Difference between revisions of "Timeline of Center for Human-Compatible AI"

From Timelines
Jump to: navigation, search
(→‎Google Ngram Viewer)
(→‎Full timeline)
 
(5 intermediate revisions by the same user not shown)
Line 51: Line 51:
  
 
| 2017 || {{dts|March}} || Collaboration || The Berkeley Existential Risk Initiative (BERI) formalizes its partnership with CHAI, providing operational support for grant management, workshops, and recruitment. This collaboration enables CHAI to expand its research capacity and host interdisciplinary events focused on AI alignment and safety.<ref>{{cite web |url=https://existence.org/ |title=What We Do - Berkeley Existential Risk Initiative |accessdate=December 15, 2024}}</ref>
 
| 2017 || {{dts|March}} || Collaboration || The Berkeley Existential Risk Initiative (BERI) formalizes its partnership with CHAI, providing operational support for grant management, workshops, and recruitment. This collaboration enables CHAI to expand its research capacity and host interdisciplinary events focused on AI alignment and safety.<ref>{{cite web |url=https://existence.org/ |title=What We Do - Berkeley Existential Risk Initiative |accessdate=December 15, 2024}}</ref>
 +
|-
 +
| 2017 || {{dts|July}} || Funding || Open Philanthropy recommends a grant of $403,890 to BERI to support its collaboration with CHAI at UC Berkeley, helping hire contractors and part-time employees for tasks such as web development and event coordination.
 +
<ref>{{cite web |url=https://www.openphilanthropy.org/grants/berkeley-existential-risk-initiative-core-support-and-chai-collaboration/ |title=Open Philanthropy Grant for CHAI Collaboration |publisher=Open Philanthropy |accessdate=December 17, 2024}}</ref>
 
|-
 
|-
 
| 2017 || {{dts|May}} || Team Expansion || Rosie Campbell joins CHAI as Assistant Director, leveraging her expertise in AI ethics and program management. She plays a critical role in organizing CHAI’s first annual workshop and fostering collaborations between academic and industry partners.<ref>{{cite web |url=https://www.bbc.co.uk/rd/people/rosie-campbell?Type=Posts&Decade=All |title=Rosie Campbell - BBC R&D |accessdate=December 15, 2024}}</ref>
 
| 2017 || {{dts|May}} || Team Expansion || Rosie Campbell joins CHAI as Assistant Director, leveraging her expertise in AI ethics and program management. She plays a critical role in organizing CHAI’s first annual workshop and fostering collaborations between academic and industry partners.<ref>{{cite web |url=https://www.bbc.co.uk/rd/people/rosie-campbell?Type=Posts&Decade=All |title=Rosie Campbell - BBC R&D |accessdate=December 15, 2024}}</ref>
Line 119: Line 122:
 
|-
 
|-
 
| 2019 || {{dts|September 28}} || Newsletter || Rohin Shah expands the AI Alignment Newsletter, transforming it into a vital resource for updates on the latest AI safety research, widely regarded as essential for researchers in the field.<ref>{{cite web|url=https://humancompatible.ai/news/2019/09/28/newsletter_update/ |title=Rohin Shah Expands the AI Alignment Newsletter |website=CHAI |date=September 28, 2019 |accessdate=October 7, 2024}}</ref>
 
| 2019 || {{dts|September 28}} || Newsletter || Rohin Shah expands the AI Alignment Newsletter, transforming it into a vital resource for updates on the latest AI safety research, widely regarded as essential for researchers in the field.<ref>{{cite web|url=https://humancompatible.ai/news/2019/09/28/newsletter_update/ |title=Rohin Shah Expands the AI Alignment Newsletter |website=CHAI |date=September 28, 2019 |accessdate=October 7, 2024}}</ref>
 +
|-
 +
 +
| 2019 || {{dts|November}} || Funding || Open Philanthropy recommends a $705,000 grant over two years to the Berkeley Existential Risk Initiative (BERI) to support its continued collaboration with the Center for Human-Compatible AI (CHAI). The funding provides resources for machine learning researchers at CHAI and bolsters initiatives focused on addressing long-term risks from advanced AI systems. By alleviating operational constraints and enabling targeted technical research, the grant strengthens CHAI's capacity to explore AI alignment challenges and develop "provably beneficial" AI systems, aligning with Open Philanthropy’s commitment to safeguarding humanity's future against existential risks.
 +
<ref>{{cite web |url=https://www.goodventures.org/our-portfolio/grants/berkeley-existential-risk-initiative-chai-collaboration-2019/ |title=Open Philanthropy Grant for BERI-CHAI Collaboration |publisher=Good Ventures |date=November 2019 |accessdate=December 17, 2024}}</ref>
 +
|-
 +
| 2020 || {{dts|May 30}} (submission), June 11 (date in paper) || Publication || The paper "AI Research Considerations for Human Existential Safety (ARCHES)" by Andrew Critch of the Center for Human-Compatible AI (CHAI) and David Krueger of the Montreal Institute for Learning Algorithms (MILA) is uploaded to the ArXiV.<ref>{{cite web|url = https://arxiv.org/abs/2006.04948|title = AI Research Considerations for Human Existential Safety (ARCHES)|date = May 30, 2020|accessdate = June 11, 2020}}</ref> MIRI's July 2020 newsletter calls it "a review of 29 AI (existential) safety research directions, each with an illustrative analogy, examples of current work and potential synergies between research directions, and discussion of ways the research approach might lower (or raise) existential risk."<ref>{{cite web|url = https://intelligence.org/2020/07/08/july-2020-newsletter/|title = July 2020 Newsletter|last = Bensinger|first = Rob|date = July 8, 2020|publisher = Machine Intelligence Research Institute}}</ref> Critch is interviewed about the paper on the AI Alignment Podcast released September 15.<ref>{{cite web|url = https://futureoflife.org/2020/09/15/andrew-critch-on-ai-research-considerations-for-human-existential-safety/|title = Andrew Critch on AI Research Considerations for Human Existential Safety|date = September 15, 2020|accessdate = December 20, 2020|publisher = Future of Life Institute}}</ref>
 
|-
 
|-
 
| 2020 || {{dts|June 1}} || Workshop || CHAI holds its first virtual workshop in response to the COVID-19 pandemic. The event gathers 150 participants from the AI safety community, featuring discussions on reducing existential risks from advanced AI, fostering collaborations, and advancing research initiatives.<ref>{{cite web|url=https://humancompatible.ai/news/2020/06/01/first-virtual-workshop/ |title=CHAI Holds Its First Virtual Workshop |website=CHAI |date=June 1, 2020 |accessdate=October 7, 2024}}</ref>
 
| 2020 || {{dts|June 1}} || Workshop || CHAI holds its first virtual workshop in response to the COVID-19 pandemic. The event gathers 150 participants from the AI safety community, featuring discussions on reducing existential risks from advanced AI, fostering collaborations, and advancing research initiatives.<ref>{{cite web|url=https://humancompatible.ai/news/2020/06/01/first-virtual-workshop/ |title=CHAI Holds Its First Virtual Workshop |website=CHAI |date=June 1, 2020 |accessdate=October 7, 2024}}</ref>
Line 124: Line 133:
 
| 2020 || {{dts|September 1}} || Staff || CHAI welcomes six new PhD students: Yuxi Liu, Micah Carroll, Cassidy Laidlaw, Alex Gunning, Alyssa Dayan, and Jessy Lin. These students, advised by Principal Investigators, bring expertise in areas like mathematics, AI-human cooperation, and safety mechanisms, furthering CHAI’s research depth.<ref>{{cite web|url=https://humancompatible.ai/news/2020/09/01/six-new-phd-students-join-chai/ |title=Six New PhD Students Join CHAI |website=CHAI |date=September 1, 2020 |accessdate=October 7, 2024}}</ref>
 
| 2020 || {{dts|September 1}} || Staff || CHAI welcomes six new PhD students: Yuxi Liu, Micah Carroll, Cassidy Laidlaw, Alex Gunning, Alyssa Dayan, and Jessy Lin. These students, advised by Principal Investigators, bring expertise in areas like mathematics, AI-human cooperation, and safety mechanisms, furthering CHAI’s research depth.<ref>{{cite web|url=https://humancompatible.ai/news/2020/09/01/six-new-phd-students-join-chai/ |title=Six New PhD Students Join CHAI |website=CHAI |date=September 1, 2020 |accessdate=October 7, 2024}}</ref>
 
|-
 
|-
 +
 
| 2020 || {{dts|September 10}} || Publication || CHAI PhD student Rachel Freedman publishes two papers at IJCAI-20 workshops. Choice Set Misspecification in Reward Inference investigates errors in robot reward inference, while Aligning with Heterogeneous Preferences for Kidney Exchange explores preference aggregation for optimizing kidney exchange programs, showcasing CHAI's practical AI safety applications.<ref>{{cite web|url=https://humancompatible.ai/news/2020/09/10/rachel-freedman-papers-ijcai/ |title=IJCAI-20 Accepts Two Papers by CHAI PhD Student Rachel Freedman |website=CHAI |date=September 10, 2020 |accessdate=October 7, 2024}}</ref>
 
