00:27:23.000 --> 00:27:34.000 Welcome everyone. Welcome to this panel events from the Government AI campus. My name is Pooja and I'm the Chief Partnerships Officer here at Apolitical. 00:27:34.000 --> 00:27:43.000 It'd be great to see who is in the room, so as you come in, please do post in the chat where you're joining us from and what organization you work for. 00:27:43.000 --> 00:27:47.000 I'll read out a few names as they come in. 00:27:47.000 --> 00:27:54.000 Today we'll be exploring the delicate balance between AI innovation and ethics with some very special guests. 00:27:54.000 --> 00:27:55.000 So. 00:27:55.000 --> 00:28:03.000 We'll be highlighting the wonderful opportunities for innovation with AI in government. As well as covering some of the ethical considerations involved. 00:28:03.000 --> 00:28:05.000 We also share practical tips, guidelines. 00:28:05.000 --> 00:28:08.000 That tried his number here. 00:28:08.000 --> 00:28:13.000 I think, I think it's Jean who's on, thank you, Jane, for muting. 00:28:13.000 --> 00:28:23.000 As I was saying, we will be sharing practical tips. Guidelines, frameworks and steps for you to use, regardless of your role-level or current use of AI. 00:28:23.000 --> 00:28:37.000 We know that we have a really diverse group registered for today's event. From the pre submitted questions, we can see that some of you are starting from square one with AI and some of you are looking for insights far beyond the basics. 00:28:37.000 --> 00:28:43.000 We'll try and do our best in the next hour to address as many of your questions as possible. 00:28:43.000 --> 00:28:45.000 So let's see who is here today. So we have. Mike from Nova Scotia, provincial government. 00:28:45.000 --> 00:28:59.000 Welcome, Mike. We have Luke from Swansea, DBSA. Claire, Policy Horizons Canada, we have Kevin from Manitoba's Department of Agriculture. 00:28:59.000 --> 00:29:15.000 We have Jessica, my colleague from a political high jess. We have Iona customer experience strategy at DWP UK one of our partners Lucy from Argentina, welcome, and so many others. 00:29:15.000 --> 00:29:22.000 We have US Department of, Labor, Andrew from DC. This is incredible. What a truly global audience. 00:29:22.000 --> 00:29:40.000 For those of you who may be joining us for the first time, welcome. Apolitical is a network and learning platform for governments used for governments used by over 200,000 public servants and policymakers for more than 160 countries. 00:29:40.000 --> 00:29:52.000 Our mission is to help build 20 first century governments that work for people and the planet. And we do this by putting the best knowledge and skills at the fingertips of public servants around the world. 00:29:52.000 --> 00:30:02.000 Our government AI campus is a really exciting collaboration between governments, academic experts, philanthropic partners, including our founding funders, Google. 00:30:02.000 --> 00:30:15.000 Org and the Rockefeller Foundation. Political Our aim at the campus is to rapidly upscale and prepare 10,000 public servants for the opportunities and challenges of AI. 00:30:15.000 --> 00:30:23.000 Since our first AI campus course went live early this year, we've had more than 1,700 public servants enrolling on it. 00:30:23.000 --> 00:30:32.000 Everyone on this call is invited as well, so please do take your free place on the campus if you haven't already done so. 00:30:32.000 --> 00:30:36.000 Before we start, I'll get a few technical notes out of the way. We're going to be recording this session. 00:30:36.000 --> 00:30:49.000 And we will host it online on the Aprilolitical website via V. We're doing this so that the participants and apolitical members can review the content on demand to further their learning. 00:30:49.000 --> 00:30:56.000 It'll also enable us to share the content with any participants who are unable to make it to the session. 00:30:56.000 --> 00:31:03.000 Please be aware that viewers of the recording may be able to see your Zoom username if you come off mute. 00:31:03.000 --> 00:31:04.000 And if you have your camera on, they may be able to see your face in addition to any information. 00:31:04.000 --> 00:31:10.000 Contained in your background. Please let me or one of my colleagues, you'll be able to see them. 00:31:10.000 --> 00:31:17.000 They'll have their names, Jenni, A-political and tacitly political. 00:31:17.000 --> 00:31:25.000 Please let us know if you have any questions or concerns about apolitical making the recording available in the manner described. 00:31:25.000 --> 00:31:33.000 And if you have any further questions, you can always find the details of our data protection officer at the footer of our homepage. 00:31:33.000 --> 00:31:43.000 So before we introduce our speakers, we're going to run a few quick polls to get a sense of what you want from today's discussion. 00:31:43.000 --> 00:31:56.000 So the first poll should be coming up soon. How often are you using AI tools in your work? And this could be chat GPT, it could be Gemini, which is to be known as BART or Microsoft co-pilot. 00:31:56.000 --> 00:32:04.000 So is it daily? Is it weekly? Is it not that often? Or I have never used them? 00:32:04.000 --> 00:32:16.000 So please go ahead and fill out the poll. If you're not able to see the poll for any reason, do use your chat, TV, not just to chat, to answer these questions. 00:32:16.000 --> 00:32:20.000 Go ahead, we'll give you a few seconds to finish the poll. 00:32:20.000 --> 00:32:28.000 I can see a few answers in the chat every single day. What's the AI accounts, right? 00:32:28.000 --> 00:32:32.000 Brilliant. Can we have the results, please? So this is, so we have about 37% that says not that often. 00:32:32.000 --> 00:32:57.000 30% that says I've never, who say I've never used them. And then there's a 14% that use daily so active adopters and but 19% that say weekly just out of interest is actually matches some of the results that we found from an inside, from a survey that we did earlier, this year as well. 00:32:57.000 --> 00:33:07.000 So the second one, is it should be coming up again in the adoption of generative AI, which issue concerns you the most. 00:33:07.000 --> 00:33:18.000 Is it the amplification of biases? Is it the copyright and intellectual property? Is it the lack of transparency and explainability of decision making? 00:33:18.000 --> 00:33:30.000 Or is it disinformation or information inaccuracy? So I'll repeat that again in case you're not able to see the poll in the adoption of gender of AI, which issue concerns you most. 00:33:30.000 --> 00:33:41.000 Amplification of biases, copyright and intellectual property, lack of transparency and explainability of decision-making, information in accuracy, or disinformation. 00:33:41.000 --> 00:33:48.000 So go ahead and give your answers in. I agree with you, Cassandra. I wish we could select 2 choices. 00:33:48.000 --> 00:33:54.000 I think we need to send that back to Zoom as feedback. The security of data use thanks, Luke. 00:33:54.000 --> 00:34:04.000 I think that's important as well. Okay, so majority of you, view information inaccuracy or disinformation as the biggest risk. 00:34:04.000 --> 00:34:18.000 I'm followed quite closely by lack of transparency and then copyright and amplification. I think this again matches some of the things that we have heard in the in our conversation to public servants in around the world. 00:34:18.000 --> 00:34:26.000 There's a few more in around the quality of results in the chat as well. Thank you. So much for taking part on those polls. 00:34:26.000 --> 00:34:39.000 They really helped to understand, where the audience is, where you were coming from. So introducing our speakers today, we're really fortunate to have 3 fantastic guests. 00:34:39.000 --> 00:34:44.000 David Leslie, Rebecca Tweed, and Giorgia Abelino joining us. 00:34:44.000 --> 00:34:51.000 Sadly, Sana Kagani, who was meant to be our fourth guest today, was a former head of the UK Government Office for AI, is unable to join us today. 