What was the ruling?
The decision states that the EPA’s actions in a 2015 rule, which included caps on emissions from power plants, overstepped the agency’s authority to regulate greenhouse gas emissions.
“Capping carbon dioxide emissions at a level that will force a nationwide transition away from the use of coal to generate electricity may be a sensible ‘solution to the crisis of the day,’” the decision reads, “But it is not plausible that Congress gave EPA the authority to adopt on its own such a regulatory scheme.
Only congress has the power to make “a decision of such magnitude and consequence,” it continues.
This decision is likely to have “broad implications” says Deborah Sivas, an environmental law professor at Stanford University. Not only is the court constraining what the EPA can do on climate policy going forward, she adds. This opinion “seems to be a major blow for agency deference,” meaning agencies beyond just the EPA could face limitations in the future.
The ruling, which is the latest in a string of bombshell cases from the court, fell largely along ideological lines. Chief Justice John Roberts authored the majority opinion, and he was joined by conservatives, including Justices Samuel Alito, Amy Coney Barrett, Neil Gorsuch, Brett Kavanaugh, and Clarence Thomas. Justices Stephen Breyer, Elena Kagan, and Sonia Sotomayor dissented.
What is the decision all about?
The main question in the case was how much power the EPA should have to regulate carbon emissions and what it should be allowed to do to accomplish that job. That question was raised by a 2015 EPA rule called the Clean Power Plan.
The Clean Power Plan targeted greenhouse gas emissions from power plants, requiring each state to make a plan to cut emissions and submit it to the federal government.
Several states and private groups immediately challenged the Clean Power Plan when it was released, arguing it was an overreach from the agency, and the Supreme Court put it on hold in 2016. After a repeal of the plan during Donald Trump’s presidency and some legal back-and-forth, a Washington D.C. district court ruled in January 2021 that the Clean Power Plan did fall within the EPA’s authority.
Deep learning can almost perfectly predict how ice forms
Researchers have used deep learning techniques to model how ice crystals form in the atmosphere with much higher precision than ever before. Their paper, published this week in PNAS, hints at the potential for the new method to significantly increase the accuracy of weather and climate forecasting.
The researchers used deep learning to predict how atoms and molecules behave. First, deep learning models were trained on small-scale simulations of 64 water molecules to help them predict how electrons in atoms interact. The models then replicated those interactions on a larger scale, with more atoms and molecules. It’s this ability to precisely simulate electron interactions that allowed the team to accurately predict physical and chemical behavior.
“The properties of matter emerge from how electrons behave,” says Pablo Piaggi, a research fellow at Princeton University and the lead author on the study. “Simulating explicitly what happens at that level is a way to capture much more rich physical phenomena.”
It’s the first time this method has been used to model something as complex as the formation of ice crystals, also known as ice nucleation. This development may eventually improve the accuracy of weather and climate forecasting, because the formation of ice crystals is one of the first steps in the formation of clouds, which is where all precipitation comes from.
Xiaohong Liu, professor of atmospheric sciences at Texas A&M University, who was not involved in the study, says half of all precipitation events—whether it’s snow or rain or sleet—begin as ice crystals, which then grow larger and result in precipitation. If researchers can model ice nucleation more accurately, it could give a big boost to weather prediction overall.
Ice nucleation is currently predicted based on laboratory experiments. Researchers collect data on ice formation under different laboratory conditions, and that data is fed into weather prediction models under similar real-world conditions. This method works well enough sometimes, but often ends up being inaccurate because of the sheer number of variables in real-world conditions. If even a few factors vary between the lab and actual conditions, the results can be quite different.
“Your data is only valid for a certain region, temperature, or kind of laboratory setting,” Liu says.
Basing ice nucleation on how electrons interact is much more precise, but it’s also extremely computationally expensive. Predicting ice nucleation requires researchers to model at least 4000 to 100,000 water molecules, which even on supercomputers could take years to run. And even that would only be able to model the interactions for 100 picoseconds, or 10-10 seconds, not enough to observe the ice nucleation process.
Using deep learning, however, researchers were able to run the calculations in just 10 days. The time duration was also 1,000 times longer—still a fraction of a second, but just enough to see the ice nucleation process.
Of course, more accurate ice nucleation models alone won’t make weather forecasting perfect, says Liu. Ice nucleation is only a small but critical component of weather modeling. Other aspects, like understanding how water droplets and ice crystals grow, and how they move and interact together under different conditions, is also important.
