AI‘s Role in Fighting COVID-19 and in Surveilling Us
40:46 Share

Hello, and welcome to Skynet Today's Let's Talk AI podcast, where you can hear from AI researchers about what's actually going on with AI and what is just clickbait headlines. I am Andrey Karenkov, a third-year PhD student at the Stanford Vision and Learning Lab. I focus mostly on learning algorithms for robotic manipulation. And with me is my co-host. I'm Sharon, a third-year PhD student in the Machine Learning Group, working with Andrew Ng.

I do research on generative models, improving generalization and neural network, and applying machine learning to tackling the climate crisis. And this week, we'll look at how AI is being used against the COVID-19 pandemic and continued concerns around Clearview and the applications of AI. Speaking of COVID-19 and crises, have you been dealing with it this week as it has really started hitting our shores? It's been interesting using Zoom so much.

for everything and also finding a certain kind of pattern and rhythm to life and making sure that I get enough exercise to keep my immunity up. - Yeah, working from home has certainly become the new normal

And my productivity has declined somewhat, but trying to get it back. One fun thing about Zoom, though, is that you can actually change your background. Probably AI, some kind of generative model filling out that background. So someone zoomed in from a cave, it looked like, and I thought they were actually in a cave. Yeah.

Yeah, I don't think so. That would be interesting. Okay, so yes, this week we're going to be talking all about AI and how it intersects with the issue of COVID-19. And we'll also touch again on the topic of facial recognition, which we discussed last week. So let's just dive straight in.

with our first article, which is titled AI could help with the next pandemic, but not with this one. And so while there's many ways in which AI does relate and is helping with the situation, this article basically makes a point that all of this is a bit nascent. And so while these things are positive and it's helping, it's not going to be a cure-all. It's not going to really resolve the situation this time.

And so the article touches on what AI has done already and also dives a little bit deeper on what

what this means and how this reflects on human efforts and how this compares to human efforts. So first with detection, it was, quote, an AI that first saw it coming, or so the story goes. On December 30th, an AI company called Blue Dot, which uses machine learning to monitor outbreaks of infectious diseases around the world, alerted clients, including various governments, hospitals, and businesses to an unusual bump in pneumonia cases in Wuhan, China.

And it was just one of several companies that do this sort of thing. So there was also this animated service called Health Map at the Boston Children's Hospital that likewise sent up some alerts. And there was also a company called Medabiotic that did as well.

Now, it wasn't just these AI models catching these concerning signs. Actually, just half an hour later, after the HealthMap system sent its own alert, a human volunteer-led program for monitoring emerging diseases called ProMed did a more detailed warning. So Marjorie Pollack, ProMed's deputy, said,

noticed the signs just four hours before the health map alert. Meaning that she claims that she had seen it as a human before the AI system was able to detect it. In fact, she received an email informing her of a Chinese social media post discussing a Wuhan virus

agency notice of unexplained pneumonia, quote unquote, and sent her team to investigate that. And that's where their analysis led them to their coronavirus alert. Of course, it was sent out slightly later when it comes to looking at different timestamps, but they made sure that it was a much more detailed and comprehensive warning. And so this suggests that perhaps human AI in combination would offer the best alerts.

Exactly. So this article makes a point that the human alert, not only did it say something's going on here, but the human was able to exercise judgment and actually provide more substantive concern and their thoughts on the situation.

This is the first major area in which AI has been applicable here, detection. But we are well past detection, of course. So it's a promising sign for future situations that these companies did catch the signs.

And hopefully they'll be refined for any new situations you might have. Something I do wonder about is how many false positives do they have? So how many times do the alerts come up and perhaps this one matched on to coronavirus? And yes, this is a true positive and this matters a lot.

But is there an alert every day, every hour, or is it every month, or is it every year, or every 10 years? These systems haven't been around for long enough to really validate all of these. And it's hard to evaluate, of course, when it's anomalous detection. Exactly. So before we get all excited that AI told us this was coming...

we should be mindful that there probably are false positives and the humans monitoring the situation also were able to detect this particular thing, which meant there are probably some strong signals.

But onto the next area in which AI has been helping and has been generating quite a bit of hype over the last few years, which is diagnosis. So being able to look at medical scans and tell whether someone actually has the infection or not. So there was actually a recent paper, things have been moving very fast, of course. So this paper titled, A Deep Learning Algorithm Using CT Images to Screen for Coronavirus Disease.

