40. Microsoft’s 6 principles for developing the best AI strategy | Jed Griffiths

23:02Those learnings have done me well in how to think about driving innovation in heavily regulated environments. So there are still things you can do, though, right? Just because something is locked down, it doesn't mean that you can't build mechanisms for innovation. Interesting. But you can't test the product. No, luckily we can't test it. Luckily, we don't test the product. I think that's probably the best way to say it. Or at least we don't. So no A-B testing, no. No, that's right. Yeah, which is probably a good thing for the world. I think it is a good thing. I'm really pleased that there's no A-B testing.

23:31But then you mentioned that you pivoted and you modestly said you were quite surprised. I'm not surprised that you'd done successfully in the new role. For someone thinking about pivoting, what should they be reflecting on? I mean, we talk a lot about transferable skills in recruitment. Yeah. But what were you thinking about and how did you make it work? My career was quite an interesting one. Being a physicist, working in research and development and a bit of manufacturing and then moving to sort of a technical policy and advisory role, you know, to policymakers in government around the sort of nuclear program.

24:04And then transitioning into organizational strategy and innovation thinking, you know, working with lots of other organizations on what we just talked about, you know, how to create innovation in a regulated space. And then moving into Microsoft, you know, that changed. The reason I looked at that was I could see where the world was going in terms of how how prevalent digital technology and, you know, AI and how that was changing how businesses were thinking around running their business and driving innovation. But this was before ChatGPT. Oh, yeah, that's before it landed, right? But actually...

24:34But you can see it coming. Yeah, you can see it coming. And, you know, being a bit of a nerd, you know, I'd read these sort of, you know, Journal of Computer Science type articles around the nature of what were then sort of adversarial, generative adversarial networks or GANs, you know, where that technology was going. Transformer technology. You know, this was sort of the 2017 mark. You could sort of see on the horizon, wow, this is pretty, this is different. You know, this is a different technology base and that's going to evolve and it's going to accelerate and it's going to be really, really important.

25:02And it has, right? And it's ultimately led to. So you wanted to be part of it. I wanted to be part of it. Yeah, that's basically what happened. And so I realized, what is it that I could offer? And it boiled down really, you know, it was my approach to thinking. And that's, if I think now where we are with AI as well. some of the core skills that you know i certainly look for in candidates or that i was trying to put forward in myself and you know when i when i made that pivot was how do you think about the world can you how do you assess information you know can you can you navigate complex environments and someone's telling you one thing how do you check that's true how would you you know how would you set about like an experimental thought right experimental thing how would i set about testing if you've told me something that's that's right that's why i'm interested you you described your approach to thinking.

25:45I want to know what that is. Yeah, it's rooted in the scientific method. So it's rooted in the scientific method. Yeah, yeah. So you don't believe anyone, you want to go and check it. Well, that's not... Or is that unfair? Yeah, that's right. I don't believe anyone. But you are always testing. It's just a mindset. I think something... Is that the scientific mindset? Yeah, I think it is. It's one of the wonderful things about why I think people who are STEM professionals who have had a career in, you know, science, tech, engineering, mathematics as well as STEM, that the scientific method is a wonderful way of framing your thinking for life, because it gives you that rigorous process of essentially critical thinking.

26:21And as we move now into an era of AI, where information could be at your fingertips, summarized or synthesized from huge volumes of other information sources, it's the number one skill, right, of how do I think through this? Is that likely to be the right answer? What might I do? How might I probe and test this? And how do I need to turn that into something useful so I can do something with it? It does seem particularly important at this point, especially critical thinking with these new emerging technologies. Yeah, I think it is. I think someone told me recently from LinkedIn, sort of the top sort of 15 skills or so on LinkedIn.

26:55I think strategic thinking, adaptability, critical thinking are now up in the top five, I think, or top 10 of skills that employees are looking for. So that suggests to me more young people should be doing STEM subjects. or they should be thinking around what i call the kind of four c's right you know the four c's collaboration communication critical thinking and creativity right how do you maximize say them again yeah you're going to say collaboration communication critical thinking and uh creativity the four c's i love the four c's right so that's good i like that many many organizations or many disciplines i should say you can leverage the four c's and they're totally transferable and i think they're really, really, really good skills to maximize on in the era of AI, because now you've got this incredible tool set that's going to be able to reach into data applications or, you know, huge fields of complex data or massive amounts of documentation, pull all this insight out, insight out, and you need to figure out what you're going to do with that.

27:54How are you going to make the value of that? What are you going to combine that with to do something useful, right? And those, that's a skill set. And I think that's where, you know, where the kind of the job market's going to shift is employers are going to be looking increasingly for people who can think a little bit bigger, bigger picture and understand how they can leverage these tools to do the more sort of lower end mundane information finding, but actually turn that into value. How do you learn those skills, do you think? Well, that's a good question. I think they can be taught for sure.

