Putting the Human in the Center of Your AI Efforts - Tiankai Feng

In this talk, the speaker delves into the human aspects of artificial intelligence, addressing the polarized views surrounding it in society. They emphasize the importance of focusing on the mindset, skill set, and behavior of individuals in AI adoption and usage, highlighting that human factors significantly contribute to the success or failure of AI initiatives in enterprises. The speaker introduces the concept of the five C's of humanizing AI strategy: competence, collaboration, communication, creativity, and conscience, which serve as essential traits for effectively integrating AI into organizations. They argue that creating a balance between technological advancements and ethical considerations is crucial for fostering a positive relationship between humans and AI. The talk also covers the need for effective governance structures and clear communication strategies to align AI efforts with business goals and ethical standards.

It's great to be here. I'm very excited to be here. Not only because this is my first time visiting India and Bangalore. So, I've been exploring the city a little bit. It's very nice to be here, but also that I get to talk to you all about um a topic that's very close to my heart, which is the human side of AI. And the reason why I'm

so passionate about it is that I feel like AI has a very polarizing aspect nowadays in society. Right? I know in this community today here at GitSt, um we're all very passionate about technology and that makes us all very enthusiastic about AI. But if you look a little more outside and into society, um there are a lot of resistance about it, right? There's a lot of people

that are against it and that feel like it's threatening their life, it's threatening their existence and their jobs etc etc. And the reason why I think that is is because in a way the extreme voices are usually the loudest in the world right and thanks to technology all the extreme voices that are either very much for or very much against AI are all screaming at us and

everyone is confused what is happening and what we should think about it. So I want to give you a little bit of a more nuanced approach especially in an enterprise context of how we can address the human side of AI in the right way. So we make human beings actually the center of AI efforts and not only a side product or something that we see as a

weakness of any AI strategy. So a quick introduction about myself. I'm currently working as a director for data and AI strategy at Thoughtworks. Uh some of you might have already seen some talks from my colleagues like um just earlier in this hall from Vana for example about the seven trends. Um I work in a global role. So I'm based in Germany originally um and I flew in

also from Germany for this one. Um and basically in my global role as being a data and AI strategist I work on the frameworks around how we solve strategic problems around data and AI right that means governance that means operating models that means architecture all these kind of things that basically helps those uh senior decision makers to make the right choices and to turn on change management

in the right way. I'm also engaged in the data management community. I'm the vice president of Dharma Germany, which is the German chapter of Dharma International, a nonprofit organization for data management professionals. So I'm there in the leadership driving the community for it. I've given various talks about AI, especially the human side of it. For example, I also given a TEDx talk about it. And I've been

writing two books about it called humanizing data strategy and humanizing AI strategy. Also in May, you see that here in the bottom right, I'm releasing a children's book about AI um called Robbie Just wants to help because that is basically reflect my passion to also help the next generation get used to and learn a little bit more about how to behave and how to basically exist in

this new world where AI is everywhere. Um I like to use my musical talents as well to make data and AI a little bit more approachable. So I write both parodies as well as original songs about data and AI and I usually use them also just to break the ice around certain topics like the digital analytics anthem and governors of data but more on the music a

little bit later on. So let's think about what is happening with AI right so um we basically as I mentioned before see all these contradicting messages around it in the news right on the one hand side AI is everywhere but on the other hand we see a lot of failures and challenges happening so first for example enterprises are adopting AI more and more we see more and

more AI generated pictures in the internet every day and tech giants are all investing lots of money into actually building AI out and making money out of it on the other hand we see a lot of news like 87% of AI projects never making it into production right that 81% of workers still not using AI tools in the day-to-day and that 4% only of companies are creating

sustain substantial value with AI right what that tells me is that we have a lot of very enthusiastic people like the ones in the room that are experimenting with it and trying it out for their own productivity reasons right but that doesn't move the needle on organizational scale yet right so we're doing it a not for software development and that is all great but something for something

to move the needle it has to actually change the entire business process of an organization to actually move the needle on the profit and loss right so why is this happening right why is AI actually failing and my theory is that actually um human factors are the root cause of most of these failures right so if we think about it we have a lack of competence often

we have often ineffective collaboration we have missing accountability and we have unevenly distributed information and all of that leading to humans making actually bad decisions and then they make technology being the scapegoat part and say AI doesn't work it's the technologies fault when in reality all of these factors have could have been avoided if people would just come together and talk more and collaborate more right so

it comes down to a little bit of a cultural aspect too and how people actually work together in organizations so to break it down what I think Human transicity is really all about is about combining three things, right? It's mindset, skill set, and behavior. So mindset is really about wanting to do the right things. To actually do that, you have to first teach people what the right