00:34:51.000 --> 00:35:01.000 But we will catch up with her after the event today and if she has any resources to share with you we will share with share those with you in the event follow up email. 00:35:01.000 --> 00:35:08.000 I'll read out the bios and just so you know their full bios we posted in the chat too. 00:35:08.000 --> 00:35:21.000 So starting with Professor David Leslie, he's a director of ethics and responsible innovation research at the Allen Turing Institute and professor of ethics, technology and society at the Queen Mary University of London. 00:35:21.000 --> 00:35:28.000 He's the author of the UK Government's official guidance on the responsible design and implementation of AI systems in the public sector. 00:35:28.000 --> 00:35:37.000 And as part of his international work, he also serves and Unesco's high-level expert group steering its recommendation on the ethics of artificial intelligence. 00:35:37.000 --> 00:35:46.000 This is adopted by 193 member states of the organization. Welcome, David. Really wonderful to have you here. 00:35:46.000 --> 00:35:55.000 Our second speaker today is Georgia Avatino. She's the Senior Director of Government Affairs and Public Policy for Google in South Europe. 00:35:55.000 --> 00:36:03.000 And has since January, 2,002 that 23 coordinated the artificial intelligence strategy for Emia region. 00:36:03.000 --> 00:36:10.000 She started working in Brussels focusing on anti-trust issues. First at a law firm and then at the European Commission. 00:36:10.000 --> 00:36:23.000 Georgia is a board member of ISPI, which is the Italian think tank focusing on geopolitical issues and she has been recently nominated amongst the 50 most influential Italian women by fortune. 00:36:23.000 --> 00:36:27.000 So lucky to have you here today with us, Georgia. Welcome. Our third speaker is Rebecca Tweed. 00:36:27.000 --> 00:36:32.000 And I love the name of her organization. She's the executive director for all tech is human. She's a leader in responsible technology. 00:36:32.000 --> 00:36:51.000 Rebecca was named one of the 100 brilliant women in AI ethics in 2,023 and as the guest editor of Springer AI and ethics journal topical collection on the social impacts of AI on youth and children. 00:36:51.000 --> 00:36:56.000 She's the co-chair of the I Tripoli Global AI Ethics Initiative Editing Committee. 00:36:56.000 --> 00:37:10.000 Rebecca worked as a project manager for New York Law F, Eisenberg and Bom, Lp's AI fairness and data privacy practice group where she examined technologies impact on society. 00:37:10.000 --> 00:37:14.000 I tell you what, I do moderate a number of panels at a political and this is one of the many impressive panelists to welcome all of you really excited to have the discussion. 00:37:14.000 --> 00:37:26.000 So before we kick off, I'd love to oppose an icebreaker to kick things off with our speakers. 00:37:26.000 --> 00:37:38.000 So you are in an elevator with an AI skeptic politician and you have 30 s to describe the biggest opportunity you using AI in government. 00:37:38.000 --> 00:37:50.000 Now this can be as far-fetched or as realistic as you like. I'm gonna start with Georgia, you've got 30 s. 00:37:50.000 --> 00:37:56.000 Good morning. Good morning, ministers. It's a pleasure meeting you again. May I ask you something? 00:37:56.000 --> 00:38:07.000 Do you think that the institution that you're leading is a future ready? Well, the reason why I'm asking you that is that artificial intelligence, as you probably know, has really the potential to revolutionize the way public sector operates and serves its mission and its citizens. 00:38:07.000 --> 00:38:17.000 Let me give an example. I'm sure that you're familiar with and it's very close to your hearth. 00:38:17.000 --> 00:38:29.000 I know that you're coming from Minnesota and most probably you know that in Minnesota 10% of the population doesn't speak English and this includes refugees and many other people. 00:38:29.000 --> 00:38:34.000 Well, one of the problem is that these people cannot have access to basic services like, you know, driving license. 00:38:34.000 --> 00:39:00.000 Well, what happened is that the Minnesota authority actually, the Minnesota Public Safety Department introduced Google Translate Power by AI and they basically resolved the problem because through this AI tool, now everyone, irrespective of the fact that they speak English or not is such a good English, can have access to basic and fundamental services. 00:39:00.000 --> 00:39:07.000 So I'm sure that you're thinking about it. I would love to have further conversation with you on that. 00:39:07.000 --> 00:39:18.000 Thank you, Georgia. You've clearly done this icebreaker before, so I'm going to move on now to David, your turn. 00:39:18.000 --> 00:39:19.000 Yes. 00:39:19.000 --> 00:39:34.000 I'm sold if I was the, I skeptic there. Right. So I would say, You know, we can think of all of the great things in the back office and operate operational efficiencies that AI and it's specifically generative AI systems can introduce for the public sector. 00:39:34.000 --> 00:39:40.000 But I really think that we need to think big. We need to think about the ways in which these large, these very complex systems are able to to serve the public interest. 00:39:40.000 --> 00:40:02.000 So thinking here of the way that in in the UK Canada we are oriented to you know nationalized medicine the way these systems can help us detect strokes earlier for better health outcomes in triaging and hospitals, the way that they can help early cancer detection, the way they can streamline various treatment and diagnosis. 00:40:02.000 --> 00:40:27.000 We can think of the way in which some of these systems are helping us address biodiversity drain and some of these systems are helping us address biodiversity drain and climate change and even some of these systems are helping us address biodiversity drain and climate change and even creating efficiencies and in energy systems by way of creating smart grids and more efficiencies in urban transportation. 00:40:27.000 --> 00:40:42.000 So one thing to remember is that government is in a unique position in holding a lot of public data. And we simply can't afford not to use that data for the public benefit. 00:40:42.000 --> 00:40:46.000 Thanks, David. Rebecca, your 10. 00:40:46.000 --> 00:40:53.000 So I would say whether you're skeptical out of fear of AIs capabilities or out of apprehension that it could take your job or out of disbelief that it's actually any good at all. 00:40:53.000 --> 00:41:06.000 You need to understand AI because government's already using it in many ways. You need to understand AI because government's already using it in many ways. 00:41:06.000 --> 00:41:10.000 AI already plays a very consequential role in some cases for algorithmic decision making with regards to accessing certain services. 00:41:10.000 --> 00:41:40.000 In law enforcement and throughout the criminal justice system in public education and hiring and many more. And increasingly because of the widespread availability of generative AI tools, it's being used by anyone and everyone in carrying out everyday functions, in, their, business, so, don't, shy, away, from, understanding, how, to, harness AI can be used in innovative ways to solve big 00:41:40.000 --> 00:41:50.000 problems, but most practically and so perhaps most impactfully in the near term it can increase efficiency, which we could always use more of. 00:41:50.000 --> 00:41:56.000 So the biggest opportunity is that it can help you get things done. 00:41:56.000 --> 00:42:02.000 Thank you all. I love the emphasis on the public good and the higher impact of this technology piece, which is so important to bring through. 00:42:02.000 --> 00:42:14.000 And before we jump into the panel, just a request for all the 3 panelists. Let's keep the answers to 2 min mainly because there are so many please submitted questions that I'd love to go through. 