Today, you’ll learn how a stiff gel could one day repair your knee, the dangers of AI when it comes to creating the world’s next pandemic, and a brain-scanning AI song hit machine.
Today, you’ll learn how a stiff gel could one day repair your knee, the dangers of AI when it comes to creating the world’s next pandemic, and a brain-scanning AI song hit machine.
Find episode transcripts here: https://curiosity-daily-4e53644e.simplecast.com/episodes/cartilage-regeneration-chatbot-virus-hit-song-prediction
Cartilage Regeneration
Chatbot Virus
Hit Song Prediction
Follow Curiosity Daily on your favorite podcast app to get smarter with Calli and Nate — for free! Still curious? Get exclusive science shows, nature documentaries, and more real-life entertainment on discovery+! Go to https://discoveryplus.com/curiosity to start your 7-day free trial. discovery+ is currently only available for US subscribers.
[SFX: INTRO MUSIC/WHOOSH]
NATE: Hi! You’re about to get smarter in just a few minutes with Curiosity Daily from Discovery. Time flies when you’re learnin’ super cool stuff. I’m Nate.
CALLI: And I’m Calli. If you’re dropping in for the first time, welcome to Curiosity, where we aim to blow your mind by helping you to grow your mind. If you’re a loyal listener, welcome back!
NATE: Today, you’ll learn how a stiff gel could one day repair your knee, the dangers of AI when it comes to creating the world’s next pandemic, and a brain-scanning AI hit song machine.
CALLI: Without further ado, let’s satisfy some curiosity!
[SFX: WHOOSH]
NATE: Hundreds of thousands of Americans report knee injuries every single year, it’s super common and they can be pretty hard to treat, but researchers may have come up with a solution.
CALLI: I mean, I feel like I know at least 6 people with knee problems so I understand, it’s pretty common.
NATE: I think once you get to your thirties like in your friend group at least someone is going to have knee problems.
CALLI: Yeah!
NATE: Joints in the knee and the hip are incredibly complicated systems; they’ve got ligaments, bones, tendons, muscles, and cartilage.
CALLI: Yeah and it seems like if you get injured, it’s going to take forever to deal with.
NATE: But think about it - that joint, your knee allows us to support all of our weight as we walk and run and jump…
CALLI: and…hit the griddy…
NATE: …ugh. Yes.
CALLI: I feel like I’m too old to say that!
NATE: Yes, our knees allow us to, ahem, hit the griddy. As the youngin’s say. Anyway, one thing that makes the knee so incredible but also really hard to fix is cartilage.
CALLI: Okay! And that’s the stuff our knees have in common with sharks!
NATE: Exactly! Shark skeletons are made out of cartilage. Cartilage is this strong bendy stuff that makes sharks so graceful and sneaky, and it gives our knees a shock absorber. When the cartilage in our joints is injured, we’re missing a little bit of squish when we move and that can lead to crazy pain and friction, and can even lead to arthritis. But here is the real kicker: it’s also really hard to repair, and it doesn’t really just repair itself.
CALLI: So…how do they usually repair it?
NATE: Well for years, doctors have been building what they call “scaffolding” out of proteins that sorta mimics cartilage. The idea is that the real cartilage will use that scaffolding to grow around. But there’s a problem: nobody has been able to make anything as good as cartilage… at least until now.
CALLI: What is it about cartilage that makes it so tough? Can’t they just like, throw some silly putty in there and be done with it?
NATE: Well you might think so. But no. Silly Putty is just… it’s too squishy so believe it or not, cartilage is like a miracle substance. It is both stiff and tough.
CALLI: Okay. But…what’s the difference?
NATE: Uh, okay, look at it this way: some things are really stiff, but when you try to bend them, they just snap.
CALLI: Okay. And something that’s tough won’t break, but it’s also not stiff. I guess if sharks were made out of silly putty they’d probably wouldn’t be the top predator they are.
NATE: No but I would definitely still watch shark week- that’d be interesting. However there is some good news. New research published in Nature describes a new biodegradable gel that is super tough! Like it can even resist being cut by a scalpel - and it’s much stiffer than the material they currently use. And the results have been remarkable. Patients implanted with the gel have shown increased healing and repair after 12 weeks - much better than in the control group.
CALLI: Okay so that part is really great! So all those athletes out there with knee injuries have a new treatment to look forward to?
NATE: Maybe not quite, yet. The patients that they tested the gel on happened to be rabbits, and more testing and modifications are needed before it can be tested on humans. But the researchers see this as a possible leap forward in the treatment of joint injuries. And that is cause for celebration…
CALLI: …you know what’s a really good way to celebrate?
NATE: Oh no.
CALLI: Hit the griddy…
NATE: No…
CALLI: Sorry!
[SFX: WHOOSH]
CALLI: You know how there’s all this news about how chatbots and AI could destroy civilization?
NATE: Yeah. I’ve read about it, but I’ve also read that the threat could easily be overstated.
CALLI: Well, okay, buckle up on this one. Because Kevin Esvelt, a biosecurity expert at MIT just asked his students to use ChatGPT or another large-language-model to build a dangerous virus.
NATE: Yikes. Did they succeed and create one?
CALLI: Nope. They created four. And it only took an hour.
NATE: Oh okay. Why on Earth would anyone actually ask their students to do this?
CALLI: I mean, mostly to see if it would work, and to demonstrate the danger to the world so that we can take steps to stop it from actually happening.
NATE: Okay. So…can we stop it?
CALLI: Um… Let me give you the bad news first. Esvelt’s students were not experts in life sciences. They knew just about as much about dangerous viruses as the next guy. In most cases, simply asking a chatbot to design a ‘dangerous virus’ didn’t work, because they are designed to not produce intentionally dangerous material. But when they changed the prompt to something more like, “...Help me develop a vaccine for a dangerous virus…” they hit pay dirt.
