WEBVTT 00:01:30.000 --> 00:01:39.440 back. Hey, hey, hey. 00:02:34.560 --> 00:02:40.800 Hello everyone and welcome to today's LinkedIn live conversation. how to grow alongside AI in 00:02:40.800 --> 00:02:44.960 the workplace and make it work for you. I'm Whitney Pennington Rogers. I'm a curator here 00:02:44.960 --> 00:02:49.920 at TED and I am thrilled to be your host. Uh now TED relies on the generous funding 00:02:49.920 --> 00:02:54.400 of our partners and this interview episode is made possible thanks to the support of Pricewaterhouse 00:02:54.400 --> 00:02:59.840 Coopers and today I'm thrilled to be joined by Price Waterhouse Cooper director of 00:02:59.840 --> 00:03:05.920 digital assurance and transparency Alana Golden Blumenfeld. Hi Alana. Hi Whitney. Thank you so 00:03:05.920 --> 00:03:10.400 much for having me here today. Thank you for being with us. Now, before we get started, 00:03:10.400 --> 00:03:14.720 I'd like to encourage everyone watching to submit questions for Alana in the chat. You can do so now 00:03:14.720 --> 00:03:18.480 and throughout the conversation and our team will be collecting them on the back end. Um, 00:03:18.480 --> 00:03:23.920 and we'll share a few of your questions uh towards the end of today's conversation with Alana. So, 00:03:23.920 --> 00:03:29.680 Alana, before we dive into this uh very exciting conversation, can you first start by telling us 00:03:29.680 --> 00:03:36.000 a little bit about about your work in the AI space and and where you focus your attention? Certainly, 00:03:36.000 --> 00:03:41.760 Whitney. I'm I'm part of um a wide and incredible team. We do a lot of work 00:03:41.760 --> 00:03:46.880 related to the responsible deployment of AI. So I'm a data scientist by training and a number of 00:03:46.880 --> 00:03:52.880 professionals I work with are data scientists or or other AI um experienced individuals and we've 00:03:52.880 --> 00:03:58.800 been building and deploying AI for a number of years and thinking in importantly about what it 00:03:58.800 --> 00:04:04.560 means to make AI work for us as organizations as people the controls the governance related 00:04:04.560 --> 00:04:09.680 to those the training and upskilling that we need to to take advantage of these systems effectively 00:04:09.680 --> 00:04:13.520 um the risks that we need to be managing toward and and other practices that help 00:04:13.520 --> 00:04:18.240 us gain confidence and trust um in in AI so that it really becomes a sustainable 00:04:18.240 --> 00:04:24.320 part of our our day-to-day and and not just a fun curiosity that we play with every so often which 00:04:24.320 --> 00:04:29.440 is really important and artificial intelligence is definitely something that we've been hearing 00:04:29.440 --> 00:04:35.200 um more and more over recent years what would you say are the defining qualities of this moment in 00:04:35.200 --> 00:04:39.280 the advancement of artificial intelligence and and how would you compare it where we are now 00:04:39.280 --> 00:04:45.760 to where we were say a year ago a year ago I I think most organizations we were working with 00:04:45.760 --> 00:04:51.520 were still in early phases of exploration. I think the business case for adopting AI was 00:04:51.520 --> 00:04:56.400 certainly very obvious. Um especially because of the messaging that we were all seeing externally 00:04:56.400 --> 00:05:01.680 about uh the need to be to be adopting AI and how it was going to be changing how we work, 00:05:01.680 --> 00:05:06.400 what we do at work, every everything about our day-to-day interactions. Uh but most organizations 00:05:06.400 --> 00:05:10.720 were still experimenting. They're trying to figure out what the technology really is, 00:05:10.720 --> 00:05:15.360 especially because it isn't just one technology. it's a a mass of different types of technologies 00:05:15.360 --> 00:05:20.880 that are are implemented in different ways for different purposes. So that experimentation was 00:05:20.880 --> 00:05:26.480 really important to help establish uh what the ultimate strategy needs to be for organizations. 00:05:26.480 --> 00:05:31.840 And I think that's the phase we're in right now is pivoting toward a more concrete strategy for 00:05:31.840 --> 00:05:36.880 the deployment and adoption of AI, building the programs, the governance around them, and really 00:05:36.880 --> 00:05:41.760 making meaningful investments and trying to scale those investments. So not just having proof of 00:05:41.760 --> 00:05:47.520 concepts but actually building um building and institutionalizing solutions and capabilities 00:05:47.520 --> 00:05:55.440 that are are rethinking how we do various forms of of work today. And as we're having this AI 00:05:55.440 --> 00:06:02.720 conversation currently, you're you're hearing this term agentic AI a lot. What what is agentic AI? 00:06:02.720 --> 00:06:06.880 The term agent's been around for a long time and it's it's been modifying over over the 00:06:06.880 --> 00:06:13.440 years. In the current form of of AI today, when people say agentic or or they refer to an agent, 00:06:13.440 --> 00:06:18.960 what they mean is an um a small AI system that's been designed to perform a specific task with a 00:06:18.960 --> 00:06:25.840 higher degree of autonomy. And it's not just a query to or a prompt to a large language model. 