Tag Archives: Shahar Avin

Hardware policies best way to manage AI safety?

Regulation of artificial intelligence (AI) has become very topical in the last couple of years. There was an AI safety summit in November 2023 at Bletchley Park in the UK (see my November 2, 2023 posting for more about that international meeting).

A very software approach?

This year (2024) has seen a rise in legislative and proposed legislative activity. I have some articles on a few of these activities. China was the first to enact regulations of any kind on AI according to Matt Sheehan’s February 27, 2024 paper for the Carnegie Endowment for International Peace,

In 2021 and 2022, China became the first country to implement detailed, binding regulations on some of the most common applications of artificial intelligence (AI). These rules formed the foundation of China’s emerging AI governance regime, an evolving policy architecture that will affect everything from frontier AI research to the functioning of the world’s second-largest economy, from large language models in Africa to autonomous vehicles in Europe.

The Chinese Communist Party (CCP) and the Chinese government started that process with the 2021 rules on recommendation algorithms, an omnipresent use of the technology that is often overlooked in international AI governance discourse. Those rules imposed new obligations on companies to intervene in content recommendations, granted new rights to users being recommended content, and offered protections to gig workers subject to algorithmic scheduling. The Chinese party-state quickly followed up with a new regulation on “deep synthesis,” the use of AI to generate synthetic media such as deepfakes. Those rules required AI providers to watermark AI-generated content and ensure that content does not violate people’s “likeness rights” or harm the “nation’s image.” Together, these two regulations also created and amended China’s algorithm registry, a regulatory tool that would evolve into a cornerstone of the country’s AI governance regime.

The UK has adopted a more generalized approach focused on encouraging innovation according to Valeria Gallo’s and Suchitra Nair’s February 21, 2024 article for Deloitte (a British professional services firm also considered one of the big four accounting firms worldwide),

At a glance

The UK Government has adopted a cross-sector and outcome-based framework for regulating AI, underpinned by five core principles. These are safety, security and robustness, appropriate transparency and explainability, fairness, accountability and governance, and contestability and redress.

Regulators will implement the framework in their sectors/domains by applying existing laws and issuing supplementary regulatory guidance. Selected regulators will publish their AI annual strategic plans by 30th April [2024], providing businesses with much-needed direction.

Voluntary safety and transparency measures for developers of highly capable AI models and systems will also supplement the framework and the activities of individual regulators.

The framework will not be codified into law for now, but the Government anticipates the need for targeted legislative interventions in the future. These interventions will address gaps in the current regulatory framework, particularly regarding the risks posed by complex General Purpose AI and the key players involved in its development.

Organisations must prepare for increased AI regulatory activity over the next year, including guidelines, information gathering, and enforcement. International firms will inevitably have to navigate regulatory divergence.

While most of the focus appears to be on the software (e.g., General Purpose AI), the UK framework does not preclude hardware.

The European Union (EU) is preparing to pass its own AI regulation act through the European Parliament in 2024 according to a December 19, 2023 “EU AI Act: first regulation on artificial intelligence” article update, Note: Links have been removed,

As part of its digital strategy, the EU wants to regulate artificial intelligence (AI) to ensure better conditions for the development and use of this innovative technology. AI can create many benefits, such as better healthcare; safer and cleaner transport; more efficient manufacturing; and cheaper and more sustainable energy.

In April 2021, the European Commission proposed the first EU regulatory framework for AI. It says that AI systems that can be used in different applications are analysed and classified according to the risk they pose to users. The different risk levels will mean more or less regulation.

The agreed text is expected to be finally adopted in April 2024. It will be fully applicable 24 months after entry into force, but some parts will be applicable sooner:

*The ban of AI systems posing unacceptable risks will apply six months after the entry into force

*Codes of practice will apply nine months after entry into force

*Rules on general-purpose AI systems that need to comply with transparency requirements will apply 12 months after the entry into force

High-risk systems will have more time to comply with the requirements as the obligations concerning them will become applicable 36 months after the entry into force.