COVID-19 made the point that they were able to use cutting edge AI algorithms to train a model to look at CT images and with fairly high accuracy predict whether they indicated that the person has the virus or not. And there was another group actually from Alibaba that made similar claims of having actually even better accuracy at 96%.

And they make a point that their system can do the diagnosis within 20 seconds, whereas a doctor would usually take between 5 and 15 minutes to do the same sort of diagnosis based on a CT scan. So I've worked previously on applying deep learning to radiology images, and I've chatted with a number of radiologists on this very topic.

And it's something that our lab is actually investigating right now for the US, as Alibaba will not release this model to other countries necessarily.

And it is almost concerning that you need a lot of cases in order to train the models. So as a result, we would need many more cases actually in order to train a model that could reach Alibaba's accuracy. But having talked to some doctors, they said it is, they think, visible. Like you can kind of tell from these CT scans that there is a difference. Like a human could definitely tell.

People have debated how accurately doctors can tell or distinguish between coronavirus and ordinary pneumonia or another type of respiratory disease. Alibaba claims that this 96% accuracy is distinguishing between ordinary pneumonia and coronavirus. But of course, there are many other respiratory diseases out there.

CT scans are, of course, also much more expensive here in the US. So that could also prevent CT scans from being done. So it's possible that there would be fewer and harder to amass given the healthcare system here. Exactly. This is an important thing to note is you actually need to go into a hospital to get a CT scan using a fairly complex machine.

So it's not ideal for doing large quantities of tests, really. It's ideal for, let's say, things seem more severe if you have a lot of symptoms, but this isn't going to make an instant test that anyone can just do in their home. So again, while useful, it's not going to kind of solve all our problems right now.

And so the last category that the article touches on is around drug discovery. So SRI International is collaborating with an AI tool that uses deep learning to generate many novel drug candidates that scientists can then assess for efficacy. And this is pretty game changing for drug discovery because it could take many months before a promising candidate becomes a viable treatment.

And there are hundreds of startups who are deploying AI tools over the last several years to speed up this drug discovery process, figuring out which candidates that we can put through clinical trial. And only recently to treat patients with OCD, obsessive compulsive disorder, there's been a successful candidate.

So in this instance, AI was able to complete in 12 months what often takes several years to reach human trials. Yeah, and this is another case where there are multiple companies working on it. So one of them is Insilico Medicine, one is Vier Biotechnology. Atomwise, there's kind of a bunch of them. Insilico made the claim that they want to have potential treatments to be testing in partnership with a pharmaceutical company in April.

So, they're trying to really move fast here. But again, as with the prior situations,

All of these companies are pretty young. We haven't had many actual treatments that we have developed with the help of AI. So hopefully they will learn a lot from this experience. And then if we do have another pandemic or another big disease, by that point, they will be more mature and more capable of rapidly responding and actually being very useful.

What's interesting to think about broadly around all of this is that data and reliable data and large quantities of data and perhaps unbiased data are essential to training an AI that would be effective in all three of these categories of detection, diagnosis, and drug discovery.

Yeah, and of course, in order to get this data, it's pretty sensitive private data related to people's health. And so it's not easy. It requires the company to be really vetted, to go through the right procedures. At least in the US, accessibility to public health data isn't large.

And so that's another reason why it will take some time for AI tools to really mature and become ready to use and to make big impact. Right. And I think that's in contrast to China and as we'll see later, Russia as well. Though, of course, there are consequences associated with perhaps using AI at that scale. And of course, as many of us have said,

experienced across the past few days and weeks, new sources and official reports often offer inconsistent accounts. So it's hard to know what to trust. So what kind of data can we use and how noisy is it? Where is it biased? It's unclear.

Yeah, exactly. So there is quite a bit of difference between these different countries. And there is the next article here from Technology Review that is about how the CDC is trying to forecast the coronavirus's spread. So the CDC is the U.S. authority on the Center for Disease Control. And it turns out that they host an annual competition who can most accurately forecast the flu.

And now they are adopting this competition to help deal with coronavirus and basically know where it's heading in terms of spread.

So this isn't just forecasting into the future. It's also doing something called nowcasting, which is predicting the number of people infected right now, given recent and historical data, including Twitter activity and Google searches. Of course, the flu follows a very different trend as coronavirus. There's lots more panic around coronavirus. People are searching for symptoms, uh,

and information about the disease, even if they don't have it, which is something that they leveraged before, where if people search flu symptoms, they assumed with some kind of probability that that person did have it or that someone around them had it.