00:06:25.840 --> 00:06:31.200 It could be actually executing a research task or asking some a system to go out and fetch certain 00:06:31.200 --> 00:06:36.480 information. The the difference here being that an agent is provided a goal and it has a variety 00:06:36.480 --> 00:06:42.000 of skills or capabilities it can use to try and fill that goal. So, it's a lot more flexible than 00:06:42.000 --> 00:06:47.520 some of the other uh maybe more task automation types of capabilities that we've seen over over 00:06:47.520 --> 00:06:53.840 the last 10 to 15 years. So, in some ways, an agent is essentially being tasked with acting 00:06:53.840 --> 00:07:00.400 on your behalf in ways that we've we've not seen before. I I would agree with that. Yeah. Um well, 00:07:00.400 --> 00:07:03.920 you know, I think there it seems like there's a lot of opportunities when you think about this 00:07:03.920 --> 00:07:10.640 like the ability to have AI do things for you that you would normally do for yourself. Um but there 00:07:10.640 --> 00:07:15.920 it also seems to present a lot of challenges when you're in entrusting this AI to make decisions for 00:07:15.920 --> 00:07:21.520 you. Uh so what are some of the challenges uh that we could face if we're not cautious with 00:07:21.520 --> 00:07:28.720 agentic AI and what role can all of us play in in creating responsible AI? Some of the value 00:07:28.720 --> 00:07:34.960 of agentic systems today are the the fact that they they're designed to work to a high degree 00:07:34.960 --> 00:07:40.480 of permissibility that they can make a lot of decisions effectively for themselves on how they 00:07:40.480 --> 00:07:44.720 complete a certain task. That's where a lot of the power comes in. That's also where some of the risk 00:07:44.720 --> 00:07:50.240 does start to come in because we need to be sure that we're allowing those agents to have access to 00:07:50.240 --> 00:07:54.560 data that they're allowed to have access to that they're not seeking out information or otherwise 00:07:54.560 --> 00:07:58.960 incorporating information that we don't expect them to to utilize within those workflows. We 00:07:58.960 --> 00:08:03.440 have to design that appropriately. So that could be like closing an agent off from from leaving 00:08:03.440 --> 00:08:09.280 the premises of your your environment. Um or uh restricting certain types of data sources so that 00:08:09.280 --> 00:08:15.680 it doesn't incorporate proprietary or sensitive information. Um there's also a degree of of of 00:08:15.680 --> 00:08:20.720 dynamicism associated with these things. Like we want them to be uh trying out different strategies 00:08:20.720 --> 00:08:25.520 until they can complete a task, but sometimes they might do things that we don't expect in the the 00:08:25.520 --> 00:08:30.320 in the service of trying to achieve that goal. All the agent knows is that it's been told to 00:08:30.320 --> 00:08:35.120 to to try and achieve a certain goal. So if we don't prescribe the constraints appropriately, 00:08:35.120 --> 00:08:39.920 it might do something that's somewhat we don't necessarily know how that'll come back on us. 00:08:39.920 --> 00:08:45.280 Plus the same risk around um large language models or other types of generative AI systems 00:08:45.280 --> 00:08:51.520 fabricating information to complete a task still exists. So again that comes into how 00:08:51.520 --> 00:08:56.800 do we define the procedures for a certain agent and what tasks it's supposed to do, 00:08:56.800 --> 00:09:01.200 what evidence it provides that it's doing that effectively. And really at the end of the day, 00:09:01.200 --> 00:09:06.560 we should be thinking of these agents as something that's accreative and supportive, augmentative to 00:09:06.560 --> 00:09:11.840 our day-to-day work, not replacing our us in that um in that day-to-day interaction. So, 00:09:11.840 --> 00:09:17.200 we are still ultimately in the driver's seat. How do we think about human and agent collaboration 00:09:17.200 --> 00:09:22.960 as an important um design mechanism for how we implement these types of systems? So, I I wouldn't 00:09:22.960 --> 00:09:27.440 think of them as being pure automations that you can just let go and and do whatever they want. we 00:09:27.440 --> 00:09:31.360 have to really think about how we're working with them and I think that that can offer a disconnect 00:09:31.360 --> 00:09:38.640 if we don't design that appropriately. Well, it it sounds like the pres the presence of a agentic AI 00:09:38.640 --> 00:09:45.040 uh sort of puts us closer to these science fiction dreams or nightmares depending on how how you look 00:09:45.040 --> 00:09:49.840 at it of of what AI can accomplish and and help us accomplish. And a part of the narrative that's 00:09:49.840 --> 00:09:55.120 being put forward by some experts um in the AI space is that AI is developing so rapidly that 