This EU initiative, like the UK framework, seems largely focused on AI software and according to the Wikipedia entry “Regulation of artificial intelligence,”

… The AI Act is expected to come into effect in late 2025 or early 2026.[109

I do have a few postings about Canadian regulatory efforts, which also seem to be focused on software but don’t preclude hardware. While the January 20, 2024 posting is titled “Canada’s voluntary code of conduct relating to advanced generative AI (artificial intelligence) systems,” information about legislative efforts is also included although you might find my May 1, 2023 posting titled “Canada, AI regulation, and the second reading of the Digital Charter Implementation Act, 2022 (Bill C-27)” offers more comprehensive information about Canada’s legislative progress or lack thereof.

The US is always to be considered in these matters and I have a November 2023 ‘briefing’ by Müge Fazlioglu on the International Association of Privacy Professionals (IAPP) website where she provides a quick overview of the international scene before diving deeper into US AI governance policy through the Barack Obama, Donald Trump, and Joe Biden administrations. There’s also this January 29, 2024 US White House “Fact Sheet: Biden-⁠Harris Administration Announces Key AI Actions Following President Biden’s Landmark Executive Order.”

What about AI and hardware?

A February 15, 2024 news item on ScienceDaily suggests that regulating hardware may be the most effective way of regulating AI,

Chips and datacentres — the ‘compute’ power driving the AI revolution — may be the most effective targets for risk-reducing AI policies as they have to be physically possessed, according to a new report.

A global registry tracking the flow of chips destined for AI supercomputers is one of the policy options highlighted by a major new report calling for regulation of “compute” — the hardware that underpins all AI — to help prevent artificial intelligence misuse and disasters.

Other technical proposals floated by the report include “compute caps” — built-in limits to the number of chips each AI chip can connect with — and distributing a “start switch” for AI training across multiple parties to allow for a digital veto of risky AI before it feeds on data.

The experts point out that powerful computing chips required to drive generative AI models are constructed via highly concentrated supply chains, dominated by just a handful of companies — making the hardware itself a strong intervention point for risk-reducing AI policies.

The report, published 14 February [2024], is authored by nineteen experts and co-led by three University of Cambridge institutes — the Leverhulme Centre for the Future of Intelligence (LCFI), the Centre for the Study of Existential Risk (CSER) and the Bennett Institute for Public Policy — along with OpenAI and the Centre for the Governance of AI.

A February 14, 2024 University of Cambridge press release by Fred Lewsey (also on EurekAlert), which originated the news item, provides more information about the ‘hardware approach to AI regulation’,

“Artificial intelligence has made startling progress in the last decade, much of which has been enabled by the sharp increase in computing power applied to training algorithms,” said Haydn Belfield, a co-lead author of the report from Cambridge’s LCFI. 

“Governments are rightly concerned about the potential consequences of AI, and looking at how to regulate the technology, but data and algorithms are intangible and difficult to control.

“AI supercomputers consist of tens of thousands of networked AI chips hosted in giant data centres often the size of several football fields, consuming dozens of megawatts of power,” said Belfield.

“Computing hardware is visible, quantifiable, and its physical nature means restrictions can be imposed in a way that might soon be nearly impossible with more virtual elements of AI.”

The computing power behind AI has grown exponentially since the “deep learning era” kicked off in earnest, with the amount of “compute” used to train the largest AI models doubling around every six months since 2010. The biggest AI models now use 350 million times more compute than thirteen years ago.

Government efforts across the world over the past year – including the US Executive Order on AI, EU AI Act, China’s Generative AI Regulation, and the UK’s AI Safety Institute – have begun to focus on compute when considering AI governance.

Outside of China, the cloud compute market is dominated by three companies, termed “hyperscalers”: Amazon, Microsoft, and Google. “Monitoring the hardware would greatly help competition authorities in keeping in check the market power of the biggest tech companies, and so opening the space for more innovation and new entrants,” said co-author Prof Diane Coyle from Cambridge’s Bennett Institute. 

