Effective Altruism Forum
EA Forum

Hide table of contents
by HaydnBelfield
Dec 2 202021 min read 0

24

AI safetyBiosecurityExistential riskPolicyCentre for the Study of Existential RiskAI alignmentAI governanceClimate changeOrganization updates
Frontpage

We send short monthly updates in our newsletter – subscribe here.

READ AS PDF

Contents

  1. Coronavirus update
  2. Upcoming Conference programme and speakers
  3. New Staff
  4. Policy and industry engagement – Impact
  5. Public engagement – Field-building
  6. Academic engagement – Field-building
  7. Publications

Overview

The Centre for the Study of Existential Risk (CSER) is an interdisciplinary research centre within the University of Cambridge dedicated to the study and mitigation of risks that could lead to civilizational collapse or human extinction. Our research focuses on Global Catastrophic Biological Risks, Extreme Risks and the Global Environment, Risks from Artificial Intelligence, and Managing Extreme Technological Risks. Our work is shaped around three main goals:

  • Understanding: we study existential risk.
  • Impact: we develop collaborative strategies to reduce existential risk.
  • Field-building: we foster a global community of academics, technologists and policy-makers working to tackle existential risk.

Our last report covered January – May 2020. This report covers the period May – September 2020 and outlines our activities and future plans. Highlights include:

  • Contributing to Cambridge University’s decision to fully divest from fossil fuels.
  • Publication of nine peer-reviewed papers on: divestment and Cambridge; our second bioengineering horizon-scan; cross-cultural cooperation in AI governance; ethics with urgency in a crisis; two AI safety papers on paradigms and degrees of generality; capital theory and sustainability; famine dynamics; and existential risk assessment.
  • Publication of four written evidence submissions, and two policy briefs, on premature procurement of AI defence systems; the EU’s proposed AI regulation; proposals for the new Advanced Research Projects Agency (ARPA) UK funding body; how to improve the UK’s risk management system; using AI to improve food supply chain security, and public mental health and COVID-19.
  • Seán Ó hÉigeartaigh joined the Global Partnership on AI.
  • Co-organising two workshops with the Institute for Public Policy Research (IPPR) on catastrophic environmental breakdown.
  • Extensive media coverage – including the front page of The Economist.
  • Welcoming six new team members.

1. Coronavirus update

The team has successfully adapted to fully-remote work, with regular use of video conferencing and Slack. We are fortunate that much of our research can be done remotely. We are currently planning a physically distanced return to the office later in the year, subject to the situation and University advice.

We have also been doing some work to contribute to the response:

2. Upcoming Conference programme and speakers

The Cambridge Conference on Catastrophic Risk ‘2020 Hindsight: Our Final Century Revisited’ (CCCR 2020), our third major international conference, will be held remotely 16-19 November 2020. It follows our ‘CCCR’ conferences in 2016 and 2018.

The programme will draw on key themes from Lord Martin Rees’ seminal 2003 book Our Final Century, to reflect on the growth and development of the field of global catastrophic risk research over the last decades. CCCR2020 is one of the few forums in the world discussing such issues, and we intend the conference to contribute to the further development of the community working on global risks, and its engagement with other relevant groups. To allow for in-depth collaboration, it is invite-only. The keynote will be by Lord Martin Rees and there will be a range of opportunities for networking and debate.

Sessions will include:

  1. Pre-2000: What global catastrophic risk researchers can learn from the history of global catastrophic risk
    Speakers: Francesco Calogero, Malcolm Potts
    Chair: Simon Beard
  2. 50/50: Assessing the chances of a global catastrophe
    Speakers: Nancy Connell, Karim Jebari
    Chair: Sabin Roman
  3. Are we nearly there yet? Taking the long and longer-term view of humanity
    Speakers: Anders Sandberg, Sheri Wells-Jensen
    Chair: Clarissa Rios Rojas
  4. Global justice and global catastrophic risk: Between error and terror
    Speakers: Bentley Allan, Ndidi Nwaneri
    Chair: Natalie Jones
  5. Threats without enemies: natural global disasters and their consequences
    Speakers: Doug Erwin, Lindley Johnson
    Chair: Lara Mani
  6. Governing science: Does it matter who is doing our scientific research and why?
    Speakers: Stuart Parkinson, Heather Roff
    Chair: Lalitha Sundaram
  7. Global catastrophic environmental risks: systemic collapse from anthropogenic environmental change
    Speakers: Jason Hickel, Veerabhadran Ramanathan, Alisha Graves, Salamatou Abdourahmane
    Chair: Luke Kemp

3. New Staff

Over the last few months we have welcomed five new postdoctoral researchers and one senior research associate:

Dr Freya Jephcott – focusing on the challenge of ‘unseen epidemics’ and the effectiveness of outbreak response systems. Freya also participates in work in the field, and policy advisory work on complex health emergencies with Médecins Sans Frontières (MSF) and the World Health Organization (WHO).