28:22You know, there's, like I said, that's why going back to a question about the scientific method, that's one of the things that it sort of gives you. You learn studying science. Absolutely. I mean, there's quite a lot online now about communication skills. You can do courses that are very modestly priced yeah yeah that's right we've got lots actually on our courses page on read.co.uk so those so those are things that anyone can go and learn you don't need to be a physicist do you no not at all i just think by by you know i guess i'm lucky really i think i never really had a life plan i just always said hey look i'm really interested in science i'm going to do that but over the course of my career i've discovered that actually a career in physics has stood me in really, really good stead for these sort of adaptable skills that I can apply virtually anywhere, really.

29:05And essentially it's thinking, it's sort of teaching you critical thought. Again, philosophy, actually, I've got a few friends who've got degrees in philosophy and the old joke when we were in university about the applied STEM or the arts degree and so on and so forth. What was the joke? Oh, dear. I always remember there was a, maybe I shouldn't say this one on the podcast, but I always remember on the hand dryer in the toilet. in the physics department. It was someone had sketched with a pen, a permanent pen, arts degree pull here. And it was just the kind of paper towel. And obviously for the purposes of listeners, no.

29:41Well, you had the Dyson one for the scientists. Yeah, Dyson one, exactly. But I think, you know, I don't prescribe to that at all, but it's really... The truth in a church. Perhaps I was a bit jealous because my course was like 30 odd hours a week. And I'd look at my colleagues in philosophy and doing three hours and spending lots of time in the students' union. Yeah, but they paid off, didn't they? They did, yeah. So where are they now sort of thing, I don't know. But actually, they're laughing now because if you think about, you know, going back to the four Cs, right, and you think about this one around critical thinking, it doesn't get much more critical thinking than philosopher, right, than the kind of philosophy, the sort of art of a philosopher, right, the way they think, the way they approach problems.

30:18So when you're looking at AI, some of the people I've seen do really well with prompting, with getting AI to do what they want and using multiple tools, or now agents, as you probably would have heard of them on, you know, sort of agentic era. They're coming from more of those communicative disciplines where they're used to structuring an argument. Yes. Right? Because that's AI likes that. If you can structure an argument... AI likes a philosopher, doesn't it? It does. AI loves a philosopher. Because that's what I studied. Pleased to hear this. I'm home. You called it agentic, the agentic age.

30:49What's that? Is that a Microsoft word? Where's it come from? I think Satya Nadella coined that one. You're a CEO. Yeah, I think he did. He's very good at coining things. I have a huge respect for Satya. I come to some of your seminars and he's always so interesting. And I like the fact that he opens up to customers and tells us what he's thinking. Yeah, he's come up with Argentic. Yeah, he has. I've had the pleasure of meeting him a few times and he has a talismanic quality to him. I think he's a fantastic leader who's able to bring that clarity of business and direction, but also really champion the importance of a good corporate culture, you know, an organizational culture that underpins good performance.

31:26And I think Satya really embodies that. I think that's why he's been such a successful CEO and why, you know, why people, he has such a strong following, right? Microsoft has turned around really under his leadership, which I find very interesting. So he's written a book, hasn't he? Yes, he has. Yeah. Hit Refresh. Hit Refresh. Yeah. Well, we should all read that, I think. Yeah, definitely. Mandatory reading for everyone. It's a mandatory for you. Yeah, that's right. When you join Microsoft, you force to read it. You have to hit refresh. Okay, well, I'm going to read it because I want to. You've used the phrase agentic, the agentic age.

31:58This has sort of stuck in my mind. So this has been the holy grail, I guess, of AI researchers for decades, which is can we create a digital system that can take action on our behalf? You know, given some instruction, then go away and compete. But you're a holiday. Well, yeah, right. I mean, you can do that now, right? But can it do sophisticated things for us, you know, given some basic instructions and then go away and complete a task and come back. And we're starting to see that now with generative AI. And it's in part because you've got this new... When you say sophisticated things, what are you thinking?

32:29Well, it might be that it goes into, let's say in a business context, I don't know, ordering a laptop in your business, right? If you break that down, I think many people out there have been in a work environment where the laptop goes wrong, so you go to IT support. And there's probably many, many steps, right, in that process to get that laptop replaced. it doesn't bode well you're not thinking this is going to be a good experience yeah sometimes that can happen yeah not in microsoft i should not not really either but um but go to it support and so what will happen in this new agentic so you can imagine you know if you think about uh you know we we certainly at microsoft we see this scale of of agents okay so at the one end you have the sort of retrieval aspect of an agent so very simply this agent might know the IT help desk process really, really well.

