Timeline of Center for Human-Compatible AI

From Timelines
Revision as of 21:37, 30 September 2024 by Hith (talk | contribs)
Jump to: navigation, search

This is a timeline of Center for Human-Compatible AI (CHAI).

Big picture

Time period Development summary More details
2016–present CHAI is established CHAI is established and begins producing research.


Full timeline

Year Month and date Event type Details
2016 August Organization The UC Berkeley Center for Human-Compatible Artificial Intelligence launches. The focus of the center is "to ensure that AI systems are beneficial to humans".[1]
2016 August Financial The Open Philanthropy Project awards a grant of $5.6 million to the Center for Human-Compatible AI.[2]
2016 November 24 Publication The initial version of "The Off-Switch Game", a paper by Dylan Hadfield-Menell, Anca Dragan, Pieter Abbeel, and Stuart Russell, is uploaded to the arXiv.[3][4]
2016 December Publication CHAI's "Annotated bibliography of recommended materials" is published around this time.[5]
2017 May 5–6 Workshop CHAI's first annual workshop takes place. The annual workshop is "designed to advance discussion and research" to "reorient the field of artificial intelligence toward developing systems that are provably beneficial to humans".[6]
2017 May 28 Publication "Should Robots be Obedient?" by Smitha Milli, Dylan Hadfield-Menell, Anca Dragan, and Stuart Russell is uploaded to the arXiv.[7][4]
2017 October Staff Rosie Campbell joins CHAI as Assistant Director.[8]
2018 February 1 Publication Joseph Halpern publishes "Information Acquisition Under Resource Limitations in a Noisy Environment," contributing to discussions on optimizing decision-making processes when resources and data are limited. The paper offers insights into how intelligent agents can make informed decisions in complex environments. [9]
2018 February 7 Media Mention Anca Dragan, a CHAI researcher, is featured in Forbes in an article discussing AI research with a focus on value alignment and the broader ethical implications of AI development. This inclusion highlights Dragan's contributions to the field and raises awareness on the importance of aligning AI behavior with human values. [10]
2018 February 26 Conference Anca Dragan presents "Expressing Robot Incapability" at the ACM/IEEE International Conference on Human-Robot Interaction. The presentation explores how robots can effectively communicate their limitations to humans, an essential step in developing more trustworthy and transparent human-robot collaborations.[11]
2018 March 8 Publication Anca Dragan and her team publish "Learning from Physical Human Corrections, One Feature at a Time," a paper focusing on the dynamics of human-robot interaction. This research emphasizes how robots can learn more effectively from human corrections, improving overall performance in collaborative settings.[12]
2018 March Staff Andrew Critch, who was previously on leave from the Machine Intelligence Research Institute to help launch CHAI and the Berkeley Existential Risk Initiative, accepts a position as CHAI's first research scientist.[13]
2018 April 4–12 Organization CHAI gets a new logo (green background with white letters "CHAI") sometime during this period.[14][15]
2018 April 9 Publication The Alignment Newsletter is publicly announced. The weekly newsletter summarizes content relevant to AI alignment from the previous week. Before the Alignment Newsletter was made public, a similar series of emails was produced internally for CHAI.[16][17] (It's not clear from the announcement whether the Alignment Newsletter is being produced officially by CHAI, or whether the initial emails were produced by CHAI and the later public newsletters are being produced independently.)
2018 April 28–29 Workshop CHAI hosts its second annual workshop, which brings together researchers, industry experts, and collaborators to discuss progress and challenges in AI safety and alignment. The event covers themes such as reward modeling, interpretability, and human-AI collaboration, promoting open dialogue on cooperative AI development and fostering connections to advance safety goals.[18]
2018 July 2 Publication Thomas Krendl Gilbert, a CHAI researcher, publishes "A Broader View on Bias in Automated Decision-Making" at ICML 2018. This work critically analyzes how bias in AI systems can perpetuate unfairness, providing strategies for integrating fairness and ethical standards into AI decision-making processes. The publication stresses the importance of preemptive efforts to address biases in automated decision-making.[19]
2018 July 13 Workshop At the 1st Workshop on Goal Specifications for Reinforcement Learning, CHAI researcher Daniel Filan presents "Exploring Hierarchy-Aware Inverse Reinforcement Learning." The work aims to advance understanding of hierarchical human goals in AI systems, enhancing AI agents' capabilities to align their behaviour with complex human intentions in varied and multi-level environments.[20]
2018 July 14 Workshop CHAI researchers Adam Gleave and Rohin Shah present "Active Inverse Reward Design" at the same workshop. Their work explores how AI systems can actively query human users to resolve ambiguities in reward structures, refining AI behavior to better align with nuanced human preferences, thereby improving decision-making in complex tasks.[21]
2018 July 16 Publication Stuart Russell and Anca Dragan of CHAI publish "An Efficient, Generalized Bellman Update For Cooperative Inverse Reinforcement Learning" at ICML 2018. The paper introduces an innovative method to enhance AI's capability to cooperate with human partners by efficiently updating beliefs and learning shared goals, which is vital for developing AI systems that safely and effectively interact with humans.[22]
