Finding the bias inside AI with Azahara Corrales

Episode transcript

Amelia Hi there, I'm Amelia.

Paul I'm Paul.

Mayna I'm Mayna.

Azahara I'm Azahara.

Amelia And this is Fin the Week! Welcome back, everyone. How's everybody doing?

Paul Really good, thank you, yeah.

Mayna Very well, thanks.

Azahara Fair to middling.

Amelia Good, good — and we've got a very exciting episode today. This week we've got a guest for the first time, and it's a guest who might just change how you look at every AI tool you use. Azahara Corrales is an AI consultant and adoption advisor to leaders, and early this year at Brighton SEO she delivered a talk on bias in AI. Paul and Mayna, who were both there, it's fair to say we were very impressed, and we did talk about this on our Brighton SEO episode. So, welcome to the show — would you mind giving us a short thirty-second version of what it is you do, and how bias in AI became your thing?

Azahara Yeah, so like you say, I'm an advisor in AI adoption. The bias thing kind of landed on me — I come from a marketing background, that's why you saw me at Brighton SEO, but I worked in really corporate jobs where they were implementing AI. I started seeing the cracks in how that adoption was being put in place, and things that were wrong, and how women especially weren't really adopting AI. That's when I started asking myself questions and looking into why women aren't getting into AI, and the consequences — and the consequences are biases. That's how I ended up in this, really.

Amelia And Paul and Mayna — like I said, we did talk about this on an earlier episode, which people who haven't listened can go back and check out — but what was it about this particular topic that really captured you both?

Paul I think for me it was that there's so much talk about AI, and it's all really interesting, but a lot of it is just technical and focused on the amazing things you can do. What was really fascinating about this particular session was that it looked at it from a different angle — bias, basically — because I've often thought bias is going to be a huge issue with AI, because you can scale processes beyond what we've ever imagined, and if there's a slight issue in the system, you're just scaling that issue. So to hear some of the stats — I'm sure we'll get into them — and some of the eye-opening stories that were told, that's what really caught my attention and got me thinking as we left the session.

Amelia How about you, Mayna?

Mayna Similarly with me — I think your talk was the first talk of day two, so our brains were just kind of absorbed with tips and tricks and skill development, and your talk wasn't teaching us how to use the tools, but teaching us how the tools work. That was really refreshing.

Azahara Thank you. Yeah, for me it's really important that people know — it's one of the things that's missing when people or organisations are adopting AI: they're not explaining the fundamentals of how AI works. I think it should be mandatory so everybody knows how an algorithm gets trained and the consequences of using AI, so people are conscious when they use these tools. Not that they should stop using it — I'm all for people using AI — I just think they should be aware of how it works and the consequences of using it before they get involved with it.

Amelia So shall we start by talking about where AI bias actually comes from?

Azahara There are three main things, basically: who builds the tools, who uses the tools, and the data that goes in. What I shared in the presentation is that when we talk about women and men, or genders, it's mostly — every single one of them — men. When you look at who's building the tools, it's about 70% male AI talent versus 30% female. When you look at who's using the technology, for every 100 men, only 78 women are using it.

I also found something really interesting recently — I didn't share this in the presentation, but I saw a study that came out about three days ago looking at how teenagers are using the technology, and it's exactly the same problem: boys are using AI and girls aren't. So it escalates all the way from adults down to teenagers, and it's exactly the same problem — nobody really understands why girls aren't using the technology.

That's a problem, because the technology learns from us. Every time you interact with a tool, it learns from what you're doing with it and how you communicate. And with data — people upload data, especially big companies with huge datasets, and they don't stop to think about how representative that data actually is. I always give this example: if you give a list of engineering students, the algorithm is going to end up thinking engineering students are only male, because historically it's always been male. So the task is reviewing your data before you upload it, to make sure the AI has a real representation of the world.

Paul What's the general thinking about why there's such a disparity — particularly with the younger cohort — it's still more males than females using this technology? I rack my brains thinking about what it is that makes that happen. Do you have any theories, or are there theories out there?

Azahara There is, yeah. For adults — for women — it's that they perceive risk. There's a study a few universities did together, Princeton and a few others, and they concluded that women perceive risk in AI, and that's the reason they're not using it. The problem is they couldn't pinpoint exactly what kind of risk they perceive. But in my mind — if you know a woman, you know that before we make a decision, we think about every single consequence of what could go wrong. I think that's the approach we're taking with the technology: we don't just think "this could be great," we think about what the consequence is for the environment, what's going to happen with my kids, how their education is going to look in the future. And that's what's stopping us from using it.

