#290 Scaling Responsible AI Literacy with Uthman Ali, Global Head of Responsible AI at BP

Hi, I'm Maggie, VP of Curriculum at datacamp, and I know firsthand how overwhelming it can be to keep up with the rapid changes in data and AI. But staying ahead doesn't have to be hard. That's why our curriculum team has curated the best learning content to help you and your team stay ahead, covering everything from data and AI literacy for busy executives to cutting edge generative AI for developers, and everything in between. Whether you're upskilling yourself or your team, we make learning practical, engaging and flexible. It just works. Power your data and AI transformation journey on datacamp.com.

At the top of that pyramid is kind of where you want your I guess you call a responsible office or your center of expertise, where it's basically a dedicated part of the organization, where these were the company's subject matter experts are that produce the guidance and training that gets filtered down. This is what makes AI ethics in my job so interesting because there are so many different people you need to work with, from an average person that just kind of uses ChatGPT now and then to an actual data scientist that's actually building potentially these solutions, right? You need appropriate training depending on what the user is and based on that Persona, essentially what does every organization want? They want all of their employees to basically use AI, but safely, with the awareness of the risks and how to mitigate them or what the company's approach is. So that in its core is basically responsible AI. The conversation has sort of reached a critical mass, but it's pretty clear now for everyone that we need to go beyond conversations into actual practical implementation. Foreign.

This is Adele, Data evangelist and educator at Data Camp. And if you're new here, Data Framed is a podcast in which we explore how individuals and organizations can succeed with data and AI. Earlier this year we ran a report asking 500+business leaders in the US and the UK to rank the most important AI skill their teams need to know. And 77% of leaders mentioned that responsible AI should be mandatory training to everyone. This isn't really surprising to me as the more AI becomes infused in the DNA of an organization, the more costly the risk becomes and the more responsible AI best practices you need. So how do you scale responsible AI across your organization? How do you scale responsible AI literacy? Enter Uthman Ali. Uthman is the global Head of Responsible AI at bp, and throughout the episode we spoke about the importance of responsible AI literacy, how it looks like in practice, how to build a culture of responsible AI, and a lot more. As always, if you enjoy this episode, make sure to subscribe to the show, give it a review and a thumbs up. And now let's get started. Uthman Ali, it's great to have you on Dataframe.

It's good to be here. Thanks for having me on.

Thanks for coming. So you've been leading responsible AI at numerous organizations. You've been in the AI ethics space for quite few a quite a bit now. Earlier this year at Datacamp we ran our State of Data and AI Literacy Report 2024 edition, where we asked more than 500 business leaders in the US and the UK on what is the most important AI skill their team needs to have. And 77% of leaders mentioned that responsible AI should be mandatory training for everyone within the organization. So maybe set the stage. Why is it so important to upskill your organizational and responsible AI?

Because if you just take it back a step, right? What does every organization want? They want all of their employees to basically use AI, but safely with the awareness of the risks and how to mitigate them or what the company's approach is. So that in its core is basically responsible AI. So if you want things scaled well and used appropriately, your organization needs to be responsible and need to train and upskill them to do that. Because for most people before ChatGPT, they probably weren't familiar with what AI is or how it works. So again, there's a huge upskilling needed across so many organizations.

Yeah, and you mentioned here ChatGPT, and from the outside looking in, it seems that the responsible AI conversation has reached a critical mass over the past few years. Right. The advent of generative AI and ChatGPT. In a lot of ways, ChatGPT put AI in the hands of the masses. And I think everyone clicked in their head what AI is capable of and why it's so important. And then also more recently, the EU AI act, which puts a lot of pressure on organizations as well to make sure that they are compliant. So how true is that that we've seen the culmination of these two forces really push the responsible AI conversation to critical mass? How do you view it? As an insider?

I think it's been a perfect storm to highlight why this field is so important. But one of the things you mentioned is the conversation has sort of reached a critical mass. But it's pretty clear now for everyone that we need to go beyond conversations into actual practical implementation. One of the key skills you have for any program, whether it's like compliance or just responsible more broadly, is you need to train people on what to do. So on the one hand you need people filling in responsible AI assessments in your organization. Collect all their use cases. But even for people reviewing this stuff, what does good guidance look like and how should it be implemented? So there's upskilling across everyone in the organization that's required.

