Healthy Human-AI Partnership

0:00Welcome back to the Peggy Smedley Show, your voice for our connected world, with your host Peggy Smedley. Hello, all listeners and welcome back to the Peggy Smedley Show. I'm your host Peggy Smedley. Today we're diving straight into one of the biggest pressure cookers I think in tech. Data center delivery, demand is exploding, timelines are tightening and teams are wrestling with mountains, a fragmented product, but something big I think is shifting that I want us to talk about today. AI is finally giving us maybe the power to rebuild institutional memory, connect the dots I think across messy schedules and spot this integration risks long before they derail a project. And we hope that can happen. But is it sci-fi, is the hype over in some cases, yes, some cases, no, but it's human centered AI that helps real-time teams build faster projects, project kind of the delivery

1:06data and kind of save tens of millions by making information actually work for them. But they've got to do it right in the right way. So if you're in the business of powering this conversation, I think it's really going to hit home. We have real-world data and digital, I would say digital information projects. So I want to welcome today, Ansar, who's the co-founder and executive chairman at Forsight Works and a professor of medical project management at the Syed Business School, the University of Oxford and at the founding director of Oxford University's program on sustainable capital intensive industries. So a chief, you've got a long president been doing. It's pretty impressive. So welcome to the show. Pleasure to be here. So in what you've been doing and trying to say all the many things, it's impressive without

2:09a doubt. You've been studying, I think, mega projects. You've been studying what are digital transformation, I think, in a big way. You've spoken. And this is what I want to really dive into. You've spoken a lot about AI needing to work with humans rather than replacing them. Now in practice, let's be honest, what needs and what does happen are two different things than a lot of companies. But what does a healthy human AI partnership look like in complex project environments? Great question. So look, there's a great degree of fear around AI and what it might do. In replacing human jobs, but what I would like to convey to my listeners is they should not worry. Here up to you, though, we've just heard Amazon again laid off all these people. So when we talk about that, if we hear these large corporations laying off people and those outside AI, it's just normally happened.

3:09How do you say there's not fear when one of the largest and most profitable companies in the world keeps laying off people? Now, I'm throwing that back at you. The data center construction volume in the US has multiplied nearly 10 to 20 times since 2020. So you know, back in 2020, there were about 500 megawatts worth of install capacity of data centers being built. It's anywhere between 5,000 to 10,000 megawatts being built in the US right now. Now supply chains have an increase 10 to 20 times. Labour pool has an increase 10 to 20 times. It's exactly where it was more or less five years ago. And if you went out today and you wanted to hire a planner or a design engineer that could build your data center, there aren't any available. We just can't train enough human beings quickly enough in order to meet the demand. The only way you can build those schedules, the only way you can do that design is by using current people that are very good at what they're doing and give them tools like AI for them to speed up their current workflows.

4:12So where the human and AI work best together is where AI does the computation and the humans provide the context and the prioritization. Now that degree of automation is inevitable, but it's actually good for people not to stand in 40 Celsius heat, diverting traffic. It's not a good use of human time. So some degree of job displacements are inevitable, but in generally technology enables better jobs that enable human creativity, particularly when used well together. I'm not arguing the point with you that job displacement is a problem. I'm arguing or maybe I'm challenging saying, are we pushing AI so fast, so furiously, that we have an upskilled or re-skilled the personnel to meet the demand that's required? It's a great question. So I think AI is a big hype at the moment, but I do think the reality of it is that it's scaling rapidly, but in not in a way that's going to blaze through the entire economy in

5:13such a fast manner that everybody's going to be up in the arms about it. So I think even in simple workflows like this podcast, I'm sure your team use video editing tools that are enabled by AI that save them a huge amount of time. And I think those are like very simple ways in which AI is helping people today. And I think that alone is a massive technological leap, but it's not a massive economic disruption obviously any listener out there that is not using AI, you do need to experiment with these tools ASAP because Donald Pink famously said AI will not replace jobs as people who use AI will replace people who don't. Are many organizations just rushing to adopt as you just said AI tools yet they still struggle with I think fragmented data? We know that there's a lot of data and we all say data data. That's was the new oil, right? We talked about it all the time. But why is institutional memory still critical? And often I think overlook and we look even more with saying foundational for effective AI.

6:17I think that's what's missing. And I think I'm going to go back to my initial comment about upskilling and re-skilling properly to even look because I don't think all this data is going to be so critical. We talk about we need good data, but are we training AI back to that point that makes it effective for foundational work? Peggy, you're putting a thing on very important issues. So I think the reason for institutional amnesia comes from, it's just simply, comes from human biases, right? And these are individual level biases at a psychological level that then translate into social biases at the organizational level. So we know from research that one in the bad price for Daniel Kahneman in 2002 that people, all human beings, a respective or cultural background, suffer from cognitive biases such as overconfidence or the ostrich effect where when we observe risk, we try to put our head in the sand rather than deal within a front, political bias.