| 2020 || {{dts|September 10}} || Publication || CHAI PhD student Rachel Freedman publishes two papers at IJCAI-20 workshops. Choice Set Misspecification in Reward Inference investigates errors in robot reward inference, while Aligning with Heterogeneous Preferences for Kidney Exchange explores preference aggregation for optimizing kidney exchange programs, showcasing CHAI's practical AI safety applications.<ref>{{cite web|url=https://humancompatible.ai/news/2020/09/10/rachel-freedman-papers-ijcai/ |title=IJCAI-20 Accepts Two Papers by CHAI PhD Student Rachel Freedman |website=CHAI |date=September 10, 2020 |accessdate=October 7, 2024}}</ref>
 
|-
 
|-
 +
 
| 2020 || {{dts|October 10}} || Publication || Brian Christian publishes The Alignment Problem: Machine Learning and Human Values, a comprehensive examination of AI safety challenges and advancements. The book highlights CHAI’s contributions to the field, including technical progress and ethical considerations.<ref>{{cite web|url=https://humancompatible.ai/news/2020/10/10/alignment-problem-published/ |title=Brian Christian Publishes The Alignment Problem |website=CHAI |date=October 10, 2020 |accessdate=October 7, 2024}}</ref>
 
| 2020 || {{dts|October 10}} || Publication || Brian Christian publishes The Alignment Problem: Machine Learning and Human Values, a comprehensive examination of AI safety challenges and advancements. The book highlights CHAI’s contributions to the field, including technical progress and ethical considerations.<ref>{{cite web|url=https://humancompatible.ai/news/2020/10/10/alignment-problem-published/ |title=Brian Christian Publishes The Alignment Problem |website=CHAI |date=October 10, 2020 |accessdate=October 7, 2024}}</ref>
 
|-
 
|-
Line 131: Line 142:
 
|-
 
|-
 
| 2020 || {{dts|November 12}} || Internship || CHAI opens applications for its 2021 research internship program, offering mentorship opportunities in AI safety research. Interns participate in seminars, workshops, and hands-on projects. Application deadlines are set for November 23 (early) and December 13 (final).<ref>{{cite web|url=https://humancompatible.ai/news/2020/11/12/internship-applications/ |title=CHAI Internship Application Is Now Open |website=CHAI |date=November 12, 2020 |accessdate=October 7, 2024}}</ref>
 
| 2020 || {{dts|November 12}} || Internship || CHAI opens applications for its 2021 research internship program, offering mentorship opportunities in AI safety research. Interns participate in seminars, workshops, and hands-on projects. Application deadlines are set for November 23 (early) and December 13 (final).<ref>{{cite web|url=https://humancompatible.ai/news/2020/11/12/internship-applications/ |title=CHAI Internship Application Is Now Open |website=CHAI |date=November 12, 2020 |accessdate=October 7, 2024}}</ref>
 +
|-
 +
| 2020 || {{dts|December}} || Funding || The Survival and Flourishing Fund (SFF) awards $247,000 to BERI to support its collaboration with the Center for Human-Compatible AI (CHAI). This funding enables BERI to provide operational and logistical support for CHAI’s AI alignment research, streamlining administrative processes and allowing CHAI researchers to focus on advancing key technical challenges in AI safety. The collaboration highlights a growing recognition of the importance of coordinated efforts to mitigate existential risks from advanced AI systems, with BERI acting as a bridge to facilitate smooth research operations.
 +
<ref>{{cite web |url=https://humancompatible.ai/news/2020/12/20/chai-beri-donations/ |title=SFF Donation to BERI for CHAI Collaboration |publisher=CHAI |date=December 20, 2020 |accessdate=December 17, 2024}}</ref>
 
|-
 
|-
 
| 2020 || {{dts|December 20}} || Financial || The Survival and Flourishing Fund (SFF) donates $799,000 to CHAI and $247,000 to BERI, supporting their collaborative efforts in AI safety research. The funding bolsters initiatives aimed at improving humanity’s long-term survival prospects through existential risk mitigation.<ref>{{cite web|url=https://humancompatible.ai/news/2020/12/20/chai-beri-donations/ |title=CHAI and BERI Receive Donations |website=CHAI |date=December 20, 2020 |accessdate=October 7, 2024}}</ref>
 
| 2020 || {{dts|December 20}} || Financial || The Survival and Flourishing Fund (SFF) donates $799,000 to CHAI and $247,000 to BERI, supporting their collaborative efforts in AI safety research. The funding bolsters initiatives aimed at improving humanity’s long-term survival prospects through existential risk mitigation.<ref>{{cite web|url=https://humancompatible.ai/news/2020/12/20/chai-beri-donations/ |title=CHAI and BERI Receive Donations |website=CHAI |date=December 20, 2020 |accessdate=October 7, 2024}}</ref>
Line 159: Line 173:
 
| 2021 || {{dts|April 11}} || Workshop || Stuart Russell and Caroline Jeanmaire organize a virtual workshop titled "AI Economic Futures," in collaboration with the Global AI Council at the World Economic Forum. The series aims to discuss AI policy recommendations and their impact on future economic prosperity.<ref>{{cite web|url=https://humancompatible.ai/news/2020/04/19/inaugural-virtal-workshop/ |title=Professor Stuart Russell and Caroline Jeanmaire Organize Virtual Workshop Titled “AI Economic Futures” |website=Center for Human-Compatible AI |date=April 19, 2020 |accessdate=October 7, 2024}}</ref>
 
| 2021 || {{dts|April 11}} || Workshop || Stuart Russell and Caroline Jeanmaire organize a virtual workshop titled "AI Economic Futures," in collaboration with the Global AI Council at the World Economic Forum. The series aims to discuss AI policy recommendations and their impact on future economic prosperity.<ref>{{cite web|url=https://humancompatible.ai/news/2020/04/19/inaugural-virtal-workshop/ |title=Professor Stuart Russell and Caroline Jeanmaire Organize Virtual Workshop Titled “AI Economic Futures” |website=Center for Human-Compatible AI |date=April 19, 2020 |accessdate=October 7, 2024}}</ref>
  
−
|-
 
−
| 2021 || {{dts|May}} || Publication || CHAI researchers co-author the ARCHES paper (Assistance Games for Cooperative Human-AI Systems), which is published in the journal Artificial Intelligence. The paper formalizes assistance games, a framework for modeling cooperative interactions between humans and AI systems. This work is significant because it provides a theoretical foundation for designing AI systems that are better aligned with human preferences and objectives, reinforcing CHAI's mission of developing provably beneficial AI.<ref>{{cite web |url=https://doi.org/10.1016/j.artint.2021.103526 |title=ARCHES: Assistance Games for Cooperative Human-AI Systems |publisher=Artificial Intelligence |date=May 2021 |accessdate=December 15, 2024}}</ref><ref>{{cite web |url=https://timelines.issarice.com/wiki/Timeline_of_AI_safety |title=Timeline of AI Safety |accessdate=December 15, 2024}}</ref>
 
 
|-
 
|-
 
| 2021 || {{dts|June 7}}-8 || Workshop || CHAI hosts its Fifth Annual Workshop, where researchers, students, and collaborators discuss advancements in AI safety, alignment, and research progress. The workshop addresses key challenges in AI reward modeling, interpretability, and scalable alignment techniques.<ref>{{cite web|url=https://humancompatible.ai/news/2021/06/16/fifth-annual-chai-workshop/ |title=Fifth Annual CHAI Workshop |website=Center for Human-Compatible AI |date=June 16, 2021 |accessdate=October 7, 2024}}</ref>
 
| 2021 || {{dts|June 7}}-8 || Workshop || CHAI hosts its Fifth Annual Workshop, where researchers, students, and collaborators discuss advancements in AI safety, alignment, and research progress. The workshop addresses key challenges in AI reward modeling, interpretability, and scalable alignment techniques.<ref>{{cite web|url=https://humancompatible.ai/news/2021/06/16/fifth-annual-chai-workshop/ |title=Fifth Annual CHAI Workshop |website=Center for Human-Compatible AI |date=June 16, 2021 |accessdate=October 7, 2024}}</ref>
Line 180: Line 192:
 
| 2022 || {{dts|January 18}} || Publication || Several papers were published by CHAI researchers. Tom Lenaerts and his co-authors explored "Voluntary safety commitments in AI development," suggesting that such commitments help escape over-regulation. Another paper, "Cross-Domain Imitation Learning via Optimal Transport," by Arnaud Fickinger, Stuart Russell, and others, discussed how to achieve cross-domain transfer in continuous control domains. Finally, Scott Emmons and his team published research on offline reinforcement learning, showing that simple design choices can improve empirical performance on RL benchmarks.<ref>{{cite web|url=https://humancompatible.ai/news/2022/01/18/new-papers-published/ |title=New Papers Published |website=Center for Human-Compatible AI |date=January 18, 2022 |accessdate=October 7, 2024}}</ref>
 
| 2022 || {{dts|January 18}} || Publication || Several papers were published by CHAI researchers. Tom Lenaerts and his co-authors explored "Voluntary safety commitments in AI development," suggesting that such commitments help escape over-regulation. Another paper, "Cross-Domain Imitation Learning via Optimal Transport," by Arnaud Fickinger, Stuart Russell, and others, discussed how to achieve cross-domain transfer in continuous control domains. Finally, Scott Emmons and his team published research on offline reinforcement learning, showing that simple design choices can improve empirical performance on RL benchmarks.<ref>{{cite web|url=https://humancompatible.ai/news/2022/01/18/new-papers-published/ |title=New Papers Published |website=Center for Human-Compatible AI |date=January 18, 2022 |accessdate=October 7, 2024}}</ref>
  