33:14And so it's got all the documentation in there. And so you, the user say, my laptop's broken and you'd press enter. And the agent would look at that and assess that and then find out and tell you, just spit back some information. Okay, you need to do the following things. So that would be a very basic sort of query based agent, right? And it might use your personal circumstance to do that. Maybe say, you know, if you're a director or you work in a field engineer, it might give you a different process. then you might go on to uh sort of a little bit more sophisticated sort of a task-based agent so in that same example i gave there you say oh my my laptop's broken so what that agent might do is they're not just give you the process but it might reach out and pull the forms that you need to complete or it might complete them for you right and it might execute a set of tasks based on your context and get you some of the way there so that you know you can start to action this problem.

34:04And then you might go on to a much more sophisticated or these kind of more autonomous agents where it says you might put in a prompt sale, my laptop seems to have stopped working, I can't do the following things and press enter. And then that agent might go, okay, what's the problem here? Looks like there's a problem with the laptop, maybe I need to look at the laptop procedures, maybe it calls another agent that goes to the stock or the warehouse, for instance, or the ordering system, orders you a new one, right? Or maybe it triggers a help desk ticket for you or sort of solves that you know with you so it will go and execute it will assess by itself and execute a number of tasks you're seeing ai do some or all of this already then that's right yeah you know across organizations um one of the key things going back to what i was saying about that personal productivity and that business value a lot of organizations now are saying well hold on a second um this means that these agents are now really really closely aligned to our business processes.

34:56So what are the business processes that I can get the agents to do so that I can get my human operators really to focus on the more value-add parts of the work, right? Because nobody really wants to spend their whole day filling in spreadsheets or doing IT service test tickets. I hate doing forms. Yeah, everyone does, right? Nobody wants to fill in forms. But you like interacting with your staff and you like talking and solving people's problems. And they are more inherently human-based activities, whereas the whole filling in the documentation in doing the ordering, well, nobody really wants to do that.

35:27That could be infinite in its application. Right. Which is why I think there's just so much value to what's going on with AI right now. I love this quote, actually. I'm not sure if I can take credit for it because... Go on. Yeah, I will. I'm going to take credit. Yeah, take credit. What is it? Yeah. So being a physicist working in manufacturing, you know, and coming out of my sort of early part of my career, if you're aware of how manufacturing processes work, they have this thing called lean, right? So if you've got any manufacturer and you think they'll have some kind of lean or the kind of Six Sigma, if you're aware of that as well, right?

35:56Process optimization, removal of waste, optimizing the factory floor, making sure everything is in the right place at the right time. And this sort of came out of the 90s really, really big. Yeah, I remember studying it. Right, exactly. It was transformational. It was indeed, yeah. So you think about what Lean did to manufacturing in the 1990s. Process, you know, measure your process, put metrics in place, eliminate waste, drive through value, all that good stuff, right? What AI now, what Genitive AI and agents are doing, in my view, for the knowledge work is basically what Lean did for manufacturing, right?

36:28So if you think about knowledge work and professional services or auditing or tax or legal consultancy, a lot of that work is in the IP of humans, okay? It's tied up in insights, in tacit information, hard to measure sometimes in documentations, in advisory, you know, complex rules. And we never really had a tool set that could access that before, right? you know it was difficult you could do it but it was difficult but now with generative ai and its ability to go through a lot of that documentation to reason with you and you can converse with it suddenly you can start to put shape around that you know you can put metrics around that and i think the the exciting thing for me is ai is going to start doing a degree of lean and process optimization now in knowledge work you know in knowledge workers that seems to be happening but But I mean, when I was listening to you, I was thinking about manufacturing and how it's changed.

37:24Not many people work in manufacturing now. I mean, lean did result in lots of people leaving manufacturing. Is that something we should be concerned about now in knowledge work? I mean, all these people at home with their laptops doing stuff. And if what you say is correct, that's going to be very disruptive to the labour market, isn't it? I think it could be disruptive. But I think, I know, you know, you read in the press around, And we're looking for headcount reductions and job losses because of AI. I actually think it won't be all that bad. In fact, if anything, I think it will allow us to really humanise work.

37:58I think a business that really thinks about this in the right way is able to push the mundane and the not engaging aspects of those knowledge tasks. Some business leaders, they're probably not the most sort of empathetic. So we're going to go from a 15 ,000 person organisation to two or something like that. And you think, well, is that really the best way to motivate your team to embrace these new technologies? Perhaps not. Probably not right. So that's odd to me that people would sort of pronounce that. Maybe they're trying to get the share price up or something. I don't know. But then there is this experience that we've all had that with more technology, there seems to be more work, not less.

38:37You know, over the, certainly my career, lots of marvellous innovations. I'm busier than ever. And our business is busy. so that's our lived experience but this technology is different isn't it maybe it will be different i'd like to think that it will be different because um you know it all goes back to what i was saying about when you give people time back in a work context what do you what is your expectations of what they do with that time and this is why it's so important i think to do hand in hand the optimization you know using ai to optimize people's workflows but then that time used productively, right, in the right way.