John Burden - focusing on the challenges of evaluating the capability and generality of AI systems. He is working on the project Paradigms of Artificial General Intelligence and Their Associated Risks. He has a background in Computer Science and is in the final stages of completing his PhD from the University of York.

Dr Tom Hobson - focusing on the militarisation of emerging technologies, particularly biological technologies. Tom has a background in International Relations, Security Studies & STS, having completed his PhD within the Centre for War & Technology at the University of Bath.

Dr Matthijs Maas – focusing on adaptive global governance approaches for extreme technological risks, especially regimes for high-stakes or destabilising uses of AI technology, such as in military contexts. Matthijs has a PhD in Law from the University of Copenhagen.

Dr Alex McLaughlin - focusing on the moral questions that arise at the intersection of global justice and existential risk. He joins our Global Justice and Global Catastrophic Risk research area. He was previously a Leverhulme Doctoral Scholar in Climate Justice at the University of Reading

Dr Jess Whittlestone – focusing on AI ethics and policy, especially on what we can do today to ensure AI is safe and beneficial in the long-term. She joins us as a Senior Research Associate from our sister centre the Leverhulme Centre for the Future of Intelligence.

4. Policy and industry engagement – impact

We have met and engaged with policy makers and institutions across the world who are grappling with the challenge of global risks. Through these personal connections and institutional advice, we have had the opportunity to reframe key aspects of the policy debate. In addition, we continued our flourishing collaboration with corporate leaders. Extending our links improves our research and allows us to advise companies on more responsible practices.

  • The Vice-Chancellor announced that the University of Cambridge will divest from all direct and indirect investments in fossil fuels by 2030. The £3.5 billion Cambridge University Endowment Fund – the biggest in Europe – intends to ramp up investments in renewable energy as it divests from fossil fuels. Our researcher Dr Ellen Quigley was appointed in May 2019 to work with the University of Cambridge's Chief Financial Officer on responsible investment. She is the lead author of a report on the advantages and disadvantages of divestment which the University released as part of the announcement, which explores divestment across moral, social, political, reputational, and financial
    dimensions. The news was covered on the University website, the Financial Times, the Washington Post, the Telegraph, the Guardian and elsewhere.
  • Seán Ó hÉigeartaigh has joined the Global Partnership on AI, an international and multi-stakeholder initiative to guide the responsible development and use of AI. The members are Australia, Canada, France, Germany, India, Italy, Japan, Mexico, New Zealand, the Republic of Korea, Singapore, Slovenia, the United Kingdom, the United States of America, and the European Union.
  • Clarissa Rios Rojas advised the S20 on foresight policy recommendations. The S20 works with national science academies to support the G20 summit. She also contributed to a collaborative exercise with the Geneva Centre for Security Policy (GCSP) on the trends that are establishing themselves and accelerating as a result of the COVID-19 pandemic.
  • Haydn Belfield, Amritha Jayanti and Shahar Avin submitted written evidence in May to the UK's Defence Committee on the risks of militaries procuring AI defence systems ‘prematurely’ – before they are fully de-risked, safe and secure.
  • Haydn Belfield, José Hernández-Orallo, Seán Ó hÉigeartaigh, Matthijs M. Maas, Alexa Hagerty, Jess Whittlestone submitted a Response to EU White Paper on AI in June, which mainly focuses on mandatory conformity assessments for high-risk AI applications carried out by independent testing centres. Our key recommendation is to keep this proposed framework and not water it down.
  • Nick Bostrom, Haydn Belfield and CSER Research Affiliate Sam Hilton submitted evidence on UK ARPA – Key Recommendations in September to the UK Parliament Science & Technology Committee, recommending a focus on high-impact risk research and global leadership in emerging technologies.
  • Our researchers submitted evidence, which is not yet online, to the UK Parliament Science & Technology Committee on the UK’s risk assessment and management system.
  • 30 July, 6 August: CSER co-hosted two remote workshops with IPPR to explore implications of destabilisation resulting from environmental breakdown for efforts to realise more sustainable, equitable, resilient societies.
  • Simon Beard spoke on a COVID-19 panel for ALDES (the Association of Liberal Democrat Engineers and Scientists) and the Young Liberals, and was the lead author on two policy briefs on Public Mental Health and COVID-19 and AI and Data Governance issues in responding to COVID-19.
  • The APPG for Future Generations held several more evidence sessions as part of their Inquiry into long-termism in policy.