00:09:55.120 --> 00:10:01.200 we're quickly approaching the point where AI might be outpacing human intelligence in potentially 00:10:01.200 --> 00:10:08.320 every way. Uh what are your thoughts on that? It's a it's definitely an interesting topic area and 00:10:08.320 --> 00:10:14.880 I think I'm a huge uh lover of science fiction so and some part of me also is looking into that type 00:10:14.880 --> 00:10:21.680 of a future and seeing where we're going to go. uh but from a real like pragmatic perspective uh 00:10:21.680 --> 00:10:26.560 AI is still fundamentally technology that we are using. People are still very much in the driver's 00:10:26.560 --> 00:10:31.920 seat here and we still need to think about what we mean when we define intelligence. There are very 00:10:31.920 --> 00:10:36.480 few common definitions of what we mean when we say intelligence and and there's a significant amount 00:10:36.480 --> 00:10:42.320 of debate and I think in some ways that debate is is distracting because what we're really looking 00:10:42.320 --> 00:10:47.520 at is how can we use technology to make our day-to-day better not how do we build technology 00:10:47.520 --> 00:10:53.280 to replace people and automate everything and ultimately change the nature of of our day-to-day. 00:10:53.280 --> 00:10:58.320 Um the vast majority of systems that we're going to be seeing are going to be embedded into our 00:10:58.320 --> 00:11:02.960 existing workflows. We are still involved. we are still defining the practices and we have 00:11:02.960 --> 00:11:08.880 ultimate responsibility and oversight for how those are used and deployed. Um so I' I' I guess 00:11:08.880 --> 00:11:15.040 the the realist in me would push people to think a bit more realistically about what's happening 00:11:15.040 --> 00:11:20.320 today and identify the issues and opportunities for enhancement that we have in our processes 00:11:20.320 --> 00:11:26.240 today rather than being fearful of a potential future where everything will be automated away. 00:11:27.200 --> 00:11:30.240 Well, let's talk a little bit about some of those opportunities because I know people are 00:11:30.240 --> 00:11:34.640 probably eager to figure out how they can act on some of the things that you're sharing here. So, 00:11:34.640 --> 00:11:39.280 how how do we ensure that we're actually growing alongside AI and using it to our benefit and not 00:11:39.280 --> 00:11:45.120 peril? Well, first and foremost, you have to be using it. If you're not if you're fear so fearful 00:11:45.120 --> 00:11:49.440 that you're not touching the technology at all, I think we're really missing missing the boat on on 00:11:49.440 --> 00:11:54.000 incorporating or identifying really testing how it can and can't work for us. There are variety 00:11:54.000 --> 00:11:59.120 of tools available, many of which have been made available across um pretty much every single 00:11:59.120 --> 00:12:05.040 organization we work with today. Uh basic tooling that you can just start to to build little custom 00:12:05.040 --> 00:12:10.480 uh agents to support your your tasks. Just experiment. Uh but experiment with it within 00:12:10.480 --> 00:12:15.440 controls. So don't build your first agent as something that's a really high critical task. 00:12:15.440 --> 00:12:20.960 Build it as a much smaller task that you have um oversight of end to end. Build that comfort. see 00:12:20.960 --> 00:12:25.440 where it works, where it doesn't work, and build kind of the internal rules and structures for how 00:12:25.440 --> 00:12:30.640 you engage with that. This is still a very new technology, and there are very few people that 00:12:30.640 --> 00:12:35.520 really are true experts in how to utilize it effectively and a lot of that is because the 00:12:35.520 --> 00:12:41.200 pace of change and the technology space is moving so quickly and are understanding of the use cases 00:12:41.200 --> 00:12:47.120 and possible applications for it are also moving very quickly. So you as an expert or domain expert 00:12:47.120 --> 00:12:53.120 in your particular space um can build a a a sense of understanding of how the technology can be 00:12:53.120 --> 00:12:57.360 useful to you if you're actually experimenting and playing with it. And then I think if you can 00:12:57.360 --> 00:13:03.360 have effective use cases for it, that's when it's it's helpful to understand what type of program 00:13:03.360 --> 00:13:07.920 is your organization moving toward. Is there an overarching strategy or a program in place for 00:13:07.920 --> 00:13:13.280 managing AI investments? um what what are they doing to try and orchestrate the appropriate 00:13:13.280 --> 00:13:17.600 tooling and technologies to make available to you? Um what's the governance program? So, 00:13:17.600 --> 00:13:21.920 don't do anything alone. Make sure that whatever you're you're building, whatever you're investing 00:13:21.920 --> 00:13:27.760 in is part of the overall enterprisewide or at least functionwide perspective on on AI. 00:13:27.760 --> 00:13:32.640 That's where we see people get really tripped up is if they do something outside of