things are. And then you need to motivate them to want to do the right things in the first place, right? Skill set is not only do they know what the right things are, they know how to do them as well. and they have the right skills to actually do it. Um to then bring it into reality and then behavior of course to bring now what um they

can do and they want to do into reality and to actually behave accordingly and not just theoretically know about what they're supposed to do in the right way. So basically it's able to do the right things and then doing the right things the right way. Right? And when we think about AI strategies um right there's a lot of technological architecture and there's a lot about uh the

newest tools that we want to use about all of this but the human aspect should be somehow reflected in this too right so how what do we do about mindset what do we do about skill set and what do we do about behavior because in many ways if we don't care of this in a positive way and we don't encourage people in the right way then we

will very likely actually get the opposite negative reaction right basically people sabotaging AI efforts right not um like basically explicitly voicing that they're against it. All these kind of things would happen if we don't positively shine a light on it and to make everyone part of the journey and make them care about why they should be part of the journey and not just give them reasons to

be quiet and then basically be against it and find their own reasons. So um first let's think what an AI strategy even is right and um just like actually with a data strategy which I've focused on before um it's very hard to find that one definition that everyone uses of what that actually is but I kind of did my research right and I've basically combined the best

of from my point of view into one statement which is that an AI strategy is a long-term plan that defines the people process and technology ies to create, process and use AI to intentionally drive value in a meaningful, secure and transparent way. And the reason why I like this definition is because it implies three things, right? First, it's about the entire AI life cycle, right? Meaning, it's

not only about the people that are using AI. It's about the people that actually brought in AI, either bought certain licenses or tokens, but also those are building AI like for example machine learning engineers, etc., etc. um everyone is part of that life cycle, right? And not just in the end who's using it and what they should do about it. The second implication is that it's about

being valuedriven, right? So we're not doing it just for fun. I mean, it is fun to work with AI, but we shouldn't do it just for the fun. It should lead to some kind of business value in the end, right? When we're doing it in organizations. So making it valuedriven means there needs always needs to be a strong alignment with business goals and not just what we're

excited about. And lastly, it's about the human perception, right? So when we think about words like meaningful, secure, transparent, those adjectives mean different things to different people, right? I mean, we all agree that it's the right things. But I might have a much stricter definition of what secure is than all of you, for example, right? And to actually move away from that perception, we need to find

agreements what that actually means, right? And not only does it mean something like when we think about semantic layers and context and all of these things that is machine readable, it starts with human agreement first, right? Because if we humans even are understanding it differently, how could we make a machine understand us if we are all confusing basically um geni for example and models about what we

define with things. So moving away from just being subjective, we need to find a way to be objective about these kind of um adjectives and then move on from there. that could be in artifacts like strategy documents, policies, guardrails, all these kind of things. But we need to basically make it explicit and find an agreement of what that means. So this leads me to the framework of

what I like to call the five C's of humanizing AI strategy. And these five Cs are actually human traits, right? Those are things that we all have in us and they would you all usually want to do. It's just that somehow in organizations especially in technology will sometimes forget to apply it in a more intentional way and those five Cs are competence, collaboration, communication, creativity and conscience.

So let's go through it one by one, right? So competence is of course starting with AI literacy, right? But AI literacy not just as a generic buzzword but as something that we really take serious and that we need to adapt to different personas and employees at the organization because not everyone needs to know everything about AI right it's just about making it work for people who use

it in a certain way or who are interacting with AI in a certain way then we have collaboration which um in this new world of AI is not only about humanto human collaboration right like how do we as human beings work cross functionally to make AI work but It's also about human to AI, AI to AI and AI to human collaboration because in all directions we are

all now working together with agents with models etc etc and how does that collaboration in the future look like um communication in a similar way now that we have generative AI and we have LLMs etc there are multiple ways how we communicate right so we use it now to basically communicate in a conversational way which before was a lot in technical languages but now that communication can

be done in this way how do we actually deal with it in the right way, right? And how do we communicate the value? How do we communicate the implicit knowledge that we have in the right way to make it happen? Creativity is actually a fun thing because not only is human creativity the um human basis for organizational innovation, right? So basically um all of the innovation we

do comes from good ideas and usually those are human original ideas. How so how do we provide the right environment especially to experiment and innovate with AI as well but also generative AI can create new things for us too right so what is creativity in this new world when generative AI can be creative and we can be creative too so what is the difference between human creativity