Still, the ability to more accurately model how ice crystals form in the atmosphere would significantly improve weather predictions, especially whether it’s likely to rain or snow, and by how much. It could also improve climate forecasting by improving the ability to model clouds, which are vital players in the absorption of sunlight and abundance of greenhouse gasses.
Piaggi says future research could model ice nucleation when there are substances like smoke in the air, which can improve the accuracy of models even more. Because of deep learning techniques, it’s now possible to use electron interactions to model larger systems for longer periods of time.
“That has opened essentially a new field,” Piaggi says. “It’s already having and will have an even greater role in simulations in chemistry and in our simulations of materials.”
How to craft effective AI policy
So to your first question, I think you’re right. That policy makers should actually define the guardrails, but I don’t think they need to do it for everything. I think we need to pick those areas that are most sensitive. The EU has called them high risk. And maybe we might take from that, some models that help us think about what’s high risk and where should we spend more time and potentially policy makers, where should we spend time together?
I’m a huge fan of regulatory sandboxes when it comes to co-design and co-evolution of feedback. Uh, I have an article coming out in an Oxford University press book on an incentive-based rating system that I could talk about in just a moment. But I also think on the flip side that all of you have to take account for your reputational risk.
As we move into a much more digitally advanced society, it is incumbent upon developers to do their due diligence too. You can’t afford as a company to go out and put an algorithm that you think, or an autonomous system that you think is the best idea, and then land up on the first page of the newspaper. Because what that does is it degrades the trustworthiness by your consumers of your product.
And so what I tell, you know, both sides is that I think it’s worth a conversation where we have certain guardrails when it comes to facial recognition technology, because we don’t have the technical accuracy when it applies to all populations. When it comes to disparate impact on financial products and services.There are great models that I’ve found in my work, in the banking industry, where they actually have triggers because they have regulatory bodies that help them understand what proxies actually deliver disparate impact. There are areas that we just saw this right in the housing and appraisal market, where AI is being used to sort of, um, replace a subjective decision making, but contributing more to the type of discrimination and predatory appraisals that we see. There are certain cases that we actually need policy makers to impose guardrails, but more so be proactive. I tell policymakers all the time, you can’t blame data scientists. If the data is horrible.
Anthony Green: Right.
Nicol Turner Lee: Put more money in R and D. Help us create better data sets that are overrepresented in certain areas or underrepresented in terms of minority populations. The key thing is, it has to work together. I don’t think that we’ll have a good winning solution if policy makers actually, you know, lead this or data scientists lead it by itself in certain areas. I think you really need people working together and collaborating on what those principles are. We create these models. Computers don’t. We know what we’re doing with these models when we’re creating algorithms or autonomous systems or ad targeting. We know! We in this room, we cannot sit back and say, we don’t understand why we use these technologies. We know because they actually have a precedent for how they’ve been expanded in our society, but we need some accountability. And that’s really what I’m trying to get at. Who’s making us accountable for these systems that we’re creating?
It’s so interesting, Anthony, these last few, uh, weeks, as many of us have watched the, uh, conflict in Ukraine. My daughter, because I have a 15 year old, has come to me with a variety of TikToks and other things that she’s seen to sort of say, “Hey mom, did you know that this is happening?” And I’ve had to sort of pull myself back cause I’ve gotten really involved in the conversation, not knowing that in some ways, once I go down that path with her. I’m going deeper and deeper and deeper into that well.
Anthony Green: Yeah.
A bioengineered cornea can restore sight to blind people
One unexpected bonus was that the implant changed the shape of the cornea enough for its recipients to wear contact lenses for the best possible vision, even though they had been previously unable to tolerate them.
The cornea helps focus light rays on the retina at the back of the eye and protects the eye from dirt and germs. When damaged by infection or injury, it can prevent light from reaching the retina, making it difficult to see.
Corneal blindness is a big problem: around 12.7 million people are estimated to be affected by the condition, and cases are rising at a rate of around a million each year. Iran, India, China, and various countries in Africa have particularly high levels of corneal blindness, and specifically keratoconus.
Because pig skin is a by-product of the food industry, using this bioengineered implant should cost fraction as much as transplanting a human donor cornea, said Neil Lagali, a professor at the Department of Biomedical and Clinical Sciences at Linköping University, one of the researchers behind the study.
“It will be affordable, even to people in low-income countries,” he said. “There’s a much bigger cost saving compared to the way traditional corneal transplantation is being done today.”
The team is hoping to run a larger clinical trial of at least 100 patients in Europe and the US. In the meantime, they plan to kick-start the regulatory process required for the US Food and Drug Administration to eventually approve the device for the market.