 +
|-
 +
| 2022 || {{dts|February}} || Funding || Open Philanthropy awards a $1,126,160 grant to the Berkeley Existential Risk Initiative (BERI) to enhance its collaboration with the Center for Human-Compatible AI (CHAI). The funding supports the creation of an in-house compute cluster at CHAI, a critical resource for accelerating large-scale machine learning experiments in AI alignment research. Additionally, the grant enables CHAI to hire a part-time system administrator, ensuring the smooth operation and maintenance of the new compute infrastructure. This investment reflects Open Philanthropy’s strategy to provide targeted resources for technical AI safety research, addressing computational bottlenecks that can hinder progress on solving existential risks.
 +
<ref>{{cite web |url=https://www.openphilanthropy.org/grants/berkeley-existential-risk-initiative-chai-collaboration-2022/ |title=Open Philanthropy Grant for CHAI Compute Cluster |publisher=Open Philanthropy |date=February 2022 |accessdate=December 17, 2024}}</ref>
 
|-
 
|-
  
Line 249: Line 264:
  
 
| 2024 || {{dts|August 7}} || Publication || Rachel Freedman and Wes Holliday publish a paper at the International Conference on Machine Learning (ICML) discussing how social choice theory can guide AI alignment when dealing with diverse human feedback. The paper explores approaches like reinforcement learning from human feedback and constitutional AI to better aggregate human preferences.<ref>{{cite web|url=https://humancompatible.ai/news/2024/08/07/social-choice-should-guide-ai-alignment-in-dealing-with-diverse-human-feedback/ |title=Social Choice Should Guide AI Alignment in Dealing with Diverse Human Feedback |website=CHAI |date=August 7, 2024 |accessdate=October 7, 2024}}</ref>
 
| 2024 || {{dts|August 7}} || Publication || Rachel Freedman and Wes Holliday publish a paper at the International Conference on Machine Learning (ICML) discussing how social choice theory can guide AI alignment when dealing with diverse human feedback. The paper explores approaches like reinforcement learning from human feedback and constitutional AI to better aggregate human preferences.<ref>{{cite web|url=https://humancompatible.ai/news/2024/08/07/social-choice-should-guide-ai-alignment-in-dealing-with-diverse-human-feedback/ |title=Social Choice Should Guide AI Alignment in Dealing with Diverse Human Feedback |website=CHAI |date=August 7, 2024 |accessdate=October 7, 2024}}</ref>
 +
|}
 +
 +
== Research Directions ==
 +
 +
{| class="wikitable"
 +
! Research Trajectory !! Current Directions !! Related Timeline Events
 +
|-
 +
| AI Alignment
 +
| Developing frameworks like assistance games, reward modeling, and dynamic reward systems to align AI with human intentions and preferences. Scalable methods include inverse reinforcement learning.
 +
| Publications: Off-Switch Game<ref>{{cite web |url=https://arxiv.org/abs/1611.08219 |title=The Off-Switch Game |date=November 24, 2016}}</ref>, ARCHES<ref>{{cite web |url=https://doi.org/10.1016/j.artint.2021.103526 |title=ARCHES: Assistance Games for Cooperative Human-AI Systems |date=May 2021}}</ref>, Dynamic Reward MDPs<ref>{{cite web |url=https://humancompatible.ai/news/2024/07/23/ai-alignment-with-changing-and-influenceable-reward-functions |title=AI Alignment with Changing and Influenceable Reward Functions |date=July 23, 2024}}</ref> 
 +
Workshops: Annual CHAI Workshops<ref>{{cite web |url=https://humancompatible.ai/news/2024/06/18/8th-annual-chai-workshop |title=8th Annual CHAI Workshop |date=June 18, 2024}}</ref>, PSBAI Workshop<ref>{{cite web |url=https://humancompatible.ai/progress-report |title=Progress Report |date=May 2023}}</ref> 
 +
Conferences: ICML 2019<ref>{{cite web |url=https://humancompatible.ai/news/2019/06/15/talks_icml_2019/ |title=CHAI Presentations at ICML |date=June 15, 2019}}</ref>, NeurIPS 2018<ref>{{cite web |url=https://neurips.cc |title=NeurIPS 2018 Conference |date=December 2018}}</ref>
 +
|-
 +
| Human Feedback and Cooperation
 +
| Improving Reinforcement Learning with Human Feedback (RLHF) using methods like active teacher selection and cluster ranking for efficient learning.
 +
| Publications: Learning from Human Corrections<ref>{{cite web |url=https://arxiv.org |title=Learning from Physical Human Corrections, One Feature at a Time |date=March 8, 2018}}</ref>, RLHF with ATS<ref>{{cite web |url=https://humancompatible.ai/news/2024/04/30/reinforcement-learning-with-human-feedback-and-active-teacher-selection-rlhf-and-ats/ |title=Reinforcement Learning with Human Feedback and Active Teacher Selection |date=April 30, 2024}}</ref>, Cluster Ranking<ref>{{cite web |url=https://humancompatible.ai/news/2022/11/18/time-efficient-reward-learning-via-visually-assisted-cluster-ranking/ |title=Time-Efficient Reward Learning via Visually Assisted Cluster Ranking |date=November 18, 2022}}</ref> 
 +
Workshops: HILL Workshop<ref>{{cite web |url=https://humancompatible.ai/news/2022/11/18/time-efficient-reward-learning-via-visually-assisted-cluster-ranking/ |title=HILL Workshop |date=2022}}</ref>, Annual CHAI Workshops<ref>{{cite web |url=https://humancompatible.ai/news/2024/06/18/8th-annual-chai-workshop |title=Annual CHAI Workshops |date=June 18, 2024}}</ref> 
 +
Competitions: NeurIPS BASALT (2021)<ref>{{cite web |url=https://humancompatible.ai/news/2021/07/09/neurips-minerl-basalt-competition-launches/ |title=NeurIPS MineRL BASALT Competition |date=July 9, 2021}}</ref>
 +
|-
 +
| AI Policy and Governance
 +
| Promoting AI regulation frameworks, such as an FDA for algorithms, and advocating for safe global AI deployment through public talks and media.
 +
| Talks: Stuart Russell TED Talk<ref>{{cite web |url=https://www.ted.com/talks/stuart_russell_how_ai_might_make_us_better_humans |title=How AI Might Make Us Better Humans |date=November 2016}}</ref>, Munk Debates<ref>{{cite web |url=https://humancompatible.ai/news/2021/02/09/stuart-russell-on-the-munk-debates/ |title=Stuart Russell on The Munk Debates |date=February 9, 2021}}</ref> 
 +
Media: WIRED (FDA for algorithms)<ref>{{cite web |url=https://www.wired.com/story/fda-for-algorithms/ |title=FDA for Algorithms |date=August 15, 2019}}</ref>, Vox articles<ref>{{cite web |url=https://www.vox.com |title=The Case for Taking AI Seriously as a Threat to Humanity |date=December 2018}}</ref> 
 +
Recognition: TIME 100 Most Influential (2023)<ref>{{cite web |url=https://humancompatible.ai/news/2023/09/18/100-most-influential-people-in-ai/ |title=TIME 100 Most Influential People in AI |date=September 18, 2023}}</ref>
 +
|-
 +
| Ethics and Societal Impact
 +
| Researching fairness, bias, and ethical risks in AI systems, including competence-aware systems and balancing AI obedience with safety.
 +
| Publications: Should Robots Be Obedient?<ref>{{cite web |url=https://arxiv.org/abs/1705.09990 |title=Should Robots Be Obedient? |date=May 28, 2017}}</ref>, Bias in AI Systems<ref>{{cite web |url=https://arxiv.org |title=Bias in Automated Systems |date=July 2, 2018}}</ref>, Competence-Aware Systems<ref>{{cite web |url=https://humancompatible.ai/news/2022/12/12/competence-aware-systems/ |title=Competence-Aware Systems |date=December 12, 2022}}</ref> 
 +
Conferences: FAT* 2019<ref>{{cite web |url=https://humancompatible.ai/news/2019/01/29/CHAI_FAT2019/ |title=FAT* 2019 |date=January 29, 2019}}</ref>, AAAI 2021<ref>{{cite web |url=https://humancompatible.ai/news/2021/03/18/chai-faculty-and-affiliates-publish-at-aaai-2021/ |title=AAAI 2021 |date=March 18, 2021}}</ref> 
 +
Recognition: Rosie Campbell Top Women in AI Ethics<ref>{{cite web |url=https://lighthouse3.com |title=Top Women in AI Ethics |date=October 31, 2018}}</ref>
 +
|-
 +
| Operational Support and Collaboration
 +
| Expanding capacity through partnerships and funding to scale research and provide compute infrastructure.
 +
| Financial: Open Philanthropy Grants<ref>{{cite web |url=https://www.openphilanthropy.org/grants/berkeley-existential-risk-initiative-chai-collaboration-2022/ |title=Open Philanthropy Grant |date=February 2022}}</ref>, SFF Donations<ref>{{cite web |url=https://humancompatible.ai/news/2020/12/20/chai-beri-donations/ |title=SFF Donations |date=December 20, 2020}}</ref> 
 +
Collaborations: BERI<ref>{{cite web |url=https://existence.org/ |title=BERI Partnership |date=2017}}</ref>, MIRI<ref>{{cite web |url=https://intelligence.org/2018/03/25/march-2018-newsletter/ |title=MIRI Newsletter |date=March 2018}}</ref> 
 +
Internal Reviews: CHAI Progress Reports (2022)<ref>{{cite web |url=https://humancompatible.ai/progress-report |title=Progress Report |date=May 2023}}</ref>
 