39:13So is it opening up new business models? Is it changing the way you operate? Or is it an opportunity to say, excellent, we can give that time back to our employees. No more 14-hour days in some, you know, certainly some professional services, you know, long, long working days and burnout in many sectors. I mean, the legal sector especially suffers from this. You know, I think it's a real opportunity to change the way we think about work. And, you know, my personal hope actually for AI it diffuses more broadly through the economy is we actually humanize work. You know, going back to what you were saying around redistributing the workforce, as it were.

39:48Very few people's jobs are just one task. You know, AI is good at task based work. Absolutely fantastic. But there's still some tasks that it's not ideally suitable. That's what humans are for. Right. So what we're really looking at is how do we rejig it so that AI does the task that it's really good at and we rehumanize work. And I would love to see that, you know, people spending and investing a lot more time in the four Cs, building those collaborations, you know, those networks with people, spending more time with customers, spending more time in employee well-being, you know, spending more, being able to spend more time, perhaps your children, your family, because you've really implemented AI well.

40:25And you've been able to strike the balance between a good business model and a work-life balance. For me, that's like. That's a really positive message. That's really where we should be going. That's exciting. That's game changing. It's a life improving offer if we can make it. Right. Come through like that. Now I'm in recruitment and been in recruitment for a long time. And there's a lot going on in the sort of AI recruitment space. What are you seeing? I mean, how do you think it might change recruitment? I'm just being cheeky here, but I want to hear what you think. Because you're in the sharp end.

40:58It's a good question because I think a lot of organizations in your sector are starting to really think seriously about this technology and what it does. And actually, there are probably three broad areas, really, that it's really having an impact on. The first is the traditional back office processes. Invoicing, finance, sort of onboarding, those kind of things, right? Being able to sort of automate and speed up those processes, being able to use much more kind of predictive analytics. And that's really important. Those are the sort of generic. In fact, they're quite generic, right? Those kind of back office processes.

41:30That was the first place we introduced AI years ago. Yeah, exactly. Cash allocation function, I think it was. Yeah. It just went through the work like a dose of sorts. Absolutely right. And those sort of processes now become supercharged. So what's happened now is, you know, one of the things that Generative AI is doing in this sort of back office process is allowing people who don't necessarily have the sort of data science or engineering skills to work with, you know, large volumes of organisational data that you would have in sort of your ERP or enterprise tooling. You can now sort of push some of those tasks out to people who don't have that skill set.

42:00So you can put AI in between. And so people much closer to the work can ask questions of maybe organizational resourcing or costing, whatever it might be, and get that data, you know, and get that insight from the kind of source of truth, the organizational data, without having to task a separate team and get them to do that analytics for them, right? So that's great, right? Because that's going to just bring those data-driven insights of an organization straight to the people who need them, you know, in the front line. For recruitment, I think in particular, one of the things I think it's going to do is really help the consultant in terms of that time for profitability.

42:34Because as I understand, there's quite a high turnover in this industry as well. You know, it's a long time to sort of onboard a consultant. You know, they bring their network and the knowledge that they have in their area. And so then you're looking at, you know, your client base, you know, what's in the market right now? What are you trying to match in terms of roles and candidates? And so there's quite a long sort of onboarding process bringing a new consultant in. And then, of course, if they leave after a matter of months, that's quite a problem. So I think one of the things that the recruitment industry is going to gain from AI is reducing that time to profitability when you're onboarding a new consultant.

43:07They'll be able to access training, candidate databases quicker, you know, job spec creation, those kind of things. They'll be able to do a lot of that much, much faster and then be much, much more effective, you know, much quicker in the organization. Which would change the economics. Right. And going back to that business model thing I was talking about, right? So you're changing the personal productivity of the consultant. The role of the business leaders is to think about how that translates into adjusted business models for value. And that's the AI strategy bit. And then this final one, I think, is the client engagement, the candidate engagement.

43:38There's a piece of anecdote. I know many recruiters lament not being able to go back to all the unsuccessful candidates and tell them, hey, look, really sorry, you didn't make it this time. Or, you know, these were the things that perhaps, you know, would have been good to improve on. well AI is a great tool there for providing a much more personalized uh set of responses or engagement right you know you can you you would be able to sort of create those uh feedback mechanisms much more readily for the candidates you might be able to create personalized sort of journeys for candidates so that they can navigate a recruitment process much more smoothly you know and have that sort of AI because there's a lot of disappointment involved right yeah apply for jobs you don't get them you go for an interview you don't hear yeah we don't get the job yeah and we You want to place everyone, clearly.