2018 August 26 Workshop CHAI students join the AI Alignment Workshop organized by MIRI. This collaborative event focuses on core challenges in AI safety, such as creating systems aligned with human values, exploring theoretical advances in AI behaviour, and generating ideas for ensuring safe AI development. Students engage with researchers from various backgrounds to discuss emerging issues and practical solutions for AI alignment.[23]
2018 September 4 Conference Jaime Fisac, a CHAI researcher, presents his work at three conferences, exploring themes in AI safety and control under uncertainty. Fisac’s research aims to enhance the understanding of robust and safe AI interactions in dynamic settings, focusing on developing methods that ensure AI systems act predictably and align with human expectations across diverse environments.[24]
2018 October 31 Recognition Rosie Campbell, Assistant Director at CHAI, is honored as one of the Top Women in AI Ethics on Social Media by Mia Dand. This recognition acknowledges Campbell’s leadership in AI ethics, her influence on responsible AI development, and her contributions to fostering discussions around the ethical challenges of advancing AI technologies.[25]
2018 December Conference CHAI researchers participate in the NeurIPS 2018 Conference, presenting their work and contributing to discussions on AI research, safety, and policy development. The event offers CHAI members the opportunity to engage with peers in the AI field, share findings on AI alignment, and explore new avenues for collaboration to enhance AI safety research.[26]
2018 December Podcast Rohin Shah, a CHAI researcher, features on the AI Alignment Podcast by the Future of Life Institute. In this episode, Shah discusses Inverse Reinforcement Learning, outlining how AI systems can better learn and adapt to human preferences, ultimately contributing to the development of safe, aligned AI technologies that act according to human values.[27]
2018 December Media Mention Stuart Russell and Rosie Campbell from CHAI are quoted in a Vox article titled "The Case for Taking AI Seriously as a Threat to Humanity." They emphasize the potential for AI systems to be developed without adequate safety measures or ethical considerations, which could lead to unintended and potentially harmful consequences. The article highlights their advocacy for aligning AI with human values and ensuring that AI safety is prioritized to prevent misuse or detrimental effects on society.[28]
2019 January 8 Talks Rosie Campbell, Assistant Director at CHAI, delivers talks at AI Meetups in San Francisco and East Bay. She discusses challenges in AI safety, emphasizing the importance of neural networks and CHAI’s unique approach to AI alignment. Her talks aim to raise awareness about creating AI systems that are reliable and safe, sparking discussions on ethical AI development and fostering community engagement in AI safety topics.[29]
2019 January 15 Award Stuart Russell, a pioneer in AI safety research and Director of CHAI, receives the AAAI Feigenbaum Prize for advancing probabilistic knowledge representation and reasoning. This prestigious award highlights his contributions to the development of algorithms for intelligent agents, and his groundbreaking work in applying AI to real-world challenges, such as seismic monitoring for nuclear test detection and ethical considerations in AI development.[30]
2019 January 17 Conference CHAI faculty and staff contribute several papers to AAAI 2019, including: "Deception in Finitely Repeated Security Games," "Robust Multi-Agent Reinforcement Learning via Minimax Deep Deterministic Policy Gradient," "A Unified Framework for Planning in Adversarial and Cooperative Environments," "Partial Unawareness," "Abstracting Causal Models," "Blameworthiness in Multi-Agent Settings," and a presentation on AI's social implications at the AIES Conference. [31]
2019 January 21 Talk Stuart Russell speaks to the Netherlands Ministry of Foreign Affairs, outlining the risks of autonomous weapons. His presentation emphasizes how AI technology, if misused, could lead to the development of weapons of mass destruction. He calls for international regulations and AI safety mechanisms to manage autonomous weaponry, framing the global need to address ethical concerns and prevent the escalation of conflict driven by AI. [32]
2019 March 20 Conference Anca Dragan and Smitha Milli have a paper accepted at UAI 2019, focusing on uncertainties within AI systems. Their research enhances the understanding of AI's interactions with uncertain environments, proposing frameworks for more reliable AI decisions when dealing with ambiguous or incomplete information, which is crucial for safe deployment in real-world applications.[33]
2019 May 20 Publication Stuart Russell’s book, "Human Compatible: Artificial Intelligence and the Problem of Control," is announced for pre-order, offering a deep dive into the challenges and ethical considerations of controlling AI systems. The book delves into the need for aligning AI behavior with human values and preventing unintended consequences as AI systems grow more advanced and autonomous.[34]
2019 May 24 Conference CHAI researchers submit the paper "Adversarial Policies: Attacking Deep Reinforcement Learning" to NeurIPS 2019. The research investigates how deep reinforcement learning models can be vulnerable to adversarial attacks, proposing strategies for increasing their resilience and robustness to ensure that AI systems behave safely under various conditions.[35]
2019 May 30 Research Collaboration Vincent Corruble, from Sorbonne University, collaborates with CHAI as a visiting researcher. During his three-month tenure, Corruble works on the risks associated with comprehensive AI systems, contributing to ongoing CHAI projects aimed at developing ethical frameworks and practical safety measures for AI applications.[36]
2019 July 5 Research Summary Rohin Shah publishes a summary of CHAI's research paper "On the Feasibility of Learning, Rather than Assuming, Human Biases for Reward Inference." The summary highlights the potential for AI systems to learn from human biases, offering a pathway to improve how AI models reward structures based on observed human behaviours and decisions.[37]