that landscape, 00:13:32.640 --> 00:13:38.640 then that's where you're you're unintentionally exposing data um creating cyber security risks or 00:13:38.640 --> 00:13:43.680 anything like that because um you're you're going outside of a formal program that that likely does 00:13:43.680 --> 00:13:49.840 exist to some extent. Well, as you mentioned, this is still all very new and evolving really quickly, 00:13:49.840 --> 00:13:53.760 but you work in a space where you're working really closely with people as they're sort of 00:13:53.760 --> 00:13:59.920 developing systems and practices around this that um are are working well. So, could you 00:13:59.920 --> 00:14:05.120 give us some examples of of some of these best practices, ways that people are doing this, 00:14:05.120 --> 00:14:10.400 the things that you're describing um in ways that really make sense for for them? 00:14:12.080 --> 00:14:17.600 So best practices for experimentation and best practices for governance I think are are somewhat 00:14:17.600 --> 00:14:23.600 related. We we see um one thing that we see a lot of organizations building is a cross functional 00:14:23.600 --> 00:14:28.720 steering committee for for governance across the the enterprise that dictates or helps at least 00:14:28.720 --> 00:14:34.720 define the practices for adoption of AI across the the organization. That has to be closely tied to 00:14:34.720 --> 00:14:39.520 an enterprise strategy. If it's not, then I think it there there ends up being a little bit of a 00:14:39.520 --> 00:14:43.680 tension between the risks and controls around AI and ultimately what we're trying to accomplish. So 00:14:43.680 --> 00:14:48.800 that's I think at the broader organizationalwide perspective. When it comes to individuals and 00:14:48.800 --> 00:14:53.840 their day-to-day roles, um this is where we see a lot of really interesting um communities 00:14:53.840 --> 00:14:59.360 of interest or other types of grassroots efforts building where individuals who are really excited 00:14:59.360 --> 00:15:03.680 about the technology are really keen to play around with it and share their learnings with 00:15:03.680 --> 00:15:08.880 one another. We saw that um especially within our organization in our early days of our our 00:15:08.880 --> 00:15:15.120 AI generative AI adoption, the deployment of that enterprisewide and it becomes really 00:15:15.120 --> 00:15:21.760 um a nice way for for staff to share insights that they're seeing and best practices or or tips and 00:15:21.760 --> 00:15:26.000 tricks that have worked really well for them. It also becomes a really great way of unearthing use 00:15:26.000 --> 00:15:32.160 cases that become uh capabilities that make sense to build more robustly and distribute 00:15:32.160 --> 00:15:37.360 across the organization. We're also seeing practices for better orchestration across a 00:15:37.360 --> 00:15:42.960 variety of different agent platforms evolving. Our firm built agent OS which is our capability for 00:15:42.960 --> 00:15:48.160 um for for creating skills across a variety of different agent tooling linking them together 00:15:48.160 --> 00:15:52.240 and and providing more robust solution and solutioning and governance associated 00:15:52.240 --> 00:15:58.320 with that as well. So there there's a mixture of things going on there. My my basic point is that 00:15:58.320 --> 00:16:05.280 we should remain curious. Um it's the space moves very quickly, the capabilities move very quickly, 00:16:05.280 --> 00:16:09.840 our understanding of what we we can do with this technology moves really quickly and it behooves us 00:16:09.840 --> 00:16:16.400 to to remain vigilant and and engaged so that we understand how and identify opportunities where we 00:16:16.400 --> 00:16:22.160 can start to incorporate it. Uh we just recently released um a new job study called the AI jobs 00:16:22.160 --> 00:16:28.960 barometer and it looked at uh I think it was over a billion job postings and thousands of uh company 00:16:28.960 --> 00:16:35.760 reports really to understand what the impact of of AI is on jobs because we hear that a lot and we're 00:16:35.760 --> 00:16:41.520 seeing that jobs all across the board are growing but importantly we're seeing most growth in spaces 00:16:41.520 --> 00:16:47.040 where um even if there's a significant degree of automation potential for a role that people are 00:16:47.040 --> 00:16:51.920 are are are identifying where AI can actually help them and we're seeing the the job and wage growth 00:16:51.920 --> 00:16:56.880 in those spaces as well. So it the the boat is not being missed by anybody whatsoever. I think 00:16:56.880 --> 00:17:01.520 the people who are missing out are the ones that are not thinking about how AI can impact 00:17:01.520 --> 00:17:06.800 their work and thinking about ways in which they can incorporate it effectively. That sounds like 00:17:06.800 --> 00:17:11.680 really smart advice that you can apply no matter what space you're in to just be really curious but 00:17:11.680 --> 00:17:17.760 smart about about this. Um, and I think to that point, some people might be wondering just very 00:17:17.760 --> 00:17:23.120 uh directly, you know, when I think about how I can use help use AI to help me moderate things, 00:17:23.120 --> 00:17:30.640 just when is it appropriate and when isn't it, are there very specific tips you can offer? It's going 00:17:30.640 --> 00:17:35.360 to vary a little bit based on your organization. I would first look at your company to see what 00:17:35.360 --> 00:17:41.760 um if any guidance has been provided around uh best practices or other types of restrictions. 