Season 4 Ep 27: Louise Humpington

Speaker A: Louise Humpington, welcome to the Data Strategy Show. Hey Louise, how are you doing?

Speaker B: Very well, thank you. Thank you so much for having me on. Samir, it's brilliant to speak to you.

Speaker A: Absolute pleasure, absolute pleasure. Um, look, for those people who don't know, uh, who Louise is, who are listening in, um, please do give us a 30,000 foot view of who you are, what you've been up to and then we'll get into it.

Speaker B: Sure. Well, I would. I'm not, I'm not what I describe as an AI ethicist because I feel like it's a bit of a difficult, um, thing to talk about ethics when you don't actually have it embedded within that system already. So I like to think of myself instead as someone who talks about the ethics and governance of AI and um, but I've come at it from a fairly sort of linear, not linear route. Um, I started off, I did a philosophy degree and then went into law as a financial services litig for an international law firm, um, during the credit crisis. That was fun. You can imagine lots of meaty ethics and governance issues going on there. Um, but then I did a fairly sort of dramatic pivot into international development and C suite leadership within the third sector. So I spent probably the last 15 years really working at the heart of diversity, inclusion, human rights, and what happens when systems are either broken or have never been developed in a way which is proportionate to the harms that are caused, but also that involves and includes and promotes the best that we can be as humanity, whether that's within business or whether that's within communities or whether that's just at home. So it's been a really sort of interesting intersection bringing all of those different lenses into the business, um, and AI space.

Speaker A: Wow. So, so, yeah, an amazing career. Uh, so this season is all about women and I think we're going to talk about diversity in that respect. And you've been doing a lot of work on diversity in the boardroom and ethics and so on. So tell me a little bit more about that.

Speaker B: Yeah, I think, you know, we've come to a point where things like diversity and inclusion have become in some senses, sort of weaponized, um, by some governments, by some business leaders in a way that really, I don't think is reflective of the role that it genuinely plays. So when we're talking about the responsibility that a board might have or the responsibility that a team leader might have, you know, within, within coaching, we talk about the golden thread, the alignment that brings everything from the organizational objectives all the way down to the individual. And actually all I think what we're missing when we start poo pooing things like DEI is that actually that's a really important part of keeping that alignment in place. It's a really important part of understanding at uh, individual level what is required for that person to perform to the best of their ability. What is needed within a team to create a culture of inclusion and safety where people can genuinely have common goals that they work towards. And if we get those things right at that level, it does lead to much better performance, much better outcomes, and ultimately that positively benefits the bottom line of an organization and makes it a stronger, more resilient, more cohesive unit that can work towards those overriding sort of business or other objectives. So I think there's a, there's a missing space. That's a really, really important part of why it is important that businesses actually come back to the table around DEI and things like that, because it does have a bottom line impact. Um, but particularly when we're talking about things like AI, I think there's a wider governance conversation that means that DEI actually can be a really important part of the guardrails and the controls and the accountability mechanisms that we have in place.

Speaker A: Can you give me an example? Because uh, I think, you know, DEI is a very big, large bucket of stuff. But can you give me an example of how that's, how that's actually impacted and more so what the challenges are that you see of not having it in place.

Speaker B: So I mean a really good example might be facial recognition software for instance. Now we know that uh, Black women are 35% more likely to be wrongly identified by facial recognition software than a white man right now, had there been more testing on black women, had there been more women involved in the design and the development and the deployment of that product before the police started using it, before federal services started using it, we might have had less harm being caused to people who were detrimentally impacted by it than we would have done otherwise. And I think that within, what I think does concern me about the way that AI is being deployed at the moment is because we don't have those governance frameworks and those accountability mechanisms in place. What we're doing is we're treating governance as a uh, mop up exercise. The harm is already being caused by the time the remediation comes into place. Rather than saying, had we included those people in the room, had we had those perspectives and those points of view considered at the pre deployment stage, those harms may not have been caused in the first place. And so I think we do need to understand why it's important, um, not just academically, but actually because real people and real human lives are being impacted by tools and systems that are being deployed at speed and at scale without those people having been involved in the conversation to begin with.

Speaker A: And is that happening more? Is that happening more here? Do you see that that's now becoming a greater risk, um, for people? Um, you talked about ethnicity there and you talked about the fact that, you know, it's actually having a detrimental effect to women, um, you know, and specifically black women, um, who are being, um, monitored and you know, recognized in terms of that facial recognition software. Um, so how do we, how do we, how do we bring these, these kinds of, uh, challenges specifically for women? Um, and I don't know how it's affecting women in the corporate world. Um, and, and what, what, what is it? And maybe it's just not, obviously it's just not going to be affecting women. It's going to be affecting a broad, uh, amount of people. But what, what is it that we need to do in order to, um, as you said, bring back those. That, that sort of safety and, and the psychological, and not having the deep psychological impact that it's having now?