00:42:14.000 --> 00:42:32.000 As many of them as possible. David, building on that icebreaker and given what we know of the current use of AI in government, what do you think the next 2 years of AI innovation in government looks like? 00:42:32.000 --> 00:43:02.000 Well, I think that we definitely are at the beginning of a much longer journey. And so in an ideal scenario we will really do a lot of work to build readiness into all areas of government at all levels so just not thinking just about those who are kind of carrying out the remits of various public bodies, but also kind of the senior management people within government so that we understand the nature and the limitations of these of these systems as well as the the 00:43:07.000 --> 00:43:22.000 real world opportunities because I do think there are various narratives out there in the world in the in the public that are not as helpful to understanding that at the end of the day these are tools. 00:43:22.000 --> 00:43:44.000 That can help us advance. Society let goals. And in order to really optimize, you know, that path of using AI responsibly and for the public good, we really do need to to take steps the next couple of years to really build, you know, more knowledge and preparedness among, the public sector. 00:43:44.000 --> 00:43:59.000 Public sector bodies, regulators, all across government. So if I could say the next 2 years and ideally would be about adoption but but about building as well preparedness to adopt responsibly. 00:43:59.000 --> 00:44:04.000 I like the building responsibility to, building preparedness to adopt responsibly. It's sort of so, important, within governments as well. 00:44:04.000 --> 00:44:18.000 George, you're building on that. Are there practical examples that you can share that have really struck you of how certain public administrations have embraced AI. 00:44:18.000 --> 00:44:29.000 Yes, definitely. Well, personal fact. I moved to Spain recently. And so I'm quite, even if I'm Italian as you can clearly understand from my accent. 00:44:29.000 --> 00:44:37.000 I moved to Spain and I saw what happened in Spain and I imagine in. Many other place, of course, or if not all placed during COVID. 00:44:37.000 --> 00:44:52.000 So during COVID, all public administration were overwhelmed with requests about information, how can I get if I'm a business you know my money because you are supporting me, me, etc, etc. 00:44:52.000 --> 00:45:00.000 Well the Spain social security administration did something quite interesting that is to work on a solution that was first of all compliant with GDPR. 00:45:00.000 --> 00:45:13.000 So with our privacy, rookie and privacy rules and secure on all side but to provide customer with all the answers that they needed. 00:45:13.000 --> 00:45:24.000 So basically they did a dialogue flow chatboat assistant in order to provide this answer to to the customers to the citizens. 00:45:24.000 --> 00:45:32.000 In the first week from the launch of this, of this chat but they had the 20,000 requests. 00:45:32.000 --> 00:45:39.000 That were registered per minute and they served at the end of the day. They provide assistance to 10 million citizens. 00:45:39.000 --> 00:45:59.000 But what I want to underline here is that This was incredibly useful, let's say for the paperwork in order to provide answers to basic question but The chat both did not provide any assessment on whether my request as Georgia was or not admissible, was or not passed. 00:45:59.000 --> 00:46:09.000 Then there were human being, let's say, doing the final step. But just by providing answers to questions and doing all the administrative stuff. 00:46:09.000 --> 00:46:19.000 Well, they basically, and I want to say solve the problem, but indeed they provide 20,000 answers in the first days. 00:46:19.000 --> 00:46:33.000 So I think that this is a clever way of using, you know, AI and it's a way where you bring together, of course, the machine, so the AI power, but then for certain most difficult or final decision. 00:46:33.000 --> 00:46:42.000 You have of course human, let's say judgment or human intervention for for the final decision. 00:46:42.000 --> 00:46:55.000 Thank you, Georgia. Rebecca, just shifting gears a bit. What when we say AI governance, what is meant by that and where is it needed and more importantly, what does it look like in practice? 00:46:55.000 --> 00:47:03.000 Yeah, so at a high level, AI governance is essentially about determining what the rules of the road are for using these AI systems and tools, as well as what characteristics we want AI to exhibit. 00:47:03.000 --> 00:47:25.000 So for instance, some of the common characteristics that apply across context would include transparent, explainable, privacy protecting, fair and things like that, as well as mechanisms for testing and measuring the output of these systems. 00:47:25.000 --> 00:47:28.000 So. When we talk about AI governance, we're, concerned about reducing risk, and with performing tests and checks to provide assurances to the public. 00:47:28.000 --> 00:47:41.000 That these systems are being used safely. AI governance is about setting up guardrails against misuse. 00:47:41.000 --> 00:47:43.000 And to ensure that we're using AI in ways that would minimize the potential harm to users because in this case we are talking about. 00:47:43.000 --> 00:47:55.000 Citizens who may depend on government and in many cases have no choice but to interact with these systems. 00:47:55.000 --> 00:48:02.000 So AI governance can provide guardrails, which we need, for instance, for AI systems like algorithmic decision-making tools. 00:48:02.000 --> 00:48:09.000 So we want to be sure that these decisions are not being made in ways that are inaccurate or unfair or even arbitrary. 00:48:09.000 --> 00:48:10.000 And we use them for AI driven surveillance tech tools to make sure that citizens rights are being protected. 00:48:10.000 --> 00:48:19.000 So AI governance can take the form of procurement guidelines for AI systems that governments buy. 00:48:19.000 --> 00:48:29.000 Or of red teaming and testing for vulnerabilities and any AI systems that governments might build, as well as algorithmic audits for any AI that got governments use. 00:48:29.000 --> 00:48:39.000 So Hey, I governance can include a variety of components. In the United States where I'm based, it often involves following a risk management framework. 00:48:39.000 --> 00:48:48.000 So, we use the NIST. AI risk management framework, but that can include a lot of other strategies as well. 00:48:48.000 --> 00:49:02.000 So we can talk about, you know, ensuring that there's proper documentation for AI systems. To utilizing an AI incident database if we're tracking problems. 00:49:02.000 --> 00:49:12.000 We can also talk about, you know, water marking synthetic images, which open AI and meta just committed to doing so it can cover a lot of ground. 00:49:12.000 --> 00:49:13.000 Thank you. Rebecca, just, to mind this lots of conversation happening in the chat. 00:49:13.000 --> 00:49:21.000 As well. Really, really glad to see all the engagement and the questions. We'll make sure we come back to them. 00:49:21.000 --> 00:49:33.000 David, coming back to you. So what are some of the emerging ethical challenges that government should prepare for as the technology itself evolves? 00:49:33.000 --> 00:49:39.000 Right, I've been keeping an eye on the chat and I think some of them have already some of the big concerns have already been. 00:49:39.000 --> 00:49:51.000 Mentioned. For instance, there is a tendency to not recognize that when, some of these predictive systems are processing. 00:49:51.000 --> 00:49:59.000 Social and demographic data, so data that's collected that has as part of the production of the data. 00:49:59.000 --> 00:50:16.000 Possibilities for bias and discrimination to kind of be baked into it. When that type of data is used in public services and social services, we are, you know, we're in peril often of replicating or augmenting those patterns of bias and discrimination. 00:50:16.000 --> 00:50:28.000 And it's very important, especially just thinking of cases where vulnerable people are really involved. So, you know, there's a example of public benefits. 