With all the hubbub around AI in the last year, all its promises and perils, I thought I would spend a video talking about a major upside to artificial intelligence – the science mysteries it is likely to solve. So here are a handful of the biggest mysteries AI could help us find an answer to.
We live in a universe teeming with unexplained mysteries. From the origins of life to the nature of dark matter to fate of Amelia Earhart, to… that thing on the bottom of my foot. It hasn’t gotten bigger or darker, but it’s gotten… angrier? That’s the best way I can describe it, it’s angrier somehow. And I think it’s judging me.
Anyway, the history of science and technology is basically one of building new tools to solve these mysteries, and in recent years we’ve created possibly the most powerful tool we’ve ever created. That is, of course artificial intelligence.
After all, if your own intelligence isn’t enough to solve a problem… make a new one!
So today we’re going to take a look at the biggest science mysteries that AI could help solve. And probably sooner than you think.
Artificial intelligence was obviously the big buzzy thing this last year, hell I actually put it at number one on my video of top science stories of 2023. And most of the buzz around it has been around how it’s really great for business or really terrible for society. Or both.
What doesn’t get talked about nearly enough is how the integration of AI is changing our understanding of the world, solving problems we couldn’t solve before. Including solving some mysteries.
Now I want to be clear I’m not talking about asking ChatGPT who was the Zodiac killer because obviously ChatGPT hallucinates and could say something crazy like Ted Cruz, which we all know that’s… not true.
But there are some pretty compelling uses of AI that have already happened and more that we could see in the near future, so let’s take a look at some interesting ones, starting with medicine.
A U.N. report in 2019 said that the death toll from drug-resistant infections could rise to 10 million by 2050.
I’ve talked about this in previous videos but yeah, we kind overdid it with the antibiotics and now there are strains of bacteria that have evolved and become resistant. The bugs have outsmarted us.
So the question is – the mystery to stay on theme – is can we come up with a universal antibiotic that is not susceptible to resistance? Because we need antibiotics, for sure. But we also don’t need the superbugs they can accidentally create.
That’s where A.I. comes in.
Back in 2020, MIT researchers used a machine-learning algorithm to find a new antibiotic compound.
In fact, this compound was able to kill several of the world’s most deadly bacteria, including those resistant to all antibiotics.
The computer model was able to screen over a hundred million chemical compounds in just a few days.
As team lead James Collins told MIT News:
“We wanted to develop a platform that would allow us to harness the power of artificial intelligence to usher in a new age of antibiotic drug discovery. Our approach revealed this amazing molecule which is arguably one of the more powerful antibiotics that has been discovered.”
Here’s how it worked:
The team had the AI model search for molecules whose physical structure could kill E. coli. They trained in it on around 2,500 molecules, including about 1,700 FDA-approved drugs, as well as 800 natural products.
They then tested the model on the Broad Institute’s Drug Repurposing Hub, which is a library of around 6,000 compounds.
And the model discovered one very promising molecule that was different from any previous antibiotics.
And they named it “halicin,” after HAL from 2001: A Space Odyssey.
So they tested halicin in the lab against a wide variety of bacteria and it killed almost everything, including many that are usually resistant to treatment. The only one it couldn’t kill was Pseudomonas aeruginosa.
I looked it up, that’s a bacteria that often causes infections in hospital patients, usually blood infections and pneumonia.
Now those tests were done in petri dishes but they also tested it on mice infected with A. baumannii, specifically a strain that is resistant to every known antibiotic.
But within 24 hours, the infection was completely cleared up so… Pretty promising.
And the way it works is interesting… as least I think it’s interesting, I’m not an expert on antibiotics but in this case anyway, it’s that structure of the molecule that does the work.
The study says that halicin kills bacteria by interrupting “their ability to maintain an electrochemical gradient across their cell membranes.”
So that molecule kinda attaches itself to the cell membrane and because of that structure it destabilizes the membrane and just frags the cell. And the cool thing about that is that makes it harder for the bacteria to develop a resistance.
In fact, they tested it against E. coli for 30 days and saw no resistance to it at all. And E. coli is kinda notorious for quickly developing resistance to drugs.
Even better, halicin could be a possible treatment for diabetes…I don’t see how exactly, I didn’t see any specifics but still, that’s cool.
Now what I don’t know, and I couldn’t find anything on this, but I don’t know if this only works on harmful bacteria or it would also wipe out good bacteria because as you know, our gut microbiome is super important and just wiping that out is no bueno… Unless you’re just really into poop transplants.
So that’s one AI solution, there’s a totally different one discovered by researchers at the University of Pennsylvania’s Machine Biology Group. I talked about this in my De-Extinction video but they basically took sequenced genome data from Neanderthals and Denisovans, and used A.I. to find potential antibiotics from that.