|}
 
|}
  

Latest revision as of 10:20, 17 December 2024

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

Big picture

Time period Development summary More details
2016–2021 CHAI is established and grows CHAI is established in 2016, begins producing research, and rapidly becomes a leading institution in AI safety. During this period, CHAI hosts numerous workshops, publishes key papers on AI alignment, and gains recognition within the AI safety community. Several PhD students join, and notable collaborations are formed with other institutions.
2021–2024 Continued expansion and recognition From 2021 onwards, CHAI continues to expand its research efforts and impact. This period includes increased international recognition, major contributions to AI policy discussions, participation in global conferences, and the establishment of new research initiatives focused on AI-human collaboration and safety. CHAI's researchers receive prestigious awards, and the institution solidifies its role as a thought leader in AI alignment.

Full timeline

Here are the inclusion criteria for various event types in the timeline related to CHAI (Center for Human-Compatible AI):

  • Publication: The intention is to include the most notable publications. These are typically those that have been highlighted by CHAI itself or have gained recognition in the broader AI safety community or media. Given the large volume of research papers, only papers that have significant impact, such as being accepted at prominent conferences or being discussed widely, are included.
  • Website: The intention is to include all websites directly associated with CHAI or initiatives that CHAI is heavily involved in. This includes new websites launched by CHAI or collaborative websites for joint projects.
  • Staff: The intention is to include significant changes or additions to CHAI staff, particularly new PhD students, research fellows, senior scientists, and leadership positions. Promotions and transitions in key roles are also included.
  • Workshop: All workshops organized or hosted by CHAI are included, particularly those that focus on advancing AI safety, alignment research, or collaborations within the AI research community. Virtual and in-person events count equally if organized by CHAI.
  • Conference: All conferences where CHAI organizes or leads are included. Conferences where CHAI staff give significant presentations, lead discussions, or organize major sessions are also featured.
  • Internal Review: This includes annual or progress reports published by CHAI, summarizing achievements, research outputs, and strategic directions. These reports provide a comprehensive review of the organization's work over a given period.
  • External Review: Includes substantive reviews of CHAI's work by external bodies or media. Only reviews that treat CHAI or its researchers in significant detail are included.
  • Financial: The inclusion focuses on large donations or funding announcements of over $10,000. Funding that supports major initiatives or collaborations that advance AI safety research is highlighted.
  • Staff Recognition: This includes awards, honors, and recognitions received by CHAI researchers for their contributions to AI safety or AI ethics. Recognitions such as TIME’s 100 Most Influential People or prestigious fellowships are included.
  • Social Media: Significant milestones such as new social media account creations for CHAI or major social media events (like Reddit AMAs) hosted by CHAI-affiliated researchers are included.
  • Project Announcement or Initiatives: Major projects, initiatives, or research programs launched by CHAI that are aimed at advancing AI alignment, existential risk mitigation, or improving AI-human cooperation are included. These announcements highlight new directions or collaborative efforts in AI safety.
  • Collaboration: Includes significant collaborations with other institutions where CHAI plays a major role, such as co-authoring reports, leading joint research projects, or providing advisory roles. Collaborations aimed at policy, safety, or alignment are particularly relevant.


Year Month and date Event type Details
2016 August Founding Vision Stuart Russell, renowned AI researcher and co-author of Artificial Intelligence: A Modern Approach, establishes CHAI at UC Berkeley. Russell emphasizes the need to shift the field toward developing AI systems that are "provably beneficial" and aligned with human values.[1]
2016 August Founding Members CHAI's founding team includes Stuart Russell, Andrew Critch (co-founder of CFAR), and Anca Dragan (expert in human-robot interaction). Their expertise lays the groundwork for interdisciplinary research in AI alignment and long-term safety.[2]
2016 August Website CHAI launches its official website to serve as a hub for resources, updates, and publications related to AI safety and alignment. The site includes research priorities, key collaborators, and recommended reading for the AI alignment community.[3]
2016 September Financial CHAI secures a $5.6 million grant from the Open Philanthropy Project. This significant funding enables early research projects, the recruitment of staff, and operational infrastructure.[4]
2016 October Collaboration CHAI partners with the Machine Intelligence Research Institute (MIRI) and the Center for Applied Rationality (CFAR). These collaborations focus on interdisciplinary research to address existential risks and foster AI alignment methodologies.[5]
2016 November Outreach Stuart Russell delivers a widely viewed TED Talk on the risks of poorly designed AI objectives, promoting CHAI’s mission to align AI with human values. The talk highlights potential misalignment issues and the necessity for AI safety measures.[6]
2016 November 24 Publication "The Off-Switch Game," co-authored by Dylan Hadfield-Menell, Anca Dragan, Pieter Abbeel, and Stuart Russell, is uploaded to arXiv. This influential paper introduces strategies for designing AI systems that allow safe human intervention, marking a pivotal contribution to AI safety.[7]
2016 December Resources CHAI publishes its annotated bibliography of recommended readings, offering curated resources on AI safety and alignment. This document becomes a cornerstone for guiding new researchers in the field.[8]
2016 December Operational Support The Berkeley Existential Risk Initiative (BERI) begins providing operational support to CHAI. BERI assists with grant management, workshop logistics, and infrastructure development, enabling researchers to focus on advancing AI safety.[9]
2017 March Collaboration The Berkeley Existential Risk Initiative (BERI) formalizes its partnership with CHAI, providing operational support for grant management, workshops, and recruitment. This collaboration enables CHAI to expand its research capacity and host interdisciplinary events focused on AI alignment and safety.[10]
2017 July Funding Open Philanthropy recommends a grant of $403,890 to BERI to support its collaboration with CHAI at UC Berkeley, helping hire contractors and part-time employees for tasks such as web development and event coordination.

[11]

