UvA recruiting PhD students in causal AI
The University of Amsterdam is advertising PhD openings in causal AI with a deadline of April 20, reflecting ongoing academic hiring in specialised areas like causality and safety. These postings show that academia remains a live path for people seeking deep, theory‑oriented training that can feed into frontier labs. For applicants, narrowly timed calls like this are concrete entry points into supervised research programs that emphasise causal methods. (x.com)
A PhD ad at the University of Amsterdam is one of those small notices that tells you where artificial intelligence research is actually hiring right now. The posting is for 2 doctoral positions, it closes on April 20, 2026, and it sits inside the Amsterdam Machine Learning Lab rather than a corporate lab. (werkenbij.uva.nl) The job is not “build a bigger chatbot.” It is about learning “causally grounded concepts,” which means training systems to connect what they see to the hidden causes behind it, the way a person learns that smoke usually points to fire rather than treating smoke as just another pattern of pixels. (academictransfer.com) That causal idea matters because standard machine learning is very good at picking up shortcuts. A model can learn that snowy backgrounds often mean wolves and then fail when the background changes, which is why researchers use causal methods to separate the real signal from the accidental clue. (amlab.science.uva.nl) The Amsterdam post ties that directly to safe artificial intelligence. The official description says the project is about improving interpretability, robustness, and safety by integrating causal reasoning, so the target is not just higher benchmark scores but systems that break less easily and are easier to inspect. (werkenbij.uva.nl) The research sits inside a project called CANES, short for “CAusal NEuro-Symbolic approach to integrating perception and abstract reasoning.” In plain English, that means combining pattern-recognition systems with rule-like structure, like pairing a camera with a notebook that keeps track of why things happen. (academictransfer.com) The funding source also tells you this is serious academic infrastructure rather than a speculative side project. The positions are backed by a Dutch Research Council Vidi grant, which is a competitive mid-career research award used in the Netherlands to build independent research lines. (werkenbij.uva.nl) The lab itself is not a niche outpost. The Amsterdam Machine Learning Lab lists causal inference alongside reinforcement learning, graph neural networks, probabilistic programming, and Bayesian deep learning, which means causality is being treated as one research pillar inside a broader artificial intelligence group. (amlab.science.uva.nl) The person leading this project is Sara Magliacane, and the vacancy says applicants will join the Causality team in that lab. Her own site repeats the same hiring call with the same April 20, 2026 deadline, which lines up with the university posting and shows the recruitment push is active across the group’s channels. (saramagliacane.github.io) The practical details look like a standard Dutch doctoral job rather than a tuition-based program. The listing gives a 38-hour work week, a salary range of €3,059 to €3,881 per month, and a location at Science Park 904 in Amsterdam, which is a reminder that many European Doctor of Philosophy tracks are salaried employee roles. (academictransfer.com) This is also not the only artificial intelligence opening at the University of Amsterdam right now. The university is separately advertising a PhD position in neurosymbolic artificial intelligence with a May 1, 2026 deadline, which suggests the institution is still expanding in theory-heavy corners of artificial intelligence rather than hiring only for product-facing engineering. (werkenbij.uva.nl) So the real story in this little vacancy is that one of Europe’s big research universities is still paying people to spend four years on the hard part of artificial intelligence: figuring out how models form concepts, how they confuse correlation with cause, and how to make them fail less often when the world changes. (werkenbij.uva.nl)