These are obviously extinct human species from hundreds of thousands of years ago, so they call this process molecular de-extinction.
Basically they trained an A.I. model to predict which molecules would be effective antibodies for humans.
They let the AI do its thing and once it found the best candidate, they created the molecules in a lab, and tested it out on mice. And yeah, it was actually really good at fighting off bacterial infections.
As biochemistry professor Jonathan Stokes told Vox in 2023:
Now let’s talk about some archeological mysteries that might get solved thanks to AI, and there’s some really exciting stuff happening here. If you enjoy history and archeology stuff. Like moi.
I did a video fairly recently about the Khipus of ancient South America and how it’s kind-of like a lost language that people are finally starting to decode well that’s just one of dozens of ancient languages that have been found and never deciphered. One of the most notorious is Linear A.
So in 1886, a British archaeologist named Arthur Evans was excavating some ancient ruins on the island of Crete. And one of the things they found were several stone tablets. And those tablets had a couple of different scripts on it.
They named the scripts Linear A and Linear B. Linear A is older – from when the Minoans were on the island during the bronze age, and Linear B came later, when Mycenean Greeks took over.
And Linear B has been decoded. It was figured out in 1952 by Michael Ventris and Alice Kober. It turns out it’s an early form of the Greek language. Obviously if you have a lot of similarities to a known language, that helps.
Linear A, on the other hand, doesn’t look like anything. And yeah, it’s older, from between 1800 and 1400 BCE.
And so far no one has decoded it. Partly because it doesn’t seem to be related to any other known language, but also because there’s only a handful of fragments, so it’s hard to find correlations or patterns.
But if there’s one thing AI is good at, it’s finding patterns. You might notice a pattern of me saying that in this video.
So, a team from MIT and Google’s AI lab created a system to see how well machine learning could decipher a language, and they used Linear B as a test.
As they told MIT Technology Review in 2019:
“We were able to correctly translate 67.3% of Linear B cognates into their Greek equivalents in the decipherment scenario. To the best of our knowledge, our experiment is the first attempt of deciphering Linear B automatically.”
Their system works by determining relationships between languages, so it basically figured it out the same way Ventris and Kober did in 1952. Only, you know, in minutes. But their next goal is to decipher languages without comparing it to an already known language, just through the patterns in the text.
Now there’s another historical mystery that AI is helping solve and it literally happened since we first started working on this script.
So you know about Pompeii, probably because of that tiny incident that occurred with the volcano and the mass death and city buried in ash – that thing?
Well, that bad day didn’t just happen to Pompeii, it also devastated the city of Herculaneum.
And among the rubble of Herculaneum are a set of papyrus scrolls that were found in a royal home. Scrolls that were not only singed by the flames of Vesuvius but smashed flat under tons of ash.
They’re in bad shape.
Also you’re welcome for pronouncing it pa-pie-riss after half of you glitched out on my cylinder seals video.
But yeah, apparently some archeologists tried to open one of the scrolls but it just disintegrated into dust so bad they were just like, “nope” and started looking for another way.
And they want to find a way to read these because this “royal house” I mentioned may have actually been the home of Julius Caesar’s father in law.
So maybe he spilled the tea about Caesar’s bad table manners or something. Or the color palate of his interior design choices.
OR… maybe something important that sheds light on a major historical event. It’s possible, this was a huge library, more than 600 scrolls have been found there and they’re being studied in museums around the world – some were in good enough condition to translate and were found to be works from philosophers like Epicurius.
But in an attempt to get inside these scrolls, a team of researchers led by Dr. Brent Seales pioneered a process called Virtual Unwrapping using X-ray CT scanners.
They successfully did this for some scrolls that were found in Israel, but the ink used in those scrolls was lead-based so it was denser than the carbon-based ink used in the Herculaneum Papyri. Meaning it was basically charcoal-based ink.
Anyway, without going into details because that’s not the point here, they basically used a particle accelerator in the same way to 3D map these scrolls with resolution down to the resolution of 4 to 8 micrometers per voxel. A voxel is a 3D pixel. Like a pixel volume… Voxel.
So they took all this data and made it public through a competition they created called the Vesuvius Challenge – which I’m pretty sure they stole from a drinking game I played in college but anyway…
They opened this up to software developers to create algorithms that can find patterns – see? – in the data to help decipher the scrolls with a series of cash prizes for the people who can figure it out.
And in October of 2023, the first of those prizes was accomplished when a college student named Luke Farritor from the University of Nebraska created a machine learning algorithm that deciphered the first full word in the scroll.
And the word in question was the ancient Greek word for… Purple. Maybe it was about Caesar’s interior design…
Actually according to the Journal Nature, purple dye was highly sought-after in ancient Rome and was made from the glands of sea snails, so it may be referring to the color or maybe snails, or maybe it’s describing someone’s rank, someone who can afford purple.