But the good news is that they have been finding ways to compensate for it. So for one, they've been using data from previous pandemics. And I think we are a bit lucky here in the US because we got hit later than some other countries. So they've been using that data. And lastly, the CDC has actually been sharing up-to-date data of the cases with the team. So all that together does mean they actually are able to at least try.

One thing we can actually say as people who live here in the US and in the Bay Area is that it is surprisingly hard to get a sense of how widespread this thing is. Testing is super limited. It's much, much less done than it has been in other countries like South Korea. And so here in the Bay Area, we have something like 200 cases per

more than 100 cases now near Stanford confirmed, but those are just the ones we know about. How about all the other unconfirmed cases? The number seems likely to be much higher. And so having some sort of accurate, confident prediction would definitely be nice.

And also, it's a common paradigm for the data distribution to change over time for deployed machine learning systems. So in this case, we see more developed surveillance systems for tracking these different cases. In healthcare, we often see new drugs come out, so people's behavior change or people's responses change. And so the model needs to be able to adapt to all these scenarios and still be useful.

for the present time and for predicting the future. And something I like to think about is that instead of a prediction, the model is actually offering up a projection. So this is a projection based on the data that we do have as opposed to a prediction. Yeah. And so as with these other things that AI can offer,

This is a very useful time in terms of getting the data and developing the techniques and later being able to see how good they were. But we are not extremely reliable right now. And really, we need to lean on more traditional, more established methods.

and basically stay home as we are. That is quite the age-old strategy. And it is the best one, I think, at least so far that more extreme measures have not been taken in this area.

Okay, so that's a few ways in which AI has been able to help with the disease itself. So being able to detect it, being able to diagnose it, to find drugs for it and to forecast its spread.

But there's a lot more to the pandemic as we've been seeing than just treating with disease. So our whole structure of society is kind of changing. Day-to-day life is being impacted. And so there's been quite a few news stories on other ways in which AI has been interacting with that. And one such article is called Adorable Self-Driving Vans Are Disinfecting Roads in China.

So a driverless delivery vehicle called Neolix is getting lots of customers in the wake of the COVID-19 crisis because cities in China are under full lockdown, meaning people can't go outside. So this is actually a pretty big time for robotics to shine and actually be very, very useful, not just in, well, disaster zones. Perhaps everything is a disaster zone now. So

They're becoming much more useful or at least being tested out right now. Yeah, exactly. Since they are self-driving, by definition, there's no person there. So they can pretty much go anywhere outside where people aren't supposed to go. And they can not only deliver medical supplies, they can deliver food. They can even spray disinfectant on the city streets, which of course should help.

as people try to get back to normal life. The funny thing with this Neolix company is actually that they were fairly young. They didn't produce that many of these cars up to now. And so they are really entering a crazy growth phase where they're trying to produce way more of these little trucks.

and getting paid a lot to do it. So this article says that the government is willing to cover 60% of the cost of each vehicle from not only Neolux, but also other similar companies.

And I think this is quite reasonable for the government to be subsidizing these dire needs. And the CEO of that startup says that new habits are formed and new capabilities are needed. The consumption pattern has shifted and this will likely be viewed as an essential set of capabilities in the new normal post-coronavirus world. Yeah, this is kind of interesting. You can draw a comparison to

really worldwide events, which have often been, as we know, world wars. And an interesting side effect is that a lot of new technology gets developed during wartime. So GPS, computers, a lot of things have been developed during wartime with a lot of investment from government that have wound up

having consequences for decades on. And so, as this article notes, it could be that this is really going to be a push for self-driving vehicles and similar AI-based technology that will stick around for much longer than we actually have to deal with this virus.

And on that topic of self-driving vehicles that are helping us out, it's not just vans that are adorable, it's also delivery drones. So there's been some news about Antwerx RA3 and TR7S drones, which have been used to center on medical samples and quarantine materials.

And the nice thing about drones, of course, is that they can just fly between different places. So they can get around much faster than land vehicles and can deliver quite a few useful materials. And this is very interesting because it increases the speed of transport by more than 50% compared to road transportation. So this is proving to be a much more efficient means of transportation for epidemic prevention and control.