00:17:41.760 --> 00:17:48.560 Um the biggest the biggest concerns I typically see are going to be ones related to the use of of 00:17:48.560 --> 00:17:52.960 non-approved technology which is important for a number of reasons. So go with whatever technology 00:17:52.960 --> 00:17:58.240 solutions are already approved within your your organization. If there are none then talk to your 00:17:58.240 --> 00:18:02.960 IT your cyber security teams to see what what tooling might be available even if in a sandbox 00:18:02.960 --> 00:18:07.760 environment so that you can play around and test with it. So the the technology piece is is 00:18:07.760 --> 00:18:12.640 certainly one that we see. The other one is is on is going to be on data. So, what information are 00:18:12.640 --> 00:18:17.280 you making available to these types of systems? What information are you allowing them to surface 00:18:17.280 --> 00:18:22.880 back? Um, there's a lot of sensitive information that's not for all eyes within company walls. And 00:18:22.880 --> 00:18:27.680 so, we have to be careful how we design aic systems so that they're not unearthing that 00:18:27.680 --> 00:18:32.320 information and uh incorporating it into something and then exposing it to someone else. So, 00:18:32.320 --> 00:18:38.080 that's another area where we see um some some concerns. But then I guess the the last one is on 00:18:38.080 --> 00:18:44.640 um um on the effectiveness. So these are probabilistic systems. They they work on uh they 00:18:44.640 --> 00:18:49.840 don't work 100% the same way every single time. That's part of their value. But you have to think 00:18:49.840 --> 00:18:56.080 about that when you are designing a a capability to to use them effect uh within a specific space. 00:18:56.080 --> 00:19:00.720 Because if something necessitates 100% the exact same type of performance every single time, 00:19:00.720 --> 00:19:04.560 then maybe an agent isn't the right approach. maybe some other type of automation is better 00:19:04.560 --> 00:19:09.920 suited or a smaller form of an agent capability coupled with other types of traditional automation 00:19:09.920 --> 00:19:14.800 tools. Just because agents are out there doesn't mean that they're the most valuable or useful for 00:19:14.800 --> 00:19:20.400 a specific type of of circumstance. It is a tool. There are sets of tooling for us to use. And so 00:19:20.400 --> 00:19:24.400 we have to think appropriately what's the right tool to solve for the problem that we have to 00:19:24.400 --> 00:19:29.920 get to the outcome that we're trying to see. Um, it's very attractive right now to make everything 00:19:29.920 --> 00:19:35.280 an enogentic problem or a generative AI problem or even an AI problem at large. Might not always 00:19:35.280 --> 00:19:41.040 be the case. It sounds like this is the approach you should take with anything is the same approach 00:19:41.040 --> 00:19:45.680 you would take with AI where just because it's there doesn't mean it's free in that instance, 00:19:45.680 --> 00:19:50.240 right? Um, not a one-sizefits-all sort of situation. Well, Alana, we're getting tons 00:19:50.240 --> 00:19:54.800 of uh questions from the audience and please keep sending your questions in. We'll try to 00:19:54.800 --> 00:19:58.880 get through as many as we we can and I'd love to start to share some of them with you now. Um 00:19:58.880 --> 00:20:04.560 so we have one uh that one person asked, can a company be successful in implementing AI 00:20:04.560 --> 00:20:09.760 without a dedicated employee or TPP, which is actually not a an acronym that I know, 00:20:09.760 --> 00:20:14.560 but maybe that means something to you. Uh without a dedicated employee overseeing its onboarding, 00:20:14.560 --> 00:20:18.640 what's the true resource allocation needed for for it to be done correctly? 00:20:20.800 --> 00:20:24.880 Every organization is going to be a little bit different and a lot of this will depend on the 00:20:24.880 --> 00:20:30.480 organization size and your current technology infrastructure technology team and then also 00:20:30.480 --> 00:20:37.200 the the the remit um within the organization. larger companies have started to I've seen 00:20:37.200 --> 00:20:44.880 um create if not a formal C AIO role so a chief AI officer role than someone who has taken on 00:20:44.880 --> 00:20:50.400 the mantle of of the both the strategy and the technology implementation associated with AI 00:20:50.400 --> 00:20:55.600 that might be a CIO it could be a co it could be really anybody within the organization but 00:20:55.600 --> 00:21:01.200 it really