And you mentioned here the upskilling that everyone needs within the organization. What does that look like in practice? I'd love if you can deep dive into bit. What does a successful responsible AI program look like and what are the skills that you need to have within the organization?

I mean for learning and development in particular. Right. This is what makes AI ethics in my job so interesting because there are so many different people you need to work with. From the average person that just kind of uses ChatGPT now and then to an actual data scientist that's actually building potentially these solutions. You need appropriate training depending on what the user is and based on their Persona essentially. So when you talk about bias and fairness with an average gen AI user, average office work is like I might just use ChatGPT to automate some stuff. It's very different high level conversation to someone actually building a system. When you're saying actually what's the appropriate metric to test for? What does the representative sample size actually look like in your data set versus like procurement versus hr? And again, these organizations are so big, some of them that people want guidance that's relevant to them. So you need to make it use case specific so you don't want people sifting through loads of documents to go. I'm working in marketing. Okay, here's what my company's approach is. You want it to be quite slick, right? You want them to kind of open up their website. The guidance is given to them. You're almost spoon feeding them. And that's almost like what good response by service should be. It's almost like customer service success. Give people what they want when they need it in the best way possible.

And there's that keyword that you mentioned here which is Personas. I wholeheartedly agree with this. Everyone's relationship within the organization with AI is different. Procurement professional maybe different from a marketing professional, different from a data scientist building models. Walk us through those Personas a bit more depth. What are common Personas that you find with an organization and how do you arrive at your own Personas? Which I think is probably the bigger question a lot of professionals have in this space.

I think within each organization I always split it into three things. Almost like you have the classic consultant like Pyramid. You've got the base layer which Is most of any typical organization of people that kind of use AI a bit need to be familiar with what it is. That's just general users that might need to know some of the risks and limitations. But then you've got that middle layer who are basically the people that might be in procurement or in hr maybe higher risk exposure areas of the company and even some of the data science and actual data teams too who are responsible for actually implementing more use case specific guidance. Right. Those are people that really have to be doing stuff. But at the top of that pyramid is kind of where you want your I guess you call a responsibly office or your center of expertise where it's basically a dedicated part of the organization. Where these were the company's subject matter experts are that produce the guidance and training that gets filtered down and how you find Personas can often be based on which parts of the organization are are you basically trying to hit like how many different parts of the company are there from different business entities and departments basically is a good one to start with and figure out what type of training or education do they need.

One thing that you mentioned, there's a spectrum of skills in a lot of ways depending on where you are and the Persona. If you're someone who's less technical maybe AI Aware would be a great term for that. The type of guidance that you would need is on how to use consumer app tools like ChatGPT or something along those lines and how to best go about that. If you're a developer or a data scientist the skills as well that you need differ. Do you find that the spectrum of responsible AI skills that you need to kind of impart within your organization also increase in complexity the more you go up that complexity Persona because maybe a data scientist needs to know something like SHAP values as part of their responsible AI toolkit. So yeah, walk me through kind of that technical nuance that nuance in the spectrum of skills they need to have.

Because this is where it gets like super interesting in terms of what skills you need to build a responsibly AI program. Because you're going to need legal expertise that's a given for compliance. Then you're going to need someone like me with an ethics expert. Right. But then the legal and ethics departments are often the ones that basically say these are things that we should be avoiding. This is how we protect the company's reputation but also be compliant regulations but also define the company stance. But when it comes down to who's actually doing a lot of this Stuff that will actually be often your data department. Right. So it's one thing having a policy saying we don't want any models that buy us to discriminatory. Right. The procedure, the way that actually gets done will be from your tech teams actually looking at data cleansing or selecting the right samples, or actually doing bias and fairness testing. So again, there's a huge technical component that you can't have without responsibil because the technology is so. The technical aspect is so important for building a program.

Yeah, I couldn't agree more. And then we're going to unpack that in a bit more depth. But there's also. We're talking about skills a lot. I think a key, equal, important aspect of skills transformation or cultural transformation in general is also behavior transformation. Maybe. What are the key behaviors that underpin an organization that has high degrees of responsible AI skills? What are the type of instincts your tech professionals need to have, for example, to be able to operationalize responsible AI effectively?

