Credit rating inequalities, making mosquito music, and better batteries
In this edition: How AI has magnified credit disparities, how malaria research was translated into sound, and how batteries can make a better world.
Play the complete podcast (above)
OR listen to individual chapters:
Magnifying disparities with machine learning – We talk to Professor Tarun Ramadorai about new research that shows how machine learning is exacerbating inequalities in credit ratings – and who gets approved for a mortgage.
Sonifying malaria research – How do you turn data about genetically modified mosquitos and their egg-laying rates into music? Target Malaria scientist Dr Federica Bernardini brought in creative composer Jamie Perera to take up the challenge.
Find out more about Sustainability Week at Imperial.
(22 February 2023)
Hello everyone, I'm Gareth Mitchell, welcome along. In this edition from an award-winning business school study, worrying conclusions about how machine learning decides who should get a mortgage. And mosquitoes, malaria and music, we listen to a composition with a difference. And it's sustainability week here at Imperial, so we are looking at batteries and a fossil free future.
Dr Billy Wu:
That's the future I foresee, that everything is going to be electrified, powered by batteries ultimately as an energy carrier and it's all going to be renewable. But to enable renewables, you need energy storage as well.
All right everybody, so let's get going. Gosh, we're in already. We're going to start, as we always do, with a couple of news stories from around the college. Maxine Myers is here to bring us the first one and this is about a relationship between our hormones and sex drive. Why is this newsworthy then Maxine?
Yeah, so this is a research study led by Imperial College London and Imperial College Healthcare NHS Trust. And it's found that a natural hormone in the body called kisspeptin, could be used to treat women and men distressed by their low sexual desire. So this is two studies, they looked at whether given an infusion of the kisspeptin can help people with low sex drive and it's a condition called Hypoactive Sexual Desire Disorder. It affects up to 10% of women around the world and 8% of men, that has huge psychological and social impact. And so they wanted to see whether it could stimulate sex drive in men and women and they found that it can. It's really important study because at the moment there aren't any treatments, effective treatments, especially for Hypoactive Sexual Desire Disorder. So it paves the way for potential kisspeptin based treatment that could be made available to men and women experiencing this condition.
All right Maxine, thank you very much indeed for that. That was Maxine Myers. Now let's speak to Hayley about black holes and dark energy. And this is either the title for the forthcoming Muse album or it's a serious science story out of Imperial College. I think it's the latter. So what's going on here?
Yeah, it's a really interesting study. So there's lots of institutions involved and we are one of them. And it's a story that posits have we found the source of dark energy and is it black holes? Which sounds like crazy physics, and maybe it is, or maybe it's a massive breakthrough. So dark energy is what we call the missing 70% of the universe, because what we observe is that the universe is expanding and it's expanding faster and faster. So things are moving away from each other faster and faster than before. And we think that they shouldn't really, because the Big Bang obviously pushed stuff out. But the universe is made of lots of things that have lots of mass. Suns are big, black holes are big and they should have lots of gravity that actually pulls things together and slows down this acceleration. So to explain the fact that it's actually speeding up, we need some kind of energy that counteracts gravity, that pushes things apart, and we don't really know what it is, which is why it's called dark energy 'cause it's just mysterious.
So what have they been studying then? What's the actual stuff they've been doing?
It's actually kind of the serendipitous result, because what they were looking at were black holes and looking at actually the evolution of black holes over 9 billion years and they wanted to see how black holes grew. Now black holes can grow in several ways. They can take in more materials. So if a star gets too close, they'll suck in that star's material or they can merge in massive events. But we thought that they might actually be growing in another way as well that couldn't be described by these processes that we know. And so they looked at these very old galaxies, called elliptical galaxies, which are sort of done growing by those normal means. There's no more stars around them to suck in, there's no other black holes around to join with. And what they found is that they're growing in a way that seems to suggest what they call cosmological coupling. As the universe itself expands, then the black holes themself expand and then they become a source of dark energy, if dark energy is just part of space-time itself.
That was beautifully, elegantly explained, Hayley. So where does this all take us then?
So it's an intriguing first result. They're suggesting that it's some first evidence that it has been a theory before, that black holes might be a source of dark energy in this way, but of course they'll need to do lots more measurements and test any other possible theories for why black holes have grown in this way. But it's a very intriguing result and has got the astrophysics community all a-flutter.
