AI Weekly: UN proposes moratorium on ‘risky’ AI while ICLR solicits blog posts


The Transform Technology Summits begin on October 13th with Low-Code / No Code: Enabling Enterprise Agility. Join Now!

The UN High Commissioner for Human Rights, Michelle Bachelet, this week called for a moratorium on the sale and use of AI systems that “pose a serious risk to human rights”. Bachelet said that adequate safeguards must be in place before development of such systems resumes and that any system that cannot be used in accordance with international human rights law should be banned.

“AI can be a force for good and help societies meet some of the great challenges of our time. But AI technologies can have negative, even catastrophic, effects if they are used without adequate consideration of people’s human rights, ”said Bachelet.

Of course, it is not always easy to define which systems pose a risk to human rights. In a new report, the Human Rights Council outlines a number of examples, including systems that “invade privacy” through increased use of personal data and “lead to discriminatory decisions”. However, as recent comments submitted to the European Parliament and the European Council suggest, definitions of “risky” can vary widely between stakeholders.

As Wired’s Khari Johnson recently wrote, some companies responding to the European Union’s AI law – which proposes oversight of “high-risk” AI – believe that legislation is going too far with rules that inhibit innovation and potentially costly. In the meantime, human rights groups and ethicists claim that it does not go far enough and leaves people vulnerable to those who have the resources to deploy powerful algorithms.

While an agreed definition of “risk” remains elusive, especially as companies like Alphabets DeepMind and OpenAI work towards universal multitasking systems that defy conventional labels, the Human Rights Council report shows ways to prevent and limit the harm through AI. For example, the report argues that AI development must be fair and non-discriminatory, with participation and accountability anchored as core parts of the process. It also claims that legality, legitimacy, necessity and proportionality requirements must be “consistently” applied to AI technologies that should be used in a way that “facilitates the realization of economic, social and cultural rights”.

“AI is now reaching almost every corner of our physical and mental life and even our emotional states. AI systems are used to determine who gets public services, who has a chance to get hired for a job, and of course they affect what information people can see and share online, ”said Bachelet. “We cannot afford to catch up any further on AI – allowing its use with limited or no boundaries or oversight and dealing with the almost inevitable human rights implications … for the benefit of all of us.”

ICLR introduces a blog post track

In other news this week, the International Conference on Learning Representations (ICLR), one of the largest machine learning conferences in the world, announced a call for submissions for the first-ever blog post track. The aim is to solicit submissions in blog format that will allow researchers to discuss previously published research that has been accepted by the ICLR.

“[Blog Post Track] recognizes and values ​​summary work as opposed to novel work, ”Sebastien Bubeck, ICLR blog post chairman and senior principal research manager at Microsoft Research, told VentureBeat via email. “For example, certain published work may contain difficult and techno-mathematical evidence for fairly abstract situations. Blog posts in this case could elaborate a specific sub-case of this general theory and distill the findings into some practical examples. Alternatively, a contribution could propose a new, simpler proof for the same result or perhaps combine the proof with ideas in other areas of computer science. “

Bubeck believes that encouraging researchers to review older peer-reviewed papers could enable them to point out the shortcomings of studies and help synthesize knowledge in the AI ​​community. He traces the initiative back to the time after the Second World War in France, when a collective of mathematicians under the pseudonym Nicolas Bourbaki decided to write a series of textbooks on the fundamentals of mathematics.

“For more applied work, blog posts could be a great way to revisit experiments with the overall goal [of helping] with the reproducibility crisis in machine learning. Indeed… unlike major conference papers, blog posts could focus on experimentation on a smaller scale to investigate whether certain phenomena are due to size or whether they are intrinsic to the architecture or the problem at hand, ”said Bubeck.

Like many scientific fields, AI has a reproducibility problem. Studies often provide benchmark results instead of source code, which becomes problematic when the thoroughness of the benchmarks is questioned. A recent report found that 60 to 70% of the answers provided by natural language processing models were embedded somewhere in the benchmark training sets, suggesting that the models often simply memorized the answers. Another study – a meta-analysis of over 3,000 AI papers – found that metrics used to benchmark AI and machine learning models tended to be inconsistent, irregularly tracked, and not particularly informative.

According to Bubeck, the conference chairs will select the reviewers for the submitted contributions for the first edition of the blog post track at ICLR. In the future, he hopes to bring blog post tracks to more computer science conferences – not just those primarily focused on AI and machine learning.

“Blog posts offer the opportunity to discuss scientific ideas informally. They offer significant value to the scientific community by providing a flexible platform to encourage open, human and transparent discussions about new knowledge or the limits of scientific publication, ”said Bubeck.

If you’d like to cover AI, send news tips to Kyle Wiggers – and subscribe to the weekly AI newsletter and bookmark our AI channel, The Machine.

Thank you for reading,

Kyle Wiggers

Author of AI staff


VentureBeat’s mission is to be a digital marketplace for technical decision makers to gain knowledge about transformative technologies and transactions. Our website provides essential information on data technologies and strategies to help you run your organization. We invite you to become a member of our community to gain access:

  • current information on the topics of interest to you
  • our newsletters
  • closed thought leadership content and discounted access to our award-winning events such as Transform 2021: Learn more
  • Network functions and more

become a member