2017 May Team Expansion Rosie Campbell joins CHAI as Assistant Director, leveraging her expertise in AI ethics and program management. She plays a critical role in organizing CHAI’s first annual workshop and fostering collaborations between academic and industry partners.[12]
2017 May 5–6 Workshop CHAI hosts its inaugural annual workshop, bringing together researchers and practitioners to discuss challenges in AI alignment. The event focuses on reorienting AI development toward systems that are provably beneficial to humans, establishing CHAI as a hub for AI safety discourse.[13]
2017 May 28 Publication The paper "Should Robots be Obedient?" by Smitha Milli, Dylan Hadfield-Menell, Anca Dragan, and Stuart Russell is uploaded to arXiv. The research explores the risks of rigid obedience in AI systems and proposes strategies for balancing compliance with ethical considerations, contributing significantly to the AI alignment discourse.[14][15]
2017 June Website CHAI updates its official website to include workshop proceedings, publications, and a dedicated section for recommended AI safety resources, enhancing accessibility for the research community.[16]
2017 July Publication CHAI researchers release a preliminary report on "Reward Modeling for Scalable AI Alignment." This internal document outlines strategies for designing reward systems that accurately capture human intentions, laying groundwork for future research.[17]
2017 October Staff Transition Rosie Campbell transitions from BBC R&D to join CHAI as Assistant Director. Her work focuses on expanding CHAI’s operations and enhancing collaborations with global AI safety organizations.[18]
2018 February 1 Publication Joseph Halpern publishes Information Acquisition Under Resource Limitations in a Noisy Environment, contributing to AI decision-making research under constraints. The paper explores how agents can optimize their decision processes with limited data and noisy environments, providing insights relevant to robust AI system design.[19]
2018 February 7 Media Mention Anca Dragan, a CHAI researcher, is featured in Forbes for her pioneering work on value alignment and ethical AI. The article highlights her contributions to ensuring AI systems respect human preferences, advancing public understanding of AI's societal implications.[20]
2018 February 26 Conference Anca Dragan presents Expressing Robot Incapability at the ACM/IEEE International Conference on Human-Robot Interaction. The presentation focuses on robots effectively communicating their limitations to humans, advancing trust and transparency in human-robot collaboration.[21]
2018 March Expansion BERI broadens its support beyond CHAI to include organizations like the Machine Intelligence Research Institute (MIRI). Despite this expansion, BERI remains a key operational partner for CHAI, ensuring smooth logistics and enabling focused AI safety research.[22]
2018 March Research Leadership Andrew Critch transitions from MIRI to CHAI as its first research scientist. Critch focuses on foundational alignment problems and helps define CHAI's long-term research direction.[23]
2018 March 8 Publication Anca Dragan and colleagues publish Learning from Physical Human Corrections, One Feature at a Time. The paper examines how robots can learn collaboratively from humans, improving performance through physical interaction.[24]
2018 April 4–12 Organization CHAI updates its branding, introducing a new logo with a green background and white "CHAI" lettering. The rebranding aims to reinforce its identity as a leader in AI alignment research.[25]
2018 April 9 Publication The Alignment Newsletter is publicly launched by CHAI affiliate Rohin Shah. This weekly newsletter consolidates AI safety updates, making it an essential resource for researchers and enthusiasts.[26]
2018 April 28–29 Workshop CHAI's second annual workshop convenes researchers, industry experts, and policymakers to discuss advances in AI alignment. Key topics include reward modeling, interpretability, and cooperative AI systems.[27]
2018 July 2 Publication Thomas Krendl Gilbert publishes A Broader View on Bias in Automated Decision-Making at ICML 2018. The work critiques bias in AI systems and offers strategies for promoting fairness and ethical standards in automated decisions.[28]
2018 July 13 Workshop Daniel Filan presents Exploring Hierarchy-Aware Inverse Reinforcement Learning at the 1st Workshop on Goal Specifications for Reinforcement Learning. The work advances understanding of how AI systems align with complex, hierarchical human goals.[29]
2018 August 26 Workshop CHAI students participate in MIRI’s AI Alignment Workshop. The event addresses critical AI safety challenges, fostering collaboration among AI researchers and practitioners.[30]
2018 September 4 Conference Jaime Fisac presents research on robust AI interactions at three conferences, focusing on ensuring AI systems behave predictably in dynamic, uncertain environments.[31]
2018 October 31 Recognition Rosie Campbell is named one of the Top Women in AI Ethics on Social Media by Mia Dand. The recognition highlights her leadership in promoting ethical AI development.[32]
2018 December Conference CHAI researchers present their findings at NeurIPS 2018, engaging in discussions on AI policy, safety, and interpretability.[33]
2018 December Podcast Rohin Shah discusses Inverse Reinforcement Learning on the AI Alignment Podcast by the Future of Life Institute. His insights advance public understanding of technical AI alignment challenges.[34]
2018 December Media Mention Stuart Russell and Rosie Campbell appear in a Vox article on AI existential risks. They emphasize the need for stringent AI safety measures to mitigate potential harm.[35]
2019 January Recognition Stuart Russell receives the AAAI Feigenbaum Prize for his pioneering work in AI research and policy. His contributions to probabilistic reasoning and AI alignment further enhance CHAI's reputation as a leader in AI safety.[36]
2019 January 8 Talks Rosie Campbell delivers public talks at San Francisco and East Bay AI Meetups, discussing neural networks and CHAI's approach to AI safety. These talks foster broader community engagement with CHAI’s research.[37]
2019 January 17 Conference CHAI faculty present multiple papers at AAAI 2019, covering topics such as deception in security games, ethical implications of AI systems, and advancements in multi-agent reinforcement learning. Their contributions emphasize AI's ethical deployment and its societal impact.[38]
2019 January 20 Publication Alex Turner, a former CHAI intern, wins the AI Alignment Prize for his work on "penalizing impact via attainable utility preservation." This research offers a novel framework for regulating AI behavior to minimize unintended harm.[39]
2019 January 29 Conference At ACM FAT* 2019, Smitha Milli and Anca Dragan present research addressing the ethical implications of AI transparency and fairness. Their work highlights the importance of accountability in automated decision-making systems.[40]
2019 June 15 Conference At ICML 2019, CHAI researchers, including Rohin Shah, Pieter Abbeel, and Anca Dragan, present research on human-AI coordination and addressing biases in AI reward inference, advancing scalable solutions for alignment.[41]
2019 July 5 Publication CHAI releases an open-source imitation learning library developed by Steven Wang, Adam Gleave, and Sam Toyer. The library provides benchmarks for algorithms like GAIL and AIRL, advancing research in behavior modeling.[42]
2019 July 5 Research Summary Rohin Shah publishes an analysis of CHAI's work on human biases in reward inference. This summary offers key insights into how AI systems can align their decision-making with nuanced human behavior.[43]
2019 August 15 Media Publication Mark Nitzberg authors an article in WIRED advocating for an “FDA for algorithms,” calling for stricter regulatory oversight of AI development to enhance safety and transparency.[44]
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 explores philosophical perspectives on aligning AI reward systems with human emotions.[45]
2019 September 28 Newsletter Rohin Shah expands the AI Alignment Newsletter, transforming it into a vital resource for updates on the latest AI safety research, widely regarded as essential for researchers in the field.[46]
2019 November Funding Open Philanthropy recommends a $705,000 grant over two years to the Berkeley Existential Risk Initiative (BERI) to support its continued collaboration with the Center for Human-Compatible AI (CHAI). The funding provides resources for machine learning researchers at CHAI and bolsters initiatives focused on addressing long-term risks from advanced AI systems. By alleviating operational constraints and enabling targeted technical research, the grant strengthens CHAI's capacity to explore AI alignment challenges and develop "provably beneficial" AI systems, aligning with Open Philanthropy’s commitment to safeguarding humanity's future against existential risks.

[47]

2020 May 30 (submission), June 11 (date in paper) Publication The paper "AI Research Considerations for Human Existential Safety (ARCHES)" by Andrew Critch of the Center for Human-Compatible AI (CHAI) and David Krueger of the Montreal Institute for Learning Algorithms (MILA) is uploaded to the ArXiV.[48] MIRI's July 2020 newsletter calls it "a review of 29 AI (existential) safety research directions, each with an illustrative analogy, examples of current work and potential synergies between research directions, and discussion of ways the research approach might lower (or raise) existential risk."[49] Critch is interviewed about the paper on the AI Alignment Podcast released September 15.[50]
2020 June 1 Workshop CHAI holds its first virtual workshop in response to the COVID-19 pandemic. The event gathers 150 participants from the AI safety community, featuring discussions on reducing existential risks from advanced AI, fostering collaborations, and advancing research initiatives.[51]
2020 September 1 Staff CHAI welcomes six new PhD students: Yuxi Liu, Micah Carroll, Cassidy Laidlaw, Alex Gunning, Alyssa Dayan, and Jessy Lin. These students, advised by Principal Investigators, bring expertise in areas like mathematics, AI-human cooperation, and safety mechanisms, furthering CHAI’s research depth.[52]
2020 September 10 Publication CHAI PhD student Rachel Freedman publishes two papers at IJCAI-20 workshops. Choice Set Misspecification in Reward Inference investigates errors in robot reward inference, while Aligning with Heterogeneous Preferences for Kidney Exchange explores preference aggregation for optimizing kidney exchange programs, showcasing CHAI's practical AI safety applications.[53]
2020 October 10 Publication Brian Christian publishes The Alignment Problem: Machine Learning and Human Values, a comprehensive examination of AI safety challenges and advancements. The book highlights CHAI’s contributions to the field, including technical progress and ethical considerations.[54]
2020 October 21 Workshop CHAI hosts a virtual launch event for Brian Christian’s book The Alignment Problem. The event includes an interview with the author, moderated by journalist Nora Young, and an audience Q&A session, focusing on AI safety and ethical frameworks.[55]
2020 November 12 Internship CHAI opens applications for its 2021 research internship program, offering mentorship opportunities in AI safety research. Interns participate in seminars, workshops, and hands-on projects. Application deadlines are set for November 23 (early) and December 13 (final).[56]
2020 December Funding The Survival and Flourishing Fund (SFF) awards $247,000 to BERI to support its collaboration with the Center for Human-Compatible AI (CHAI). This funding enables BERI to provide operational and logistical support for CHAI’s AI alignment research, streamlining administrative processes and allowing CHAI researchers to focus on advancing key technical challenges in AI safety. The collaboration highlights a growing recognition of the importance of coordinated efforts to mitigate existential risks from advanced AI systems, with BERI acting as a bridge to facilitate smooth research operations.