The point is, that’s just the first word but it’s a proof of concept, that an AI could help unlock the contents of old scrolls. There might be a day when any scroll that gets found can be scanned and interpreted just like any other document. This would be a massive boon to the entire field of archeology.
Let’s move on to astronomy, which I don’t think it’s any surprise that AI can help in a lot of ways in this department.
Especially when it comes to exoplanets.
There are now more than 5,000 confirmed exoplanets since the first one was discovered in 1992.
Those are the easy ones to find. Big planets that pull on their stars and transiting planets, we can do that without AI, though AI will help that tremendously as well.
One thing we’ve always struggled to visualize though is exoplanets in the formation stage, mostly because they’re covered by thick clouds of gas.
But A.I. could help overcome these challenges.
A recent study from the University of Georgia showed that machine learning could help find these exoplanets that are still forming.
It’s a proof of concept where researchers use “exclusively synthetic telescope data” that a computer has simulated and use it to train A.I. They then apply it to data for a real telescope.
Which, as study co-author Cassandra Hall said:
“This has never been done before in our field, and paves the way for a deluge of discoveries as James Webb Telescope data rolls in.”
One of the other researchers described the experiment as making a better person. Currently, scientists analyze data from hundreds of images looking for a specific disc.
And sometimes it’s easy to overlook a wiggle in the images because they’re tiny or the image isn’t clean.
But using A.I. is faster and more accurate. It also saves money.
Or, as lead author of the study Jason Terry, puts it:
“There remains, within science and particularly astronomy in general, skepticism about machine learning and of AI, a valid criticism of it being this black box – where you have hundreds of millions of parameters and somehow you get out an answer. But we think we’ve demonstrated pretty strongly in this work that machine learning is up to the task.”
And it’s not just exoplanets that A.I. can help find. It can also discover things, like when astronomers used it to find an Odd Radio Circle that they named SAURON.
That stands for a Steep and Uneven Ring Of Non-thermal Radiation. (react) There’s like 3 words in there that aren’t even reflected in the acronym but…
What the astronomers did was they created a coding framework called Astronomaly to help go through data from the MeerKAT radio telescope.
Yes, MeerKAT found SAURON! Science!
As you can probably guess the purpose of astronomaly is to find astronomical anomalies, so basically instead of having to go through 6,000 images themselves, their framework helped narrow their choices to 200 images of anomalies.
And within the first 60 images, this… thing… they named SAURON was found.
Now, they just have to figure out what exactly they found. And who knows what else they’ll find.
Last but not least is one of the biggest possible mysteries in science and that is the mystery of how consciousness works.
This is a complicated topic, but just for simplicity let’s talk about the easy problem and the hard problem of consciousness.
The easy problem of consciousness basically explains how and why we do things. It’s how brains perform cognitive tasks and run our meatsuit.
Then there’s the hard problem of consciousness. That’s explaining how and why we feel. There are no physical laws to explain this.
And as AI becomes more and more advanced, we’ll see them start to encroach on both the easy and hard levels of consciousness. And that’s going to teach us a lot about how our own brains work.
And maybe even at some point they will come to grapple with the hard problem. And being that they can calculate, analyze, and perform at a much higher rate than us… They might figure it out before we do.
But hey, it wouldn’t be science if it didn’t bring up as many questions as it answers so yeah, there’s a lot of mysteries around AI itself.
For example, we teach learning algorithms the same way we teach children. Feed the system examples of something you want it to recognize and over time, it’ll create its own “neural network” to categorize new things.
But just like human learning, we don’t understand how deep machine learning works. It ends up losing track of inputs that helped it with decisions. Or may it never keep track of them.
Not understanding this is known as the “black box problem.” And there are some reasons why it’s important.
One, it makes it hard for us to fix deep learning systems if they put out outcomes we don’t want.
Then there’s ethics. I mean, deep learning systems are used to make decisions about humans for things like medical treatments, bank loans, and job interviews.
But they can also reflect biases from the humans who work with them. The thing is, the systems can’t explain why they do what they do, so that’s why they don’t seem fair.
Some think the black box is exaggerated.
As Princeton computer scientist professor Arvind Narayanan tweeted in 2023:
“We have fantastic tools to reverse engineer them. The barriers are cultural (building things is seen as cooler than understanding) and political (funding for companies vs for research on societal impact.)”
And some say that claiming we don’t know how it works is just a way to avoid responsibility.
Either way, A.I. is becoming more enmeshed in our daily lives. And while it may solve some of our greatest problems and mysteries, it may just end up being the biggest mystery of all.
Now… if you’ll pardon me… I’m out of coffee.