matters that there is someone who has been at the seuite assigned the responsibility 00:21:01.200 --> 00:21:05.520 of overseeing what the adoption strategy of AI is going to look like at the enterprise. Again, 00:21:05.520 --> 00:21:10.240 that doesn't necessarily mean it's a new net new role. Uh really depends on the organizational 00:21:10.240 --> 00:21:17.280 structure. And then the supporting roles for how we effectively use AI will end up being a shared 00:21:17.280 --> 00:21:21.760 sense of responsibilities across governance which is when when you might have that cross 00:21:21.760 --> 00:21:26.640 functional steering committee or even a governance team that's completely dedicated to AI. Uh it 00:21:26.640 --> 00:21:31.200 again it takes different forms from different companies. The importance being that there is 00:21:31.200 --> 00:21:36.400 a group of people who have been tasked with even if it's their side job thinking about what does 00:21:36.400 --> 00:21:42.960 it mean to be uh to be enabling the organization with the appropriate structures, processes, risk 00:21:42.960 --> 00:21:48.800 uh appetite statements and assessments etc. um to meaningfully adopt and move forward the 00:21:48.800 --> 00:21:53.440 technology. And then on the complement of that are our teams that are are thinking about the 00:21:53.440 --> 00:21:58.560 uh the use cases and the capabilities and overseeing the the functional implementation. 00:21:58.560 --> 00:22:04.400 So in some sense a lot of this becomes um augmentations to to people's current roles. 00:22:04.400 --> 00:22:08.640 Um in large companies you might see new staff that are completely dedicated to this because of 00:22:08.640 --> 00:22:16.080 the scale that's necessary for it. Uh but it but it really does vary unfortunately and I the next 00:22:16.080 --> 00:22:20.560 question here kind of connects to this sort of thinking about how our work changes and it sounds 00:22:20.560 --> 00:22:28.320 like that it may also vary but how does agentic AI reframe the way we work dayto-day? I I think 00:22:28.320 --> 00:22:35.280 it's an interesting point in time for us because um one immediate knee-jerk reaction is okay, 00:22:35.280 --> 00:22:40.560 what am I spending a lot of time on? What are the things that I um I feel like are some semi-re 00:22:40.560 --> 00:22:45.360 repeatable with some degree of nuance or uh a way that I could have a little assistance supporting 00:22:45.360 --> 00:22:50.320 me throughout my my whatever my my tasks are for that particular day. And that's a a great 00:22:50.320 --> 00:22:55.280 way to think about how to incorporate agentic systems just just on the surface. But it's also 00:22:55.280 --> 00:23:00.240 an opportunity for us to take a step back at some of our existing processes and see um which ones 00:23:00.240 --> 00:23:05.680 could we just completely rethink. There are there is one approach which is let's take the existing 00:23:05.680 --> 00:23:10.640 process and then peacemeal it right add little automations here and there maybe um some RPA 00:23:10.640 --> 00:23:15.280 like of the traditional types of automation tools plus some agentic systems coupled with some other 00:23:15.280 --> 00:23:19.840 stuff but then the other perspective is what's really not working here and could we actually 00:23:19.840 --> 00:23:27.200 find a faster path from point A to point B and actually wholesale eliminate or re re- um envision 00:23:27.200 --> 00:23:33.120 uh what happens in the middle and there's a real opportunity to use aic systems for that um it can 00:23:33.120 --> 00:23:38.560 facilitate a lot of collaboration. Um we've seen um clients in a number of spaces using these types 00:23:38.560 --> 00:23:43.920 of systems to support the the facilitation and collaboration across existing processes. There's a 00:23:43.920 --> 00:23:49.280 um a healthcare company that we've been working with um that was redesigning their process 00:23:49.280 --> 00:23:55.120 for obtaining insights from clinical trials and translating that into additional materials further 00:23:55.120 --> 00:24:01.200 down the line. And um the use of agents helped them rethink that process and actually collaborate 00:24:01.200 --> 00:24:06.000 across borders a lot better, across teams a lot better. Um but getting first drafts of different 00:24:06.000 --> 00:24:11.840 types of of uh insights generated earlier and they saw some some significant savings across 00:24:11.840 --> 00:24:16.960 that life cycle. But they did rethink the life cycle that the endto-end process. They didn't just 00:24:16.960 --> 00:24:23.520 um apply it as is. And I think that's part of the value that we see. H um well we have a question 00:24:23.520 --> 00:24:31.040 here um about AI hallucination. Um does AI hallucination still occur or still occur as much 00:24:31.040 --> 00:24:34.960 when you provide the data sets or perimeter to work within? And before you answer, if you could 00:24:34.960 --> 00:24:42.000 just maybe explain what AI hallucination is for those who uh aren't familiar. Um AI hallucination 00:24:42.000 --> 00:24:48.320 is a term that refers to