[57]

2020 December 20 Financial The Survival and Flourishing Fund (SFF) donates $799,000 to CHAI and $247,000 to BERI, supporting their collaborative efforts in AI safety research. The funding bolsters initiatives aimed at improving humanity’s long-term survival prospects through existential risk mitigation.[58]
2021 January 6 Podcast Daniel Filan debuts the AI X-risk Research Podcast (AXRP). The podcast focuses on AI alignment, technical challenges, and existential risks, bringing insights from experts and researchers working on ensuring AI safety.[59]
2021 January 25 Podcast Michael Dennis appears on the TalkRL podcast, discussing reinforcement learning, AI safety, and the challenges in creating safe and reliable AI systems. Dennis addresses the complexities of reward design and behavior modeling to align AI with human objectives.[60]
2021 February 5 Publication Thomas Krendl Gilbert releases the paper "AI Development for the Public Interest: From Abstraction Traps to Sociotechnical Risks" at IEEE ISTAS20. The paper critiques limited abstraction in AI research and advocates for integrating social context and ethical considerations into AI systems development.[61]
2021 February 9 Debate Stuart Russell debates Melanie Mitchell on The Munk Debates. Russell discusses AI safety concerns, the urgency of AI alignment research, and the risks associated with unregulated AI development. He stresses the importance of international governance to control AI technologies safely.[62]
2021 March 18 Conference At AAAI 2021, CHAI faculty and affiliates present multiple papers focusing on AI alignment, safe reinforcement learning, and improving AI-human interactions. Contributions include research on scalable reward modeling and methods for better interpretability of AI behavior.[63]
2021 March 25 Award Brian Christian's book "The Alignment Problem" is recognized with the Excellence in Science Communication Award by Eric and Wendy Schmidt and the National Academies. The book explores AI alignment challenges and ethical dilemmas in designing AI systems that behave as intended.[64]
2021 April 11 Workshop Stuart Russell and Caroline Jeanmaire organize a virtual workshop titled "AI Economic Futures," in collaboration with the Global AI Council at the World Economic Forum. The series aims to discuss AI policy recommendations and their impact on future economic prosperity.[65]
2021 June 7-8 Workshop CHAI hosts its Fifth Annual Workshop, where researchers, students, and collaborators discuss advancements in AI safety, alignment, and research progress. The workshop addresses key challenges in AI reward modeling, interpretability, and scalable alignment techniques.[66]
2021 July 9 Competition CHAI researchers contribute to the launch of the NeurIPS MineRL BASALT Competition. The competition aims to promote research in imitation learning, focusing on AI systems learning from human demonstration within the open-world Minecraft environment to improve behavior modeling.[67]
2021 August 7 Award Stuart Russell is named an Officer of the Most Excellent Order of the British Empire (OBE) for his significant contributions to artificial intelligence research and AI safety. This recognition highlights his impact on AI ethics and governance.[68]
2021 October 26 Internship CHAI announces its 2022 AI safety research internship program, with applications due by November 13, 2021. The program offers 3-4 month mentorship opportunities to work on AI safety research projects, either in-person at UC Berkeley or remotely. The internship aims to provide experience in technical AI safety research for individuals with a background in mathematics, computer science, or related fields. The selection process includes a cover letter or research proposal, programming assessments, and interviews.[69]
2022 January 18 Publication Several papers were published by CHAI researchers. Tom Lenaerts and his co-authors explored "Voluntary safety commitments in AI development," suggesting that such commitments help escape over-regulation. Another paper, "Cross-Domain Imitation Learning via Optimal Transport," by Arnaud Fickinger, Stuart Russell, and others, discussed how to achieve cross-domain transfer in continuous control domains. Finally, Scott Emmons and his team published research on offline reinforcement learning, showing that simple design choices can improve empirical performance on RL benchmarks.[70]
2022 February Funding Open Philanthropy awards a $1,126,160 grant to the Berkeley Existential Risk Initiative (BERI) to enhance its collaboration with the Center for Human-Compatible AI (CHAI). The funding supports the creation of an in-house compute cluster at CHAI, a critical resource for accelerating large-scale machine learning experiments in AI alignment research. Additionally, the grant enables CHAI to hire a part-time system administrator, ensuring the smooth operation and maintenance of the new compute infrastructure. This investment reflects Open Philanthropy’s strategy to provide targeted resources for technical AI safety research, addressing computational bottlenecks that can hinder progress on solving existential risks.

[71]

2022 February 17 Award Stuart Russell joins the inaugural cohort of AI2050 fellows, an initiative launched by Schmidt Futures with $125 million in funding over five years. The goal of AI2050 is to address the challenges of AI development. Russell’s focus is on enhancing AI interpretability, provable safety, and performance through probabilistic programming.[72]
2022 May 2022 Internal Review CHAI releases a progress report detailing the growth, research outputs, and engagements from May 2022 to April 2023. This includes 32 papers on AI topics like assistance games, adversarial robustness, and social impacts. It also covers CHAI’s work on advising on AI regulation and policy. Additionally, CHAI's research on safe AI development in large language models and AI vulnerabilities is emphasized.[73]
2022 October 7-9 Workshop CHAI holds the NSF Convergence Accelerator Workshop on Provably Safe and Beneficial AI (PSBAI) to develop a research agenda for creating verifiable, well-founded AI systems. The workshop gathers 51 experts from diverse fields such as AI, ethics, and law, to ensure safe AI integration into society.[74]
2022 November 18 Publication A paper titled “Time-Efficient Reward Learning via Visually Assisted Cluster Ranking” was accepted at the Human-in-the-loop Learning (HILL) Workshop at NeurIPS 2022. Written by Micah Carroll, Anca Dragan, and collaborators, the paper addresses improving reward learning efficiency through the use of data visualization techniques, enabling humans to label clusters of data points simultaneously rather than individually, optimizing human feedback usage.[75]
2022 December 12 Publication CHAI researcher Justin Svegliato publishes a paper in the Artificial Intelligence Journal on competence-aware systems (CAS). CAS are designed to understand and reason about their own competence and adjust their level of autonomy based on interactions with human authority, optimizing autonomy in varying situations.[76]
2022 December 29 Publication CHAI researcher Tom Lenaerts co-authors a publication in Nature titled "Fast deliberation is related to unconditional behaviour in iterated Prisoners’ Dilemma experiments." The research investigates the relationship between cognitive effort and social value orientation, analyzing how response times in strategic situations reflect different social behaviors.[77]
2023 March 10 Publication CHAI begins contributing to discussions on AI takeover scenarios, focusing on maximizing objectives like productive output, leading to outer misalignment and potential existential risks. Research examines AI's impact on critical resources for humans and challenges in safely deploying transformative AI due to competitive pressures, emphasizing robust AI alignment and regulation.[78]
2023 June 16-18 Workshop CHAI hosts its 7th annual workshop at Asilomar Conference Grounds, Pacific Grove, California. Nearly 200 attendees participate in discussions, lightning talks, and group activities on AI safety and alignment research. Casual activities like beach walks and bonfires build community within the AI safety field.[79]
2023 September 22 Workshop CHAI holds a virtual sister workshop on the Ethical Design of AIs (EDAI) that complements the Provably Safe and Beneficial AI (PSBAI) workshop. The sessions focus on ethical principles, human-centered AI design, governance, and addressing domain-specific challenges in implementing ethical AI systems.[80]
2023 September 7 Award Stuart Russell, founder of CHAI and professor at UC Berkeley, is named one of TIME's 100 Most Influential People in AI. Recognized as a leading thinker in AI safety, Russell is acknowledged for his contributions to responsible AI development and advocacy for AI safety, including support for pausing large-scale AI experiments.[81]
2023 May 2023 Award Stuart Russell, Professor at UC Berkeley and founder of CHAI, receives the ACM’s AAAI Allen Newell Award for foundational contributions to AI. The award honors career achievements with a broad impact within computer science or across multiple disciplines. Russell is noted for his work, including the widely used textbook "Artificial Intelligence: A Modern Approach" and his focus on AI safety.[82]
2023 November 10 Presentation Jonathan Stray, CHAI Senior Scientist, presents a talk titled “Orienting AI Toward Peace” at Stanford's "Beyond Moderation: How We Can Use Technology to De-Escalate Political Conflict" conference. He proposes strategies for AI to avoid escalating political conflicts, including defining desirable conflicts, developing conflict indicators, and incorporating this feedback into AI objectives.[83]
2024 March 5 Publication CHAI researchers publish a study addressing challenges with partial observability in AI systems. The study explores issues related to AI misinterpreting human feedback under limited information, which can result in unintentional deception or overly compensatory actions by the AI. The research aims to improve AI's alignment with human interests through more refined feedback mechanisms.[84]
2024 March 28 Publication Brian Christian, CHAI affiliate and renowned author, shares insights from his academic journey exploring AI that better reflects human values. Christian's work focuses on closing the gap between AI systems' assumptions about human rationality and the complexities of human behavior, offering a more nuanced approach to AI ethics.[85]
2024 April 30 Presentation Rachel Freedman, a CHAI PhD graduate student, presents new developments in reinforcement learning with human feedback (RLHF) at Stanford. The research addresses key challenges in modeling human preferences, proposing solutions like active teacher selection (ATS) to improve AI's ability to learn from diverse human feedback.[86]
2024 June 13-16 Workshop CHAI holds its 8th annual workshop at Asilomar Conference Grounds, with over 200 attendees. The workshop features more than 60 speakers, covering a range of topics such as societal impacts of AI, adversarial robustness, and AI alignment.[87]
2024 July 23 Publication CHAI researchers, including Stuart Russell and Anca Dragan, publish a paper on the challenges of AI alignment with changing and influenceable reward functions. The study introduces Dynamic Reward Markov Decision Processes (DR-MDPs), which model preference changes over time and address the risks associated with AI influencing human preferences.[88]
2024 August 7 Publication Rachel Freedman and Wes Holliday publish a paper at the International Conference on Machine Learning (ICML) discussing how social choice theory can guide AI alignment when dealing with diverse human feedback. The paper explores approaches like reinforcement learning from human feedback and constitutional AI to better aggregate human preferences.[89]