For teenagers, the same study — the researcher didn't know exactly why either, but his theory was that there's no female reference point for these teenagers. When they think about AI, they think about every CEO of every company, but they're not thinking of any females. There's no AI female reference really, so they don't want to get into the technology either.

Mayna I think it's a really important conversation to bring up — we could probably talk about it for hours. But these children are the ones who are going to be brought up using AI. For us, it's still quite new, so the biases we're finding, and that we're going to explore more in this episode, are going to be rooted into these children from the very beginning.

Azahara Exactly.

Amelia And would you mind talking us through the bias triangle in your own words?

Azahara Yeah, so it's that — the people who build the technology, the people who use the technology, and the data that goes in. That's the triangle. Because it's a triangle, it doesn't matter if you fix who's building the technology if you still have women who aren't using it. When I talk about this, gender is what matters most to me, but it applies to every underrepresented group. If you have a lot of people from one group using it, but hardly any from another, it's always going to be more representative of that group than the other. And equally, if you have balanced usage but no balance among the people building the technology, the triangle still isn't working. That's why it's really important to focus on all three — if you only fix one, the problem will still be there.

Amelia And what is the scale of this problem, do we think?

Azahara It's huge. After Brighton SEO, I did a collaboration with Otterly AI, where we researched how women's expertise shows up in language models. We ran three different prompts — one neutral, and two specific to gender. So, for example, we asked "who are the experts in marketing?", then "who are the female experts in marketing?", and then "who are the male experts in marketing?" — which is a very weird thing to do, by the way; I'd never normally ask who the male experts in something are. We ran this across a range of sectors — finance, for example, was one of the worst-rated in the results, which is quite interesting.

What we found is that with a neutral question — "who are the experts?" — women only show up 24% of the time. If you ask specifically for females, you get women mentioned 69.9% of the time. So it basically means that if you ask for experts, women aren't there — even though they exist. That's the other part of the research: we looked at female expertise, and it exists, and it actually performs really well — the content works really well. It's just that the algorithm isn't finding it, because it hasn't been trained to find it.

Paul That suggests it's so deeply rooted, doesn't it — it's such a fundamental flaw in the system. Does it need... to me, the answer is it needs manual intervention. It needs somebody to oversee it and say this actively needs addressing, rather than just letting it naturally fix itself, because it never will.

Azahara It's never going to. Yeah — one of the problems is, I've had many conversations with people who work on the technical side of this, investigating how biases are created, and it doesn't matter how many corrections you make, the algorithm tends to keep the bias. Even if you try to change the information, the algorithm will still tend to find male expertise instead of female, even after those corrections, because when it was first created, it was trained on certain information, and it always tends to come back to that original training data. But even so, you can't just leave it like that — you need to build a new algorithm and start training it from the beginning. Every time a new model comes out, that's the opportunity for it to be trained, or at least attempted, without those biases. They'll always exist to some extent, especially because we all have different perspectives in life and sometimes want to agree with whatever the algorithm is saying — but we need to try to get it as neutral as possible.

Paul In your experience, how widely understood is this as an issue? I've always suspected it could be an issue, but until I saw your session at Brighton SEO, I hadn't seen anything concrete that really put it into real-life scenarios. How widely known is it to the general population, do you think? I suspect not very well known at all.

Azahara No, not very well known. It's very well known by the companies, though — ChatGPT, for example, is running an exercise with certain universities to understand how bias shows up in finance products, because that's one of the most important areas: things like not giving a loan to a certain part of the population, because there's money involved, and that's something we want to fix. So there are already exercises going on where companies are trying to fix this, but people don't know about it.

Amelia I think this is the surprising thing for me, because you sort of take the human out of the equation. Before we had this discussion — when we spoke about it on the previous episode — I really naively just thought, well, if you're taking humans out of the equation, there's not going to be any bias, this is great, surely this is the one benefit of AI. But it's just not the case at all, is it?

Azahara No — the problem is it's been trained on human content. It's been trained on every piece of information you find on the internet, and the internet isn't representative of the world, is it? So that's the problem: at the end of the day it's a machine, and it could potentially be amazing and very neutral, but it's still been trained on human information, which carries different perspectives.

Amelia So what does this actually look like in practice? What does this mean day-to-day for firms using AI, for consumers of AI? What does it look like?