a model generating a essentially factually incorrect response. Um, 00:24:48.320 --> 00:24:53.280 and it's important to know why that happens. U, generative models in particular, they they are 00:24:53.280 --> 00:24:57.200 prone to generating factually incorrect information if they don't don't have the 00:24:57.200 --> 00:25:02.240 appropriate set of information because they're basically instructed to provide an answer and 00:25:02.240 --> 00:25:07.600 then convey that answer authoritatively. And so if you don't know uh where that information is coming 00:25:07.600 --> 00:25:15.600 from, it could be quite challenging to to even see that it's it's not not correct or it's erroneous 00:25:15.600 --> 00:25:21.280 in some respect. Um, hallucinations can can be there are common examples of like lawyers uh 00:25:21.280 --> 00:25:26.880 submitting briefs to to um uh to judges and like a bunch of the citations were not actually real 00:25:26.880 --> 00:25:30.080 and that's because they were using a variety of generative tools without appropriate constraints. 00:25:30.080 --> 00:25:36.880 So it they they occur all the time. Again, this is just a model producing something in a space 00:25:36.880 --> 00:25:43.280 that it doesn't necessarily have the appropriate uh data set for. constraining the operations of 00:25:43.280 --> 00:25:48.560 the model, requiring it to look at certain data sources, requiring it to surface its information, 00:25:48.560 --> 00:25:54.880 explain where it's getting information from are all ways to reduce hallucinations or erroneous 00:25:54.880 --> 00:26:00.160 responses. Um, it's also a great way for to keep the human in the loop or as part 00:26:00.160 --> 00:26:04.400 of the team with the agent. Um, so that that we are still ultimately responsible for going 00:26:04.400 --> 00:26:09.840 down and factchecking. But I think we do have to appreciate that this is an artifact of of AI. It's 00:26:09.840 --> 00:26:15.200 in some sense part of the design that we want these systems to be creative to think outside 00:26:15.200 --> 00:26:20.480 the box to to push boundaries and and provide information from a variety of sources. So it's 00:26:20.480 --> 00:26:25.120 also important for us to understand and define what is a hallucination in the context of the 00:26:25.120 --> 00:26:29.840 operation of a particular system because sometimes we want that creativity somewhat unconstrained 00:26:29.840 --> 00:26:34.960 and sometimes we want it a lot more restricted when there is an actual real answer to be had. 00:26:36.320 --> 00:26:41.360 U well so we have time for a couple more questions. Um so one person asked what is 00:26:41.360 --> 00:26:48.640 the recommended team structure way of working to grow along a AI inside of an organization. I wish 00:26:48.640 --> 00:26:57.840 there was like one one answer to that. U so team structure from incorporating AI. I think of AI as 00:26:57.840 --> 00:27:05.680 a team member is probably the maybe the answer to that. Um in tactically speaking um it depends on 00:27:05.680 --> 00:27:10.960 the variety of different AI skills or utilities you're trying to incorporate. Um we see some 00:27:10.960 --> 00:27:17.360 organizations mixing a a degree of of technical uh pe technical talent. So people who understand 00:27:17.360 --> 00:27:23.280 a bit more about how models work. Um with uh more teams that are specialized within the specific 00:27:23.280 --> 00:27:29.360 processes or functions that they work with to try and identify use cases and and work on aligning on 00:27:29.360 --> 00:27:34.880 expectations and val and um establishing what we think good practice would look like in relation to 00:27:34.880 --> 00:27:40.240 to a specific system. And I think that's certainly a good approach. Have someone be the champion 00:27:40.240 --> 00:27:47.040 within your group. have someone be responsible for pulling ideas in, thinking critically, pushing you 00:27:47.040 --> 00:27:51.360 on how you could use AI a bit more effect to to make your work a little bit more effective. But 00:27:51.360 --> 00:27:57.280 I would also push you to think about AI generally as a team member. If you had another team member 00:27:57.280 --> 00:28:03.280 that you could offload tasks to or that could help you rethink how you do something or could um could 00:28:03.280 --> 00:28:08.960 provide you with a sounding board for you to to to get new ideas that could be a very good use for 00:28:08.960 --> 00:28:14.240 for AI within your team. It's it's so interesting because I think in thinking of it that way, 00:28:14.240 --> 00:28:20.960 it sort of takes it up a level from just being a tool to some other thing. Um maybe not a human 00:28:20.960 --> 00:28:25.840 resource, but some other resource that's closer to to to human. And and that sort of leads to this 00:28:25.840 --> 00:28:31.200 next question, which is whether work is being done to equip AI with emotional intelligence, a soft 00:28:31.200 --> 00:28:39.280 skill set that humans have, and if not, should it be? I think we there is certainly work being done. 00:28:39.280 --> 00:28:46.480 That's a very