And it becomes increasingly interesting to see drones being leveraged now as prior to this, my perception was that Zipline, the company Zipline, which has drones delivering medical supplies and blood in Africa, was very successful. But there were few drone companies outside of Zipline that were successful in doing delivery due to regulation. But now it's kind of changing people's perception around.

them. Yeah, this article makes a point that this is especially important because otherwise you would need to have people driving ambulances to transfer all these different materials and one of the most stretched resources are medical professionals in hospitals. So having more truly autonomous drones and cars do that work frees up people to do more essential activities.

It's even possible that if something is tele-opt, so not fully autonomous, that could be helpful too. Exactly. I think if people are stuck from home, they can actually help by tele-opt mining robots. That would be interesting. Gig economy. Gig economy. It's in our future.

And let's move on to our last little cute article about other ways in which AI is helping, which is titled "Artificial Intelligence Applications Surge as China Battles to Contain Coronavirus Epidemic." This is a slightly older one, but it's of course still relevant. And it listed even more applications of AI that have been useful.

So for instance, as we've already mentioned, there have been robotic cleaners that have been spraying disinfectant. But we have also been AI voice assistants that have been calling people to give advice. And largely this is to reduce human-to-human contact in hospitals to prevent a disease from spreading so that doctors and nurses can communicate with patients effectively.

through these AI voice assistants. Yes, and another application listed here is that there are automated temperature sensors that can direct sensing to the foreheads of moving people to check if they seem to have symptoms or not, which can be problematic if someone is carrying coffee and has false positives, but still...

In this really dire scenario, it is useful to have more sensing, more data to kind of inform people of they might have symptoms. My sense from these articles is that it's not necessarily AI that is helping a lot of these situations. It's technology and digital means in general.

And in large part, I think a lot of physical technology, particularly in robotics, that is empowered and made more autonomous by AI. These are the types of technology that are shining in this moment. I think AI is...

perhaps not quite the right word, especially in this last article, to reflect on all of these things. Because I think the technology itself, having that digital layer, is really what prevents human-to-human contact. Yeah, it's interesting. AI is often conceived in popular fiction as being some sort of completely independent agent,

completely separate entity, but my impression is most actual uses of it are kind of more system-based. So you have some applications that use some data and are optimized, but this is part of a larger program or a larger tool that someone has built that leverages AI in some way, but it isn't just AI, as you said. It's actually a combination of various things that then makes it useful.

And so our next two articles are looking at how China and Russia are using facial recognition technology for the coronavirus right now. And so in Moscow, Sergei Sobyanin, the mayor of Moscow, said that some 2,500 people who had landed in the city from China had been ordered to go into quarantine.

But to prevent them from leaving their apartments, the authorities are actually using facial recognition technology in the city to catch any offenders. So you cannot leave your home and we can see if you have left your home. So there has been a case at

least of surveillance footage showing a woman who had returned from China leaving her apartment and meeting her friends outside. And the authorities were able to track her down and use that video footage as proof. Yeah, and the interesting thing is this whole use of facial recognition

for surveillance and for policing was only announced last month for Moscow. And so this is following up on that very quickly. And it's hard to know if this is meant to be sort of a sign that it's useful and it should be kept.

Or if it really is essential to quarantine and necessary. Yes, it could be a foot in the door for something worse. And this could be a nice excuse. But it could also be very, very effective right now. Yeah, and to that point of it leading to other things, our next article is titled China's facial recognition giant says it can crack mask faces during a coronavirus. So basically, if you're wearing a mask,

Now you can still be recognized by facial recognition technology. And this is technology coming from SenseTime, which is the world's most highly valued AI startup, who said that they were rolling out a facial recognition product that incorporates thermal imaging cameras to help spot people with elevated temperatures and send pop-up alerts.

And I believe that they were able to be able to detect people even with occlusions before. And that's been shown before to be very effective. And now it's very officially with masks, which is interesting. Now, this article notes that...

There are applications where this is very useful. So for instance, if you're going to your office, some companies actually have you enter with facial recognition. So to be able to come into the office while still being safe, it's good to have it. But it does mean that in this phase of crisis,

The surveillance and use of facial recognition will be even more expanded. So one user on the social media platform Weibo noted that every day we are seeing more tech companies in China roll out fancy surveillance combinations such as facial recognition plus big data and remote monitoring plus real-time alerts in order to control the outbreak. This is an example of how tech advancements sneak into our lives using an excuse like a virus. Where is the boundary for people's privacy?