and I and AI generated creativity and lastly conscience right which is our moral compass it's really about us in our hearts and in our brains knowing what the right things are our gut feeling about what the right things are to do and that makes it not only about compliance, right? Of course, we have a lot of regulations and a lot of laws about that we need to

comply with. It's also about how can we be proactive about it more, right? Because there are so many consequences that we are still not sure about what's might happen and especially thinking about malicious actors nowadays, right? How they can use to scam people at scale suddenly, right? For example, how do we do that? Like how do we deal with that, right? And the way we have to

think about it is that we need to use all of our expertise and think about what might happen. Not only what can go right, but also what can go wrong. And we find a balance between preventing things and making sure that things that already happened that were bad for them to not happen again. And that is what conscience is all about. So I'm going to go a

little bit deeper into each of those and what that means. Right? So let's start with competence. AI literacy, right? Um, again a buzz word that we have not only in regulations nowadays, but everyone talks about how everyone should be more AI literate and that's a great thing, but it does put pressure on people, right? Everyone suddenly feels like they need to read all the news and read

all of the tech blogs and be upskilled every day and it's just too much. No one can learn everything that's out there about AI, right? So when we think from an organizational context and not just from our excitement then we should think about what is actually the necessary amount of knowledge that everyone and different employees in our company needs to have right and what helps with that

to basically shape those different interaction types is having already established literacy frameworks right so this one is actually a European Union funded framework called the AI literacy framework that is basically suggesting that there are four main types of interacting with AI which are engaging with AI, creating with AI, managing AI and designing AI, right? So engaging with AI could be for example a marketing team need that

is using CRM tools with AI features more competently. Uh creating with AI might be a customer service agent that's using Genai to summarize tickets, right? Managing AI could be a compliance officer that's ensuring responsible and auditable usage of AI and designing AI which might be a machine learning engineer building AI models and evaluating the user experience itself. Right? And across everything of course you also have leadership.

They need to basically have a slice through all of those. They need to understand at least high level what happens in all of these different ways of interacting with it. But um all in all right the more you can think about how different roles are interacting with AI differently the more you can sharpen and you can tailor what they need to learn and what the upskilling program

looks like in the right way right so in the end I think it comes down to like basic AI literacy of like the dos and don'ts and what generally is possible and not possible and then all the way to have more detailed and more in-depth kind of trainings in a specific way to make sure that they are actually fulfilling the task in this world in the right

way and they're not distracted by all of the news and by all of the public things that we have about AI and are focused on what actually brings value to the organization. And that leads me to human in the loop, right? Basically another buzz word that I want to decode here a little bit because human in the loop is often positioned as this way of being um

is that still even necessary, right? Like if we have agents and everything, do we still need these human in the loops? And I think of course the human loop will evolve but I put it and classify it under competence because you need to have the right person at the right time right just a matter you have the wrong person right of course the human doesn't bring anything

right it's just it doesn't even know what it's reviewing and what is intercepting so what's the point and not the right time means you might make it too much right so then why even automate if you have a human reviewing everything you do it too little and then some things basically pass by and you don't know what happens and that risk becomes reality and some failures happen,

right? So in a way what I think about it is that we should just think about mobile structurally, right? So when we have processes, right, that are enabled by AI, we have basically a cascading path of how that works. The first one is a trigger point, right? What are the trigger points when a human review is actually necessary, right? And that can be threshold- based, that can

be um rulebased and all these kind of things that we imagine. Um but this is where basically we think okay this is when a mandatory human review is needed. So for example when fraud detection AI is flagging a transaction with a high risk score right then the risk is so high and it surpassed that um kind of risk score so a human reviewer now is needed because

AI is not sure actually what to do about it. it's put that process on hold right from there um there's a decision note right so the human reviewer it judges now based on the context and by that context I don't mean the context in an AI context window sense but in a way of in in basically human knowledge and institutional knowledge right people that work in fraud

detection they have a certain gut feeling and certain experiences they are not explicitly writing down they kind of just know already in what direction to investigate further and they kind of get to that uh review faster. Still sometimes um that requires more expertise, right? So uh for example, once the fraud analyst checks the contextual customer infos, it realizes that it needs to involve some other people as

well, right? So there's an escalation path, right? So if specific cases require expertise, you need to bring those people in as well. So for example, a compliance team now checks the suspicious customer address from that transaction that we have seen and how that works. And through all of that, you need to have a fallback mechanism too, right? Because you need cannot just stop the process entirely. You

might have an alternative. In this case, it might be about asking the customer to buy it through a different way and not through a digital transaction, but maybe go to a physical store to actually buy that product there in a safer way. Either way, right, all of those things are again good to train AI and to make better decisions there. So when we have basically all of