active research space. Uh but I I think the question about AI ultimately replacing 00:28:46.480 --> 00:28:50.880 us, which humans have emotional intelligence. I think that's one of the big distinguishing 00:28:50.880 --> 00:28:56.640 factors between us and the technology solutions that we have. Um I don't think that we should be 00:28:56.640 --> 00:29:01.440 thinking about how AI will ultimately replace us. It's more about how we work with AI to to 00:29:01.440 --> 00:29:08.000 make more effective teams. Um, and so to that end, emotional intelligence, I mean that there's these 00:29:08.000 --> 00:29:13.840 are functionally still technical systems. Um, and I, this is my opinion, but I think that sometimes 00:29:13.840 --> 00:29:18.640 when we want to impart things like emotional intelligence onto technical systems, we're 00:29:18.640 --> 00:29:22.960 anthropomorphizing them and that actually can put us in somewhat of a risky space where we think 00:29:22.960 --> 00:29:29.120 that these these are thinking machines that can can solve lots lots more problems for us than they 00:29:29.120 --> 00:29:36.000 actually can. Um I do think there is interesting spaces where AI is being used to support um from 00:29:36.000 --> 00:29:41.440 like a coaching perspective or from uh even like a therapy perspective that that's happened in 00:29:41.440 --> 00:29:46.800 certain spaces. But those are systems that are designed for those purposes. And so we just we 00:29:46.800 --> 00:29:51.920 have to come back to what's the purpose of of this this set of capabilities that we're building. How 00:29:51.920 --> 00:29:57.680 are we using it most effectively? Is it what is it being designed for? How do we want to work 00:29:57.680 --> 00:30:02.960 with it? And I think those types of inputs will help us define what we really need from systems. 00:30:02.960 --> 00:30:07.680 The AI research space will continue to evolve. We'll definitely see more and more functionality 00:30:07.680 --> 00:30:13.120 coming out all the time. But we're we should also think about what makes us unique, what makes us u 00:30:13.120 --> 00:30:18.800 interesting as people and and how AI can really be supporting us so that this is really a human-led 00:30:18.800 --> 00:30:24.160 techpowered or agent powered universe rather than um a circumstance where we're deploying a lot of 00:30:24.160 --> 00:30:30.720 technology and and hoping that humans will just fill fill the gaps here and there. Well, Alana, 00:30:30.720 --> 00:30:36.800 you've painted a really comprehensive picture of where we are in this moment. Uh could you share 00:30:36.800 --> 00:30:42.160 a little bit about your vision for the future? Where do you see things heading in a year or 10 00:30:42.160 --> 00:30:53.040 years? In in a year, I think we will have um more structured programs across most organizations for 00:30:53.040 --> 00:30:57.040 thinking about and incorporating AI and I'm very excited about that. Part of that would be in my 00:30:57.040 --> 00:31:02.800 mind that's um also thinking about the governance and controls and maybe instituting broader 00:31:02.800 --> 00:31:08.880 uh practices across organizations to surface um capabilities where we can facilitate better 00:31:08.880 --> 00:31:15.600 relationship building with AI. Where I see us in 5 to 10 years u I honestly think that in five 00:31:15.600 --> 00:31:21.200 years we'll be thinking of AI as normal. It's just something that we work with that is there. 00:31:21.200 --> 00:31:26.640 It's it's um it's a set of tools and a set of capabilities that we rely upon when we need it. 00:31:26.640 --> 00:31:31.280 I'm I'm I think the hype will really have I hope the hype will have died down to some degree and 00:31:31.280 --> 00:31:36.240 we'll have a bit more rationality around what it makes sense to to to incorporate this technology 00:31:36.240 --> 00:31:43.200 for, but it'll be really really useful what what remains and I'm very excited for that. Well, thank 00:31:43.200 --> 00:31:49.600 you so much Alana for for all of this. I feel like you have um we could talk for for hours about all 00:31:49.600 --> 00:31:53.040 of this. There's so many questions that we haven't gotten to and that uh we could discuss, but we 00:31:53.040 --> 00:31:57.520 reach the end of our time. So, thank you so much for your time here with us and um and thank you 00:31:57.520 --> 00:32:01.520 to all of you for your questions and for joining us today. Um and of course, thank you to Price 00:32:01.520 --> 00:32:06.000 Waterhouse Coopers uh for making this LinkedIn episode possible through their generous support. 00:32:06.000 --> 00:32:11.440 This was actually the second episode presented in partnership with PwC. You can watch the first how 00:32:11.440 --> 00:32:15.840 do you stay ahead as AI evolves on demand on our LinkedIn channel, which is where this episode will 00:32:15.840 --> 00:32:20.480 also be available to rewatch. Uh, we hope you'll join us again for another live conversation in the 00:32:20.480 --> 00:32:39.040 future. Thank you again, Alana, and enjoy the rest of your day, everyone. Thank you, Annie.