Richard Sutton and Andrew Barto Win 2024 Turing Award

Overall reaction to the award

• Broad agreement it’s “well deserved,” especially for foundational work in reinforcement learning (RL).
• Several commenters highlight their RL textbook as a classic: free, influential, and beautifully written, though opinions split on how accessible it is.
• Some note this recognition is “a long time coming,” given how much modern RL and agents build on their work.

The Bitter Lesson and black-box AI

• Many link and revisit “The Bitter Lesson”: scaling compute and general methods outperform hand-crafted knowledge.
• Some lament the move from understandable symbolic systems to opaque “big black boxes,” expressing distrust for using them in high‑risk domains.
• Others argue this is the reality of progress and point to AI‑assisted theorem proving and formal verification as a way to get mathematical guarantees even if the model’s internals are opaque.

Formal verification, safety, and “provably safe” AI

• Practitioners describe real-world use of formal methods in embedded systems, money handling, healthcare, aviation, and hardware design.
• Discussion of “provably safe AGI” notes a core difficulty: formally specifying human values or well‑being.
• Some suggest splitting “logical” systems (for safety‑critical tasks) from “feeling” systems for other domains; others point out that even formal logic rests on unprovable axioms.

Deep learning vs human-understanding goals

• Commenters contrast two AI research goals: building systems that perform tasks vs understanding human cognition.
• Some argue the Bitter Lesson is great if the goal is performance, but less helpful for explaining human intelligence.
• Debate over whether Go/Chess systems are “brute force”: consensus that they use heavy search plus learned evaluation, with human-crafted heuristics now largely replaced by learned ones.

Ethics, “successionism,” and public views

• A substantial subthread debates one awardee’s publicly stated “successionist” view (AI eventually replacing humanity’s role).
• Critics see this as dangerous or disqualifying; defenders say the award is for technical work and characterize his view as utopian/inevitable rather than genocidal.
• Several argue that technical achievement and views on AI governance should be evaluated separately, but both are relevant to policy discussions.

RL, education, and future impact

• Multiple commenters expect RL to grow in importance, particularly due to test-time compute and flexibility.
• Experiences with the RL textbook vary: some find it highly accessible; others find it too formal and recommend more hands-on “grokking” style books first.
• There’s nostalgia from early CV and classical NLP practitioners about being overtaken by deep learning, and a meta-lesson about not resisting new approaches that clearly work.