7:18So on so forth. So all of that means that human ability and even willingness to deal with data in a cold-hearted impartial manner is fairly limited. Machines in that sense have an edge over us because they deal with data very impartially. So a lot of the fragmentation of that data tends to be because it's uncomfortable for people to remember that data inside organizations. So for previous project failed, it's much more convenient for people involved to bury that information and not recall it than to constantly be reminded of it. There was a famous professor at Stanford University in March. He used to call it the failure myopia. So we suffer from these myopias like we don't like learning from distant geographies. So projects built in Australia, people keep asking, oh, just show me data from Australia. I don't want to learn from the data in the US. But if you're building a road in Australia, it's just as relevant to look at data from India or from Nigeria or from the US. But people find that very difficult. Similarly, Jim March talked about time, myopia, where things that happened 10 years ago,

8:22we begin to believe are no longer relevant. Psychologists call that recency bias. So we overly anchor into what happened yesterday, then what happened 10 days ago, 10 years ago. So all of these are human barriers to data being used. So that fragmentation in institutions is as much a technical database problem as it is a human bias problem. Where AI can help is both at a technical level and a human bias level. Technically, because of AI agents, we can now stitch together these fragmented data in far better ways than was previously possible through very deterministic systems. It's just made it a lot easier and a lot less expensive. At a human level, AI can provide deep bias information. So for example, just to take a completely different example, I really like this tennis flag called John Xenai. It's a young tennis player, I'm a skier, he's a skier from Italy. He lost to Novak Djokovic the other day in Australian Open

9:22for anybody who follows tennis among the listeners. And I was quite upset about it. And I just asked Chachibiti to give me historical data on across various dimensions of where different tennis players like Djokovic, Federer, Nadal or John Xenai Rank and turns out Djokovic is hands down the best empirical player that's ever been. So he is the goat. And once I learned that I could overcome my bias and be and hats off to Djokovic for being the best player in his break-point discipline in his serves, in his win matches, so on and so forth. So I think that's the kind of de-biasing that AI can bring that's very hard for human beings in the absence of AI. First of all, I'm still Roger Federer fan, so I'm just saying he's much more graceful than Djokovic, but well, that's a conversation for another day. But I guess what bothers me here in this conversation a little bit and what concerns me is you talk about these cognitive biases. It then leads me to think if we're not looking at all the data

10:23and we're training AI, if I look 100 years from now, even 50 years from now, what are we really providing the information for future generations to understand? And that's what I think we have to look at. You often emphasize AI is only good as it's training and the usage, because what's the most common mistakes we see leaders make when deploying these large-scale models or these large-scale projects, we could be building the weakest buildings, the most unsecure buildings, and that catastrophically fail, because we're not looking at the best data. We're looking at inferior data, because we're not looking at all the data we need globally. And knowledge is power, and we're not getting the best knowledge, because we're not, because now you just said these biases, this myopic kind of view of things, I guess that's my question. I think your question is spot-on, Peggy, and I don't want to disagree with you,

11:24because I think what you're saying is asking for a much wiser use of AI, and these biases, we've known this from hiring markets, for example, can be inherited by the algorithms as well. A skeptical view of all data, whether human or AI generated, is a useful default stance by people. That's a wise strategy, is be a very skeptical consumer of any information from any source, in general, and it's the philosophical entry point in any of this. Having said that, I think it expert people tend to be better users of AI than complete novices, and I think that's when novices can end up really hurting themselves. So give an example, if you're a writer, and you know how to write well, you can use models like Claude to improve the speed and even quality of writing far better than if you don't know the basics of rhetoric or basics of structuring an argument. So in a way, people should think about AI tools as an augmentation. It does help to be skeptical.

12:26So again, it does take an ability. Kahneman called it the inside versus the outside view. So often these biases emerge, because people make a movie reel of the future in their inside mind. So if I'm building a project, I might think this time is different. I'll get this project done in 10 months, which previously took 20 months. Where AI and big data sets become useful is just reminding us to ask for that outside view. And again, I think experts, people with a degree of knowledge will outperform when using AI will outperform people who are complete novices and just end up making a hash of it, hope everything. So that now leads to a much bigger problem in society. Are we going to have the halves and halves? And a bigger chasm of who has the best knowledge in society? Those who really understand AI will be smarter. So society is going to change dramatically

13:27because those who are able to learn, know how to use it. So we already talk about how, because of COVID, we have students that aren't reading, that are behind here in the United States, in other countries, as you just described, novices. When those who are able to teach themselves learn are going to be smarter, are going to be able to be more efficient, self-sufficient. So I look at this, the experience of being able to work, know how to work with AI, is going to lead to much more societal problems than we already have. So I think we could think of it in two scenarios. So scenario one, which is a positive scenario, is that AI becomes much more democratic. People with knowledge and who are ideally based on a good education system are able to deploy these tools to have fulfilling careers with much more career freedom. Even basic consumer tools like Lovable, that is allowing 14, 15, 16-year-old children,