It certainly has, and me. Great story and, well, brilliantly explained as well. Thank you. That's Hayley Dunning.
Right, now you are applying for a mortgage but you fail the credit check, yet someone broadly like you gets the loan. So why? Could it be because you are from a minority ethnicity? Well, a paper from the business school has recently won a prestigious award for finding that loan decisions made by machine learning, compared to traditional credit assessments, could be more likely to disadvantage those from ethnic minorities. In this case, the disadvantaged groups are Black and Hispanic borrowers applying for mortgages in the United States. The paper concludes that machine learning increases disparities between and within groups. The study has just won the 2022 Brattle Group first prize for the best paper in corporate finance, published in the Journal of Finance. The authors are Dr Ansgar Walther and Professor Tarun Ramadorai, both of the Imperial College Business School. Tarun is a professor of financial economics.
Professor Tarun Ramadorai:
It does seem like the machine learning algorithm, in general, declares more people to be credit worthy than under the old technology. So there's a little bit of an aspect of a rising tide lifts all boats. More worryingly, some boats get lifted a little bit more than other boats do. In particular what you find is that white, non-Hispanic populations had a greater improvement in their perceived credit worthiness under the new model than did Hispanic or African American groups in the population. And so this was quite worrying because even though everybody gets a little bit of an improvement, it widens the disparity in access. In the words of George Orwell, "Some animals are more equal than other animals."
Yeah, I was just thinking about that quote as you said it. Given that it seems as if some of the existing biases in society are being exacerbated by the algorithms. There's an issue here, with the training data, is that what is biasing the algorithms?
Professor Tarun Ramadorai:
What machine learning is able to do is just simply pick up the fact that certain groups had levels of income that looked different from other groups and that those differences in levels of income should translate into differences in credit worthiness just according to the logic of the model. That turns out to account for most of our effects. Now this might seem innocuous until you realize that it isn't, because this begs the question, why are those income levels so different in the first place? And all you're doing then, is you're taking a preexisting disparity that exists in income and magnifying that through the credit system. We want everyone, regardless of their race, ethnicity, or gender to be given an equal shot at getting credit.
Now it turns out that under the old technology, the machine is not very good at picking up the fact that someone may belong to a particular ethnic group on the basis of their income level. The machine learning model is much better able to triangulate which ethnic group you might belong to on the basis of your income and then use that information in an end-run around the regulation, to then penalize certain people with worse credit scores. And that accounts for about 15% of the effect. So that's a less innocuous reason.
Given the algorithms are, I say "only" in inverted commas, reflecting what they see, what they've been trained from, from society and relationships between income levels and availability of credit. It's about fixing society, he said in a slightly naive utopian way, isn't it?
Professor Tarun Ramadorai:
No, of course it is. I think that's absolutely correct. I think there's sort of two very high level findings that one can take away from this. The first one is that if you put regulations in place and new technologies come in, regulations like you should be blind to the identity of people when you're allocating credit and so on, those regulations have got to be updated to reflect the fact that new technologies are essentially going to come in and undo whatever legal standards you put in place. This is about the triangulation aspect, which is about the 15% or so. And this is a different way of saying that regulators need to be upgrading their technology at the same time that credit providers are upgrading their own technology.
So the other aspect of this is, you're quite right, machines are going to unveil things that exist in society. It isn't as though the machines are necessarily only going to be fabulous at leveling out the problems that you have. They're also going to expose the problems that you have in society. And in some sense that's what we find with the 85%. The machine is just better able to pinpoint the fact that there are these historical discrepancies and in some sense it's just pushing those wedges even further. So I agree with you, we have some introspection to do here about the way to do these things right.
It shows almost that the algorithms are too good. I mean we know that the outcome is bad, but they're doing their job, they're doing what they're programed to do, aren't they? They're doing it very well, too well.