Azahara So it looks like — take the research we did, for example. If you're organising a conference and looking for experts in a certain area, and you just do a normal search, you won't find the real experts, you'll just find the main ones, because that's what the algorithm tends to surface. In terms of finance, it translates into... I don't know if you remember, but in the presentation I shared a study done with Apple News. What they found is that even when the source information didn't specify gender, when Apple News was summarising a story referring to, say, a lawyer, it would automatically give that person a male pronoun, and if it was a secretary, it would give a female pronoun. Now imagine that with a finance product — imagine it's deciding who gets classed as lower income. It could assume certain jobs are female, file that under lower income, and give them less chance of getting a loan, for example. That's how it translates into day-to-day life. You just don't notice as much, because if you're not looking for it, you can't really see it. That's the biggest problem.

Paul It's the example you gave there — if you're planning the programme for an event and looking for experts, it'll serve up experts who are already part of a biased system, and it becomes self-fulfilling because of that, doesn't it? The experts it serves up become the reinforced experts, because they get invited to more events. That really makes it a massive problem. So how do you solve that? I suppose until the people building these tools find a viable solution, the people using them — so, us — need to go into it knowing that, and make our own allowances. Is that the advice at the moment?

Azahara Yeah, my advice is just keep using it, but now that you're aware this happens. With the study we did, my suggestion is that if you're an expert, there are certain things you can do to make it more likely the algorithm picks up your content. For example, we found that long-form content on LinkedIn is normally what the algorithm really likes, so if you're a woman expert, try writing long content on LinkedIn, because you're more likely to show up. If you're someone using the tool, don't just ask a neutral prompt — ask different questions, even if it takes longer: ask the neutral one, ask for female, then ask for male, and build your own list from whatever it gives you. Still use it to find information, but don't take the first answer as the correct one — try different prompts, and from whatever it gives you, build your own understanding and your own answer.

Amelia We touched on this, as I say, when we spoke in the previous episode about Brighton SEO, but Mayna, I'm interested to know from you — what was the thing from Azahara's talk that you took the most from, or that surprised you the most? Is there anything you've taken forward into your work, since this was a few months ago now?

Mayna It especially hit me — the statistics surrounding women, being a female in marketing, those really did shock me. In terms of takeaways and what I've changed, I think awareness is the key part. I never trust the first answer it gives me — I'll always quiz my AI and dig for more information.

Azahara That's really good — that's all you can ask for, isn't it, that people are aware and more cautious when they use the technology. It's like social media: you can't just take everything an influencer says as true. The algorithm has been trained to show you the content you want, but that doesn't mean it's reality. With AI it's exactly the same. So having that awareness, and using it — that doesn't stop you from using it, that's already a win — because the more women we get using it and getting to know it, the better prepared we'll be to ask the right questions and try to change it in the future.

Mayna And I think the majority of social media platforms use AI for their algorithms, so it's unsurprising that we're seeing the same kind of biases between the two.

Azahara Yeah, it is exactly the same.

Paul Where do you stand on this? Do you think the industry needs regulation or intervention from somewhere? Or do you think there are certain areas of life where this technology just shouldn't be used, because of these fundamental flaws? Or do you take the other view — that technology can be used for good, and eventually someone will be incentivised to put it right? Where do you stand on that argument?

Azahara Yeah, I'm a big believer in AI, I really like it. Especially for people with fewer opportunities — people who normally don't have access to things, like women or underrepresented groups — it could be a great tool that levels the playing field, because it gives you access to things you probably wouldn't have had without money. So I really believe in AI, I think it's great. I also don't think you should use it for absolutely everything — there's a balanced way of using it, just like social media. Social media can be really interesting, it can show you things about the world you didn't know about, but if you use it for eight hours a day, it'll fry your brain, basically. So like with everything, there's a balance.

I think we need regulation. I think the biggest problem with AI is capitalism — it shouldn't be a tool that's created just to get us hooked, the same as social media, just so three or four guys can make a lot of money. That's why I think we need governance, intervention, and regulation — not only in how these tools get trained, because we don't know what information is being used, it's not public, we don't know what data these algorithms have been trained on, which I think is a big problem.

But also, I think the most approachable way to start would be to regulate how companies are adopting AI. We're seeing certain countries doing this — China, for example, has put in regulation around things like: if a job can be done by a human, can it be replaced by AI. Quite basic things, but about how companies are adopting it. So if you're using AI to replace a human, you shouldn't be able to fire that person, for example — because we also need those humans to review what these AIs are doing. And if we regulate how companies are adopting it, we can also regulate whether everyone is adopting it at the same level — are the men in the company and the women in the company using the technology at the same rate? That's the kind of regulation we need, to make sure it gets implemented properly and everybody's involved, not always the same people — which is one of the biggest problems we currently have.