44:21So maintaining good relations is very important. You could see AI supporting that. Yeah, I really could. I think, you know, the recruitment industry has actually got, there's some real innovation that could be made in this sector, right? You know, if you think, I like to think sort of blue sky. So, you know, let's imagine a world where everyone has their own kind of personal AI career tooling, right? So, you know, the recruitment industry will create a kind of this competitive world where you're offering to, you know, I pay, I don't know, a subscription a month or something like that. And I get this like personalized AI tooling that's guiding me through my career, advising what's coming up.

44:57What should I be thinking about? What skills? I mean, that'd be amazing, right? So there's real opportunity to create these personalized candidate and career experiences based on the knowledge and the experience that recruiters have in the market and job market. We're seeing people use AI to create their CVs, shall we say, and it's used for creating job descriptions. and then the job description goes online and someone applies. You get this situation where AI is talking to AI. Yeah. You're smiling. I am, because I've read a few of these perfect CVs recently where we put out a job spec and the spec, then the candidates replies, this is an outstanding CV.

45:33Yeah, and the AI thinks, wow, that's the perfect person. I mean, are we kidding here? I mean, maybe the AI is better at choosing people. I don't know, but it's sort of, I mean, it's quite easy to game it, isn't it? Well, this is really interesting. So this is like, you know, going back to things that I talked to boards about, this has been kind of the recruitment industry's number one pain point at the moment in that, you know, it's very, it's, you have to say one head, one step ahead of the game in terms of how you're doing candidate preselection and what you're looking for now and be quite creative in that because we have a screening business.

46:04Exactly. That's really busy. Yeah. Because more and more people want to have people double check. Yeah. And I think this is just one of the, you know, one of the things that happens when a technology like this hits the market. There are waves of disruption for a little while while everyone reconfigures. Okay, how are we going to - Updates their CV. Yeah, right. How are we going to deal with this? How are we going to think about that? We've got a perfect CV. But we used to say, I mean, I've written a book about CVs and it was considered fine to tailor your CV. So if you're going for a sales job, you'd emphasize your sales experience.

46:33Yeah, of course. If you're going for an admin job, you'll emphasise your admin. But now I think that people just say, this is the job I'm applying for. Chat, what should I say? Yeah, that's right. And then it comes back and it's sort of... And it's become difficult, right? And I think that's where the innovation needs to come from is, okay, how can you then use the technology to have a much more engaging interview process, right? How do you change the selection interview process with this technology, right? So really, for me, you know, the things that I would be looking for is how do you find, has this candidate got the four C's I'm looking for?

47:06You could get AI to check for that. Well, I don't know. Maybe you could. Maybe you couldn't. Creativity, I don't know. Right. Well, this is the innovation, I think, that the sector's got a huge opportunity to invest and think about, well, what does that look like? I'm not an expert. But some people are using it to screen people out as well, aren't they? They use AI to assess people on a video interview, ask set questions. Right. Looks like you're perspiring a bit or your eyes are dilating. No, you know, this is what I'm hearing. I don't know if it's true, but it frightens me that that might be happening.

47:38And I've heard similar things too. The sort of randomness and craziness of that. Yeah. It doesn't bode well, but you've heard similar things. I have, I have. What do you think about that? In other industries as well, you know, you hear people saying, well, I could use AI for this and, you know, I could use AI to, you know, make decisions about whether someone's eligible for this program or not. And then I always, my little internal alarm bell goes off at that point, because one of the things that's exceptionally important when you think about AI is how you approach it from an ethical and responsible point of view.

48:08You know, where will you use AI? Where won't you use AI? That's where you need your philosopher. That's right. Well, and that's what you need for an organisation. And I think it's, you know, going back to. So how does that work? How do people deal? How do companies address that? For your listeners, if there's one piece of advice I could give, whether you're an entrepreneur, you know, setting up a business or you're, you know, an established business, set about creating some form of ethics or governance board for AI so that you really have this multidisciplinary senior team of people whose role it is to dictate this is where we're going to use it.

48:40And these are the reasons why. And this is where we're not going to use it. You know, these use cases are deemed too sensitive or prohibited in our industry. And we just don't feel comfortable. and then build a degree, you know, a set of principles and a degree of transparency and accountability around that. That is, if you're embarking in the world of AI and you're dealing with people's personal and private data and you're making decisions about, you know, the kind of future prospects, you need something like that. Well, you need to be able to justify a decision, I suppose. Well, not just that, but I think people will expect, now society will expect organisations to operate with degrees of transparency and trust, and they will need to, you know, certainly at Microsoft, that's something we treat extremely seriously around how our six AI principles then turn into how we build.

49:23You have six AI principles. Yes, yeah. It would be helpful just to hear what they are, just in summary. So if other people are thinking how they want to ensure, protect, secure their business, what they might consider. Sure, yeah. So the six principles of AI principles at Microsoft, they came from our Ether Committee, which is AI ethics in Microsoft, how we think about what's the current state of the art is with AI and how we should think about it. And sort of back in 2017, 2018, when we began this journey, we said, look, if we're going to do this generative AI thing, or if we're going to do AI, and we're going to serve these products to the world, we need to make sure that we're doing in a responsible way.