Professor Tarun Ramadorai:
Absolutely. This paper allowed me to think a little bit more about where I think the constraints lie. One approach to this would be to say, "Oh, maybe we should just slow down a bit in terms of the kinds of technologies that we're allowing to proliferate in society." But I'm not convinced that that is the answer. I think the answer is that if we are moving very fast in terms of the science, we should move equally fast in terms of the social science. And so this is one of the reasons why economics can be an incredibly useful compliment to basic technology. And another reason why a place like Imperial is right at the heart of this kind of investigation where we have both excellent science as well as excellent social science to be able to understand how we might translate technology into impacts and consequences for society.
Professor Tarun Ramadorai, the award-winning paper, by the way, is called Predictably Unequal: The Effects of Machine Learning on Credit Markets. And you can search for it online in the Journal of Finance.
Well now let's listen to a musical composition. Well this piece is called Swarm, but to call it music or a composition doesn't quite do this sound art justice. What you're actually listening to is data. And the data is from a 2018 paper published in the Journal of Nature by Imperial College researchers. The landmark paper describes CRISPR-Cas gene-drive technology in suppressing mosquito populations. The influential work has been a big contributor to the global battle against malaria. The two-minute track, Swarm, aims to raise awareness of the threat from malaria and of the gene drive technology working against that threat. Jamie Perera made the track. He's a sound artist and composer who uses sound in innovative and provocative ways, including through sonification. He's collaborated with Dr Federica Bernardini, a senior post-doctoral researcher at Imperial. Hayley Dunning has been speaking to both the composer and the researcher.
Federica, it was about a specific paper that Jamie was making the sonification. What's the science? Can you tell us about the study?
Dr Federica Bernardini:
So the sonification is based on genetic technology that we have developed based on gene drive. A gene drive is essentially a natural phenomenon that causes a genetic trait to spread very quickly within a target species, generation to generation. This means that if you modify a gene that has to do, for example, with the fertility of female mosquitoes, you can couple this genetic modification with a gene-drive system and make sure that this modification is inherited by all the offspring. What we did in this caged experiment at Imperial, we released these genetically modified mosquitoes in this wild-type population within a number of cages. And what we were doing throughout the whole experiment was to measure, first of all, the frequency of the genetic modification from a generation to the other one. And at the same time we were also taking count of the eggs that were laid by all the females present in those cages, at every generation. And accordingly, the population within each of these cages was collapsed in very quick time. Eventually all these female were sterile and no eggs were collected for any of these cages.
So how did you go about translating all of that into sound?
It was quite technical, this sonification, but the overall feeling that we wanted to get was to evoke the experience of some of the people that are on the hard end of malaria and what we discussed was, well how do we bring some of that threat to our ears? The thing that you hear at the beginning that's very striking is the sound of thousands and thousands of mosquitoes, and that's representing the hatching rate, which is a 100% at the beginning.
And then as you go through the sonification, there's a sound that starts welling up from almost silence, which is a synth that is representing the gene modification rate, which goes up as the hatching rate decreases.
And all of this is broken up by rhythmic markers. There's a sort of pulse going on that's keeping exact time, one beat per day, and then there's a little rhythm for each week. And then you basically hear the mosquitoes diminished to a point where you can actually hear the music that's representing that gene modification come through.
The only thing that is not a direct sonification is at the end where you just hear almost near silence to provide a sense of relief from that frenetic beginning where you are kind of being inundated by the Swarm, so to speak.
Federica, what did you feel when you first heard the sonification?
Dr Federica Bernardini:
It was strange. I didn't really know what to expect because it doesn't really happen often that someone takes data that are, again, generated in a molecular lab and translate them in music. I think I had a smile throughout the whole track. As Jamie said, it gives you that sense of relief that is in a way mimicking what we hope.
Federica Bernardini and Jamie Perera, both speaking to our very own Hayley Dunning. Federica is part of the Target Malaria Consortium. And if you'd like to hear Swarm in its entirety, search for it on the Target Malaria webpages. There's a great blog post from Jamie about it all too.
Well finally, it's Imperial College Sustainability Week this week, five days of workshops and interactive events where Imperial researchers tell us about their work on the transition to zero pollution and it's a chance to find out what the college is doing to reduce its own environmental impact. In any conversation about shifting from carbon, it's not long before the subject of batteries comes up. So our reporter, Julie Hoeflinger, has been speaking to Dr Billy Wu, a senior lecturer in the Dyson School of Design Engineering.