00:50:28.000 --> 00:50:41.000 We could also think of predictive risk modeling in children's social care. There are other, there are other other use cases where you have potential over collection of data amongst vulnerable communities. 00:50:41.000 --> 00:50:57.000 That can bias data sets and create these kind of cycles of, you know, discrimination and that are that that actually come through in the use of predictive risk modeling for those populations and, that, actually come through in the use of predictive risk modeling for those populations. 00:50:57.000 --> 00:51:22.000 And so I would say first and foremost, we need to in that sense, just be aware that there are a bias mitigation practices, end to end bias mitigation practices that can improve those those data sets and those AI or predictive risk model life cycles but that it will only in at the end of the day be able to manage the bias we won't be able to eliminate bias from data sets and we have to 00:51:22.000 --> 00:51:36.000 be really aware and this is why the point about having thinking of it as if you have automated decisions assistance so that human judgment can be involved at the end of the day to sort of contextualize and corroborate. 00:51:36.000 --> 00:51:57.000 The predictions of the given model with the understanding of its limitations. That's extremely important. I'll clearly point out maybe one, more, which is I think we, are in a transitional moment where we're in parallel of in a sense, moments of overreliance and, and even over compliance with some of these systems in their operation. 00:51:57.000 --> 00:52:18.000 Where operators could either those who are using these systems could either become descaled or have some of their professional kind of judgment, dolled a bit and and I do think that we need to really be aware that at the end of the day these are these systems even the like very impressive conversational generative AI systems. 00:52:18.000 --> 00:52:26.000 These are if you will, supports or evidence-based human judgment, especially in the public sector for the public interest. 00:52:26.000 --> 00:52:37.000 And so I think we really need to be aware that, these systems should augment our capacities from a human centered point of view rather than displace or, the skill. 00:52:37.000 --> 00:52:39.000 Our agency. 00:52:39.000 --> 00:52:46.000 Yeah, that's such an important point, David, and there's a lot of conversation happening in the chat around the importance of the human review and human judgment and at what stage of the process does it does it play through. 00:52:46.000 --> 00:52:54.000 Rebecca, you spoke about transparency, in, in your previous, answer. 00:52:54.000 --> 00:53:12.000 Now, how can governments ensure transparency and accountability in AI algorithms use for public decision making. And especially from a, you know, what policies of mechanisms can be put in place to do that. 00:53:12.000 --> 00:53:35.000 Yeah, so this is incredibly important. So algorithmic decision-making tools are used in very consequential areas of our lives and echoing some of what David was just mentioning, you know, the the people most at risk of harm are the ones who are the most vulnerable and these tend to be some of the same people who would be underrepresented in some of the training data for these large language models. 00:53:35.000 --> 00:53:36.000 So it's especially important that we concern ourselves with ensuring transparency and accountability and fairness. You know, and I think about in ensuring transparency and accountability and fairness. 00:53:36.000 --> 00:53:47.000 And I think about in the EU AI Act, there's such an emphasis on and accountability and fairness. 00:53:47.000 --> 00:53:56.000 You know, and I think about in the EU AI Act, there's such an emphasis on requiring additional governance, there's such an emphasis on requiring additional governance mechanisms and safeguards for AI that would be used in high risk areas like accessing public services and requiring additional governance mechanisms and safeguards for AI that would be used in high risk areas, like accessing public services 00:53:56.000 --> 00:54:07.000 and benefits and things like law enforcement, migration and border control management, and in the United States criminal justice system, for instance, you know, these automated risk assessment instruments have been used to generate scores. 00:54:07.000 --> 00:54:37.000 To predict the risk of recidivism in order to determine pre-trial detention decisions and things like, you know, making, decisions, around, sentencing, or, parole, or, predicting, hotspots, for, future, crime targeting policies and services and providing services based on RIF profiles in child protection and in providing employment services or these automated decision systems are used in predictive 00:54:42.000 --> 00:54:58.000 analytics in social services like the predicting risks, to children from abuse and neglect, in child protection, or predicting welfare or tax fraud in compliance systems or in employment services, you know, predicting long term unemployment risk, things like that. 00:54:58.000 --> 00:55:01.000 So while these are important, it's it's crucial that we use extreme care and how we utilize these tools. 00:55:01.000 --> 00:55:20.000 And, you know, in order to minimize harm. So in addition to auditing these AI systems, the reasons for decisions need to be explainable, there should be a transparent process for appealing a decision and decisions. 00:55:20.000 --> 00:55:33.000 Should be subject to human review. So in the US, President Biden addressed a lot of these types of concerns with October's executive order on safe secure and trust for the AI. 00:55:33.000 --> 00:55:40.000 And then we have frameworks like the NIST AI risk management framework that do provide helpful guidance for dealing with these types of risks. 00:55:40.000 --> 00:55:49.000 But honestly, I also think this is a case where it's very important. A strategy for addressing algorithmic harm is ensuring that you have a diverse mix of people at the table in the first place. 00:55:49.000 --> 00:56:04.000 You know, that the people designing, developing, deploying, procuring or governing, procuring or governing these tools have a diverse set of backgrounds come from a diverse set of backgrounds, these tools have a diverse set of backgrounds, come from a diverse set of disciplines and have a diverse set of perspectives. 00:56:04.000 --> 00:56:09.000 So that's one of the central goals of all tech is human, the organization where I'm an executive director. 00:56:09.000 --> 00:56:23.000 So we do develop the responsible tech talent pipeline and we include public interest technologists and responsible AI practitioners and that's because it's so important to attract a diverse array of talent. 00:56:23.000 --> 00:56:32.000 And we like to talk about empowering the problem finders instead of just the problem solvers. So these are some things that are really important. 00:56:32.000 --> 00:56:40.000 Thank you. Reckon this is an interesting, analogy given to them spreadsheets, came up and sort of how that evolved. 00:56:40.000 --> 00:56:56.000 I think go down to sort of the the practical implications of these 4 public servants, right? So, Georgia, are there checklists or ethical guidelines or questions that public servants can refer to when they are applying AI solutions in their work. 00:56:56.000 --> 00:57:05.000 And more importantly, are there tools that help you interrogate the analysis of decisions produced by AI tools. 00:57:05.000 --> 00:57:13.000 And, thank, thank you very much for the question. I really would like to, underline and I agree with what Rebecca said. 00:57:13.000 --> 00:57:28.000 Public servants, so public sector absolutely need to be aware of the opportunity and risk that whatever system or whatever AI project they're doing implies. 00:57:28.000 --> 00:57:40.000 So I saw a lot of comments into the chat about awareness, education, and I think that awareness, education, and I think that awareness, education, and I think that awareness, education, and I think that awareness, education, and I think that awareness, education, and I think that awareness, education, and I think that awareness allocation and David, you were mentioning this is important for the final, see the final user, so 00:57:40.000 --> 00:57:50.000 the citizen, but is also very, very important for the public servant. They are the one that then finally are going to use this system with with citizens. 