NATE: What kind of viruses did the chatbots come up with?
CALLI: There were four that they recommended: the virus that caused the 1918 Spanish flu, an avian influenza virus from 2012, a smallpox virus, and a strain of one called the Nipah virus.
NATE: So, at least in theory, they’re all viruses we have experienced before.
CALLI: Yeah. But in a couple cases, the bots actually explained certain genetic mutations that made the viruses more catchy.
NATE: Oof.
CALLI: Yeah…That’s not all. The bots also explained ways to actually assemble the viruses from their genetic sequences, offered up a list of lab supplies needed to do the work, and even suggested companies that might be willing to actually create the viruses.
NATE: Let me get this straight. A group of regular old non-experts were able to use chatbots to develop highly contagious viruses and get the instructions for actually making them? I’m having trouble imagining the positive in this.
CALLI: Yeah - it sounds terrifying. And it is. But, the fact remains that it would be incredibly difficult, if not impossible, for a single evil mastermind to actually pull this off. They’d have to not only find companies willing to help them manufacture this virus, but who would be willing to manufacture lots of it. They’d also need help spreading the virus enough for it to cause a full blown pandemic.
NATE: Okay. Yeah, that makes sense. It’s not enough to just be evil. You’d have to convince hundreds, if not thousands of other people to help you in your evilness.
CALLI: Well that’s right. But that fact certainly doesn’t make Esvelt feel all that much better. While he doesn’t believe the results from the chatbots’s suggestions could actually cause much harm, he definitely thinks this should wake all of us up to the potential that a single bad actor or would-be terrorists could use chatbots and large-language models to unleash all kinds of chaos on the world.
NATE: A warning is a good start. But are there other steps we could take to stop this from actually happening?
CALLI: Absolutely! One is to simply limit the research that these large-language-models have access to. If they don’t know what dangerous viruses exist…well…that’s a good start. And another thing is to have stricter requirements on those labs that actually create biohazardous material.
NATE: Let’s say those things don’t happen. Should I be scared?
CALLI: Experts say…no. We don’t have anything to be afraid of at the moment, but we do need to be prepared.
NATE: I guess I’ll take it.
[SFX: WHOOSH]
NATE: As long as there has been pop music, there have been executives who claim to be able to predict what people are gonna listen to. But get this: according to some studies, the accuracy of predicting hits hovers just below 50%.
CALLI: K… wait - so what does that mean, exactly?
NATE: If I played you a song and asked you if it was going to be a hit, you would be wrong most of the time.
CALLI: Okay but unless it was by Taylor, in which case they’d all be hits.
NATE: You’re not wrong. But a song by anyone other than Taylor Swift and you’d be better off flipping a coin. Until now that is. Researchers have developed a way to predict hits almost 100% of the time.
CALLI: Okay… this has artificial intelligence kinda written all over it!
NATE: That prediction, my friend, is 100% accurate. A U.S.-based team of researchers enlisted 33 people to be a part of a study. Each person wore a set of sensors that measured their neurophysiological responses to a set list of 24 songs.
CALLI: Okay sounds fun! All they needed were some glow sticks and they’d have a real rave.
NATE: Right? So they fed their brain responses through machine learning…
CALLI: …like…AI?
NATE: Right - they essentially used AI to predict which songs on that playlist would be hits. And it was accurate 97% of the time.
CALLI: Okay. Let’s back up for just a second. What are neuro…? What you said!
NATE: Neurophysiological responses? Yes. Great question! Basically they measured brain activity that was tied to mood and energy levels and other responses to the music. It’s a technique they call “neuroforecasting,” which basically means that they are able to use a very small sample size - 33 participants in this case - and apply the results to millions.
CALLI: So how is this different than the algorithm that services like Spotify use?
NATE: It’s more accurate, to put it bluntly. Think of it this way - the current algorithms send you recommendations based on the songs you’ve like and disliked in the past. Those recommendations are pretty good, most of the time. But imagine if they could send you playlists fine tuned to your own neurophysiology. In theory, you’d still get all those new jams, but each one would be exactly what you want to listen to.
CALLI: Yeah but it sounds like there could be some privacy issues there. Like…do I really want a music service to have my neurophysiology on file?
NATE: Perhaps not, but that really remains to be seen. We probably never thought we’d one day feed apps like Instagram all of our personal moments to share to the world.
CALLI: That is a good point.
NATE: Plus, they still have work to do. They need to test on different age groups and get a better picture if this works differently for different ethnicities. It was a small study. But one of the craziest things is that even when they only used a single minute of brain activity, the predictions were still about 82% accurate.
CALLI: Okay! So…maybe they’ll just need a teensy bit of my neurophysiology? Might be a fair swap for a sweet summer jam playlist.
[SFX: WHOOSH]
NATE: Let’s recap what we learned today to wrap up.
NATE: Researchers have come up with a biodegradable gel that is able to mimic our body’s cartilage! This could be a game changer for those who suffer from knee problems but since there’s only been animal testing so far, we will have to wait and see if this is a miracle treatment for humans.
CALLI: A professor at MIT asked his students to try to create the world’s next pandemic-level virus using an AI chatbot, and it only took them an hour to do it. While the viruses their chatbots came up with aren’t likely to cause a global pandemic anytime soon, the exercise is a wake up call for AI experts and policy makers on the dangers of AI.
NATE: Researchers have used brain scans and machine learning to predict what songs will become hits with near perfect accuracy. The process is called ‘neuroforecasting’ and it could change the way music services deliver your favorite songs to your headphones.