Speaker B: Yeah, I, um, think that exactly what you're doing at the moment, which is getting more women's voices into the tech space, is a really important. Able to represent not, not entire communities because obviously every person is different, but at least have perspectives whether that is from a black ethnic minority perspective, whether that is another underrepresented group, whether that's a female point of view. If you have those voices represented within the design team, within the development team, within the testing team, within the board team, then you are going to be managing to bring those points of view into that space without actually having to do very much at all. So that's where things like how you recruit and how you possibly, you know, positively discriminate in some areas where you recognize that it's a governance issue actually can help you to prevent and mitigate the risk of harms being caused. But also, you know, potentially businesses ending up with a legal liability if they get it wrong, you know, particularly where we're talking about, say, financial services institutions. You know, there's been a lot of, um, talk around postcode lotteries that are associated with the way the AI is looking at, um, credit risk, for example. Now if you are an individual who feels that they have been negatively discriminated against because of a characteristic which is protected as a board, you're at liability for breaches, potentially of the Equal Equalities act, but also the FCA guidelines themselves, which require you to treat customers fairly. So there is actually a legal liability and a risk liability that sits alongside it just being the right thing to do. And actually an important part of that is making sure that you've got those different perspectives and points of view embedded within your teams and your culture, and you've got the psychological safety within those teams for people to feel comfortable saying, hold on a second, I think we might have a problem here without, without being treated as being disloyal or that they're sort of, you know, hindering progress or anything like that, which is lot of about. So, you know, those are some of the really positive things that can be done quite easily to avoid those legal liabilities, but also to make sure that we're genuinely opening up the conversation to everybody when everybody is affected by these tools and these systems.

Speaker A: And do you think the challenge is that most organizations right now is that they just don't know what this stuff is? They don't know what the AI will eventually spit out. And this is a big problem, right, because the transparency in these models isn't there yet. Nobody's asking the risk type or ethical type questions up front. Everybody's just literally, let's build it and see what happens. Um, and that is the liability that you're talking about. And so I, I don't know whether, as you said, um, do those people know that they have been impacted by these models? And how would they know? Um, specifically, if I'm in a particular postcode, and I see that there is some negative impact to me, how would I know the first things to do? What would be, what would be my recourse? Um, and so that's kind of. How do I start? What would I do?

Speaker B: I think that's a really good question. And you know, don't forget that we have got subject access requests and freedom of information request, um, opportunities available to all of us. Every single person who, um, has data held by another person is entitled to ask for that, to say, you know, what is it that you. Hold on me? How are the, how is this information used in making the decision that was made? And you have a right to do that. Everybody has a right to do that. And so I think part of the sort of, if you like digital literacy that everybody needs to know about, is also about understanding what, what rights already exist, what you're entitled to and what you're entitled to say no to as well. And so that's a really important part of all of us taking responsibility for how our own data is used, but also holding those companies to account and saying, hold on a second, um, I'm not okay with this. And you know, a little bit like, I don't know, you know, the way that all of a sudden every, everybody had to have a sort of a green agenda a few years ago. Um, this is very much the same kind of progress, if you like, that actually consumers have a huge part to play in the values that they hold and the messages and the signaling that they give to those businesses. Now if you think about carbon disclosure, for example, um, CDP Carbon Disclosure Project was a third sector organization. It started sort of early 2000s to gain some traction because it was demanding that companies disclose their carbon emissions and their carbon data in a way which was transparent, but also where you could compare company to company. And what was really interesting about that was that you then move from a place where these big FTSE 100 and FTSE 500 companies were saying, all right, okay, we do want to be taken seriously within this space. Our stakeholders, our investors are requiring that from us, so we will sign up to your voluntary disclosures. We're now at a point where those big businesses are also demanding that their customers are subject to the same controls. And the company's house legislation has also changed so that some of that reporting is now required by law. And that just shows how powerful the role of consumers and the role of investors at, ah, both ends of that scale is on influencing the behavior of businesses. And I think that with the AI space, we're seeing a very analogous situation emerging where actually consumers have a lot more power than they realize and so do investors. Both of those accountability portals will require businesses to do things that maybe they're not voluntarily doing at the moment, but will have more and more need to do if they're to be taken seriously within their space.

