Expert opinions on Research Policy

The Republic of Korea declared our vision for establishing global AI hub together with 9 participating UN and related organizations to contribute to strengthening of global AI capacity.

But the main regret I have with this talk is that I wish I had prepared a more for the section on a “CERN for AI.” The idea behind such a project is to pool many countries’ and companies’ resources into a single (possibly physically decentralized) ci...

There are many conceivable versions of a “CERN for AI.” But the version that seems politically realistic (a new catch-up lab) probably would not do much for safety, while the versions that would materially improve safety (e.g., pause + merge of all c...

Dedicated AIRR capacity supports the AI Security Institute and the Sovereign AI Unit, including guaranteed resource for mission-critical AI safety research, and strategically selected UK AI startups.

Expanding access to AI compute capacity, and promoting sustainable, resilient and environmentally responsible digital infrastructure, and supporting innovators while enhancing Canada and Finland’s shared economic security and prosperity. Exploring co...

Building on the findings of a report on international AI verification, I contend that verifying behavior in the domain of artificial intelligence is possible, but it is not trivial. The problem of AI verification generally looks less like overseeing ...

A one-size-fits-all approach to AI governance will not be effective. We all face different risks, priorities, levels of digital development, and regulatory environments. Effective governance requires shared principles that can be implemented flexibly...

We think that the smart way to balance these two truths is to allow AI developers to write their own safety standards, but require them to be transparent about what they are, so we can hold them to their own promises.

Our view is that AI is a general-purpose technology with a wide range of applications, and also that regulators understand well the sectors they are regulating. They understand the risks that are present and understand, as many noble Lords have raise...

A more tractable near-term proposal is mandatory pre-release evaluation against a defined set of uplift benchmarks, particularly for chemical and biological capability, with results published before weights ship.

However, it is important to differentiate regulating applications (which we need) vs. regulating the technology (which is ill-advised).

The AI Act Whistleblower Tool empowers individuals to securely submit a report and contribute directly to make AI in Europe safe transparent, and trustworthy. [...] Alongside with the secure tool, the AI Office commits itself to a high standard of co...

Connecticut already has strong whistleblower laws that protect employees from retaliation. [...] creating a separate, AI-specific whistleblower regime could lead to confusion, duplication, and unintended consequences for employers operating in highly...

When we integrate AI capabilities into our platform, we use API endpoints and vendors that do not feed your data back into model training. Opt-in vs. opt-out matters more than you think. Even with opt-in as our policy, we’re not rushing to flip the s...

Restricting access to training data, whether through mandatory compensation, expanded opt-out rights, or onerous compliance requirements, can have the effect of restricting AI development itself.

Plan A is not that crazy, but it’s close: the algorithms are mandated to be open source, and while the weights aren’t (in fact the weights are banned from being OS), access to the weights is mandated to be open; members of the public can do evaluatio...

Open innovation is foundational to AI progress, and broad access to models, data, and tools is essential to accelerating breakthroughs. By adopting the Linux Foundation’s OpenMDW framework for NVIDIA open model families, we’re helping establish a sim...

Such technological advancement will provide society with invaluable tools to address critical global challenges, including drug discovery, economic growth and climate change.

This is going to be the most productive decade in the history of our species. But in order to truly reap the benefits of AI, we need to learn how to contain it. Paradoxically, part of that will mean collectively saying no to certain forms of progress...

There’s no such thing as AGI. There is human-level AI, but human intelligence is extremely specialised. So calling it general intelligence is complete nonsense.

It always seemed entirely plausible to me that Covid escaped from a laboratory, but that doesn’t mean that it was deliberate or the result of nefarious activity.

This acknowledgement should not lead to the dismissal of the possibility (e.g., accidental leakage from a BSL-4 or lower biosafety level laboratory) or the legitimate concerns surrounding biosecurity and biosafety in general.

The story of lab leak in Wuhan is a fiction and it is as ridiculous as the counterclaim that Sars-CoV-2 comes from lab leaks in the US.

The problem is that argument doesn’t work — all evidence points to Covid starting around the same time the market outbreak started.

All he is saying in this Science article is that we need better, more comprehensive patient case and dated viral sequence data – at the time of first appearance of these viruses – to make this viral phylogenetic/evolutionary history approach more acc...

The data from the paper support the notion that Huanan market may have been just an amplifier of COVID-19 transmission, rather than the place where COVID-19 originated.

Much of the evidence used to shape the recommendations comes from epidemiological studies—large population surveys that look for associations between diet and disease. While these studies can generate hypotheses, they cannot prove cause and effect. T...

The evidence that cholesterol in one egg a day is safe for most people comes from huge studies — many conducted here at Harvard Medical School — that have followed hundreds of thousands of people over decades.

Bayesian reasoning serves as a bridge between quantitative and qualitative research that allows one to analyze any and all kinds of data within a single, unified inferential framework. This paper demonstrates the strength and flexibility of the Bayes...

The arguments around covid origins seem largely to fall into an inference trap Andrew has termed premature collapse of the wavefunction. Rather than maintaining multiple hypotheses with different levels of credibility, there seems to be a desire to c...

Determining the origin of an outbreak or pandemic can be divided into four stages: detection, identification, characterization, and attribution. Understanding how a specific pathogen entered and spread in a population to cause an outbreak is a multid...

Creating even a microscopic black hole or wormhole would require an accelerator the size of the whole universe, say researchers.

Scientists’ conclusions about risk cannot be taken at face value. Their assurances must be discounted by the probability of their own error. [...] It does not follow that LHC risk is zero: An accurate assessment of risk must include the possibility t...

Theoretical calculations show that the LHC could produce two kinds of dark matter—black holes and strange, ultradense quark matter—that are extremely dangerous, as both have been theoretically proven to swallow in a chain reaction the entirety of Ear...