00:57:50.000 --> 00:57:58.000 But so going to your question, yes. There is a set of questions that public servant should ask when they decide what AI system to use or whether or if they are assessing an AI project. 00:57:58.000 --> 00:58:13.000 I give you some. For instance, they need to ask themselves whether they have consent to use the data for AI training. 00:58:13.000 --> 00:58:25.000 Does the training data contain elements that may lead to bias? Outcomes and of course is a very difficult question because not always the set of data is well known by by by the public servant that used this. 00:58:25.000 --> 00:58:45.000 Another very important question that then connects to another point has the senior board of director whatever whatever the body within this institution is or name is has enough knowledge about this AI project, does they have all the information that they need in order to assess whether we want to go in that direction. 00:58:45.000 --> 00:58:54.000 Or not. And then has this AI model been rigorously, let's say, treated in analyzed, how many time is going to be reassessed. 00:58:54.000 --> 00:59:05.000 So there are a set of questions. Then there is a second point in my view. That I mutate or I take from what I see at Google. 00:59:05.000 --> 00:59:17.000 So at Google, the company decided rightly so in 2,018 to have a set of AI ethical principle. 00:59:17.000 --> 00:59:34.000 That basically inform all the decisions about the product, any decision that is AI related to Google. Well, any institution should have this same or other but a set of principles that they abide to and Rebecca was mentioning this when she was talking about procurement and other things. 00:59:34.000 --> 00:59:40.000 But it's important that then this guidelines or this principle are applied at all level of the institution. So from the board of director. 00:59:40.000 --> 00:59:56.000 You know, let's imagine that we are talking about local office to the last local office. Because it's the entire machine that needs to go according to those guidelines. 00:59:56.000 --> 01:00:11.000 And third, but not less relevant. When the institutions interact or decide to interact with, a cloud operator or a AI operator is very, very important that they ask themselves. 01:00:11.000 --> 01:00:20.000 With whether this operator has security and privacy as And utmost, let's say, in importance or level. 01:00:20.000 --> 01:00:31.000 And so the services that they buy. A buy to the same principles that they apply to the institutions. 01:00:31.000 --> 01:00:32.000 The the last mile thinking about sort of this thing, how does this play across different levels of government? Thinking about sort of this thing, how does this play across different levels of government? 01:00:32.000 --> 01:00:38.000 It's not just a central city, you know, beyond that and how does this play across different levels of government, not just in the central city, you know, beyond that and how does that actually impact? That's such an important thing to consider. 01:00:38.000 --> 01:00:48.000 Now I'm very mindful that there is, an amazing number of questions and chats. 01:00:48.000 --> 01:00:55.000 I'm gonna try and do 2 more questions and then kind of move straight to what's being on the chat but David this is something that came up in a couple of comments on the chat as well about you know AI. 01:00:55.000 --> 01:01:19.000 I exaggerate exasperating the impact on vulnerable populations and so on. So the question for you is how should public servants approach the use of AI in sensitive areas such as social services or justice, where the ethical stakes are particularly higher. 01:01:19.000 --> 01:01:28.000 I mean, I think one first and foremost for me, it's having and going into any potential use case. 01:01:28.000 --> 01:01:58.000 Even at the horizon scanning stage so at the very early kind of consideration of even using an AI systems at that stage really thinking about the context of the populations that will be affected by the potential use case and considering also you know, the nature of the data is is the data likely to replicate, you know, patterns of bias is the will the use, unfairly target certain populations that might have 01:02:03.000 --> 01:02:07.000 vulnerabilities or might be historically marginalized. I think for me the important thing is to start at the very beginning and to put in to considering the use case. 01:02:07.000 --> 01:02:30.000 Also consideration of the ecosystem dynamics and and how much you know are there longer term patterns of socio historical injustice structural injustice and discrimination that will likely be kind of baked in or replicated in the system. 01:02:30.000 --> 01:02:36.000 And I think that's that's I would say a society centered approach to thinking about AI. 01:02:36.000 --> 01:02:39.000 You know oftentimes that you know I think it's called Maslow's Hammer. 01:02:39.000 --> 01:02:43.000 It's a bias where it's, you know, that every hammer always looks for a nail. 01:02:43.000 --> 01:02:49.000 Solutions always looking for a problem. And, and actually this, this lines up with this, like a comment or 2 about the kind of FOMO driven, perspective on using right AI systems. 01:02:49.000 --> 01:03:04.000 No, we should actually when we think about that that early decision point, we have to think about yes, what are the public's needs? 01:03:04.000 --> 01:03:05.000 How can this serve the public interest in an optimal way? But we also have to think, is is AI the right solution? 01:03:05.000 --> 01:03:20.000 Like should we be kind of either automating this process or using us to statistically based system to influence decision making in this particular area. 01:03:20.000 --> 01:03:26.000 So I think the I think I would say early entry for me is the primary kind of stopgap there. 01:03:26.000 --> 01:03:38.000 Yeah. And that's probably somewhere where, I think it was Rebecca mentioned about ensuring diversity of, you know, behind people who are designing it, delivering it, reviewing it can potentially help with that as well. 01:03:38.000 --> 01:03:53.000 Rebecca, a question for you before we move on to opening the floor. And what level of training should be provided to government staff on how to use these tools in their work and this is something that we wrap up with every day as well as be political. 01:03:53.000 --> 01:03:55.000 So I'd love to hear you answer. 01:03:55.000 --> 01:04:05.000 Yeah, it is so important. I think basic AI literacy is crucial, but, as is, you know, understanding how to accurately prompt, do the prompt writing. 01:04:05.000 --> 01:04:11.000 But I think the most important thing that, that needs to be considered when it comes to training. 01:04:11.000 --> 01:04:26.000 Especially when you're providing this training to government staff is specific and clear guidance on how to use generative AI tools responsibly because you will have some people who will be very reluctant to try a new tool or hesitant. 01:04:26.000 --> 01:04:28.000 They don't want to make a mistake or use it correctly. And then other people are already very comfortable and have been independently exploring these tools for a long time. 01:04:28.000 --> 01:04:50.000 But in ways that could potentially be risky to the institution or the organization. So we use the term shadow AI to refer to the employee use of their own personally accessible, to the employee use of their own personally accessible generative AI tools or apps on their own personally accessible Generative AI tools or apps on their own phones or devices. 01:04:50.000 --> 01:05:08.000 And that is used outside the view of their employer. And that's a risk no matter what sector it's being done in, but compounding this risk is that employees will tend to keep that to themselves and not disclose the use of the tools if the workplace culture either discourages it or doesn't have clear guidance around how to use these tools. 