I think in terms of you hit the nail on the head on culture, and this is really what the job is more than like risk mitigation. It's with AI ethics. One of the things is not just saying this is the ethical risk of introducing a new product or service. It's also like, what is the risk of not innovating now? Basically, what is the risk of actually not keeping up with the times, particularly if there are things that it could improve, like, for example, safety and operations, depending on which industry you're in. But in terms of those key skills, it's incredibly important that again, like, you have the right technical expertise, you have the right legal and ethics expertise to filter down guidance. But the cultural change that you want to see is people that actually want to fill in these assessments and learn more about the topic, learn more about what the company's approach is. They actually want to be upskilled and learn what to do. And again, for companies, this is one of the key drivers because if it's seen too much as like restrictive compliance, people are only human. They'll look for ways to not get around it, but they'll be like, look, I really want to put off doing this thing right. But if they view it as, look, I speak to these responsibly people, they're actually adding value to our product or service or finding ways that this could be improved. Actually, even commercially, it's actually improving the thing we're building as well. They're more likely to come to you again, like a customer success philosophy. But how you build your Program and your team structure. Right. The cultural change again. For lots of companies, they've had AI ethics principles about fairness, transparency, accountability. You need to make these lived reality that everyone believes in, in the organization from when they walk in, for when they're onboarded. They're taught what the company's approach is, who to speak to for when they exit and they ask, how do you think we did in responsepli? How should we be improving this?

Yeah, I couldn't agree more. And I'd love to kind of focus on that last point here on lived reality because a lot of times ethics departments, responsible AI departments get a bad rap because organizations may want to, for example, ethics wash their existing AI practices. And you're making that point here of lived reality, I think allows that responsible AI department to go beyond that perception. So maybe what are best practices that you advise here for these types of functions to go beyond that perception that we just discussed?

Yeah, I think one of the biggest things is actually being deeply involved in the business where you operate. So for example, when you explain what AI ethics is and why it's important, actually going through use cases with various product teams and explain what the field is, I think goes a long way because there's often a perception that ethics is that this is like this barrier to innovation. The ethics person might be the police officer, they're just going to shut down things. But that's actually not what this role is in industry. The role is about helping people make informed decisions and actually know the benefits and the burdens and the trade offs. And when you explain these things, it can be at a very technical level. Right. But if you're a data scientist and you say, look, if there's a classic conundrum between accuracy and bias, if we slide more on one scale versus the other, this is what the unintended consequences could be. A lot of them actually see very, I say, almost always see value in this because, hey, you're actually explaining again, the product and what we're trying to build and where this could be going next.

Yeah, that's wonderful. And I love connecting that to the business objective and the business function. And you mentioned something earlier when we were discussing kind of the upskilling component. Right. There's the legal team, the ethics team, the technical teams. Right. In a lot of ways this is uncharted territory, managing kind of a responsible AI upskilling program or even a responsible AI program at large. Right. Who should be in charge of driving this agenda? Especially who should be in charge if you don't have someone who's owning the responsible AI and ethics agenda. So yeah, walk me through kind of your best practices here for who should take the charge and how do you organize this effort within your organization?

What I'm going to say is going to sound like the classic corporate cliche, right? You know, they say responsible AI, it's a team sport, but it's so much of a team sport it's hard to pin down who's the captain.

Right.

Who's actually the captain on this team. But I'd say you've got to break it down into different components. If you're specifically looking at AI compliance, that should be your legal department. And also I'd say even the ethics should also be in there because this is often the part of the company that says, what can we do? Especially from legal, it's like what can or we cannot do? What are the clear legal red lines? Ethics is the question of should we be doing this? But then your technology teams are the ones of like, okay, how do we actually go about putting principles into practice like transparency, Whether it's things like watermarking, avoidance of bias, discrimination. Right. And in terms of ownership, there has to be a top down effort because if you start from a bottom up effort you can have some success. But eventually you're going to run into issues over what's the mandate for what responsible AI means at an organization who's basically the working group or the people accountable for creating basically the ethical principles or the values or requirements, or basically creating the company's global approach. But that needs to come from the highest levels, the C level at the company to go, this is on the priority list, this is where we think this is going. Or even to delegate downwards to basically your responsible AI office or your AI governance office or committee to say, look, the C level might say, we're not experts, you guys are. Explain to us clearly what this field is or what the options are for implementation. I will be the ones to sign off on what this means for us as a company.