49:59So our six principles are fairness, reliability and safety, privacy and security, inclusiveness, and transparency and accountability. So those are the six principles. And those guide everything we do. those principles are then translated into operational things that we do at Microsoft to make sure that the products and the services that we build, our software development, you know, how we partner, how we build solutions, we can make sure that we're adhering to those principles in how we want to bring AI to the world. And they're sort of constant. They haven't changed. Yes, those principles are constant.

50:31I think that's actually a good point. The principles have been constant. How we've operationalised them has shifted and changed in feedback and, you know, from our own learnings and also from the market. So we actually have something called the Responsible AI Standard that we had a first iteration of this in 2019 and where we talked about the principles in AI. And actually the market gave us some very good feedback and saying, look, you know, essentially these are great principles, but they're not very actionable. And so in 2022, we had a second release of the AI Standard. And what we did there was go a little bit deeper into those principles.

51:03So they break down into a number of goals that sit underneath each of those and some sort of guidance to help build some metrics and guidance for engineering teams. So, OK, if I'm looking at fairness and I'm building a system, how can I make sure that I'm aligning to that goal of fairness? I think when so much is changing, which it is at the moment, it's good to have some constants. It really helps people. Definitely. Yeah. Anchor to certain key things. Are you saying just so I understand that any company that's doing this at any scale should have a sort of policy of that sort with a board or sub board that oversees this?

51:36I think that's important. I think, yeah, you know, obviously, depending on what you're doing and what industry you're in, you know, will dictate the level at which you go to. But I think it's very, very important because because of the nature of generative AI and it's how how very powerful it is in creating these human like responses. you know there's a duty of care that if you're dealing with a if you're if you're a b2c company you're dealing with uh you know you know you're going out to the mass market you need to make sure that the people interacting with your product are clear where you're using ai you know they should know it's ai i think that's right yeah you know we we certainly we make that clear and also if you're using the microsoft products you'll see occasionally ai does occasionally make mistakes you know generative ai by its very nature is a probabilistic tool set so it does occasionally return.

52:21It does make mistakes. Yeah, people say hallucinate. I don't like the term hallucinate. I think people hallucinate. AI makes errors, but it will occasionally make mistakes. Why does it make mistakes? Oh my goodness. How long have we got? I don't know. I'm interested, but you're all the scientists. Why is it? Because it seems like it. It's almost making stuff up. If we look at the color-core technology, the transformer-based technology that things like GPT, for instance, generative pre-trained transformer, what it really does in in this sort of base form is it's kind of predicting the next word in in a sentence you know pretty much like autocomplete on your phone but it's doing so on on steroids right so it's doing that but it has a great deal of context about that sentence or that paragraph or the you know the say the prompt or the input that you've put in so it's just got a lot more context than your standard sort of autocomplete would but in at its heart what it's doing is it's it's predicting the likelihood of the next words or the next string of the sentence you know much more innovation has gone on since the sort of first of the transformer technology landed you know a couple of years ago and then now it does quite a lot it's actually quite complicated and a lot of the what we call reasoning at inference time so that how the models are able to make decisions around that that context when you ask it something but in in essence that's what it's doing so occasionally when you've got sort of subjects where it doesn't have a huge amount of data on or, you know, where it perhaps hasn't got that, let's call it statistical background of information to rely upon.

53:51Occasionally it will hallucinate, as they say, it will create the next word or next string of words or create references perhaps that aren't true. Now, that's the model in its sort of raw form. How can you check it? Right, exactly. So that's the model in its raw form. What you do, if you think about how you use generative AI in practice, you actually surround those systems with a huge number of safety systems so other sort of technical measures that you put around the model to check right so you know groundedness you know you know how how well rooted is this response in the kind of the user's query and you can look at sort of harmful or biased information as well you can do a lot behind the scenes to make sure that the output of that model is actually getting pretty consistent and safe and everything that said occasionally it does make errors but you can do quite a lot in the safety systems to cut that down.

54:39And that's where actually we're a tremendous amount of innovation at Microsoft is spent in doing that. Yeah. So the cost of AI, I mean, it started off pretty expensive, you know, to run these great computer systems. It's coming down, isn't it? Yeah, that's right. Yeah. So where do you see that? Yeah. But quite dramatically and quite surprisingly. Yeah. I mean, it's dropping, you know, massively. And I think, I think that's what it is. Where does that leave you at Microsoft? Because you've got this huge infrastructure, haven't you that support yeah and and we're improving that in you know infrastructure sort of every year absolutely i mean in the uk we've got massive you know multi-billion pound investments in building new data centers in the uk to make sure that we're providing that you know the right you know the sovereign capabilities in in the data centers in the uk to to serve the uk market you know high is that what people want they want a data center well it gives you that low latency and performance you can deploy the models closer where the people are Ultimately, you still have to move electrons over cables to provide this capability.