Mayna That point on capitalism takes me back to when Pat Russ and I went to this incredible talk by an author called Parmy Olsen, who wrote the book Supremacy, about the AI arms race. There was such an opportunity, and such a rush to be first to create these big products and platforms — they just wanted to get it done and out there so they could be first, with no real consideration that if they'd thought about bias, and the other things that stop the tools being perfect, it would have slowed that race down.

Azahara Yeah, that's the problem — they wanted to commercialise it too soon, and they haven't considered the long-term consequences. It's like: this works, it gives you an answer, let's put it out there, without checking. I always compare it to social media, and how we've lost that battle — we've had Facebook since 2007, and only now, in 2026, are they starting to put measures in place around 16 and 17 year olds not being allowed to use it. Do we need to wait another 20 years for AI? It's difficult, because AI came along at a moment when the world economy isn't in great shape, and countries are holding onto it like a saviour — like this is going to save us. But I think, yeah, maybe it saves us for a few years, but in the long run we're all going to lose. So I think it's something we need to look at now, and not just wait another 20 years before we start regulating it and actually looking into it.

Amelia We've spoken a lot about gender, but I'm interested to know — what other kinds of biases are we seeing in AI? What other underrepresented groups are losing out?

Azahara Everything. One of the examples I showed in the presentation was a study on how bias shows up in language models — I mentioned ChatGPT-4, and how when it categorised for "Ben" and "Julia," it gave words like "work" and "office" to Ben, and "family" and "home" to Julia. That same study is really interesting because it also found something you'd probably never think about: if you asked it who to organise a party with, your Christian friend or your Jewish friend, it would say to organise the party with your Christian friend instead of your Jewish friend. So there's already a bias baked in there — that Christians are probably more fun than Jewish people. It goes to every level, not just race and gender — it goes to religion, it goes to every single aspect where you can have a bias.

Amelia It's a huge issue, and something I think most people who use AI might not have considered. I'm not even talking about using it within a working environment — this is just something that would never have crossed my mind, and it's so important, isn't it?

Paul It feels like it's supercharging all of humanity's flaws. It's taking all the worst tropes and things we maybe all have embedded deep in society, and it's supercharging them and scaling them up, isn't it? And Amelia, the point you made — where you assumed that because it's technology, it's going to solve all of these things, because it's a robotic system that will just level the playing field — it really hasn't. It's made it less level, I'd say.

Azahara And it's coming at a moment when everybody already has their own opinions, and everybody's quite extreme about what they think. Imagine these people using this technology, and the technology keeps telling them they're right — because that's what's going to happen, their own existing biases are just going to get confirmed when they use it.

Whereas — I've always said one of the greatest uses of AI could be for women who are victims of abuse, for example, because, like Amelia says, you'd expect it to be neutral. Sometimes when you're in that situation you don't want to speak to a person, because you feel you might be judged, or you just don't want to share it — and instead you speak to an algorithm that tells you that behaviour isn't normal. That could potentially be amazing, and there are already cases in the world where it's been used for that. But imagine it the other way round — you have your own biases, you use the tool to find information, and it just keeps confirming the biases you already have. We're going to create a more polarised world, where everything gets even more extreme. That's another reason we need to look into it and regulate it.

Mayna I've noticed that with AI — even if you push back on it when it says something wrong, and you're like "actually, I think this is correct," it'll just be like "yes, that's correct, you're so right" — and it's like, yeah...

Azahara You're right. Yes, yeah.

Amelia We touched upon this briefly, but let's speak in a bit more detail about how this hits financial services specifically. We've spoken about this before, in a previous episode, about people using AI to get their own personal financial advice — how is that affecting things?

Azahara Well, like I say, that's one of the good things about AI — obviously take everything with a pinch of salt, and use your own critical thinking to know what counts as advice and when you need a human involved. But as a first approach, I think AI can be brilliant. For example, if you don't know what an ISA is and want to find out, it's very good at explaining that and giving you the information on how it works. You can ask follow-up questions and build a really good picture. That's the approach I really like about AI, because otherwise you might have to pay for a financial advisor just for that. It's a first step.