01:05:08.000 --> 01:05:28.000 So some people might feel, Like they can't disclose that they're using them or it's a crutch or they're cheating by using it so they don't disclose unless there's this clear expectation by the organization that we do want to use these tools and have clear guidance on how to use the tools, in what ways we use these tools. 01:05:28.000 --> 01:05:38.000 So by providing very explicit instructions on how to use them and which tools are sanctioned and in what instances employees can be empowered to use them in ways that are going to be helpful for the organization. 01:05:38.000 --> 01:05:53.000 And are more likely to be transparent if anything does go wrong. So the guidance that we need to provide should be transparent if anything does go wrong. 01:05:53.000 --> 01:05:58.000 So the guidance that we need to provide should be explicitly empowering public servants in the workforce to use Generative AI on how to manage hallucinations and accuracy issues, for instance. 01:05:58.000 --> 01:06:24.000 So every statement of fact needs to be verified. We need to talk about how we understand bias when it comes to generative AI and public servants should know that the foundation models underlying these chat bots were trained on massive data sets that were scraped from large swaps of the open internet and and should understand how that will impact the content. 01:06:24.000 --> 01:06:34.000 And then also, generative AI tools specifically are built to surface the most possible outputs and that will default to norms and can tend to pull in stereotypes. 01:06:34.000 --> 01:06:42.000 So we have to keep an eye out for inadvertent bias in these types of outputs. But then also how do we deal with privacy and sensitive information that we put into prompt? 01:06:42.000 --> 01:06:46.000 So employees and public servants need to understand the difference in the privacy settings on their chat GPT pro on their phone that they pay 20 bucks a month for. 01:06:46.000 --> 01:07:02.000 Versus you know an enterprise version of an LLM that their organization might have API access to and has fine-tuned with the institution's own proprietary data. 01:07:02.000 --> 01:07:15.000 So we need to understand what info can we put into the prompts. And these are the types of very practical things that government staff should be trained on for use in their own departments and on their own teams. 01:07:15.000 --> 01:07:16.000 Super valuable and I think these are the words that we hear quite often in terms of guidelines like are they clear? 01:07:16.000 --> 01:07:25.000 Are they simple? Are they specific? And are they timely? Because they come too late, too slow, it's not sort of super valuable for them as well. 01:07:25.000 --> 01:07:38.000 Thank you so much for answering the first round of questions. What an amazing audience. Please keep sharing any questions you have for today's speakers in the chat. 01:07:38.000 --> 01:07:42.000 We've got a number of P submitted ones as well, but it's always great to mix them up with the with the live questions. 01:07:42.000 --> 01:07:50.000 We're going to head into Q&A now and we'll give you a minute or 2 to type your questions. 01:07:50.000 --> 01:07:51.000 In the meantime, you wanted to remind you of a minute or 2 to type your questions. In the meantime, you wanted to remind you of our AI in government community. 01:07:51.000 --> 01:08:07.000 Communities are really exciting new feature on our apolitical platform. You can use them to engage, share and learn the public servants interested in similar topics or working in counterpart roles around the world. 01:08:07.000 --> 01:08:13.000 AI and government was our first community, not surprisingly. And it now sits at about 700 members strong and it's really vibrant. 01:08:13.000 --> 01:08:18.000 So you'll be able to find the community on the whole page and my colleague should be posting a link and we'll send it to you in the event follow-up email as well. 01:08:18.000 --> 01:08:43.000 So jumping straight to the questions, starting with a few that we've received today. Rebecca, are there practical approaches that promote collaboration between government departments that are responsible for regulation and those that are responsible for innovation, how do we make sure that these don't get siloed? 01:08:43.000 --> 01:08:54.000 Yeah, this is an excellent question. I think this Tyling tends to happen in regardless of the sector that you're involved in and it's something that at all tech is human, we do grapple with this a lot. 01:08:54.000 --> 01:08:59.000 And one of the things that we try to do as an organization is provide an agnostic space for different stakeholders to be able to come together and share ideas. 01:08:59.000 --> 01:09:17.000 I think practically speaking it's very difficult to do that. And I think, you know, building that in, to the structures of the organizations that we work in is really important. 01:09:17.000 --> 01:09:38.000 Finding ways to build specific collaborations maybe having you know a committee that can be formed pulling people from different spaces together but i think one of the most important things you can do you know government itself might not be the place to provide that kind of collaboration and I think you know there are civil society. 01:09:38.000 --> 01:09:50.000 And I think, you know, there are civil society organizations like all tech as human that are built specifically you know, there are civil society organizations like all tech as human that are built specifically organizations like all tech as human that are built specifically to bring stakeholders, and that are built specifically to bring stakeholders together from across different sectors and across different siloed, to bring stakeholders together from across different sectors and across different, siloed, 01:09:50.000 --> 01:09:57.000 components within each of those sectors. So an a political is another good one finding a space to be able to, to collaborate in that way. 01:09:57.000 --> 01:10:04.000 And then also just doing a lot of reading, understanding what the innovators are doing and what the policy makers are doing. 01:10:04.000 --> 01:10:14.000 And, just making sure that you're, taking a proactive, role in, understanding, each, of, these, different, stakeholders. 01:10:14.000 --> 01:10:22.000 Thank you, Rebecca. I feel like I'm in I'm in a, I'm determined to get through these questions, it's like a rapid fire round. 01:10:22.000 --> 01:10:23.000 David, there's a question I think it was by Scott who's doing some incredible work in South Africa as well. 01:10:23.000 --> 01:10:35.000 How do we stimulate governments, especially in lower and middle income countries to be responsive in AI regulations and engagement. 01:10:35.000 --> 01:10:45.000 At the moment the responsiveness will always remain behind the curve as the tech develops faster than the governments will ever respond. 01:10:45.000 --> 01:10:51.000 Well, I mean, I think we're definitely at a at a point in time where there are stronger networks that are being built. 01:10:51.000 --> 01:11:01.000 In terms of the international community that, are promoting more proactive approaches to responsible AI innovation. 01:11:01.000 --> 01:11:31.000 So just to mention one right now we're right at the point of doing a big campaign to do readiness assessment methodology, the readiness assessment methodology of UNESCO, which is this is in this is a program that UNESCO launched this part of this ethics recommendation that it had approved in 2,021 where countries all over the world right now it's over 50 countries in in 01:11:31.000 --> 01:12:00.000 Africa, South and Central America, Europe, I mean just all over are basically undergoing some quite diligent, socio technical analysis of what are the conditions across government and, and in, in the environments in which government is embedded in the economy with regard to education in all of these worlds, like how how are the conditions within a particular country enabling responsible uptake. 