Speaking of boundary of people's privacy, we have a little bit more to talk about with Clearview. So a quick background from our last week's podcast. Clearview AI is a secretive company that's claimed to have scraped more than 3 billion photos from social media and the web.

And last month, a BuzzFeed News investigation found that people at more than 2,200 organizations have tried Clearview's facial recognition technology, including federal entities such as Immigration and Customs Enforcement, the FBI, and private companies like Macy's, the NBA, and Bank of America. And of course, all of these organizations have tried Clearview's facial recognition technology.

it has also been known that Clearview has not necessarily been used strictly with these entities and organizations but also individuals. Yeah, this article notes that

It seems that the technology has been shared with political connections, potential investors, and entities just known or designated as a friend, which is quite a vague and funny description. This is following up on a leak that happened last month that gave us this client list.

And now we are seeing some more of the details. So for instance, some of the organizations deemed a friend is Ace HW Partners LLC, a company founded by Jason Miller, a former Trump campaign senior communications official and one-time nominee for the White House communications director.

And it wasn't really explained why this political-seeming person, who now co-hosts a podcast with Steve Bannon, had access to facial recognition pathology. In addition to two conservative think tanks, other entities on Cleavage's list with this friend designation included the Samarian Group, which is a private equity firm, Gross Rennie, Dallas Commercial Architecture Company,

and Tor Eklund, which is Clearview's outside law firm. So really quite a varied set of companies designated as a trend here. It seems like Clearview is basically just trying to gain favor by giving their facial recognition capability to anyone they want.

What's perhaps even more concerning is that on top of the thousands of organizations with accounts, with named accounts, there are other unknown individual users who ran searches with the app. So on his list, the startup actually maintained an entry for, quote, Clearview test users, which included more than 220 accounts that

had run more than 30,000 searches over a 19-month period. So, quite creepy to say the least that just random people that you see or like seem to be able to use facial recognition. To be concrete,

These people would be able to just take a photo of your face and then find your name if it is in that huge database of theirs. And this really speaks to how sometimes when you have a startup and you're trying to grow, especially in Silicon Valley, there can be a problem of basically skirting any ethics, any notion of privacy in order to continue growing.

And what might be interesting to our Bay Area listeners is that some notable customers of Clearview are the Founders Fund, which is a venture fund founded by Peter Thiel, who's invested $200,000 in SmartChecker, which is the Clearview CEO's predecessor company. Other customers include individuals with ties to Iconic Capital, which manages money for Mark Zuckerberg and other tech billionaires.

And finally, the Oculus Rift co-founder, Palmer Luckey. It's also on that list. Yeah, quite the social club this company has. So once again, as we've

last week it's pretty clear this is going too far and this company does need some restrictions in place to just not share it with any random person or company clearly some legal action will need to be done on the federal level but maybe not for a while as we are dealing with something of a

a little more significance right now. It's pretty crazy that they have been able to even expand this much and share it with this many people. Yes, I'm fairly surprised at that too. And following up on that, there was another article from OneZero titled, "'This is the ad Cleave UAI used to sell your face to police.'"

And it says that Clearview AI emailed advertisements to police departments in August of 2018 with the subject line "How to solve crimes instantly with face search technology" using the Fraternal Order of Police's online platform FOP Connect. So in that email, they pitched their product as Google Search for Faces. All it takes is a photo of a suspect's face,

which they can upload to Cleaview's app and they should be able to find a match of an actual person and their name with that photo.

So this was strategic marketing on Clearview's part. FOP Connect, which again stands for Fraternal Order of Police Connect, is operated by 911 Media, which has previously advertised an email list with more than 60,000 recipients, including police departments around the country.

So the Fraternal Order of Police is a U.S. law enforcement labor organization with around 350,000 members. And on a phone call with OneZero, a member of 911 Media said the Clearview email was an advertisement but declined to comment further.

And beyond the email to FOP Connect, Cleary was also sponsored a part in the police magazine on facial recognition and was later to speak in September on two panels at security conferences.

The reception to these emails and Clearview in general has been mixed. So some law enforcement officers were not very receptive. One sergeant from Irving, Texas, was concerned that things like what Clearview's doing will get facial recognition shut down for law enforcement entirely. And since BuzzFeed's story that we talked about last time in our previous podcast, that exposed the scale of Clearview's operations and the number of companies using its app,

Clearview has received cease and desist letters from Google, LinkedIn, Twitter, and Facebook demanding that Clearview stop scraping their websites for images. And so now Clearview is facing a lawsuit that cites Illinois' Biometric Information Privacy Act that prohibits the collection of biometric information in that state without consent.

