Computer Science
the default, most direct routeProvides the algorithms, systems and, at most universities now, dedicated machine-learning coursework that forms the technical base for almost every AI research specialty.
There is no license to practice AI research, and no professional board decides who may call themselves a researcher. The field's real gatekeeping happens later and less formally, at peer review — whether NeurIPS, ICML or ICLR accepts a paper — which is why a PhD, while not legally required, remains the path nearly everyone still takes.
That informality has not stopped a fairly rigid pipeline from forming in practice: a small number of universities and industry labs produce a disproportionate share of the field's most-cited work, and getting into one early still shapes a career more than almost any single later decision.
The route in
-
1
Undergraduate foundations
3–4 yrsA degree heavy in mathematics, statistics, computer science and, increasingly, coursework in machine learning itself, alongside early research assistantships in a professor's lab.
The filterUniversity admission and access to a lab willing to take on an undergraduate.
-
2
Master's or research role
1–2 yrsMany go straight into a PhD; a growing share instead spend a year or two as a research engineer or master's student, building a publication record before applying.
The filterGraduate admissions, or a hiring bar at a lab willing to co-author with a junior researcher.
-
3
PhD
5–6 yrsYears of coursework followed by independent research under an advisor, ending in a dissertation defended before a committee.
The filterA qualifying exam, and finding an advisor with funding and interest in taking you on.
-
4
Building a publication record
3–5 yrs (within the PhD)Papers accepted at NeurIPS, ICML or ICLR function as the field's real currency, mattering more to future employers than the university's name on the diploma.
The filterPeer review at a top-tier venue — the closest thing the field has to a licensing exam.
-
5
Postdoc or first research-scientist role
1–3 yrsSome continue in academia through a postdoctoral position; most now move directly into an industry lab's research organization.
The filterA multi-stage interview process at a lab, or a postdoctoral offer from a specific principal investigator.
-
6
Independent researcher or lab lead
5+ yrsSets an independent research agenda, mentors junior researchers or PhD students, and is judged on the field's direction their work has influenced rather than any single paper.
The filterTenure and promotion committees in academia; sustained impact and internal promotion in industry.
PhD in AI/ML (US, funded route)
$0–$40k
Most US PhD programs in computer science or machine learning fund students through a stipend combined with teaching or research assistant duties — commonly $35,000–$48,000 a year — so direct tuition cost is often close to zero; the larger cost is five to six years of reduced earnings relative to an industry job. A master's degree taken first, common as a stepping stone, can add $40,000–$100,000 in tuition, and PhD programs outside the US vary widely — many European doctorates are salaried research positions rather than tuition-charging degrees.
What to study
Mathematics or Statistics
for the theory underneath the modelsLinear algebra, probability and optimization are the language modern machine learning is written in; researchers who understand why a method works, not just how to run it, often come from here.
Physics
an unusually common feeder into AI researchPhysicists bring strong intuitions for optimization, statistical mechanics and high-dimensional systems; John Hopfield, who moved from physics into building an early neural-network model, shared the 2024 Nobel Prize in Physics for it.
Cognitive Science or Neuroscience
for the biological inspiration behind the modelsNeural networks were named for, and loosely inspired by, biological neurons; researchers with this background often work at the boundary between machine learning and how brains actually learn.
Electrical Engineering
for hardware-aware and signal-processing workUseful for researchers working close to the hardware training runs depend on, or on the signal-processing foundations behind speech and audio models.
Where it is taught best
Stanford University
United StatesHome to Fei-Fei Li's ImageNet project and the Stanford Institute for Human-Centered AI; one of the field's most consistently cited departments.
University of Toronto
CanadaGeoffrey Hinton's academic home for decades and the birthplace of the AlexNet team that restarted the field in 2012.
Université de Montréal / Mila
CanadaBuilt around Yoshua Bengio's lab into one of the world's largest single concentrations of deep-learning researchers.
University of Cambridge
United KingdomA strong machine-learning and theoretical computer-science tradition feeding directly into nearby Google DeepMind and the UK's broader AI research base.
Tsinghua University
ChinaConsistently ranks among the world's top computer-science departments by publication volume and feeds China's largest AI labs directly.
ETH Zurich
SwitzerlandA leading European machine-learning and robotics research center, with close ties to Google's and Meta's Zurich research offices.
Technion – Israel Institute of Technology
IsraelAnchors Israel's dense AI research and startup ecosystem, alongside corporate research labs opened nearby by Google, Meta and Intel.
National University of Singapore
SingaporeA fast-rising machine-learning research program and a regional hub for Southeast Asia's growing AI research investment.
Licences and exams
PhD in Computer Science, Machine Learning or a related field
GlobalThe field's de facto credential; with no licensing board, a completed dissertation and an advisor's endorsement function as the closest thing AI research has to certification.
Peer-reviewed publications at NeurIPS, ICML or ICLR
GlobalNot a formal credential, but the actual gatekeeping mechanism hiring committees weigh most heavily — a strong record at these venues routinely outweighs any certificate or the university's name.
ACM or AAAI Fellow
Global (US-headquartered bodies)A career-recognition status rather than an entry requirement, awarded by the Association for Computing Machinery or the Association for the Advancement of Artificial Intelligence for a sustained body of influential work.
CITI human-subjects research certification
United States (equivalents required elsewhere)Required by most university ethics boards before a researcher can run any study involving human participants or their data — relevant to AI researchers doing human-evaluation studies, user studies or crowdsourced labeling.
The other way in
Industry AI residency programs
Programs such as the Google AI Residency accept strong engineers or scientists without a PhD for a one-to-two-year mentored research placement, sometimes producing co-authored papers that substitute for a doctorate on a resume.
Open-source and preprint route
A smaller number of researchers build a public reputation through widely used open-source machine-learning code or well-cited arXiv preprints, occasionally reaching research roles at labs that hire on demonstrated ability rather than credentials.
Keep exploring
More in Engineering & Technology
💻
Software Engineer
Writes, tests and maintains the code that runs modern life — and is one of the first professions watching AI automate its own daily work.
AI-resistant 35 🛰️Aerospace Engineer
Designs, analyzes and certifies the aircraft, rockets and spacecraft that leave the ground, working to safety margins that leave no room for guessing.
AI-resistant 74 🔌Semiconductor Engineer
Designs and fabricates the transistors inside every computer, phone and weapon, using machines precise enough that only a few factories on Earth can run them.
AI-resistant 60