition is producing work that demonstrates research ability, not just engineering ability. Specifically:
- Implement and extend published papers — reproduce a paper's results, then add an experiment the authors didn't do. Write it up clearly. This demonstrates you can engage with research literature at a productive level.
- Develop ML fundamentals depth — not enough to have "used PyTorch". You need to understand backpropagation, attention mechanisms, optimisation theory, and statistical learning theory deeply enough to debug novel problems.
- Publish or produce research-quality outputs — even a workshop paper or a detailed technical blog post analysing an underexplored aspect of a model demonstrates research thinking. NeurIPS, ICML, and ICLR workshops are accessible entry points.
The fastest path in: research internships
Research internships at industry labs are the single fastest route to a research career. Google DeepMind, Microsoft Research Cambridge, and other UK labs hire research interns from PhD programmes and, selectively, from exceptional industry candidates. Strong intern performance leads to return offers. Apply 6–9 months before your intended start date — the process is long.
Building the Skills: From "I Understand Papers" to "I Can Produce Research"
There's a large gap between understanding papers and producing research-quality work. To close it:
Phase 1 — Paper comprehension: Read 2–3 papers per week in your area. Use tools like Semantic Scholar and Papers with Code. Keep a reading log with your notes on each paper's contributions and limitations.
Phase 2 — Implementation: Reproduce paper results from scratch in PyTorch or JAX, without looking at the authors' code first. This builds deep understanding that reading alone doesn't.
Phase 3 — Extension: Take a reproduced paper and run experiments the authors didn't. Ablations, different datasets, modifications to the architecture. Write up what you find, even if the result is null.
Phase 4 — Novel contribution: Identify a gap in the literature or an unexplored direction. This is the hardest step and the one that distinguishes researchers. It requires enough domain knowledge to recognise what hasn't been done yet.
UK Universities Worth Targeting for AI Research
If you're planning a PhD, these institutions have particularly strong AI research communities with good industry connections and publication track records in competitive venues:
- Edinburgh: Historically one of the UK's strongest ML departments. Strong in NLP, RL, and probabilistic methods.
- Cambridge: Machine Learning Group and nearby DeepMind creates good ecosystem for PhD students.
- UCL: AI Centre and DeepMind collaboration. Strong in RL (DeepMind co-founders have UCL connections).
- Oxford: Strong in NLP (Future of Humanity Institute), computer vision, and AI ethics.
- Imperial College London: Growing AI research capacity, particularly in healthcare AI.
See the full AI Researcher career guide
Salary data, required skills, UK research labs, and career path from PhD to principal scientist.
Frequently Asked Questions
Do I need a PhD?
For Research Scientist roles at top labs, yes — it's very strongly expected. For Research Engineer roles, strong ML engineers with demonstrated research contributions can be competitive. The PhD is the realistic route if you want to do independent research and publish.
Can I get a research role from an industry ML background?
Yes, via the Research Engineer path. The key is producing work that demonstrates research ability: implementing and extending papers, contributing to research-quality experiments, and producing written outputs that demonstrate scientific thinking.
Which UK university is best for AI?
The supervisor matters more than the institution. Among institutions with strong AI research communities: Edinburgh, Cambridge, UCL, Oxford, and Imperial College London are consistently strong.
How long does the path take?
Academic: PhD (3–4 years) plus often a postdoc (1–3 years). Industry route: 3–5+ years of strong ML work plus demonstrated research output. Research internship to full-time is the fastest path: 6–12 months.
What's the difference between Research Engineer and Research Scientist?
Research Scientists lead research directions and are evaluated on scientific contributions and publications. Research Engineers implement research ideas and build research infrastructure. Research Scientists almost universally have PhDs; Research Engineers frequently do not.