The report provides “sketches” of possible directions for compute governance, highlighting the analogy between AI training and uranium enrichment. “International regulation of nuclear supplies focuses on a vital input that has to go through a lengthy, difficult and expensive process,” said Belfield. “A focus on compute would allow AI regulation to do the same.”

Policy ideas are divided into three camps: increasing the global visibility of AI computing; allocating compute resources for the greatest benefit to society; enforcing restrictions on computing power.

For example, a regularly-audited international AI chip registry requiring chip producers, sellers, and resellers to report all transfers would provide precise information on the amount of compute possessed by nations and corporations at any one time.

The report even suggests a unique identifier could be added to each chip to prevent industrial espionage and “chip smuggling”.

“Governments already track many economic transactions, so it makes sense to increase monitoring of a commodity as rare and powerful as an advanced AI chip,” said Belfield. However, the team point out that such approaches could lead to a black market in untraceable “ghost chips”.

Other suggestions to increase visibility – and accountability – include reporting of large-scale AI training by cloud computing providers, and privacy-preserving “workload monitoring” to help prevent an arms race if massive compute investments are made without enough transparency.  

“Users of compute will engage in a mixture of beneficial, benign and harmful activities, and determined groups will find ways to circumvent restrictions,” said Belfield. “Regulators will need to create checks and balances that thwart malicious or misguided uses of AI computing.”

These might include physical limits on chip-to-chip networking, or cryptographic technology that allows for remote disabling of AI chips in extreme circumstances. One suggested approach would require the consent of multiple parties to unlock AI compute for particularly risky training runs, a mechanism familiar from nuclear weapons.

AI risk mitigation policies might see compute prioritised for research most likely to benefit society – from green energy to health and education. This could even take the form of major international AI “megaprojects” that tackle global issues by pooling compute resources.

The report’s authors are clear that their policy suggestions are “exploratory” rather than fully fledged proposals and that they all carry potential downsides, from risks of proprietary data leaks to negative economic impacts and the hampering of positive AI development.

They offer five considerations for regulating AI through compute, including the exclusion of small-scale and non-AI computing, regular revisiting of compute thresholds, and a focus on privacy preservation.

Added Belfield: “Trying to govern AI models as they are deployed could prove futile, like chasing shadows. Those seeking to establish AI regulation should look upstream to compute, the source of the power driving the AI revolution. If compute remains ungoverned it poses severe risks to society.”

You can find the report, “Computing Power and the Governance of Artificial Intelligence” on the University of Cambridge’s Centre for the Study of Existential Risk.

Authors include: Girish Sastry, Lennart Heim, Haydn Belfield, Markus Anderljung, Miles Brundage, Julian Hazell, Cullen O’Keefe, Gillian K. Hadfield, Richard Ngo, Konstantin Pilz, George Gor, Emma Bluemke, Sarah Shoker, Janet Egan, Robert F. Trager, Shahar Avin, Adrian Weller, Yoshua Bengio, and Diane Coyle.

The authors are associated with these companies/agencies: OpenAI, Centre for the Governance of AI (GovAI), Leverhulme Centre for the Future of Intelligence at the Uni. of Cambridge, Oxford Internet Institute, Institute for Law & AI, University of Toronto Vector Institute for AI, Georgetown University, ILINA Program, Harvard Kennedy School (of Government), *AI Governance Institute,* Uni. of Oxford, Centre for the Study of Existential Risk at Uni. of Cambridge, Uni. of Cambridge, Uni. of Montreal / Mila, Bennett Institute for Public Policy at the Uni. of Cambridge.

“The ILINIA program is dedicated to providing an outstanding platform for Africans to learn and work on questions around maximizing wellbeing and responding to global catastrophic risks” according to the organization’s homepage.

*As for the AI Governance Institute, I believe that should be the Centre for the Governance of AI at Oxford University since the associated academic is Robert F. Trager from the University of Oxford.

As the months (years?) fly by, I guess we’ll find out if this hardware approach gains any traction where AI regulation is concerned.