Yeah, that's wonderful. And you mentioned here building guidance for organizations globally kind of the AI ethics stand as well. You've built this for quite a few organizations. Walk me through the process of finally reaching that end state product. What goes into it? What is the resources, the inspirations that you go through and what are the inputs that come into play that come out when you see that final finished product?

I think one of the biggest things that I'd advise companies to do is take a real case law approach to this. I Think in my field now, this is becoming more common practice where it's one thing to say we should be avoiding bias or discriminatory outputs as a common example, but you need to guide people through clear examples of where this could happen in your company. The CV screening one HR algorithm that discriminates against people based on the race background, where they live, socioeconomic status is a classic example. But what I advise companies to do is once you have that mandate and you have your AI governance committee or ethics committee or whatever you want to call it, basically the team captain that thing is set up is to run through a load of even fictional products that you can see your company building from co pilots and digital assistants to basically think through what are the unintended consequences? Actually, what are the ethical risks that we can see from things like job displacement or company reputation or whatever it happens to be? The key thing is that it needs to be harmonized with what your company's guidance already is. So for example, the ethics team will say, we're against discrimination, right? Okay, everyone's against discrimination. It's like a legal mandate to be against discrimination. But when they say we want to do the right thing and be equitable, it's like, how do you know that? That's your company stance. The only way to do that is to actually look at your existing policies, procedures, public statements, to actually infer from what your company's position is and even a strategy. Where do we stand ethically as a company? So a good example I think is in even bias and fairness in hiring. You might say, as an organization, we'll do enough to not basically be legally penalized. That's legal, right? They'll be like, that's what we're going to do. But ethically, people, they kind of be like, okay, what's the ethical approach? Even when it comes to things like affirmative action programs. But if you take a step back and you say, look, our company is a target by 2027, we need x amount of female employees in this department. That is actually even the commercial goal of the company, which is linked to diversity, equity, inclusion. Right? If we use this discriminatory tool, we're not going to get there. So that's how you can anchor your guidance to basically what is the company stated it wants to do.

I love that. And it comes back to that point that we discussed as well about how to make responsible AI actionable and operationalized. One thing as well, when you've been managing these programs, Othman, what are some common challenges that you can expect along the way we mentioned how really important it is to get C level executive sponsorship, I think lack of executive sponsorship is probably one of the biggest death nails for a responsible AI program. So, yeah, walk me through some of these challenges that you can expect managing these types of programs and where do you start if you have that sponsorship?

I think one of the biggest challenges again is even defining what AI governance, responsible AI means for your organization. Because there's the ethics and compliance angle, which is a lot about risk mitigation and even ethics, if done well, can have commercial benefit as well, right? But then there's the angle which tech teams often come to, which is like AI governance for me is how do I actually build stuff well and do it quickly, right? They're thinking, what's the commercial aspect of this? But also just how do I work with AI in my company? And again, it's being clear, again, like from the top down over what is the mandate for this? What do we mean by responsible AI? Which key teams actually need to be involved? The key personnel? Because if you leave it too open or too vaguely, you can run yourself around in circles where basically using the same words like even AI governance, if you ask a cybersecurity person versus a data scientist versus a lawyer and ethicist, they might come up with four different definitions of what AI governance means, right? So you might be thinking, well, you're using the same words, why don't they get it? But because you're not being clear. So even to start with, to say in this response by program, who's involved? But even write down what are your objectives, right? What are you trying to get out of this? What does good look like for you to even draw a line over? This is actually what this governance team is trying to do. This is what we mean. And then after that you can think through what guidance do we need? How do we structure a program from all the things like learning and development to third party risk. Then you have to think about, okay, what are even some KPIs ways we're going to measure success for each of these actions. And when you slowly start doing this, you're like guidance. And that sort of thing comes under this as well. You start actually building a full program, then you look at how many teams need to be involved, who's important. You can build the racing matrix, but that's how you basically will start to be able to scale this sensibly. But that is also a great resource to give to C level executives to go. This is actually what it takes to embed this at the company to also not just be legally compliant, but also meet our other strategic objectives at the company as well.

And one thing that stuck with me here is defining the metrics and KPIs that matter to be able to measure the success of the program. What are some of those KPIs and metrics and are they organization specific or can you find common ground with different organizations?