5. Public engagement – Field-building

We have also started a new Meet the Researcher series on our website. First up is Luke Kemp - here.

We’re reaching more and more people with our research

9,347 website visitors over the last two months
7,615 newsletter subscribers
10,450 Twitter followers
3,001 Facebook followers
 

6. Academic engagement - fieldbuilding

We continued our weekly work-in-progress series:

  • 27 April Lara Mani Results and Discussion of CSER Vision Staff Survey
  • 4 May Natalie Jones Procedural Rights of Minority Groups and Indigenous Peoples in International Decision Making
  • 18 May Nathan Sears Existential Security: Towards a Security Framework for the Survival of Humanity
  • 1 June Science of Global Risk updates
  • 8 June Sabin Roman Chaotic Dynamics in the Chinese Dynasties
  • 15 June Matthew Adler Risk Regulation
  • 22 June Andrew Futter Towards a Third Nuclear Age
  • 29 June Ellen Quigley Universal Ownership in the Age of COVID-19: Social Norms, Feedback Loops, and the Double Hermeneutic
  • 13 July Tom Moynihan
  • 20 July Natalie Jones The SDGs and COVID-19: Reporting back from the UN High-Level Political Forum
  • 7 September Shahar Edgerton Avin, Lara Mani, Jess Whittlestone, Tom Hobson COVID Implementation Failures project update
  • 21 September Seth Baum The Global Catastrophic Risk Institute’s recent work

7. Publications

Journal articles:

  • Luke Kemp, Laura Adam, Christian R. Boehm, Rainer Breitling, Rocco Casagrande, Malcolm Dando, Appolinaire Djikeng, Nicholas Evans, Richard Hammond, Kelly Hills, Lauren Holt, Todd Kuiken, Alemka Markotić, Piers Millett, Johnathan A Napier, Cassidy Nelson, Seán Ó hÉigeartaigh, Anne Osbourn, Megan Palmer, Nicola J Patron, Edward Perello, Wibool Piyawattanametha, Vanessa Restrepo-Schild, Clarissa Rios Rojas, Catherine Rhodes, Anna Roessing, Deborah Scott, Philip Shapira, Christopher Simuntala, Robert DJ Smith, Lalitha Sundaram, Eriko Takano, Gwyn Uttmark, Bonnie Wintle, Nadia B Zahra, William Sutherland. (2020). Point of View: Bioengineering horizon scan 2020. eLife.

“Horizon scanning is intended to identify the opportunities and threats associated with technological, regulatory and social change. In 2017 some of the present authors conducted a horizon scan for bioengineering (Wintle et al., 2017). Here we report the results of a new horizon scan that is based on inputs from a larger and more international group of 38 participants. The final list of 20 issues includes topics spanning from the political (the regulation of genomic data, increased philanthropic funding and malicious uses of neurochemicals) to the environmental (crops for changing climates and agricultural gene drives). The early identification of such issues is relevant to researchers, policy-makers and the wider public.”

“Achieving the global benefits of artificial intelligence (AI) will require international cooperation on many areas of governance and ethical standards, while allowing for diverse cultural perspectives and priorities. There are many barriers to achieving this at present, including mistrust between cultures, and more practical challenges of coordinating across different locations. This paper focuses particularly on barriers to cooperation between Europe and North America on the one hand and East Asia on the other, as regions which currently have an outsized impact on the development of AI ethics and governance. We suggest that there is reason to be optimistic about achieving greater cross-cultural cooperation on AI ethics and governance. We argue that misunderstandings between cultures and regions play a more important role in undermining cross-cultural trust, relative to fundamental disagreements, than is often supposed. Even where fundamental differences exist, these may not necessarily prevent productive cross-cultural cooperation, for two reasons: (1) cooperation does not require achieving agreement on principles and standards for all areas of AI; and (2) it is sometimes possible to reach agreement on practical issues despite disagreement on more abstract values or principles. We believe that academia has a key role to play in promoting cross-cultural cooperation on AI ethics and governance, by building greater mutual understanding, and clarifying where different forms of agreement will be both necessary and possible. We make a number of recommendations for practical steps and initiatives, including translation and multilingual publication of key documents, researcher exchange programmes, and development of research agendas on cross-cultural topics.”

