AI research scientist
Also known as: ML research scientist, research scientist, AI researcher
AI research scientists form hypotheses about how to improve AI systems, design experiments to test them, and publish or apply the results. At organizations like OpenAI, Anthropic, Google DeepMind, and Meta AI, they work on foundational problems: new training algorithms, more efficient architectures, better alignment techniques, and ways to make models safer or more capable.
The role almost always requires a PhD or equivalent research track record. Publications in top venues like NeurIPS, ICML, or ICLR are a typical signal of qualification. This separates it sharply from the AI engineer or ML engineer tracks, which are much more accessible without graduate research credentials.
For most TNB builders, AI research scientists are upstream of their work: the people whose papers show up in model release notes, whose techniques become the fine-tuning methods and safety frameworks that builders use. Understanding what research scientists are working on is still valuable signal, though. The direction of research today predicts what tools and capabilities will exist in products 12 to 24 months from now.