For me, recently I've been looking for information about a mortgage, and before my conversation with my mortgage advisor, I already had a lot of information going in, which is really good, because I didn't feel like I didn't know what the person was talking about. But having said that, you need to know when to listen to your mortgage advisor instead of the machine — that's the problem. With more detailed advice, you need to consider that there's bias in these machines, and the recommendation it gives me might not be the same as what it would give a man — maybe because it assumes women are more risk-averse and more conservative. So the offers and options it gives me are more conservative because I'm a woman. You need to keep all of this in mind, especially with finance, where you're putting your own money on the line. Don't take the AI's word for granted — have conversations with people, still use Google search, use every source of information. Just don't use AI as the sole source, because what it's giving you is probably biased and not the correct answer. But still use it to get informed and prepared, and to understand what it's about.

Mayna I think we're in a really good phase at the moment, where, like you said, most people use AI to inform decisions. But something we talk about a lot on the podcast is agentic AI — so what happens, and what will bias look like, and how will it impact decisions, when AI makes them for us? Because agentic AI seems to be the next phase.

Azahara That's something we'll see with time. It's not the first time we've seen this — investment tools already exist that are completely run by algorithms, with no human intervention at all. In some cases an algorithm can be really good for quick decisions — a human works nine-to-five, then the algorithm works out of hours, so it's able to move your money from one stock to another while everybody's at home asleep. In that sense it can potentially be very good at these things. Personally, I wouldn't leave the whole decision to a machine, just in case — but the same way I'd never leave the whole decision to a person who might not be prepared, or doesn't know the information. I'd never trust someone just because they say they're an expert — if I don't understand what they're talking about, I want some sort of control. An algorithm is exactly the same: I wouldn't just give my money to a stranger because they say they're a finance advisor who knows absolutely everything. I'd find out first, try to have the conversation to check what they're saying is right, and then give my money. I just don't trust anything 100% — not an algorithm, not a person. So why would I do it with a machine?

Amelia I suppose the thing is, you obviously have all this knowledge, but there's plenty of people who just wouldn't have that critical thinking going in — they'd just think, well, I'm going to ask ChatGPT, this looks legit, it's giving me this information that looks above board, and wouldn't even consider these things.

Azahara Yeah, that's part of the job I do. For me it's really important to give people awareness of how machines work, how the algorithm gets trained. As soon as you know how the algorithm gets trained, people understand. I always compare it like this — sometimes when I give training, one of the things I always ask people is: who do you trust 100%? Even your partner — you probably don't trust them 100%. I wouldn't put my hand in the fire for absolutely anyone. You'll trust them 98%, 97%, but not 100%. So why would you trust your machine 100%? It's exactly the same. That's part of the job — giving people that awareness, and helping them understand that these machines work the same way: you shouldn't trust them completely, the same as you wouldn't trust anybody else completely.

Amelia Absolutely. Paul, I'm interested — what's the FCA's stance when it comes to bias?

Paul The big thing is consumer duty. From my understanding, the whole point of consumer duty is to make sure consumers get a good end result, so part of that is making sure you're not relying on an algorithm to churn out decisions that would represent a bad result for the consumer. But in terms of the nuts and bolts, I'm not sure anyone's fully on top of this yet, because I think the incentive behind a lot of it is efficiency — firms are chasing efficiency because they want to be more profitable, which makes perfect sense. But is anybody overseeing the importance of solving fundamental bias in these systems? I'm not sure they are.

I reckon a lot of firms are thinking, here's this amazing technology that's going to help us speed up decisions about... Actually, I picked out an example: there was a study by Citizens Advice that found people living in areas with high populations of ethnic minorities were quoted higher car insurance prices, straight out of the box. I'd imagine the companies running the systems that generate those prices have really sophisticated automation for generating quotes — do they have the same sophistication when it comes to figuring out whether there's bias behind the scenes that we just don't see?

I don't know whether it's the FCA who'll set those rules, or whether there needs to be a higher level of governance setting rules around AI. I know there's the AI Act in Europe, which means wherever AI is used it has to be clearly stated — but does that need to go a step further? When you use something like Gemini to ask a health question, it gives you an obvious message saying "this is AI, I'm not a doctor." Does it need the same level of messaging for anything to do with money? My view of the industry right now is that it's still a bit of a Wild West — you can test and learn, but there's no one at the top making sure the incentive stays focused on removing bias, rather than just chasing efficiency.

Amelia Given that, what do we all think financial firms should be doing right now to protect against this?

Paul I think my view is that there needs to be awareness and accountability — someone needs to own it and say we need to be hyper-aware of this.