“Artificial intelligence tools can help save lives in a pandemic. However, the need to implement technological solutions rapidly raises challenging ethical issues. We need new approaches for ethics with urgency, to ensure AI can be safely and beneficially used in the COVID-19 response and beyond.”

“AI safety often analyses a risk or safety issue, such as interruptibility, under a particular AI paradigm, such as reinforcement learning. But what is an AI paradigm and how does it affect the understanding and implications of the safety issue? Is AI safety research covering the most representative paradigms and the right combinations of paradigms with safety issues? Will current research directions in AI safety be able to anticipate more capable and powerful systems yet to come? In this paper we analyse these questions, introducing a distinction between two types of paradigms in AI: artefacts and techniques. We then use experimental data of research and media documents from AI Topics, an official publication of the AAAI, to examine how safety research is distributed across artefacts and techniques. We observe that AI safety research is not sufficiently anticipatory, and is heavily weighted towards certain research paradigms. We identify a need for AI safety to be more explicit about the artefacts and techniques for which a particular issue may be applicable, in order to identify gaps and cover a broader range of issues.”

“The landscape of AI safety is frequently explored differently by contrasting specialised AI versus general AI (or AGI), by analysing the short-term hazards of systems with limited capabilities against those more long-term risks posed by ‘superintelligence’, and by conceptualising sophisticated ways of bounding control an AI system has over its environment and itself (impact, harm to humans, self-harm, containment, etc.). In this position paper we reconsider these three aspects of AI safety as quantitative factors –generality, capability and control–, suggesting that by defining metrics for these dimensions, AI risks can be characterised and analysed more precisely. As an example, we illustrate how to define these metrics and their values for some simple agents in a toy scenario within a reinforcement learning setting.”

“The question of how to sustain human development in the current ecological and institutional landscape is arguably one of the utmost scientific and administratively challenging contemporary dilemmas. In response to this issue, the concept of Sustainable Development was proposed by the United Nations to inform policies for societal and human development. However, for national governments, the prevalent sustainability schemes summon more confusion than coherence. This is due to the frequent and inconsistent ways the concept of sustainability is put into practice, and consequently, difficulties in measuring and managing sustainability. The ability to evaluate how sustainable public projects are, will remain deficient if sustainability remains a notion open for interpretation. This perspective article maintains that the capital theory approach to sustainability stands out as the most rigorous framework on the topic. The capital theory is a state-centric system of ideas where national governments oversee a portfolio of capital stocks of four families: natural capital, economic capital, human capital, and social capital. It is the duty of the government to act on the capital flow between different stocks to allow sustainable long-term development. This perspective paper underscores several envisaged gains from the application of the capital theory approach in public policy. Considering these expected gains, policy makers should be encouraged to experiment with the approach.”

“Civilization has steered itself onto a vicious spiral. The modern system of agriculture, upon which global food security hinges, devours the planet’s scarce supplies of fertile lands, fresh water, productive fisheries and forest ecosystems. 821 million lives hang in the balance, already suffering famine and all forms of malnutrition, while early signs of an even larger catastrophe begin to transpire. Instead of perpetuating self-undermining dynamics, the international science and policy communities should radically reform the methods of food production and provision. Systems thinking should lend insight.”

“We welcome Seth Baum's reply to our paper. While we are in broad agreement with him on the range of topics covered, this particular field of research remains very young and undeveloped and we think that there are many points on which further reflection is needed. We briefly discuss three: the normative aspects of terms like 'existential catastrophe,' the opportunities for low hanging fruit in method selection and application and the importance of context when making probability claims.”

Reports and preprints:

“The University of Cambridge holds assets of approximately £3.5 billion, the largest university endowment in Europe. Within the University there is broad agreement about the urgent need to reduce carbon emissions. However, whether full divestment of University funds from fossil fuel assets is the best way to make that happen has been the subject of intense debate. Based on a review of the academic literature, interviews and focus groups with relevant stakeholders inside and outside the University, records of University and college discussions, and some further primary data collection, this report explores the advantages and disadvantages of a policy of fossil fuel divestment across its moral, social, political, reputational, and financial dimensions, ending with a summary of costed divestment scenarios for the University.”

  • Haydn Belfield, Amritha Jayanti, Shahar Avin. (2020). Premature Procurement. UK Parliament Defence Committee, Written Evidence - Defence industrial policy: procurement and prosperity.