01:12:00.000 --> 01:12:16.000 Of AI innovation. And the more attention that we pay to this. Through this sort of. Knowledge transfer lens where, you know, different governments are sharing information, learning from each other, from the same regions and even different regions across the world, the more proactive governments can be. 01:12:16.000 --> 01:12:36.000 And I do see really strong networks building, in all parts of Africa right now. You know, it's it's a I think we are at a kind of inflection point when it comes to the kind of concerted effort to not sort of, if you will fall behind on this. 01:12:36.000 --> 01:12:50.000 And, and I do think it's gonna be, has to be a kind of inclusive and cooperative effort among many, different nation states. 01:12:50.000 --> 01:12:57.000 And so I do see, I think one of the for for that is definitely this. 01:12:57.000 --> 01:12:58.000 And is it? No, no, please come in. 01:12:58.000 --> 01:13:16.000 Fuja, do you mind if I add something to what David? Because it was very interesting and, you know, with the, I'm working, as I said, on with the, accountress and I've been amazed, read the, by how emerging markets are ready. 01:13:16.000 --> 01:13:32.000 Interested. To, embrace the eye opportunity but also aware of the risk that are let's say present and so very willing to embrace AI tools in the best possible way. 01:13:32.000 --> 01:13:47.000 So there is a difference that I see for instance when you talk to European public sector and when you talk to emerging market public sector or institutions that is the latter are much more about yes let's understand how to use it. 01:13:47.000 --> 01:13:58.000 Let's understand what are the opportunities. Let's understand how to work and cooperate for instance through UNESCO on, you know, making this something that is positive for everyone. 01:13:58.000 --> 01:14:13.000 But there is a lot of activity. There is a lot of attention and interest. So I don't think that is that is that is it that is a fact that emerging markets will adopt this later or in a different way. 01:14:13.000 --> 01:14:19.000 I really don't think so. Potentially will be the opposite. 01:14:19.000 --> 01:14:27.000 I couldn't agree more, Georgia. We just came back from the, the world. Government summit and our CEO had the same thought that the whole approach to it was just so much different than what we hear from our colleagues here in this side of the balls. 01:14:27.000 --> 01:14:49.000 Georgia, actually while you're still on mute, unmuted. And the one question for you and this is by does, I think, while I'm intrigued by the possibilities of AIs capabilities, however AI hallucinations or disinformation and further AI sentience are red flags. 01:14:49.000 --> 01:14:56.000 How can we guarantee these problems will not create chaos in government? 01:14:56.000 --> 01:15:16.000 And this is a very good question and not just a very good question, but a very timely question because we all know that we are at least in Europe getting closer to the European elections, which will be a moment in which of course this information misinformation will be particularly important. 01:15:16.000 --> 01:15:27.000 And there are several, excuse me, activities that companies Google, but everyone that is then adopting AI can can put in place. 01:15:27.000 --> 01:15:33.000 Definitely the banking campaign are very important. So try to provide as much as possible with the wider set of information. 01:15:33.000 --> 01:15:47.000 Is is very important. Second point is, and I think that Rebecca, you touch on this before, there is a responsibility for companies such as Google to provide 01:15:47.000 --> 01:15:55.000 So I've received a letter on the survey from. Okay. 01:15:55.000 --> 01:15:56.000 You can go ahead. 01:15:56.000 --> 01:16:06.000 Okay, to provide tools to identify with our fully with our synthetic, content with our non synthetic content. 01:16:06.000 --> 01:16:19.000 I give it a stupid example, but today when you are on a Gemini before it was called barred you can click and ask what is about this image to understand what that what would that image is about where it comes up from and have more context. 01:16:19.000 --> 01:16:49.000 So I think that of course people need to use their critical thinking and in the chat there were a lot of people talking about critical thinking but there are tools that operators can put in place in order to help people understand at least the roots of a certain content, especially if they are scientific contents or not. 01:16:49.000 --> 01:16:58.000 Thank you, Julia. As you was speaking, I was thinking I wish we had an AI to mute and unmute people having not how you school without me, for public good. 01:16:58.000 --> 01:17:05.000 But just going on to the question, this one, Rebecca, perhaps for you, and please be able to jump in David and Georgia if you would like to as well. 01:17:05.000 --> 01:17:19.000 Reha how should we give the public servants the information to defend their decision based on an AI model? So it's evidence, and this one's from Steve. 01:17:19.000 --> 01:17:29.000 Yeah, lots of good questions today. Yeah, I think It is just incredibly important as we're using these tools. 01:17:29.000 --> 01:17:40.000 We just always need to understand what the limitations are. So that when it comes to defending a decision or defending something that you put out there that you have already done the work to understand. 01:17:40.000 --> 01:17:45.000 Just where in which instances these tools are going to be useful and then what they're not able to do well right now. 01:17:45.000 --> 01:18:04.000 And I think of instances like, you know, There's just a lot of cases where people use these tools and and maybe overestimate what they are able to do at this point. 01:18:04.000 --> 01:18:23.000 I've seen a number of instances where for instance a lawyer will use chat VPT and it will hallucinate cases and you know there was there was a lawyer that was fined thousands of dollars because you know he had utilize these cases to a judge and had to had to pay fines for it. 01:18:23.000 --> 01:18:29.000 In this instance, you know, they would not be able to defend what the AI tool came up with. 01:18:29.000 --> 01:18:41.000 So I think first and foremost, it's understanding what these tools can do. And then I would say also, just ensuring that the organization that you're working for has. 01:18:41.000 --> 01:18:57.000 Guidelines to to back up. When you're using these tools and in what ways it's appropriate and happy to have either David or Georgia jump in on how you might see that as well. 01:18:57.000 --> 01:19:08.000 Yeah, I'm happy to jump in quickly just say to add on on top of that that it's also really helpful to have a even a high level understanding of how these systems work. 01:19:08.000 --> 01:19:19.000 So For instance, a lot of these machine learning models are just simply working off of large data sets and and generating statistical inference. 01:19:19.000 --> 01:19:30.000 So these are basically correlations, they're identifying correlations out there in the world and and then we know that correlation doesn't equal causation. 01:19:30.000 --> 01:19:39.000 And so if there are kind of lurking factors that haven't been picked up by a dataset that are that are kind of shielded from a prediction model. 01:19:39.000 --> 01:20:03.000 Those models are not gonna sort of effectively predict in certain circumstances. And if you understand that these are not kind of, they're not causally reasoning, they're taking us a kind of a statistical average basically and extrapolating, then I think we can, you know, handle a little bit more responsibly how to use those that information it's inductive it's not you know gonna be a causal 01:20:03.000 --> 01:20:27.000 base but quickly the other thing is on these large language models these these you know chachi peas these these frontier models you know these are very sophisticated pattern matchers they're predicting word sequences and they've been trained using reinforcement learning with human feedback to want to please us or to to be oriented to trying to answer your questions the best way possible. 