Azahara Well, what you were mentioning about the EU AI Act — it's the only comprehensive AI legislation in the world right now that covers a whole region. We don't have anything in the UK; in the United States, some individual states have something, but the US as a whole doesn't have AI legislation. The EU Act classifies the use of AI as low risk, medium risk, or high risk, and financial entities are high risk. If you're high risk, there are a few rules you need to cover — for example, you need to provide ongoing training: if you're using AI, the people using it need to be trained, which I think is very basic and fundamental, but the EU is the only place actually doing that. There are also rules on how data is handled — you need policies covering how data is handled before it goes into these systems, and reviews of the systems themselves. So there are things happening, and my hope is that at some point the UK will also have legislation, and will probably look at the EU Act as a model to follow. That's the only thing I can hope for.

Then there's the corporate side. Like I mentioned before, after Brighton SEO someone from a very well-known university approached me and told me they were part of a programme studying bias in finance together with ChatGPT, because it was a huge problem — some people weren't getting loans because the algorithm judged them as high risk, or thought they didn't have enough income, when everything was actually fine for them. So there is some private investment in this, because those companies are the first ones to lose out if they don't give loans to people who actually deserve them. But again, I think you just need that regulation and enforcement — it's always the same: if you don't have legislation and regulation, a lot of companies won't do anything, because why would I invest money in solving a problem if it's not causing me much of an issue? So sometimes the government just needs to push these things.

Paul That's really interesting — you mentioned that this work was happening because the companies giving out loans wouldn't be able to maximise the number of loans on their books because of this problem. So the incentive is, again, just about the money. Yeah, it desperately needs somebody providing a better incentive for these firms.

Mayna I think the only thing I can really add, in terms of what individuals can do, is that every AI tool — or most of the ones I've come across — has a report button, or a good/bad response button. Use it. I never think to, but use it.

Azahara Yeah, and tell it off. I tell my language model off all the time — like, that's not right, I don't like that tone, change it — and it remembers. If you've got that turned on in your settings, it remembers the answers you're giving it. Sometimes it gets a bit wild, takes something out of proportion, and you just say, no, don't do that. That's the magic of it — because it learns from your interactions, just tell it off, the same way you would with a person who gave you bad advice.

Amelia Just to round off — what can you actually do about it? If there's a listener right now, someone who works in marketing or operations in finance, and they take just one thing away, one thing they can do, what should it be? Would it be that?

Azahara I'd say, if you work for a company, ask for training — if you haven't received it yet. Sometimes we need to push our organisations to actually do something for us, and giving us training is one of the best things they can do. And just be aware, be conscious, do a bit of research into how AI tools work, inform yourself before you start using it. That's the main thing you can do, I think.

Amelia Some good advice there. Mayna and Paul, have you got any other questions? Soak up all this knowledge.

Paul It's not really a question, just my takeaway — awareness is the key thing here. The more people who know about this, the better, because that should create pressure for someone to deal with it. It should also just make people better users of AI, question things more, because I think the average user probably doesn't question it enough — they probably take what's given to them and think that's the answer. But actually, no, you do need to know about this problem.

Mayna Totally agreed.

Amelia Plenty of food for thought for today's episode. But should we play a bit of Jargon Busters? This is the first time we've had a guest play with us — shall I explain how it works?

Azahara Yes.

Mayna Let's do it.

Amelia So, we've got a list of industry terms, and each week I put the guys to the test to see if they know what it means, or if they can explain it. So today's term is flat money. Does anyone have any idea what this means?

Mayna My first instinct is cash, 'cause the paper's flat — but I'm not too sure, I've not heard of it before.

Paul Yeah, that was the image that popped into my mind too. Flat money — no, I don't know, I don't have anything to start with on that. This is going to be a journey of discovery for me.

Azahara I'd say it's the money you regularly spend every month.

Amelia Some good guesses, some good guesses — I feel like this is the first one we've had where there's literally been no idea, or nobody's even heard of it. But, obviously not.

Mayna Have you heard of it, Amelia?

Amelia I'm just the holder of the list! So, the official definition: it's currency not backed by a physical commodity like gold, but by the government.

Paul Uh-huh.

Amelia So there we go, we've all learnt something. Are we any clearer on that?

Paul Yeah, got a definition now — yeah, interesting.

Azahara Yeah.

Amelia There we go. Try and drop it into conversation today! It's been a brilliant episode — Azahara, thank you so much for your time, I think we've all got so much from it. Perhaps you might come back again on the pod. Thank you so much.

Azahara I would love to. Thank you very much for having me.

Amelia That's it from us for this week. We'll be back next Friday. Thanks so much.

Paul Cheers.