Research Directions

Research Trajectory Current Directions Related Timeline Events
AI Alignment Developing frameworks like assistance games, reward modeling, and dynamic reward systems to align AI with human intentions and preferences. Scalable methods include inverse reinforcement learning. Publications: Off-Switch Game[90], ARCHES[91], Dynamic Reward MDPs[92]

Workshops: Annual CHAI Workshops[93], PSBAI Workshop[94] Conferences: ICML 2019[95], NeurIPS 2018[96]

Human Feedback and Cooperation Improving Reinforcement Learning with Human Feedback (RLHF) using methods like active teacher selection and cluster ranking for efficient learning. Publications: Learning from Human Corrections[97], RLHF with ATS[98], Cluster Ranking[99]

Workshops: HILL Workshop[100], Annual CHAI Workshops[101] Competitions: NeurIPS BASALT (2021)[102]

AI Policy and Governance Promoting AI regulation frameworks, such as an FDA for algorithms, and advocating for safe global AI deployment through public talks and media. Talks: Stuart Russell TED Talk[103], Munk Debates[104]

Media: WIRED (FDA for algorithms)[105], Vox articles[106] Recognition: TIME 100 Most Influential (2023)[107]

Ethics and Societal Impact Researching fairness, bias, and ethical risks in AI systems, including competence-aware systems and balancing AI obedience with safety. Publications: Should Robots Be Obedient?[108], Bias in AI Systems[109], Competence-Aware Systems[110]

Conferences: FAT* 2019[111], AAAI 2021[112] Recognition: Rosie Campbell Top Women in AI Ethics[113]

Operational Support and Collaboration Expanding capacity through partnerships and funding to scale research and provide compute infrastructure. Financial: Open Philanthropy Grants[114], SFF Donations[115]

Collaborations: BERI[116], MIRI[117] Internal Reviews: CHAI Progress Reports (2022)[118]

Numerical and visual data

Google Scholar

The following table summarizes per-year mentions on Google Scholar as of November 27th, 2024.

Year "Center for Human-Compatible AI"
2016 2
2017 11
2018 20
2019 21
2020 34
2021 34
2022 27
2023 40
Center for Human-Compatible AI gscho.png