Yeah, so I think we can both agree that it makes some sense for police departments to be able to use tools like this in investigations. But at the same time, this is a private company. They were not sanctioned to collect this data. They were not sanctioned to make a product that gives you this power. So it's quite creepy and problematic.

that they have grown this much and it's nice to see following this reporting that they are getting some blowback

It's interesting as well that the users of this product are going to be pretty silent about it naturally, inherently based on the product itself. So it's almost this mutual agreement to be quiet about things. I have friends who are dealing with various lawsuits and that they could leverage a piece of technology like this if they had the means to really protect themselves. And so it's not for something...

terribly negative. But of course, these boundaries are very fine and it's hard to define, right?

Yeah. And again, it's interesting to connect back to the previous point of how now Moscow and China are using facial recognition in this crisis and how they have a state-sponsored kind of machine to actually be able to do mass surveillance. Right now, we do not have that in the US as far as we know, but clearly it seems to be trying to build that. And it's a real question of if it benefits outweigh the potential negatives of

For my part, I would rather not...

Not to give the government or police departments that much power. Maybe having it be a little more limited would be nice. It's interesting because Clearview is a private company. This is almost the American instantiation of what China and Russia are doing, right? Under a more capitalist economy. So it's interesting that this is a very similar instantiation as state-run surveillance.

perhaps giving it to people who have better means right under capitalism. So it is in a way similar and it's just a different instantiation. It's a matter of is there one more ethical than the other? I'm not sure if I'm comfortable with either. I maybe even prefer the Clearview one, but I can't say that confidently at all.

Yes. And I think we can also mention in that comparison, Europe, which we haven't mentioned thus far. And I think a big part of why is that, of course, they have far better protections for privacy. So they just recently passed the GDPR, which part of it is about giving you control of your data.

And so all the stuff that Clearview is doing from scraping LinkedIn and Twitter and Facebook, that would be against the law in Europe. And that would be very problematic. And I don't think Clearview could even try to do that given the regulatory protection. We don't have as much regulation right now in the US for such things. And people have been talking about doing something like GDPR here.

So maybe now that Clearview's work has been kind of more exposed, that will gain some steam. Yeah, I think that could be very possible. And I wonder if the timing is now or perhaps given the current crisis, it's not now. So it remains to be seen.

Yeah, I think right now the time is not now for a lot of things, including some of our research. But hopefully we can all get back and get used to this new situation soon enough. According to Stanford, now is always the time for research. Yeah.

Yeah, I guess we could mention, since we're discussing COVID, what's going on at Stanford, which is that everything is shut down. Classes we need to do online. You cannot go to the gym anymore. We are not supposed to come into our offices. We are supposed to work from home. So that's been an interesting change. I think undergraduate students are thankful because their finals are now optional.

So we're sort of getting out early, but it'll be very interesting next quarter when all the classes have to start online, how things will go. I think it's also been quite alarming for many students as Stanford asked them to leave.

and they have to immediately evacuate and that's quite a bit to ask for, especially if it's not easy for them to leave or if there isn't necessarily an easy extra home for them to go to, if they're not in this country or if they're just generally scared of taking a flight right now, you know? Yeah, this is especially hard for international students. And on the one hand, it makes some sense for Stanford to recommend this because, of course,

Student housing tends to be very dense, so a lot of people are next to each other. The dining halls are likewise very dense, so there's a lot of crowds. So it makes sense to try and reduce the crowding, but on the other hand, where do people go? Especially now that flights are starting to be limited, it's quite a tricky situation.

Right, the students are still rolling their eyes that they have to pay for tuition. Yeah, maybe we are. It would be nice to have a bit of a discount given all the classes are now online. I think I've been rolling my eyes at tuition for a while, even prior to this crisis. Yes, that is a fact of life at Stanford.

Well, and with that, thank you so much for listening to this week's episode of Skynet Today's Let's Talk AI podcast. You can find the articles we discussed here today and subscribe to our weekly newsletter with similar ones at skynettoday.com. Yep. Subscribe to us wherever you get your podcasts. And of course, don't forget to leave us a rating if you enjoyed the show. Be sure to tune in next week.

That's so cute. Yeah.