55:34You have to keep these centers nice and cool. Yes. Yeah, absolutely. So the UK is good for that. Well, you know, not as most. We could serve the whole world. It could be a new USP. Sadly, sadly, no. Oh, that's a shame. But you're right. Actually, talking about data center technology, there's a huge amount of innovation in that space. You know, people look at the sustainability of data centers. What I love about what we've done at Microsoft and how we've done that is the type of data centers you'll see will vary depending on their geography, right? So they will make use of the natural environment.

56:06You know, in the Nordics, for instance, we've got data centers that use geothermal power and then they use much more of a natural environment to cool. And obviously that's different in hotter climates where, you know, they haven't got that and they need different sort of cooling requirements. And there's a huge amount of innovation in terms of chip cooling and things and even using AI within the data center to optimize the chip usage as well. So there's a tremendous amount of innovation going into data centers. So this reduction in cost and the huge infinite number of applications, from what we're hearing, means that the agentic age is truly upon us, doesn't it?

56:39I mean, it's going to be huge. This is going to be huge. It's going to be massive. Yeah, I think it really is going to be massive. It's going to change everything, isn't it? Well, it'll change a lot of things. I mean, I think to a degree where we won't, it's very difficult to forecast exactly what... Impossible, I don't know. Yeah. I mean, when we started out on the World Wide Web, you know, in the early 90s, I mean, no one thought of social media, did they? Yeah. And the consequences of that are huge. So there'll be a lot of things that I guess no one's even thought of. I think that's right, yeah.

57:07And that's like with any, you know, AI is a general purpose technology. You know, what that means is it can be applied in, you know, many different industries in many, many different ways. and some of that's sort of unforeseen. But like many general purpose technologies, it creates new jobs and new roles that nobody had thought of before, right? So if you're a young person listening to our conversation, how would you advise them to sort of get to grips with AI? What sort of things should they be doing? Because it seems to me pretty obvious. I mean, you left a very good job at the Atomic Weapons Institute to get into AI, pivoted successfully, because you could see this is where the action is.

57:40I mean, I think people listening to this will be doing the same thing. This is where the action is going to be for, I'd have said decades to come, but certainly the next couple of decades. How do you get started? Because this is all new. Yeah, I think. And it's really interesting because maybe the answer to that a number of years ago would have been, you know, go straight into sort of AI development and code development. But actually, that might not necessarily be the answer. Yeah. And I'm not saying that we're not going to need coders, by the way. AI can write good code. AI can write code. Yeah.

58:13But you still need coders. You still need to know what good looks like. You know, just a quick aside on that. I am a friend who's an author. Obviously, author's getting very worried around, you know, around AI's capability to sort of write, write copy and text. And but you still need authors, right? You still need people who know what good writing is to make sure that we're still, you know, doing that. You know, that that role won't disappear. It will change, but it won't disappear. No, I'm just writing a book. Yeah, exactly. That's what you've got to have something to say. These things are important.

58:41I mean, look at what, you know, what sort of mass manufacture did over the centuries. We still have artisan furniture makers, right? You can make every piece of furniture with machinery. Not as many. Right, not as many. Things change. Not many will, right? Yeah, exactly. But a lot of these, a lot of the roles and the jobs will change. So the will could be a continuation. But I'm thinking, you know, if I want to get started, so I should be using these apps a lot. I should be thinking about how to interrogate them and learning by trying. Yeah. Going back to your question, I've got side trends in.

59:14The sort of skills that you really need to think about are how to use the technology, right? So unless you, I guess it depends. If you want to go into sort of AI coding program, absolutely. Those traditional routes are still absolutely right, completely valid. But if you're thinking more around the application of AI, then perhaps it's a bit different, right? And perhaps you need to be thinking more around, well, what do I know about particular industries or problems out there? You know, how do I get more immersed in the capabilities of this technology? So a lot of that will depend on you being much more experiential.

59:43So the number one thing that I say to anyone who perhaps hasn't ventured into AI is get hold of some tooling and start playing. Because it's the kind of skill set that is learned from experience. You know, you don't sort of sit down and read a manual about generative AI and then get good at using it. You have to use it to get good at using it. And that's because of the nature of the tooling. So if you're looking at, you know, you've got some great business ideas or you think that you can really disrupt your industry or create new value with this tool, then the best advice is to try and do that, right?