Speaker A: Yeah, and I think people often forget the power that they have, um, in order to do because, you know, it's, it's, it's, it's almost like we've been beaten into submission of, you know, oh, no, I can't go and ask those questions. That's, that's got nothing to do with me. And, and so how, you know, let's just shift the uh, the focus on women for a second because I'd like to know how are these policies, um, uh, how are they challenging women in the workplace? How, how are they limiting or or affecting the work or that representation, um, for women in tech and you know, and in companies at large. Really.

Speaker B: Yeah. Well, I think there's been a lot of noise, um, recently, I think in the last couple of months around the adoption gap and um, this idea that AI is being adopted more slowly by women than other cohorts or demographics, um, and it's often sort of you know, treated as being that women don't understand it or they don't feel able to engage in these things. I would like to push back on that narrative slightly and say actually maybe that slower adoption, maybe that more cynical approach, maybe that more reticent approach is actually pretty rational and pretty wise given what we do and what we don't know about how AI works and the tools that, um, the, the power that those tools have. Um, so actually I would, I would push back on this idea that adoption is something which is problematic. I would say that actually, surely your strongest skeptic is the person that you really want to win over. Ah. And actually if your tool is not being adopted by the most skeptical, the most, um, sort of, you know, cynical people within a room, those are exactly the people that you need to be asking why. Why is it that this feels scary to you? Why is it that you don't trust this? What is it about this that makes you feel nervous and not wanting to engage? Because I don't think it's about ability and I don't think it's about capability. There are plenty of intelligent, capable, absolutely choosing not to do this. Women and other, other people as well. Um, but I think the other aspect of it is that, um, when we're talking about, you know, all the hype around jobs being lost, um, and things like that, you know, why would people voluntarily opt into using a system that is being framed as something that's going to replace them? That's not rational. So you've got to actually also understand what do people understand by AI? What do they mean by adoption? What do they expect to happen if they do engage with these tools? And I think it's also really important at this point to distinguish between things like LLMs, which are, relatively speaking, fairly static. You know, they do evolve and they are moving quite fast, but they're prompt, driven, so they work on instructions as opposed to agentic AI, which sits there without a human in that loop running off in the background. And those are where we're really hearing lots of this horror stories and lots of the harms being caught, caused even in circumstances where the AI has been given express instructions and has chosen to ignore them and do as it pleases. So.

Speaker A: Sounds like many humans I know.

Speaker B: Absolutely. I've got three small humans who push me on that on a regular basis.

Speaker A: Oh yes, yeah, yeah. It's an interesting one because like you said, LLMs are pretty static and we know that they're just tokenized based on you know, looking at the next, next um, word or, or next phrase or whatever it might be. Um, or building an image that, that actually has come, you know, leaps and bounds or um. But, but agentic AI, um is, is what everyone's talking about now.

Speaker B: Yeah.

Speaker A: Um, it's on everybody's lips. Um, and we do hear some cases of O. Um, the agentic AI went off and completely uh, I think deleted the entire database and we've lost all of our information. Well, I think those are isolated cases. I think people are still in the early stages of the whole agentic AI because let's face it, most organizations are not there yet. They are probably going to find it's quite difficult to embed that, that sort of unwieldy um, entity into their organization.

Speaker B: Um, you say that but Microsoft in the last few weeks has come out saying that Copilot now has agentic capability. So I wouldn't say that it's necessarily as hidden or as um, not accessible as maybe some people would like to. Yeah, and there are varying degrees. So uh, it's certainly, you know there's, there's different, there's a difference between automation and, and. Yeah, exactly.

Speaker A: And that's the. Yes. Yeah.

Speaker B: But there are, but there are lots of small, small players I would say who are not necessarily have, who don't necessarily have the technical ability to build um, who are using lovable and things like that to build agents or coaches, you know, Claude Copilot or whatever and then selling them on as being products that somebody could embed within their systems. And I think that's where there is a bit of a danger zone actually.

Speaker A: Well you don't have the wrapper of guardrails. You don't have the right. Yeah, yeah, yeah.

Speaker B: But you also are buying from somebody who doesn't uh, necessarily have the technical ability to solve it when it goes wrong. So you know, be very careful about who you pay your hard earned cash to. Uh, particularly if you're a small medium sized business.

Speaker A: I think a lot of this is still R D. You know, I think a huge amount of it is. And um, even if you're buying from Microsoft, it's still going to be in some part R and D, you know, because it's trial and error at the moment and you know, we've seen the open claw and you know, and lots of people went straight into that and realized that oh my goodness, it's got access to every bit of information about me. Uh, and that's another thing that people need to worry about. Um, although you know, half of the people in the world who signed up to Facebook gave, gave away all their information anyway.