2019 August 15 Media Publication Mark Nitzberg, Executive Director of CHAI, writes a WIRED article advocating for an “FDA for algorithms.” The article argues for the establishment of a regulatory body to oversee the development of AI technologies, ensuring their safety, transparency, and ethical use to mitigate potential harms caused by unchecked AI deployment.[38]
2019 August 16 Talk Michael Wellman, a principal investigator at CHAI, presents his work at the ICML Workshop on AI and Finance. His talk, "Trend-Following Trading Strategies and Financial Market Stability," examines how the use of AI in algorithmic trading impacts market stability and raises questions about the ethical implications of autonomous financial decision-making.[39]
2019 August 28 Paper Submission Thomas Krendl Gilbert submits "The Passions and the Reward Functions: Rival Views of AI Safety?" to FAT*2020. The paper delves into philosophical perspectives on AI safety, debating different approaches to constructing reward functions that align with human emotions and societal values, and offering insights into balancing technical and ethical considerations in AI design.[40]
2019 September 28 Newsletter Rohin Shah expands and professionalizes the AI Alignment Newsletter, transitioning it from a volunteer-based effort to a team-driven project. This enables regular and detailed updates on AI safety and alignment research, serving as a comprehensive resource for the AI community on developments and discussions in AI safety.[41]
2019 October 8 Book Release Stuart Russell publishes "Human Compatible: Artificial Intelligence and the Problem of Control," exploring the potential risks of advanced AI systems and proposing a vision for their ethical design and control. The book examines the broader implications of...controlling advanced AI. The book discusses the dangers of unaligned AI, the need for ethical design, and strategies for ensuring AI systems serve human interests, providing a roadmap for the safe development of AI technologies.[42] 2020 January 28 Course Launch A new course, "Foundations for Beneficial AI," is co-taught at UC Berkeley by Stuart Russell, Lara Buchak, Wesley Holliday, and Shachar Kariv. The course focuses on fundamental AI concepts with an emphasis on ensuring AI technologies align with beneficial societal outcomes, bridging computer science, philosophy, and economics.[43]
2020 February 21 Paper Accepted Thomas Gilbert’s paper "Hard Choices in Artificial Intelligence: Addressing Normative Uncertainty through Sociotechnical Commitments" is accepted at AIES 2020. The work explores strategies for managing ethical uncertainty in AI systems, offering a framework to integrate sociotechnical considerations into decision-making processes.[44]
2020 March 5 Podcast Anca Dragan appears on Lex Fridman’s "Artificial Intelligence" podcast, where she discusses the intricacies of human-robot collaboration and control, emphasizing the design of AI systems that can understand and respond to human behaviors, particularly in semi-autonomous scenarios like driving.[45]
2020 April 19 Workshop Stuart Russell and Caroline Jeanmaire organize the "AI Economic Futures" virtual workshop in partnership with the Global AI Council at the World Economic Forum. This inaugural event explores the economic implications of AI, discussing strategies for aligning AI development with sustainable and inclusive growth.[46]
2020 April 30 Book Reissue Stuart Russell’s paperback edition of "Human Compatible: AI and the Problem of Control" is published in the UK. The book, originally released in 2019, critically examines the challenges of controlling advanced AI, emphasizing ethical frameworks for its safe deployment.[47]
2020 May 5 Publication Vael Gates, Anca Dragan, and Tom Griffiths publish "How to Be Helpful to Multiple People at Once" in Cognitive Science. The paper examines how AI systems can effectively assist multiple people simultaneously, proposing decision-making models that optimize AI's helpfulness in multi-person interactions.[48]
2020 May 30 Paper Published Andrew Critch and David Krueger publish "AI Research Considerations for Human Existential Safety (ARCHES)." The paper addresses existential risks posed by AI, proposing frameworks for developing safe AI that do not threaten humanity's future.[49]
2020 June 1 Workshop CHAI hosts its first virtual workshop, bringing together researchers to discuss progress in AI safety, alignment strategies, and challenges in AI systems development. The format encourages collaborative problem-solving and community building in a digital space.[50]
2020 June 15 Fellowship Alyssa Dayan, a PhD student advised by Stuart Russell at CHAI, is awarded the Hertz Fellowship. The fellowship recognizes her potential contributions to AI safety, supporting her research efforts to align AI systems with human values.[51]
2020 July 15 Student Placement Rohin Shah, Jaime Fernández Fisac, and Dylan Hadfield-Menell, CHAI PhD students, secure positions at MIT, Princeton, and DeepMind, respectively. Their placements highlight CHAI's impact on cultivating future leaders in AI safety and alignment research.[52]
2020 August 3 Publication Joseph Halpern and Xinming Liu publish "Bounded Rationality in Las Vegas: Probabilistic Finite Automata Play Multi-Armed Bandits" at the UAI Virtual Conference. The research explores AI decision-making under bounded rationality using probabilistic finite automata, contributing to models for understanding AI behavior under limited rational capacities.[53]
2020 September 3 Award Stuart Russell receives the 2020 AAAS Scientific Freedom and Responsibility Award for his work in advocating ethical AI and promoting the responsible development of AI technologies that align with human safety and societal values.[54]

