TLDR;
This sketch floats the idea of a conglomerate of AI-Labs and AI-Safety-Orgs. It is inspired by the seemingly functional Roundtable for Sustainable Palm Oil (RSPO). It also lists differences to the already existing Partnership on AI (PAI), some reasons why a Roundtable for Safe AI (RSAI) might not work/have adverse effects, and questions that might be interesting.
Epistemic Status: very draft-y shallow dive. I’m busy with other things at the moment, but maybe this could be useful for someone somewhere.
Past Discussion about safety standards
Jade Leung, Cullen O’keefe and Markus Anderljung outline the importance and process of institutionalized technical safety standards for AI in their post and emphasize the importance of “deepening engagement with relevant standard setting organizations (SSOs) and AI regulation, translating emerging TAI safety best practices into technical safety standards, and investigating what an ideal SSO for TAI safety would look like.” A Roundtable for Safe AI could be a mixture of SSO and Credibility Label for Regulators.
What is the RSPO?
The RSPO is an organisation founded in 2004 and describes itself as a “a not-for-profit that unites stakeholders from the 7 sectors of the palm oil industry: oil palm producers, processors or traders, consumer goods manufacturers, retailers, banks/investors, and environmental and social non-governmental organisations (NGOs), to develop and implement global standards for sustainable palm oil.” NGOs and conscientious corporations should right the wrongs of a faulty industry through certificates and membership duties. This proposal rests firmly on the belief that NGOs and participating corporations have common goals, or at least the same means to reach their goals. The RSPO proclaimed its first “Principles and Criteria”(P&C). This document, among other things, defines the term “sustainability” and the criteria to be met by applying members. The interpretation of the P&C varies internationally and was revised three times so far.
The success of the RSPO in getting a considerable number of big corporations to comply with sustainability standards sparked similar initiatives in other areas like the Roundtable on Sustainable Biofuels, Roundtable on Sustainable Biomaterials,[22] Roundtable on Sustainable Forests,[23] Roundtable on Sustainable Development,[24] Roundtable on Responsible Soy,[25] and Roundtable for a Sustainable Cocoa Economy.[26]
What would such an organisation look like in AI?
A Roundtable for Safe AI (RSAI) would provide a forum through which AI Safety and AI Industry Leaders could collaborate and work on benchmarks together. It would institutionalize frequent conversations and provide an industry centred and -supported seal of quality and credibility. The roundtable could have a few well sounding “goals”, but could also have a clause where industry members need to allow audits by the RSAI or have to publish reports on how they comply with RSAI Standards. How deep and intrusive these audits/standards could be depends, of course, on the success of AI Safety organisations to lobby within the RSAI, the credibility and public attention of the RSAI, etc.
How is this different from the Partnership on AI (PAI)
In short, it would include binding conditions for members. Whoever would like to be a member of the RSAI would have to prove to comply with the standards the RSAI sets itself. The RSAI would provide a seal of credibility, rather than just “Convenings” and “High Quality Resources.” A successful RSAI would both set standards and enforce them by threatening any actor who doesn’t comply with them with expulsion. Although the existence of an institution like the PAI does lower the chance of something like the RSAI having a lot of positive impact, mainly because there seem to be less of an incentive for industry actors to join it and comply with its standards.
How likely could this change the behaviour of AI Industry Actors?
This question seems to depend largely on whether the RSAI can build enough pressure to get AI Labs to join and comply. In the RSPO, Membership is voluntary, but comes with a bunch of benefits. For example, In the Swiss-Indonesian free trade agreement, it was agreed that only RSPO Members would be included. An RSAI Membership, conversely, could provide an indication for regulatory bodies that certain AI Labs care more about safety than others. That doesn’t seem to exist yet. Crucially, it could, from the beginning, suggest internationally applicable benchmarks and auditing processes.