14:27effectively to build million dollar businesses for the first time by literally just taking an idea and using natural language to start building a website, start selling something and being very entrepreneurial. And we're seeing that sort of happen in middle-class Swedish armories. There's something I was reading about, very young people doing very successfully. So there's one vision of the future where these technologies become incredibly democratic, but they do rely on, as an assumption, some social infrastructure, a good education system, for example, the other scenario for what you paint, although a depressing one, is not completely unlikely. So it's something that we have to work together to prevent, which is that the benefits of AI, which would be very large, are captured by just a few extremely successful people and do create this massive network effect of some incredibly successful entrepreneurs and others who are not so successful. I don't buy, that's the likely outcome partly,

15:28because I do think that the knowledge about AI is now quite pervasive. Even a general person on the street is using trap GPT, is using tools like Lovable. They're familiar with it. They're using, even for simple use cases, like doing search for restaurants, or booking, finding courses, et cetera. And I also have enough faith in people's own ability to look after their best interest in terms of understanding that this is a competitive industry, the competition is sharpening. So I have enough faith that the market mechanism works, that as people see their peers making lots of money, because they've invented a new app, using Lovable, or using trap GPT, or using a tool like Foreside, they're delivering projects way faster, that they're gonna keep up with the Joneses, and learn very quickly what they need to learn to compete. I think I'm not trying to paint a negative picture. I'm trying to look a little bit more realistic than not everybody's entrepreneurial. The other side of this, the flip side,

16:29is culturally businesses going to recognize that they're going to help constantly educate their employees. It's not just going to be left to the employee to find the time to constantly educate themselves. Society's going to have to change, that they're constantly going to have to help educate, to give them the time to constantly want to educate. So I think that upskilling is not gonna have to be once in a while, it's going to have to be constant. So I think it's a kind of different type of world we're gonna have to see than we're used to in the past for businesses, to have to constantly invest in their people. Absolutely, look, I'm an educated at heart. So in fact, even the AI tools I'm building, the point is to educate project managers to confront their own biases and do better. It's not to misjudge their own scalar on project management, because they're dealing with a very hot contextual environment, but to give them the cold calculations they're on the AI can perform to help them bridge their gaps.

17:30So I completely agree with you that the upskilling is the right path. And I think in that technology tools are very useful, but obviously morally ambivalent, right? Like you can use YouTube to learn, to code, or you can waste your time doomscrolling. It's really ultimately your choice as an individual and requires huge amount of discipline to be able to do that. So I do think that all of that has constantly got to be, we have to be reminded that they have this choice and it's easy to do doomscroll rather than to learn. But we see very inspiring examples of people who teach themselves great skills. So from your experience, how does standardizing what works and changing the way teams plan and deliver, let's say major infrastructure or digital projects? Absolutely. So I think part of standardization is repeatability. So human beings, every time we do something slightly differently, it can introduce bias again.

18:31It's an opportunity for bias when we view of tracks. So standardization is not a GANS innovation. It's actually a way of encoding innovation. Again, to give an example from a different context, a friend of mine worked at a top flight restaurant in London. And one of the things he learned was that restaurants are very funny businesses because you have to innovate a lot in order to get your finish line stars. But at the same time, every single meal has to be exactly identical. So it's this very strange paradox between delivering the same outcome again and again, but a very rigorous process of innovation and discipline that goes before that. Projects are the same. You need a period of tinkering and then you need a period of standardization type of thing. So I think people should embrace standardization and that's standardization in design, in big data centers, for example, trying to fit sites to design and the other way around. Or figuring out designs are much more modular in nature. Computing industry did that extremely well back in the 60s.

19:32Tom Watson had built a very large set of, actually data centers back in 1950s with MIT for a US defense project called Sage. And those computers never worked. And around 1960, he basically locked up his entire team in a warehouse in upstate New York. And he said, look guys, we're not coming out until we've redesigned the computer. And these guys and gals were locked out for nearly five years. And out of it came the system 360, which was essentially an ancestor all the modern day computers, including your Apple Watch that you might be wearing or something like that today. So I think that level of modularity hasn't happened to the built environment, that the grip standardization hasn't happened yet. But it's necessary to create both much less variance around algorithms and provide much more flexibility. So now we can assemble very complex computers by using the same basic components. That sort of thing needs to happen to the data centers, which are effectively very large computers at the top of the value chain. I think when we look at design and project delivery