I think a lot of them will be organization specific, right? So classic ones around just compliance training and that sort of thing is how many people have done the modules, the learning development and workshops. You can quiz them, did they understand it? You can run through use cases, even fictional with them. Did the people actually understand the guidance or do you actually know how to implement that? Right. But then you've got the other side, which makes this interesting is that companies might often have innovation metrics where we want to basically make X amount of money or revenue or increase value by this amount by this date. Right? Using AI. So you need to design them in a way that's kind of harmonized that your responsibil program and your basic innovation metrics aren't always conflicting. Because if you just have metrics in one part of the company which is like zero risk, zero risk at all, they might turn around and say look, we shouldn't be doing anything at all. But if you have other part of your company which are just pure cowboy approach of make as much money as you can in the shortest amount of time. There's no cohesive strategy here. And this is why it needs to be coming down from basically C level executives and most senior people. You have to say it's almost like a negotiation, but also a well thought through strategy to say even in terms of how many high risk use cases or models or products can we have, how many are we resourced to review? You kind of want to put them in a way that like it's sensible where you can say we're well resourced to do this. No one's going to complain saying look, we're getting shut down all the time because no one to review stuff. But also your responsibility team aren't saying, look, we're getting overwhelmed with so many use cases because now there's basically like a coherent plan.

And then one thing that you mentioned here that I wanted to ask you about actually that segues perfectly is you mentioned that you know the IT component here. How many tools do we have in our organization that are proliferating? There's got shadow IT phenomenon where a lot of organizations have Grassroots usage of tools like ChatGPT or MidJourney or something along those lines. And we've seen organizations weigh this differently, right? Some creating really strict compliance about what type of tools that you can use, some outright banning ChatGPT within the organization. What is the balance that you need to strike here if you want to balance between responsibility and innovation? Should you ban chatgpt? Should you not?

I think again, this is what it comes down to is that the reality is the genie's out the bottle. You can't really ban it the way you do. People will find a way, just go on my phone, I'll just do on my personal device. They'll find a way.

I literally know people that have their laptop open and then their personal laptop open to use.

Yeah, yeah, there you go. They're just multi screening. All that's the problem is if your metrics aren't tailored, is that you might be presenting perfect metrics saying we mitigated all risk, but there's no usage in the company. But again, it's not lived reality because you know, people are doing it on their personal devices. So that's why you've got to know your own organizational culture, but also be sensible with this. And this is where culture is so important in upskilling people, is that you want to say, I can trust you to use this tool, but not use sensitive company data or do anything that might put us at a reputational risk. Right? I'm going to trust you to basically do this. I'm going to empower you that I don't have to basically watch you, I don't have to be a police officer. And then afterwards you can say, look, these are like our list of approved tools that we've been through. Please use these ones. These are the ones that we vetted even for security considerations. These are the ones like, look, these are the guardrails. Stay within these walls and you'll be fine, basically. And that's basically the sales pitch for AI. Ethics should always be that this is like an enabler for digital transformation, if done well.

And how do you create monitoring mechanisms to be able to monitor any misuse or irresponsible use of AI?

Yeah, I mean there could be monitoring in ways that you can monitor even for using security tools to see websites where basically blacklisted. Just general use AI. Right. You can monitor usage. So how many people are using those sorts of things. But then if you go into like actual prompt auditing, there are ways you can audit like even were there toxic or harmful prompts or outputs produced. And you can actually have ways to filter for those. There are tools you can use like shadow AI detection to actually see how many models do we have potentially in production that we weren't aware of. But there are also tools that you can have over actual monitoring for things like technical risks like use case or model drift to actually see did this fall below basically acceptable thresholds. So monitoring again is a really important one to basically when you have a huge organization, you need some way to basically have a good understanding of where risks are proliferating across the organization. Again, why is that?

And then you mentioned something as well about that element of creating trust that we trust you, we are empowering you to use these tools effectively. And I think a big part of that by which you get that trust, at least as program leaders doing assessments. So you assess people's skills, you assess people's capabilities. Understanding how should you go about creating assessments for responsible AI? What should you be testing for? And how do you communicate the use of assessments in a way that inspires confidence from your audience and not just as a the police is here type.