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 December 15, 2024. 
  2. ↑ "March 2018 Newsletter - Machine Intelligence Research Institute". Machine Intelligence Research Institute. March 25, 2018. Retrieved December 15, 2024. 
  3. ↑ "Center for Human-Compatible AI Official Website". Retrieved December 15, 2024. 
  4. ↑ "Open Philanthropy Project Awards Grant to CHAI". Retrieved December 15, 2024. 
  5. ↑ "March 2018 Newsletter - Machine Intelligence Research Institute". Machine Intelligence Research Institute. March 25, 2018. Retrieved December 15, 2024. 
  6. ↑ "How AI Might Make Us Better Humans". TED. November 2016. Retrieved December 15, 2024. 
  7. ↑ "[1611.08219] The Off-Switch Game". Retrieved December 15, 2024. 
  8. ↑ "Center for Human-Compatible AI". Retrieved December 15, 2024. 
  9. ↑ "March 2018 Newsletter - Machine Intelligence Research Institute". Machine Intelligence Research Institute. March 25, 2018. Retrieved December 15, 2024. 
  10. ↑ "What We Do - Berkeley Existential Risk Initiative". Retrieved December 15, 2024. 
  11. ↑ "Open Philanthropy Grant for CHAI Collaboration". Open Philanthropy. Retrieved December 17, 2024. 
  12. ↑ "Rosie Campbell - BBC R&D". Retrieved December 15, 2024. 
  13. ↑ "Center for Human-Compatible AI". Archived from the original on February 9, 2018. Retrieved February 9, 2018. 
  14. ↑ "[1705.09990] Should Robots be Obedient?". Retrieved December 15, 2024. 
  15. ↑ "2018 AI Safety Literature Review and Charity Comparison". Effective Altruism Forum. Retrieved December 15, 2024. 
  16. ↑ "Center for Human-Compatible AI Official Website". Retrieved December 15, 2024. 
  17. ↑ "Research at CHAI". Retrieved December 15, 2024. 
  18. ↑ "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. 
  19. ↑ "Information Acquisition Under Resource Limitations in a Noisy Environment". February 1, 2018. Retrieved September 30, 2024. 
  20. ↑ "Anca Dragan on AI Value Alignment". February 7, 2018. Retrieved September 30, 2024. 
  21. ↑ "Expressing Robot Incapability". ACM/IEEE International Conference. February 26, 2018. Retrieved September 30, 2024. 
  22. ↑ "March 2018 Newsletter - Machine Intelligence Research Institute". Machine Intelligence Research Institute. March 25, 2018. Retrieved December 15, 2024. 
  23. ↑ "March 2018 Newsletter - Machine Intelligence Research Institute". Machine Intelligence Research Institute. March 25, 2018. Retrieved December 15, 2024. 
  24. ↑ "Learning from Physical Human Corrections, One Feature at a Time". March 8, 2018. Retrieved September 30, 2024. 
  25. ↑ "Center for Human-Compatible AI". Archived from the original on April 4, 2018. Retrieved May 10, 2018. 
  26. ↑ Shah, Rohin (April 9, 2018). "Announcing the Alignment Newsletter". Retrieved May 10, 2018. 
  27. ↑ "Center for Human-Compatible AI Workshop 2018". Archived from the original on February 9, 2018. Retrieved February 9, 2018. 
  28. ↑ "A Broader View on Bias in Automated Decision-Making". ICML 2018. July 2, 2018. 
  29. ↑ "Exploring Hierarchy-Aware Inverse Reinforcement Learning". Goal Specifications Workshop. July 13, 2018. 
  30. ↑ "MIRI AI Alignment Workshop". Machine Intelligence Research Institute. August 26, 2018. 
  31. ↑ "Jaime Fisac AI Safety Research". CHAI. September 4, 2018. 
  32. ↑ "Top Women in AI Ethics". Lighthouse3. October 31, 2018. 
  33. ↑ "NeurIPS 2018 Conference". NeurIPS. December 2018. 
  34. ↑ "AI Alignment Podcast: Rohin Shah". Future of Life Institute. December 2018. 
  35. ↑ "The Case for Taking AI Seriously as a Threat to Humanity". Vox. December 2018. 
  36. ↑ "Stuart Russell Receives AAAI Feigenbaum Prize". CHAI. January 15, 2019. Retrieved December 15, 2024. 
  37. ↑ "Rosie Campbell Speaks About AI Safety and Neural Networks at San Francisco and East Bay AI Meetups". CHAI. January 8, 2019. Retrieved October 7, 2024. 
  38. ↑ "CHAI Papers at the AAAI 2019 Conference". CHAI. January 17, 2019. Retrieved October 7, 2024. 
  39. ↑ "Former CHAI Intern Wins AI Alignment Prize". CHAI. January 20, 2019. Retrieved October 7, 2024. 
  40. ↑ "CHAI Papers at FAT* 2019". CHAI. January 29, 2019. Retrieved October 7, 2024. 
  41. ↑ "CHAI Presentations at ICML". CHAI. June 15, 2019. Retrieved October 7, 2024. 
  42. ↑ "CHAI Releases Imitation Learning Library". CHAI. July 5, 2019. Retrieved October 7, 2024. 
  43. ↑ "Rohin Shah Summarizes CHAI's Research on Learning Human Biases". CHAI. July 5, 2019. Retrieved October 7, 2024. 
  44. ↑ "Mark Nitzberg Writes in WIRED on the Need for an FDA for Algorithms". WIRED. August 15, 2019. Retrieved October 7, 2024. 
  45. ↑ "Thomas Krendl Gilbert Submits Paper on Philosophical AI Safety". CHAI. August 28, 2019. Retrieved October 7, 2024. 
  46. ↑ "Rohin Shah Expands the AI Alignment Newsletter". CHAI. September 28, 2019. Retrieved October 7, 2024. 
  47. ↑ "Open Philanthropy Grant for BERI-CHAI Collaboration". Good Ventures. November 2019. Retrieved December 17, 2024. 
  48. ↑ "AI Research Considerations for Human Existential Safety (ARCHES)". May 30, 2020. Retrieved June 11, 2020. 
  49. ↑ Bensinger, Rob (July 8, 2020). "July 2020 Newsletter". Machine Intelligence Research Institute. 
  50. ↑ "Andrew Critch on AI Research Considerations for Human Existential Safety". Future of Life Institute. September 15, 2020. Retrieved December 20, 2020. 
  51. ↑ "CHAI Holds Its First Virtual Workshop". CHAI. June 1, 2020. Retrieved October 7, 2024. 
  52. ↑ "Six New PhD Students Join CHAI". CHAI. September 1, 2020. Retrieved October 7, 2024. 
  53. ↑ "IJCAI-20 Accepts Two Papers by CHAI PhD Student Rachel Freedman". CHAI. September 10, 2020. Retrieved October 7, 2024. 
  54. ↑ "Brian Christian Publishes The Alignment Problem". CHAI. October 10, 2020. Retrieved October 7, 2024. 
  55. ↑ "Watch the Book Launch of The Alignment Problem in Conversation with Brian Christian and Nora Young". CHAI. November 5, 2020. Retrieved October 7, 2024. 
  56. ↑ "CHAI Internship Application Is Now Open". CHAI. November 12, 2020. Retrieved October 7, 2024. 
  57. ↑ "SFF Donation to BERI for CHAI Collaboration". CHAI. December 20, 2020. Retrieved December 17, 2024. 
  58. ↑ "CHAI and BERI Receive Donations". CHAI. December 20, 2020. Retrieved October 7, 2024. 
  59. ↑ "Daniel Filan Launches AI X-risk Research Podcast". Center for Human-Compatible AI. January 6, 2021. Retrieved October 7, 2024. 
  60. ↑ "Michael Dennis". TalkRL Podcast. January 25, 2021. Retrieved October 7, 2024. 
  61. ↑ "Tom Gilbert Published in IEEE ISTAS20". Center for Human-Compatible AI. February 5, 2021. Retrieved October 7, 2024. 
  62. ↑ "Stuart Russell on The Munk Debates". Center for Human-Compatible AI. February 9, 2021. Retrieved October 7, 2024. 
  63. ↑ "CHAI Faculty and Affiliates Publish at AAAI 2021". Center for Human-Compatible AI. March 18, 2021. Retrieved October 7, 2024. 
  64. ↑ "Brian Christian's "The Alignment Problem" wins the Excellence in Science Communication Award". Center for Human-Compatible AI. November 1, 2022. Retrieved October 7, 2024. 
  65. ↑ "Professor Stuart Russell and Caroline Jeanmaire Organize Virtual Workshop Titled "AI Economic Futures"". Center for Human-Compatible AI. April 19, 2020. Retrieved October 7, 2024. 
  66. ↑ "Fifth Annual CHAI Workshop". Center for Human-Compatible AI. June 16, 2021. Retrieved October 7, 2024. 
  67. ↑ "NeurIPS MineRL BASALT Competition Launches". Center for Human-Compatible AI. July 9, 2021. Retrieved October 7, 2024. 
  68. ↑ "Birthday Honours 2021: Overseas and International List". GOV.UK. June 11, 2021. Retrieved October 7, 2024. 
  69. ↑ "CHAI Internship Applications are Open". LessWrong. October 26, 2021. Retrieved October 7, 2024. 
  70. ↑ "New Papers Published". Center for Human-Compatible AI. January 18, 2022. Retrieved October 7, 2024. 
  71. ↑ "Open Philanthropy Grant for CHAI Compute Cluster". Open Philanthropy. February 2022. Retrieved December 17, 2024. 
  72. ↑ "Schmidt Futures Launches AI2050 to Protect Our Human Future in the Age of Artificial Intelligence". Center for Human-Compatible AI. February 17, 2022. Retrieved October 7, 2024. 
  73. ↑ "Progress Report". Center for Human-Compatible AI. May 31, 2023. Retrieved October 7, 2024. 
  74. ↑ "Progress Report". Center for Human-Compatible AI. May 31, 2023. Retrieved October 7, 2024. 
  75. ↑ "Time-Efficient Reward Learning via Visually Assisted Cluster Ranking". Center for Human-Compatible AI. November 18, 2022. Retrieved October 7, 2024. 
  76. ↑ "Competence-Aware Systems". Center for Human-Compatible AI. December 12, 2022. Retrieved October 7, 2024. 
  77. ↑ "Fast Deliberation is Related to Unconditional Behaviour in Iterated Prisoners' Dilemma Experiments". Center for Human-Compatible AI. December 29, 2022. Retrieved October 7, 2024. 
  78. ↑ "Distinguishing AI Takeover Scenarios". AI Alignment Forum. March 10, 2023. Retrieved October 7, 2024. 
  79. ↑ "Seventh Annual CHAI Workshop". Center for Human-Compatible AI. June 20, 2023. Retrieved October 7, 2024. 
  80. ↑ "Progress Report". CHAI. 2023. Retrieved October 7, 2024. 
  81. ↑ "100 Most Influential People in AI". Center for Human-Compatible AI. September 18, 2023. Retrieved October 7, 2024. 
  82. ↑ "Stuart Russell receives ACM's AAAI Allen Newell Award". Berkeley Engineering. May 4, 2023. Retrieved October 7, 2024. 
  83. ↑ "Orienting AI Toward Peace". Center for Human-Compatible AI. November 21, 2023. Retrieved October 7, 2024. 
  84. ↑ "When Your AIs Deceive You: Challenges with Partial Observability of Human Evaluators in Reward Learning". CHAI. March 5, 2024. Retrieved October 7, 2024. 
  85. ↑ "Embracing AI That Reflects Human Values: Insights from Brian Christian's Journey". CHAI. March 28, 2024. Retrieved October 7, 2024. 
  86. ↑ "Reinforcement Learning with Human Feedback and Active Teacher Selection". CHAI. April 30, 2024. Retrieved October 7, 2024. 
  87. ↑ "8th Annual CHAI Workshop". CHAI. June 18, 2024. Retrieved October 7, 2024. 
  88. ↑ "AI Alignment with Changing and Influenceable Reward Functions". CHAI. July 23, 2024. Retrieved October 7, 2024. 
  89. ↑ "Social Choice Should Guide AI Alignment in Dealing with Diverse Human Feedback". CHAI. August 7, 2024. Retrieved October 7, 2024. 
  90. ↑ "The Off-Switch Game". November 24, 2016. 
  91. ↑ "ARCHES: Assistance Games for Cooperative Human-AI Systems". May 2021. 
  92. ↑ "AI Alignment with Changing and Influenceable Reward Functions". July 23, 2024. 
  93. ↑ "8th Annual CHAI Workshop". June 18, 2024. 
  94. ↑ "Progress Report". May 2023. 
  95. ↑ "CHAI Presentations at ICML". June 15, 2019. 
  96. ↑ "NeurIPS 2018 Conference". December 2018. 
  97. ↑ "Learning from Physical Human Corrections, One Feature at a Time". March 8, 2018. 
  98. ↑ "Reinforcement Learning with Human Feedback and Active Teacher Selection". April 30, 2024. 
  99. ↑ "Time-Efficient Reward Learning via Visually Assisted Cluster Ranking". November 18, 2022. 
  100. ↑ "HILL Workshop". 2022. 
  101. ↑ "Annual CHAI Workshops". June 18, 2024. 
  102. ↑ "NeurIPS MineRL BASALT Competition". July 9, 2021. 
  103. ↑ "How AI Might Make Us Better Humans". November 2016. 
  104. ↑ "Stuart Russell on The Munk Debates". February 9, 2021. 
  105. ↑ "FDA for Algorithms". August 15, 2019. 
  106. ↑ "The Case for Taking AI Seriously as a Threat to Humanity". December 2018. 
  107. ↑ "TIME 100 Most Influential People in AI". September 18, 2023. 
  108. ↑ "Should Robots Be Obedient?". May 28, 2017. 
  109. ↑ "Bias in Automated Systems". July 2, 2018. 
  110. ↑ "Competence-Aware Systems". December 12, 2022. 
  111. ↑ "FAT* 2019". January 29, 2019. 
  112. ↑ "AAAI 2021". March 18, 2021. 
  113. ↑ "Top Women in AI Ethics". October 31, 2018. 
  114. ↑ "Open Philanthropy Grant". February 2022. 
  115. ↑ "SFF Donations". December 20, 2020. 
  116. ↑ "BERI Partnership". 2017. 
  117. ↑ "MIRI Newsletter". March 2018. 
  118. ↑ "Progress Report". May 2023. 
Retrieved from "http://timelines.issarice.com/index.php?title=Timeline_of_Center_for_Human-Compatible_AI&oldid=83107"