20:34and we think of using such tools as 4D, 5D, 60, as we talk about in, and where AI comes in. And the built environments, the ability to share information that they've never did in tear down those silos, changes in an entirely different way of taking our world somewhere completely different that they never thought about. I think it's truly exciting for that built environment. That's right Peggy. Give you one example, a two-year data center project, it takes six months to baseline the original schedule. So if you're trying to build a schedule, it usually takes three human planners and simply the planning. With an AI tool like four-site, we have brought that down to three hours. And the idea that you can use that to then share out on the next project to be able to see something that maybe you don't have to reinvent the wheel is exciting. So I guess where I want to leave, because we've got a few minutes left, but what should leaders start thinking about

21:35when they look at data as a long-term strategic asset? And we think about this, then as a byproduct of delivery, or the bigger things that we're talking about now, we just talk about, where I get excited about the digital twin in the built environment, we get excited about the ability to replicate save time as you just said, because it takes so much time. And we're now thinking about data centers everywhere, we're thinking about how this is, where we're going to change our world. But we also have to think about what's next, not just about what's now, but what's next. Some people try to wait it out, like you're basically losing very much. Those are diners, yeah. So first of all, take a plunge and start somewhere. Ideally, you would do something big from the beginning. So being overly incremental and just keep testing the waters is also not great, but at some point you do have to take a plunge. And you can still create some guard rails. So I'm totally cool with guard rails, but blocking any form of innovation in the name of data confidentiality

22:35or mistrust or fear is ultimately hurting the very foundation and the businesses and can even cause them to go extinct because other players are moving really quickly. So I think the first thing is, take the plunge. Second thing is know that everybody has biases, the individual level and at the institutional level. And the only way to correct those biases is by looking at the data. So this very simple thing, if you've built 50 data centers in the past 10 years, down just the rest on your laurels the fact that you've been doing this for 10 years, collect the historical data and ask us a simple question, how long did we think this particular project was going to last? How long did it actually last? And even structuring a very simple Excel spreadsheet will reveal some really mind-blowing information about the magnitude and frequency of delay, about where that delay came from. And don't use that to create a blame game, but use that to really learn from historic projects

23:37to create that learning loop of where this took place. Emotionally, that's way easier for AI because they don't bring emotion to it than human beings, who inevitably want to defend their past record, but somebody has to do that work inside the organization of learning from the past. This is known as referenceless forecasting. So create a referenceless forecast of your past projects. And from that, many decisions will emerge like clear as day in terms of where time is lost, for example, in the permitting process or in the design stages in the scheduling process. And from that evidence, the path emergence emerges very clearly, but you just have to put in place a degree of human discipline on top of that AI discipline. We typically call that the three lines of defense. So you need a governance model that goes along with the AI. And that basically means you have to separate people who do the work for people who review the work, from people who approve the work. So those are the three lines. And generally in construction, people who do the work mark their own homework.

24:38And that can become quite complicated. And so it's not to be unfriendly to people doing the work. They have a noble role in this, but making sure that there's at least three layers of governance of people doing work, reviewing work, and approving work, back by the same source of data. The chief, is there anything that you've seen from doing this, that you've seen some shocking, that really surprised you, time savings or cost savings, that you say, if you've seen an organization, really say, look, this really surprised them that they didn't expect. Ultimately, life boils down to dog discipline. And so if you're trying to get fit, like there's very few things you can do apart from get your sleep, get your nutrition, get your training right. And then do it every single day. So similarly with AI, what it points to is multiple moves you need to get right in order, and which all pointed dog at discipline. They don't point to any single magical, one moment that makes the problem go away. And what we've typically seen in projects that delay are, people try and blame it on one issue.

25:41All those bus bars arrived late and they were incorrect. And as a result, the whole project was delayed. Unfortunately, it's not just the bus bars, although they played a role. It's also the fact that design reviews were not done on time. And as a result, the POs for the long lead items were not issued on time. And the permits were not received on time. So there's multiple moves that provide the coherence that the AI reveals. And it allows people to prioritize. One prioritization is completely free. It doesn't require applying any additional resource to it. What if there's one big magic bullet that we found again and again in this, that in high pressure environments, the teams start to work in a headless chicken mode where they are working out of priority. And actually, they're making a lot of moves. They're wasting a lot of their energy and morale. And if you could just use a system like an AI, foresight focuses their attention on what matters most in that day, in that week. And just deliver that with great dog at discipline.

26:42Suddenly, the flow improves tremendously and they can get their projects done much earlier than they anticipated. Discipline's so much the key in the build environment. Absolutely. Thank you to keep my approach straight all the time today. Thank you, Peggy. It's great to be here with you. OK, listen, that's all the time we have for this segment. Make sure to share and subscribe to our episodes each week. Share your thoughts with me on XX Connected World or follow me on LinkedIn and YouTube and continue the conversation there. This is the Peggy Smeadly Show. Your voice is for our Connected World. And remember, great technology problems, great responsibility.