I love the phrasing like inspiring confidence because a lot of this is what people are looking for, confidence that we actually know what we're doing moving forward with AI. Right. But again something you mentioned is really important is even defining the scope of what your responsible AI assessment is. Because if you're not careful in figuring out what your mandate is early, some people would say the commercial value or even alignment to business strategy should be part of responsible AI. Other people at ethics or legal compliance department say we're solely more about the risk mitigation or reputational risks. Your technical teams might be like, look, we want to make sure people are following best practices. So again you need that clear mandate to show if there's almost imagine a responsible questionnaire, you know, you're definitely going to have legal in that. That's like non negotiable ethics. You should be having in that as well. And ethics often precedes things like technical assessments as well. Like even when you look at bias and fairness, disparate impact assessments, how that gets conducted is often quite technical assessments. But then if you want that commercial aspect in there, does this use case generate value? Again, that's the company wide decision to go. Is that within the scope of responsible AI? And to each company it's quite different. But those are usually the core tenets that you want to have.

Maybe when you working to assess these people as well, kind of coming back into the messaging and kind of the communication program around the use of the upskilling program and the assessments, et cetera. What are some best practices that you've learned around how to get people excited about being part of the agenda and not just willful participants?

Yeah, I think the biggest thing is that you want people to actually want to do this. You don't want people to say look, this is more assessments I need to do on top of other assessments. So I think one of the first things that companies should do is actually when you look at responsible AI and actually AI more broadly is that usually people right now are super excited about AI and what it could do. So you already have a lot of buy in. I think generally I've yet to find someone that doesn't find AI ethics in particular as interesting because this is like a really, it's almost like an existential topic that we're all facing now as well, even in the future of how are we going to work with machines in the future. But in terms of building that confidence and wanting people to do it again, it's a good opportunity for companies to actually look at their existing assessments and actually figure out how can we just make this more palatable for people. If you've got three to four different teams asking three to four different questionnaires, if you know you need to introduce more stuff for responsible Inow, is this not a great opportunity to just solidify everything into one simple format or have all teams speak together at once? So just reduce the time taken to do reviews. I think that in a lot of ways inspires confidence to people that like, look, there's dedicated resource here, there's a team here, but also they're actually trying to make my life easy to go through this process.

And you mentioned something here, kind of that long term conversation about what working with AI means and I'd love to actually switch gears to discuss that more long term perspective. If the AI space keeps moving at the same speed that it's moving in today and if the same pace of improvement keeps up now there's a debate on that. What does this mean for the responsible AI agenda? How do you think the next few years will evolve?

You know, it's so funny because I was actually at a digital ethics conference recently and this is one of the roundtables I led was, well, it was basically unstructured. You could ask anything. Right. And my question was what will our jobs and responsibly look like in five years? You know, we've never really thought about it. So let's Sit down and actually think about what will the future job roles be. Is it going to be like a new form of chief compliance officer for responsible AI is actually linked to more sustainability and well being. But then we also thought about new jobs you might see is actually like the role of almost AI in storytelling and also building company narratives and branding, learning and development. There might be bespoke roles specifically for responsible AI, given how complicated this is actually with the map.

We already see this in the data literacy and the data literacy space. We already see data literacy officers literally bespoke people.

But the one I like is because we're actually entering an era right, where you're having AI enabled crime. Deepfakes are a great example. And even a future role would be almost expert witness. Even for people working in responsible AI to actually be giving expert testimony to even talk about, oh wow. Specific almost in criminal trials, it's like this is basically an AI ethicist perspective of how this AI was used or why this basically. Basically the scenario came about. So I think almost like when you look at other fields, it's almost going to be that AI specific niche revolving around. We need an ethics expert or someone involved in responsible AI to talk about this.

Yeah. And how do you view organizations having to adapt their responsible AI programs to increasingly powerful AI systems as well? Like what are kind of the pressures that you imagine organizations will be under?

I think it's almost like building the plane whilst flying it. Right. That's the reality of where everyone's at, from big tech developers to consumers. I think for most companies, what they're trying to get to grips with is that many of them still don't really understand what AI ethics is and what this means. A lot of them aren't sure. Is this just some sort of vague activism or is this some sort of new form of business professional ethics? So a lot of them are struggling to really understand what is responsible AI. What does it actually mean for my company if I embed this at scale? What does this look like? So you've got that issue, but then you've got other ones where they're like, is responsible AI a standalone thing or do we embed that within existing compliance programs? Does it follow the same processes? Should it be different? So companies need to take approach on that. But then you have all this wave of multimodal gen AI and all this new stuff coming. And even with AR VR metaverse, companies are like, who's actually looking at what might disrupt us next? Right. Like who's actually keeping their eye on the ball because the next ChatGPT might happen. Right. And it's like when that happens, the risk is that if we're too narrow in our approach, even in our guidance for AI and what this program looks like, if the new thing comes up, we might have to tear this down and start all over again. So when you look at digital ethics more broadly, you need to have it in a way that's quite adaptable, that when the next disruptive tech comes, you're like, actually we kind of know how we can just basically consume that with whatever we're doing now. Right. We don't need to create a whole new team or process or program for this. We're actually setting our company up for success in the future.