Visual data

Wikipedia Views

CHCAI WV.jpeg


Meta information on the timeline

How the timeline was built

The initial version of the timeline was written by Issa Rice.

Funding information for this timeline is available.

What the timeline is still missing

Timeline update strategy

See also

External links

References

  1. ↑ "UC Berkeley launches Center for Human-Compatible Artificial Intelligence". Berkeley News. August 29, 2016. Retrieved July 26, 2017. 
  2. ↑ "Open Philanthropy Project donations made (filtered to cause areas matching AI risk)". Retrieved July 27, 2017. 
  3. ↑ "[1611.08219] The Off-Switch Game". Retrieved February 9, 2018. 
  4. ↑ 4.0 4.1 "2018 AI Safety Literature Review and Charity Comparison - Effective Altruism Forum". Retrieved February 9, 2018. 
  5. ↑ "Center for Human-Compatible AI". Retrieved February 9, 2018. 
  6. ↑ "Center for Human-Compatible AI". Archived from the original on February 9, 2018. Retrieved February 9, 2018. 
  7. ↑ "[1705.09990] Should Robots be Obedient?". Retrieved February 9, 2018. 
  8. ↑ "Rosie Campbell - BBC R&D". Archived from the original on May 11, 2018. Retrieved May 11, 2018. Rosie left in October 2017 to take on the role of Assistant Director of the Center for Human-compatible AI at UC Berkeley, a research group which aims to ensure that artificially intelligent systems are provably beneficial to humans. 
  9. ↑ "Information Acquisition Under Resource Limitations in a Noisy Environment". February 1, 2018. Retrieved October 14, 2024. 
  10. ↑ "Anca Dragan on AI Value Alignment". February 7, 2018. Retrieved October 14, 2024. 
  11. ↑ "Expressing Robot Incapability". ACM/IEEE International Conference. February 26, 2018. Retrieved October 14, 2024. 
  12. ↑ "Learning from Physical Human Corrections, One Feature at a Time". March 8, 2018. Retrieved October 14, 2024. 
  13. ↑ Bensinger, Rob (March 25, 2018). "March 2018 Newsletter - Machine Intelligence Research Institute". Machine Intelligence Research Institute. Retrieved May 10, 2018. Andrew Critch, previously on leave from MIRI to help launch the Center for Human-Compatible AI and the Berkeley Existential Risk Initiative, has accepted a position as CHAI’s first research scientist. Critch will continue to work with and advise the MIRI team from his new academic home at UC Berkeley. Our congratulations to Critch! 
  14. ↑ "Center for Human-Compatible AI". Archived from the original on April 4, 2018. Retrieved May 10, 2018. 
  15. ↑ "Center for Human-Compatible AI". Archived from the original on April 12, 2018. Retrieved May 10, 2018. 
  16. ↑ Shah, Rohin (April 9, 2018). "Announcing the Alignment Newsletter". LessWrong. Retrieved May 10, 2018. 
  17. ↑ "Alignment Newsletter". Rohin Shah. Archived from the original on May 10, 2018. Retrieved May 10, 2018. 
  18. ↑ "Center for Human-Compatible AI Workshop 2018". Archived from the original on February 9, 2018. Retrieved February 9, 2018. 