01:20:27.000 --> 01:20:33.000 And that's why they can fabulate. That's why these systems will, you know, they're not they don't have access to ground truth, right? 01:20:33.000 --> 01:20:48.000 And so they will, they'll produce references if you ask them for it because they want to make you, you know, you've been trained to satisfy you and so we really have to take you with a grain of salt that there will be hallucination, there will be the generation of disinformation from these systems because they simply aren't. 01:20:48.000 --> 01:21:00.000 Designed to generate consistent truth as such. They're they're they're being trained to make accurate predictions, but those predictions are not linked. 01:21:00.000 --> 01:21:07.000 As such to a kind of accessible ground truth. 01:21:07.000 --> 01:21:17.000 That's, it's, just the whole thing about there was a conversation about the system wanting to please you and hence doing these things, such a insight into just how these things emerge as well. 01:21:17.000 --> 01:21:31.000 There's a really interesting question here. I'm gonna leave it open to which one of you would like to answer this but how can citizen feedback be incorporated into AI policy and practice and I think this will become more and more important. 01:21:31.000 --> 01:21:37.000 Any of you would like to have a take a take a step in it? 01:21:37.000 --> 01:21:39.000 Good. Rebecca first, but I have something to say too. 01:21:39.000 --> 01:21:48.000 Okay, I briefly, I'll just say, you know, one thing that I think is wonderful about, you know, in the United States, the, there are plenty of opportunities to provide feedback. 01:21:48.000 --> 01:22:09.000 Often, you know, there will be public calls for response and individuals or organizations can can provide feedback and a lot of times that's the best way that the governments can access information from a lot of different stakeholders. 01:22:09.000 --> 01:22:21.000 So, you know, there will be calls for the public to get involved. And so I think keeping an eye out for those types of opportunities and there are some organizations that will track those opportunities. 01:22:21.000 --> 01:22:25.000 The Center for AI and Digital Policy is one that really is helpful for, for pulling together responses from various people. 01:22:25.000 --> 01:22:39.000 The law firm that I used to work for, Eisenberg and BOMB in New York, we we put together all kinds of comments when it came to things around, you know, New York City and surveillance tech tools. 01:22:39.000 --> 01:23:02.000 And different things like that. So I just want to say those opportunities are out there for different stakeholders to get involved into and to to either utilize research or you know surveys or just their own from their own perspective how is how is the policy going to impact them and that can help shape the policy going forward. 01:23:02.000 --> 01:23:05.000 Thank you. Rebecca. There's a lot of love for all the speakers in the in the chat about how valuable this has been. 01:23:05.000 --> 01:23:16.000 And as we get closer to time for goodbyes, we'd love to get some closing guidance for today's attendees. 01:23:16.000 --> 01:23:22.000 Now, I'm gonna this is gonna be a 10 s round. So we're sharpening your answers for faster responses. 01:23:22.000 --> 01:23:27.000 But what is the one closing piece of advice you'd give to public servants who are conscious of using AI tools safely, responsibly, and ethically. 01:23:27.000 --> 01:23:38.000 We will get 15 Si think it's, you need to, that at least. Jojo, why don't we start with you? 01:23:38.000 --> 01:23:39.000 After the elevator pitch, the closing remarks in 15 s. This is a challenge, I mean, I'm swaying. 01:23:39.000 --> 01:23:48.000 All right. 01:23:48.000 --> 01:23:52.000 I think that the best piece of advice is to make sure that you have very clear understanding of the system that you're using. 01:23:52.000 --> 01:24:05.000 The data that you input and the principle and the guidelines that you want to follow, the ethical guidance that you want to follow in order to ensure that you provide a good AI product. 01:24:05.000 --> 01:24:15.000 Or, service to your citizens. 01:24:15.000 --> 01:24:18.000 Thank you. Georgia. David. 01:24:18.000 --> 01:24:30.000 So just I want to just close with the importance of social license, democratic governance and public consent and just say this follows on less root thing Rebecca said. 01:24:30.000 --> 01:24:32.000 So, you know, there's this, there's this, There's this dictum, nothing of nothing about me without me, right? 01:24:32.000 --> 01:24:43.000 And this is the idea that the unique, you know, the public especially in public sector use cases, needs to be involved. 01:24:43.000 --> 01:24:49.000 We need to climb the ladder of participation, which is share sharing Ernest's latter participation. 01:24:49.000 --> 01:25:04.000 And really offer opportunities for stakeholder participation in the code design and consultation. And as you go across from the very beginning to very end of the AI design development, employment life cycle, and there's been some great work done on really taking a structured approach to this. 01:25:04.000 --> 01:25:14.000 I put out really quick, I'm gonna shamelessly plug the, I put a, I put a link to our public sector, AI ethics and governance and practice guidance. 01:25:14.000 --> 01:25:21.000 And, and we have AI sustainability in there with a whole stakeholder engagement process codified. So and that's also used in the Council of Europe and UNESCO now. 01:25:21.000 --> 01:25:30.000 So I would say go for a go for stakeholder engagement first and foremost. 01:25:30.000 --> 01:25:35.000 Thank you, David. We love shameless place. Rebecca, you turn. 01:25:35.000 --> 01:25:43.000 Yeah, I could not agree more with David. Yeah, I would say you don't have to reinvent the wheel and you don't have to go it alone. 01:25:43.000 --> 01:25:53.000 Make sure that you are engaging with the work that's already been done in the space like David was was talking about and that you can join communities of people who grapple with these issues who have deeply thought about this already. 01:25:53.000 --> 01:26:20.000 And, and so that would be, you know, all tech is human. Alan Turing Institute, you know, we, we do have these communities available for people to join forces and to understand the work that's already been done in this space and there are many resources we actually have hundreds of resources posted on all tech and humans website that you can dig into and and understand more about. 01:26:20.000 --> 01:26:22.000 The space. 01:26:22.000 --> 01:26:25.000 Thank you so much. Thank you to all our speakers and we're a wonderful audience. It's the great, great, great discussion. 01:26:25.000 --> 01:26:36.000 The team is going to pull up a very quick poll to find out how you felt about today, which will help us to continue improving our content on AI in government. 01:26:36.000 --> 01:26:38.000 And we really appreciate your responses. So the poll is going to come up now on a scale of one to 10 rank agreement with this statement. 01:26:38.000 --> 01:26:52.000 This event was a valuable use of my time. One being completely disagree, hopefully not, and 10 being completely agreed. 01:26:52.000 --> 01:27:00.000 Lastly, we'd like to remind you about the AI and government community on apolitical. There are some great questions being asked on there and insights being shared. 01:27:00.000 --> 01:27:04.000 So we'd love for all of you to be getting involved and give you thoughts and answers to them. 01:27:04.000 --> 01:27:11.000 We're going to post one right now asking for your number one takeaway. So please do go to the community and post your answers. 01:27:11.000 --> 01:27:18.000 That's all we have time for today. We hope you found it helpful. And we want to say a huge thank you once again to everyone who's made today happen. Thank you everyone. 01:27:18.000 --> 01:27:23.000 Have a great day.