Speaker B: Absolutely. You know, but what is genuinely alarming about it is that this R D platform is playing out in real time. This is not going on in a sandbox. That's behind the scenes. This is actually we are the guinea pigs ultimately and um, we have to be alert to that. But I think probably with the LLMs, I would like to just come back to one point. Yeah, we've talked about it as being maybe less gentic AI, but we've also got to remember that the base data that the LLMs were trained on actually is pretty biased and pretty assumptive and pretty foundationally flawed to begin with. So there is still some reticence around being able to spot what comes out of um, an LLM, even if it's prompt based and being able to interrogate the veracity of that and be sort of skeptical and cynical to an extent about what you are being told. Because we do know that hallucinate all the time and we also know that they're trained on data sets which don't necessarily represent all of the marginalized voices and communities that we're talking about. So you know, this still needs to be a little bit of skepticism around outputs as well.

Speaker A: I think, I think you're right. I think it's getting better. Because if you go back to 2002, I remember speaking to a uh, researcher um, in the area and she said that um, she put in uh, the prompt in 2002. Actually it was before GPT came out. Um, and they launched November 20, 2002. Um, she was, she was uh, doing research and testing it and she, she put in um, you know, what uh, is a man and it came back with very positive statements about men. She put in what is a woman and it came back with extremely negative comments about women. Um, so but she said she's done that, you know, through the years and it seems to be getting better. Um, so there are obviously some, some basis for these things improving. Um, and you know to come back to your point, I hear a lot about hallucinations. I hear a lot about, you know, I know plenty of Human beings that lie, you know, and, and then, uh, you know, and they think it's true. And I know a lot of people use information out there that they don't really understand, but they use it to their advantage. And, and so I think that there is. Yes, you're right. It's been trained on a data set that is, is heavily biased. And you know, uh, but that's down to us then using it. At the end of the day, you know, these things have been built for the purpose of. And you look at it, you know, you look at now, you know, students, you look at, uh, people in the corporate world. You know, they are using these things to accelerate their work, whether that is to do with research or whether that is to just to, you know, write an email or, you know, uh, dare I say, I don't know why people are using it to write an email. I still, I still find that hilarious. If you can't bloody write an email, then, you know, gosh, what you're doing in a corporate job. Um, so it's an, it's an odd thing that, that, that we constantly go on about this hallucination stuff and even bias. You know, I'm not saying that's an odd thing, but I'm saying that there is an inherent bias in everything.

Speaker B: Yeah.

Speaker A: Um, you know. Yeah. And it's gonna, it's gonna continue. Yeah, I get, you know, and I think we've got to take these things with a pinch of salt and, and say if we're put off by bias and, and these, these LLMs, which are pervasive, and they are absolutely going to be, you know, still going to be around in the next year, two years, three years. I'm sure they will be. There'll be a new iteration of them, no doubt. Um, we've just got to learn how to be okay with certain things, um,

Speaker B: and be able to challenge them.

Speaker A: Well, there's that as well.

Speaker B: And actually that's a really positive way. Yeah, yeah, that's a really positive way of looking at it. That we are seeing the amplification of these biases in a way that perhaps hasn't existed. There are more of them and they are far, further reaching and they are impacting on more people. And they're really kind of shining quite a strong spotlight on some of those underlying things that have bumped, bumbled between beneath the surface, um, of the way that we all speak to each other and interact and, um, judge and stigmatize and have our own sort of personal biases as well. I think where we're seeing that really sort of scaled version of it actually is making people feel uncomfortable in a way that I'm not unhappy about, actually. Because if you're seeing it at scale and it's making you go, oh, that doesn't quite seem right. That's an opportunity for us to all have a conversation, how we treat each other.

Speaker A: Yeah. And that's one of the things that I think generally as societies, we. We sort of need to go through anyway in some way. Um, because we are evolving all the time. Yeah. Um, and we should be having these conversations. I think it's when the conversations get shut down that we can't talk about those kind of things. That's the bit which, you know, is the bad thing that we cannot, um, have that, you know, discourse and debate and try and work through, you know, our own, uh, parameters or difficulties or whatever it is. But having an honest conversation is what. What we should be doing. Um, I feel like we're getting more into a philosophical, you know, world here now.

Speaker B: Sorry, that's the philosopher in me. I can't help.