“Our submission is on the risks of militaries procuring AI defence systems ‘prematurely’ – before they are fully derisked, safe and secure. We therefore make the following recommendations to protect against premature and/or unsafe procurement and deployment of ML-based systems:
— Improve systemic risk assessment in defence procurement.
— Ensure clear lines of responsibility so that senior officials can be held responsible for errors caused in the procurement chain and are therefore incentivised to reduce them;
— Acknowledge potential shifts in international standards for autonomous systems, and build flexible procurement standards accordingly.
— Update the MoD’s definition of lethal autonomous weapons - the Integrated Security, Defence and Foreign Policy Review provides an excellent opportunity to bring the UK in line with its allies.”

“The submission mainly focuses on mandatory conformity assessments for high-risk AI applications carried out by independent testing centres. Our key recommendation is to keep this proposed framework and not water it down. In this submission we 1) support the Commission’s proposed structure, defend this approach on technical, policy, practical, and ethical grounds, and offer some considerations for future extensions; and 2) offer some specific recommendations for the mandatory requirements.”

“Supply chains of staple crops, in developed and developing regions, are vulnerable to an array of disturbances and disruptions. These include biotic, abiotic and institutional risk factors. Artificial intelligence (AI) systems have the potential to mitigate some of these vulnerabilities across supply chains, and thereby improve the state of global food security. However, the particular properties of each supply chain phase, from "the farm to the fork," might suggest that some phases are more vulnerable to risks than others. Furthermore, the social circumstances and technological environment of each phase may indicate that several phases of the supply chains will be more receptive to AI adoption and deployment than others. This research paper seeks to test these assumptions to inform the integration of AI in agricultural supply chains. It employs a supply chain risk management approach (SCRM) and draws on a mix-methods research design. In the qualitative component of the research, interviews are conducted with agricultural supply chain and food security experts from the Food and Agricultural Organization of the UN (FAO), the World Bank, CGIAR, the World Food Program (WFP) and the University of Cambridge. In the quantitative component of the paper, seventy-two scientists and researchers in the domains of digital agriculture, big data in agriculture and agricultural supply chains are surveyed. The survey is used to generate assessments of the vulnerability of different phases of supply chains to biotic, abiotic and institutional risks, and the ease of AI adoption and deployment in these phases. The findings show that respondents expect the vulnerability to risks of all but one supply chain phases to increase over the next ten years. Importantly, where the integration of AI systems will be most desirable, in highly vulnerable supply chain phases in developing countries, the potential for AI integration is likely to be limited. To the best of our knowledge, the methodical examination of AI through the prism of agricultural SCRM, drawing on expert insights, has never been conducted. This paper carries out a first assessment of this kind and provides preliminary prioritizations to benefit agricultural SCRM as well as to guide further research on AI for global food security.”

  • Nick Bostrom, Haydn Belfield, Sam Hilton. (2020). UK ARPA – Key Recommendations. UK Parliament Science & Technology Committee, Written Evidence - A new UK funding agency.

Recommends focussing on high-impact risk research and global leadership in emerging technologies. Also makes recommendations relating to structure and culture, mechanisms to identity and start projects, testing and evaluation, recommended tools and hiring.

  • Sam Hilton, Haydn Belfield. (Not yet online). The UK Government’s risk management system. UK Parliament Science & Technology Committee, Written Evidence - UK Science, Research and Technology Capability and Influence in Global Disease Outbreaks.

Recommends several improvements to the UK Government’s risk management system.

“The mental health impacts of COVID-19 are expected to be very significant and could be the single greatest health burden caused by this pandemic in the long term. However, at present they are not being assessed at the population level, while not enough is being done to study these impacts and how to respond to them. This briefing note explores the issues raised around mental health in the context of the COVID-19 pandemic, and encourages a genuine and robust compassionate response to COVID-19 which also requires ambitious and creative thinking in response to many social and economic problems.”

“AI and data-driven technologies have the potential to support pandemic prevention and control at many stages, including improving response and recovery for COVID-19. As well as having direct benefits, the increased use of AI and data technologies could increase social understanding of their benefits and public trust in them. However, the speed at which new technologies are being considered and implemented raises a number of concerns. This briefing note discusses these concerns, including those specific to the NHS contact tracing app and also discusses some of the broader governance issues surrounding AI ethics.”We send short monthly updates in our newsletter – subscribe here.

24

0
0

Reactions

0
0
Mentioned in

More posts like this

Comments
No comments on this post yet.
Be the first to respond.
More from HaydnBelfield
View more
Curated and popular this week
Relevant opportunities
View more