Dr Billy Wu:
If you look at where we use batteries right now, they basically power all our consumer electronics. So your mobile phone, your laptop, that's powered by the lithium-ion battery. Right now, electric vehicles and electric bikes, they are now being powered by batteries and you'll find them in everything, so your AirPods and all of those other devices. Without it, essentially, I just don't think those devices would exist. Now hypothetically you can have a blackout where the electrical grid goes down and if you have no alternative there, then essentially you have no electricity. And right now electricity is so ubiquitous to our life that, okay, anecdotally you can't use your TV or whatnot, but there are other more critical services. So in a hospital, if you've got a respirator, which is keeping someone alive, if you have no electricity there, then that's terrible.
Specifically in regards to making more sustainable batteries, why is that so important? Why should the everyday person care about that?
Dr Billy Wu:
Yeah, so we are going through a once in a generation technology transition, where we're moving away from fossil fuel powered internal combustion engine vehicles. So around the world, countries have acknowledged that global warming or climate change is happening and a major cause of that is emissions from things like road transport, air transport, and how we use energy in general. So I think there's an undeniable transition towards decarbonization. So what we need to do is develop batteries, which is one way to power low-carbon vehicles. But the problem here is that the batteries that we have right now are still quite expensive. They don't last as long as they could do, and the recharge times could be shorter as well.
So to have mass market adoption, we need to improve and address all of these problems. And the problem with lithium is that most of the world's lithium is found in places like Chile and Argentina and Australia, that then gets shipped over to China for processing, they then make it to batteries and then they ship it over. And lithium is really expensive as an element and we're only going to be using more of it. And that's why we're looking at alternative battery chemistries for the future as well. There's a type of battery chemistry called sodium-ion, which uses sodium and you will find sodium in literally the ocean, which is full of salt there, and those have the potential to be significantly cheaper. And if you have cheaper batteries, it basically makes electric vehicles a lot more accessible.
Why is this research important to you?
Dr Billy Wu:
So for me it's really about transitioning away from fossil fuels and we can see the impact of it globally, but you can also see it locally as well. So I cycle around London, and one of the things you notice is the air quality. If you go down Oxford Street, which has often loads of buses that kind of stop-start, the air quality is actually really bad and you can feel that. And if you are behind a large truck as well, you can actually feel the dirt that you breathe in and so on. So that has a detrimental impact on our health as well. So for me, developing these low carbon technologies is essential for us to move away from that future.
What's sort of the end goal for all of this? Say like 50 years down the road, what would we want the world to look like with this research?
Dr Billy Wu:
Yeah, so in a utopian future, we're using a 100% renewables. So the challenge with renewable energy, such as solar power and wind is that it's actually quite abundant. However, it's intermittent. That basically means the sun shines and the wind blows. Not all the time, as we know. However, when you look at how we use electricity, that is misaligned with how we generate electricity and there's a big balancing act. And if that balance is disruptive, let's say if everyone boils their kettle at the same time, then what can happen is that you can get blackouts or brownouts where you lose electricity because the power plants can't spin up fast enough and that problem only gets worse as you have more renewables on the network.
So in that future where we are mostly using renewable energy, we also need batteries on the electrical grid to store that excess electricity when we have the sun shining and the wind blowing, and then for it to discharge when we are actually consuming that electricity. More and more things will be powered by lithium-ion batteries, we've seen things like E-Bikes and E-Scooters, but they will also be used in larger scale items. So things like not just cars, but trucks and also integrating with solar and wind. And that's the future I foresee, that everything is going to be electrified, powered by batteries ultimately as an energy carrier and it's all going to be renewable. But to enable renewables, you need energy storage as well.
Billy Wu talking to Julie Hoeflinger. And you can see the TikTok version of that interview, guess where? Yes, on our very own Imperial College TikTok account.
Well that's it for this edition. Thank you very much indeed for listening. Just before I go, a reminder that we are on all your favorite pod platforms and if you don't want to sit through the whole thing, which would be a terrible shame, but if you don't, you're busy. Why not? If you want to just listen to chapterised versions of each of our items, you can find them on the Be Inspired webpages on the Imperial website. All right, until next time, take care and goodbye.