  19. ↑ "A Broader View on Bias in Automated Decision-Making". ICML 2018. July 2, 2018. Retrieved September 30, 2024. 
  20. ↑ "Exploring Hierarchy-Aware Inverse Reinforcement Learning". Goal Specifications Workshop. July 13, 2018. Retrieved September 30, 2024. 
  21. ↑ "Active Inverse Reward Design". Goal Specifications Workshop. July 14, 2018. Retrieved September 30, 2024. 
  22. ↑ "An Efficient, Generalized Bellman Update For Cooperative Inverse Reinforcement Learning". ICML 2018. July 16, 2018. Retrieved September 30, 2024. 
  23. ↑ "MIRI AI Alignment Workshop". Machine Intelligence Research Institute. August 26, 2018. Retrieved September 30, 2024. 
  24. ↑ "Jaime Fisac AI Safety Research". CHAI. September 4, 2018. Retrieved September 30, 2024. 
  25. ↑ "Top Women in AI Ethics". Lighthouse3. October 31, 2018. Retrieved September 30, 2024. 
  26. ↑ "NeurIPS 2018 Conference". NeurIPS. December 2018. Retrieved September 30, 2024. 
  27. ↑ "AI Alignment Podcast: Rohin Shah". Future of Life Institute. December 2018. Retrieved September 30, 2024. 
  28. ↑ "The Case for Taking AI Seriously as a Threat to Humanity". Vox. December 2018. Retrieved October 14, 2024. 
  29. ↑ "CHAI News January 8, 2019". Retrieved September 30, 2024. 
  30. ↑ "CHAI News January 15, 2019". Retrieved September 30, 2024. 
  31. ↑ "CHAI News January 17, 2019". Retrieved September 30, 2024. 
  32. ↑ "CHAI News January 21, 2019". Retrieved September 30, 2024. 
  33. ↑ "CHAI News March 20, 2019". Retrieved September 30, 2024. 
  34. ↑ "CHAI News May 20, 2019". Retrieved September 30, 2024. 
  35. ↑ "CHAI News May 24, 2019". Retrieved September 30, 2024. 
  36. ↑ "CHAI News May 30, 2019". Retrieved September 30, 2024. 
  37. ↑ "CHAI News July 5, 2019". Retrieved September 30, 2024. 
  38. ↑ "WIRED Article by Mark Nitzberg". August 15, 2019. Retrieved September 30, 2024. 
  39. ↑ "CHAI News August 16, 2019". Retrieved September 30, 2024. 
  40. ↑ "CHAI News August 28, 2019". Retrieved September 30, 2024. 
  41. ↑ "CHAI News September 28, 2019". Retrieved September 30, 2024. 
  42. ↑ "CHAI News October 8, 2019". Retrieved September 30, 2024. 
  43. ↑ "CHAI News January 28, 2020". Retrieved October 14, 2024. 
  44. ↑ "CHAI News February 21, 2020". Retrieved October 14, 2024. 
  45. ↑ "CHAI News March 5, 2020". Retrieved October 14, 2024. 
  46. ↑ "CHAI News April 19, 2020". Retrieved October 14, 2024. 
  47. ↑ "CHAI News April 30, 2020". Retrieved October 14, 2024. 
  48. ↑ "CHAI News May 5, 2020". Retrieved October 14, 2024. 
  49. ↑ "CHAI News May 30, 2020". Retrieved October 14, 2024. 
  50. ↑ "CHAI News June 1, 2020". Retrieved October 14, 2024. 
  51. ↑ "CHAI News June 15, 2020". Retrieved October 14, 2024. 
  52. ↑ "CHAI News July 15, 2020". Retrieved October 14, 2024. 
  53. ↑ "CHAI News August 3, 2020". Retrieved October 14, 2024. 
  54. ↑ "CHAI News September 3, 2020". Retrieved October 14, 2024. 
Retrieved from "http://timelines.issarice.com/index.php?title=Timeline_of_Center_for_Human-Compatible_AI&oldid=82275"