Reasons for Scepticism
There are not nearly as many incentives for AI industry actors to join as there are for palm oil firms because the public perception of AI Labs doesn’t seem to be as bad. Industry Actors don’t want to cooperate because it heightens the risk of losing valuable intellectual property. With the PAI already existing (where lots of important Non-Profits like the Future of Life Institute are already involved), something like an RSAI might have very little to no impact. It could even diminish the impact of Non Profits in PAI. The project might be net negative because of the “greenwashing” effect it could have, if too many influential AI Industry Leaders get involved early on.
Questions to research for people interested
- How promising is the work of the PAI? How successful has it been in the past?
- What are the differences/similarities between Palm Oil and the AI Industry that would make an RSAI more or less promising?
- How likely are “greenwashing”-effects for an organisation like this?
Sources and Further Reading:
Primary Sources
Minutes of the preparatory meeting. Hayes (London), September 20, 2002 Reinier de Man Judit Juranics Round Table on Sustainable Palm Oil, Leiden, 15 October 2002.
Greenpeace Activists Prevent Sinar Mas Palm Oil Tanker from Loading in Indonesia: Greenpeace Challenges RSPO To Stop Green-Washing Member Companies. Press release. November 14, 2008.
Article: Unilever Takes Stance Against Deforestation
Literature
Bailis, Robert, Baka, Jennifer: Constructing Sustainable Biofuels. Governance of the Emerging Biofuel Economy, in: Annals of the Association of American Geographers 101 (4), 2011, S. 827–838.
Burns, Tom R. et al.: The Promise and Limitations of Partnered Governance. The Case of Sustainable Palm Oil, in: Corporate Governance: The International Journal of Business in Society 10 (1), 2010, S. 59–72.
Butler, Rhett, et al.: Improving the Performance of the Roundtable on Sustainable Palm Oil for Nature Conservation, in: Conservation Biology 24 (2), 2010, S. 377–381.
Corley, R.H.V, Tinker, P.B.: The Oil Palm: Fifth Edition, 52016.
Cramb, Robert, McCarthy, John F.: The Oil Palm Complex: Smallholders, Agribusiness and the State in Indonesia and Malaysia, Singapore 2016.
Daviron, Benoit, Djama, Marcel; Managerial Rationality and Power Reconfiguration in The Multi-Stakeholder Initiatives for Agricultural Commodities: The Case of The Roundtable for Sustainable Palm Oil (RSPO), Paris 2010.
Fouilleux, Eve, Loconto, Allison: Politics of Private Regulation. ISEAL and the Shaping of Transnational Sustainability Governance, in: Regulation & Governance 8 (2), 2014, S. 166–185.
Gelder, Jan Willem van, Kusumaningtyas, Retno,: Toward Responsible and Inclusive Financing of The Palm Oil Sector, Bogor 2017.
Nesadurai, Helen E. S.: Food Security, The Palm Oil–Land Conflict Nexus, and Sustainability: A Governance Role for a Private Multi-Stakeholder Regime Like the RSPO? in: Pacific Review 26 (5), 2013, S. 505–529.
Pye, Oliver, Deconstructing the Roundtable on Sustainable Palm Oil, in: Mccarthy, John, Cramb, Rob (Hrsg.): The Oil Palm Complex: Smallholders, Agribusiness and the State in Indonesia and Malaysia, 2016, S. 409-44
Ruysschaert, Denis, Salles, Denis: Towards Global Voluntary Standards: Questioning the Effectiveness in Attaining Conservation Goals: The Case of the Roundtable on Sustainable Palm Oil (RSPO), in: Ecological Economics (1) 107, 2014, S. 438–446.
Glasbergen, Pieter, Schouten, Greetje: Creating Legitimacy in Global Private Governance: The Case of the Roundtable on Sustainable Palm Oil, in: Ecological Economics 70 (11), 2011, S. 1891–1899.
Big thanks to: Tobias Häberli, Tobias Pulver, Morgan Simpson and Michel Justen for looking over this and giving Feedback!