Speaker A: No, that's fine. No, no, no. And it happens with me. But let's get back to how this is. You know, I, I really want to understand how an. Examples of where this is impacting women and impacting maybe just, you know, minority, uh, groups and. And what. What we need to do, as, you know, because I understand the principles of DEI and, you know, and sometimes, you know, having that in the back of the mind. Um, and I've said, you know, I grew up here in the 70s and 80s, uh, which wasn't really a great place for somebody who looks like me. Um, but certainly I had to work through all that stuff. And you do, you know, and, you know, um, so. So maybe this is just another iteration of that that we're having to, you know, reassert ourselves and re. And. And re. Understand the world in the way of, um. You know, whether that's gonna be the world in the way of AI Or. Or whatever it is. But, you know, let's go back to the practical things. What would you say to people out there who are, um. Who are. Who are scared of this stuff? You know, specifically, as you said, more, more. More women might get more affected by what. What would you say to them? What are the things that they should do that. That will help them understand and clarify and maybe, perhaps, um, do a little bit more, uh, get a, you know, do some more investigation and get a deeper understanding of this and the way it's going to impact, you know, they themselves and maybe their children and you know, and I guess now we're talking about AI for good as well in some way.

Speaker B: Yeah, absolutely. And you know, actually you can use AI in order to help you do that.

Speaker A: Um, yeah.

Speaker B: You know what, ah, what AI is really, really good at is finding all those different points of view and um, perspectives, provided you prompt it in the way.

Speaker A: Yeah.

Speaker B: In order to get that. So, you know, there's some great free courses, masses and masses of free courses. The University of Helsinki does one called Elements of AI, also has an Ethics of AI course. Go on to Coursera. And there are bundles and bundles and bundles of, of AI from every different viewpoint and angle you can possibly think of. You know, if you're in an industry, maybe have a look at some things that are, uh, particular to your industry and see how it's being used currently in your industry. Um, so that you can get an idea of practical contexts in which it's using. Because I think that when we start actually putting things into real life contexts, whether that's work or personal, that actually gives us a point view and a structure to focus on that helps it make more sense. You know, you can have all these academic and philosophical arguments about how it's, how it's layered and what it's leveraged on, but actually, you know, what you want to know is how is it being used in real life? What, how does that affect me? And there are lots and lots of these courses that will give a more objective approach than the sort of newspaper articles and because obviously they want to present the most salacious version of whatever is going on. Um, but you know, learning is your friend at the end of the day. And AI actually can be quite useful for this kind of thing, provided you prompt it in the way so it can learn about your voice, it can learn about your perspective, and then you can ask it to give you the opposite point of view as well. So quite, quite often when I'm researching my articles, I will, um, I'll say, right, okay, well this is the point of view that I'm coming from. These are the, the arguments I'm going to write about. Now give me the still person response. Now give me the opposite point of view or now give me the point of view from someone who doesn't look like me or doesn't have my characteristics so that you can see that from different perspectives. And that's a really great way of understanding why other people might not agree with you. But also it's helping you to use AI in a way that informs your critical thinking and helps you to problem solve. So we can use AI in a way and this is what schools are really concerned about and probably lots of parents as well that, you know, kids are just writing into LLMs and saying give me the answer to this and being spoon fed answers. But actually there's a really, really positive way that you can use it to actually stop that cognitive atrophy and help it to actually inform your perspective. So, you know, understanding things from lots of different perspectives is the best way of not getting stuck in an echo chamber, but also ensuring that you are being given all the information you need to make your own choice. Find out what your rights are around, what information is being held by who have a. Do a little ecosystem map if you like. Which platforms are you signed up to? Have a look at their terms and conditions. Maybe if it sounds a bit too complicated and you don't come from a kind of, you know, contract or a legal point of view. Exactly. Bung them into Gemini or Claude and say Claude, explain this to me like I'm a 12 year old.

Speaker A: Yes.

Speaker B: And um, then understand what it is you've given and how you've given it so that you can make informed choices about who you choose to engage with and the information that you provide to them. And that's really empowering. And once you start understanding how those things work and what you're giving away, you then feel more empowered to take control of them and to do something about it and to talk to your mates and to tell the other parents in the playground and empower them too. And it's this sort of, you know, AI is pervasive, but so is the influence we all have in our daily lives. And it's that networked approach, that ecosystem approach, I call it the Mycelium network, that everybody is impacted and therefore everybody has a role to play in how we control and govern and hold AI and the companies using it to account.

Speaker A: Amazing. And on that note, thank you very much Louise. I've had such a great conversation. Yeah, likewise, likewise. So where can people find you?

Speaker B: You can find me on Substack. I have a blog on there called Intent from Intentions to Impact. You can find me on LinkedIn and you can find me on my website, kairosynthesis.com cool, great.

Speaker A: Well, thank you very much. Had a fantastic conversation with you. Enjoy your day and ah, I'll see you soon. Take care.

Speaker B: M. Bye. Bye.

Speaker A: Done. Excellent.

Speaker B: How was that?