Yeah. So that jump to being proactive is really essential. And one additional point on we discussed this slightly early in the episode when we talked about the EU AI Act. Even looking at the current landscape, and I can only project this into the future, there seems to be strong disagreement within the AI and tech community, for example, about the need for regulating AI. Some people think that the UAI act is not strong enough. Some people think it's an overreach and it hinders innovation. How do you see this debate happening over time, especially as the technology gets stronger? What do you think are modes of regulation that will have to be conjured up? Because it's a global market, its impact is a technology that has global implications as well?

I think this is a really fascinating topic actually for lots of companies even designing the responsibly program. A key conundrum is do we make the EU AI act that benchmark? Because if this regulation is the strictest in terms of documentation and even the prohibited categories list where it could be 5, 7% turnover for this being used in Europe, is that going to be a global ethical stance that this prohibited uses will? Actually, that's a company wide approach. Right. And many of them, I think quite sensibly are saying, we're not going to make a decision on this yet because it's quite too early. But we'll be able to have a framework that's both flexible. But you're prepared for the EU AI act, essentially. And in terms of people working in tech, I actually know lawyers even that said, they're not convinced by the UA Act. I know others that are fully for.

It saying this is in what way? Is it too much of an overreach or is it too much of a limited regulation?

I think what it comes down to there are a few things. One of them Is that the question, the need for AI specific regulation? Because the argument is that it's arguably a distraction because you already have lots of regulation already, like data privacy, IP anti discrimination laws, consumer protection laws. We already have lots of stuff.

Yeah. You can't do libel against anyone because there are libel laws, even if it's AI generated or human generated. Yeah.

So we already have existing stuff already. So to say, what is it about AI that mandated the need for a specific regulation dedicated to the technology? And I'd say one of the most controversial points about the EU AI act is the regulation over the size of the models itself and the technical aspects. Well, this was the thing was should you be regulating the use case, which I think most people agree with is basically high users where risks occur. But when you start talking about regulating the size of the model itself and tech development, that's when it's okay, how far are we going beyond the scope of originally what this is intended to be? And also does regulating the things of the size of models, is that even a good proxy for risk given that tech developers are trying to make them smaller? Actually. So again, is that actually a long term solution? So again, I think that it remains to be seen, but I think what I'm hoping we'll see is that a global sort of harmonized approach where we kind of agree on this, is what best practices look like for what you should be documenting for things like AI risk management, even AI ethics in practice. Different cultures might have different perspectives of ethics which might inform which prohibitive practices look like globally. But hopefully companies will be in a position to say that this is where we're willing to comply with or take a stricter stance. But I'd say again, for most people building their programs, remember, AI is not the Wild west in terms of regulation. There's already a lot of stuff that you should be complying with.

As it stands now as we wrap up today's episode, Uthman, do you have any final call to action or maybe one single piece of advice you have for listeners today that want to get started with their responsible AI agenda within the organization? What's your single piece of advice?

It's hard to narrow it down to one, but I'd say a single piece of advice, right, is actually put together a plan because you know that this is something your company has to do for even legal compliance reasons. Right. Actually get together with a few experts. Most companies might have a volunteer network of people interested in this to actually sit down and start thinking through even just vision pain two years from now if you were to embed your vision or responsible at the company, what does this actually look like? Even take the compliance out of it because we all agree that this is quite important even for the future of work. When you enter your workforce, when you get up in the morning and you start going into work or you're doing a daily commute or whatever, what does that actually look like? What does it feel like to work at this company? What do the processes look like? I really start having a vision of what that is. Then work backwards from this and go, what do you need to start doing now to basically achieve this?

Okay, work backwards from the plan. That's probably the best advice in any single context. I have to say that is always the best place to start. Uthman, it was great to have you on dataframe. Couldn't thank you enough.

Thank you. Thanks so much for having me.