1:00:13Try and spend time playing with the technology. Look at maybe some of the solutions that are out there. Maybe look to create some new solutions with, again, it's lowered the barrier, right? The skills barrier. That's what it's done, you know, in terms of writing code. I mean, I certainly, as a physicist, I wish I had GitHub Copilot, you know, 10, 15 years ago when I was writing some code. I would have killed career. I wasn't a great coder as it was, but now it would have been so much easier to just to use the tool to accelerate my learning. So there's all these, you know, it's an amazing learning tool.

1:00:42People coming into the job market, what tools might you recommend for people who want to teach themselves a few things? Actually, at the moment, there's just so much out there, you know, that's actually free to access. On Microsoft's own pages, actually we have something called Microsoft Learn. We've had that for a very, very long time. And actually what people don't realise is that it's free. You know, you can get a lot of this content, and learning about our tools and products, but also we're generally about AI products. Additionally, as part of our sort of commitments in the UK and supporting the growth of the UK economy and the diffusion of AI, we had something called the Get On Skills programme.

1:01:17And we pledged to sort of skill over a million and a half people in AI skills over the last few years. And I think we've hit our target now, which is amazing. And we're continuing to do that. If you want to get on, get on skills. How do you do that? You just go online? If you go online and look up, yeah, exactly, you know, Get On or AI Skills Learning, you'll see a number of sort of free programs that are out there. There is actually a tremendous amount of free learning on AI out there. Even LinkedIn do some amazing courses as well that are free. There's a really good one for business leaders on, you know, it's two years in.

1:01:46And if you're embarrassed as a senior leader to say, hey, look, I haven't really been paying much attention to this. But there's a great grounding course on LinkedIn for specifically for leaders. It's about, I think, about four or five hours long. It's a free course. And yeah, there's so much great content out there to give you a good starting point for how to think about this technology. And I think that's the key part. Yeah. So one of our guests said, you know, that AI wasn't going to take your job, but someone who knew how to use it might. Yeah. And so it's about learning how to use it. And I think that's true.

1:02:15But again, you know, I like to, the way my mind works, I like to think by kind of big picture stuff. And if you look back over time and the diffusion of these general purpose technologies, that's always been true. you know whenever the new technology comes along do you think of it as a general purpose it is very much so yeah what are the general purpose technologies well you know electricity yeah right electricity you know the internet those kind of things you know very general way you can apply them in many different ways i think jeffrey ding has written a great book on this actually the diffusion of general purpose technologies over time it's a fantastic read uh you know i would like to do a recommendation but um that's good yeah try and get jeffrey yeah and it's a great it's a great read because it talks about this right you know one thing history can teach us is how these sorts of big technology inflection points affect society and the kind of things we need to do and and what we can learn from that right and it's the same yeah yeah well it's exciting to be living in this moment and you've certainly lifted my spirits around all of this and i think it's yeah the opportunities are vast so thanks so much for coming to talk to me today yeah pleasure thank you so much for having me thank you so much i'm going to ask you two questions though okay which i ask all my guests well and the first question um is what gets you up on a monday morning because we love so what gets you up on a monday morning well it's normally one of my young children that's a good answer but uh but no really what time do they get up oh varying yeah normally normally before 6am some of them um so they'll be entrepreneurs But no, what gets me up on a Monday morning is I feel very privileged to be in the position I'm in now to be able to work with business leaders and really clever technical leaders from all across the UK and to be able to help and steer and advise how they navigate AI in their industries as their experts in their industry.

1:04:09So that's an amazing privilege, I think. And I know I want to make the most of that to make sure that I'm adding as much value as I can in helping them do that. So that definitely gets me out of bed in the morning, knowing that, you know, a piece of advice or some guidance that I can give can help members of an industry change a business and do some great work. So that's, you know, that's pretty cool. Yeah. And the last question is, where do you see yourself in five years time? Oh, my goodness. This is literally interesting in this space. I think earlier I said, you know, I've never really had a life plan.

1:04:40I've always sort of just gone, oh, that's interesting. I'll do that. I have that kind of a mind. I'm sort of attracted to things I find interesting in the moment. And so never really had a life plan and sort of looked at things and said, oh, that's interesting. I'd like to be part of that. I don't actually know. I don't know where I'll be in five years. I'd like to think that I'd be able to, you know, learn from the experiences I've had now in this career pivot I've had in the last few years and then move on to something that cashes that in and provides more value somewhere else and hopefully doing something interesting.

1:05:08Yeah, I've got a feeling you'll be somewhere interesting in five years time. We might have to invite you back to find out where that is. Okay, let's do that. Let's put a date in the calendar. What the next five years. We'll get AI to all the newspapers. Thanks very much, Jed. It's a pleasure talking to you. Thank you. Thank you, Jed, for joining me on All About Business. I'm your host, James Reid, chairman and CEO of Reid, a family-run recruitment and philanthropy company. If you'd like to find out more about Reid, Jed, or Microsoft, all links are in the show notes. See you next time.

1:05:44Thank you.