those things being logged then we can all make it part of traceability which again improves decision-m in an automated way. So in the future it can become less and less but nonetheless basically when we think about all the institutional knowledge and expertise we have we can think about markdown files and documenting all of it but it's very hard for us human beings to put all of the

implicit knowledge in our heads down on paper right like that when we think about it then we can do it right but if you're not thinking about it at this point in time you might not know that you should have even written it down so basically just automating and taking the human out of loop completely doesn't make sense if you have not actually transitioned there from having

the implicit knowledge from human beings that were before the experts to actually guide that AI more and more and that leads me also to career paths um in AI right so um I mean we all talk about it jokingly or half jokingly that AI is going to replace all of us right but I believe that with every technology that and every innovation we had um there might

be that sense of replacing people immediately. But actually, right, tasks are evolving, right? And in a sim, very simple way, we are moving away from human beings doing things to orchestrating and guiding things more, right? Because if AI can actually do everything for us and be automatic in that way, then we as human beings still need to be the ones who are orchestrating, guiding, and helping to

make it all happen, right? So, in that way, I actually think that AI pushes us to be more human, right? Because all of the let's say thoughtless tasks of doing repetitive stuff all the time and turning our brains down and just kind of do things all that is now basically delegated. It it leaves us with critical thinking right suddenly we are thinking about actually what should we

do what is the right thing to do and how can we guide these things in the right way and this is why I think there will be new jobs ideally right new jobs that actually are AI adjacent and are needed in this new world for example having AI product managers right just that when we have product owners for everything else of course we need that still as

human accountability for AI products as well right AI trainers might be also there right so those are subject matter experts that actually help to fine-tune models and to be better in instructions because again to the point before you need all of that implicit knowledge to be translated into instructions so AI can actually does better and reflects better what we as human experts have done before then there

might be ethics leads right because being responsible is not as easy as it sounds when you don't know what for what malicious acts intentional or unintentionally it might be used so having someone who's actually rooted in that ethics part I think can be really helpful to actually think a little bit ahead on how to make that work. And AI translators, right? Those that are bridging business and

technology needs, right? I think nowadays the gap is even bigger because technologically there's so much possible and we it's so advanced but the business is lagging behind but by not even knowing what is possible. So you have that um basically mismatch of business asking for ridiculous things when you know it can be much easier or it can be much more advanced than what is. So having this

bridges and translators actually is really important for the future as well. All right, that leads me to collaboration. So collaboration first let's talk about the humanto human collaboration, right? As with any technology and making value out of technology um it's always about how do humans cross functionally actually work and when you bring specific functions together there's often often tension field, right? And I feel like it's basically

between two main tangible ways we have either is very transactional right so you treat one function only as a service function right typically let's say there's a central AI team and there's a business team and the business team treats the AI team only as a service function that builds AI stuff for them then throw it over and then we're done right you have basically transaction finished and

then you have the coordination where everyone is decentralized and let's say all business funers have their own AI expert and they had the great idea to say, "Oh, we should know more about what we're working on." So, they all meet once per month and just talk about who's working on what, but they're not really working together, right? They're just keeping each other informed and keep on doing

their own thing. The middle ground of all of this to actually make it valuable is, I think, co-creation, right? Meaning that it's not only about I give you and you give me or let's keep each other informed and then we're good enough. It's actually let's work together and create things together in the right way because it actually implies three things that need to happen, right? One is

that we have shared goals and a shared purpose, right? We actually want to use AI for the same goals and it helps us both functions to actually drive value. Then we have clear contributions because every team and every individual has different expertises and responsibilities to contribute to those goals and you have the strong commitment right you formalize it as well. basically either then OKRs and product plans

and performance reviews everything there you make it part of the work to basically be co-creational and with that um you never basically go in that situation anymore hopefully where you only complain about what's going wrong and why it's the other side's uh fault of it not working you actually are focused on solutions right we want to achieve that same goal so instead of fingerpointing whose fault it

is that it's not working it's actually about how can we achieve this together better and you basically are driven by solutions and not just problems one second right and that leads me to governance right governance which I know is a triggering word for many because governance always means it's the police coming and now no one move and everyone please stop what they're doing but I mean it

actually in a way of being more lean right and of it being balanced between centralized guardrails and automated guardrails versus actually autonomy of business functions to drive their own value with AI, right? And what I mean by that is similar as we have in data mesh for example implemented federated governance there. I think we should learn from it also for AI since AI is one of the

main use cases nowadays for data right and in that sense basically you have of course the central function of operations and monitoring that is central for everyone but you want to empower different business domains to have their own views and their own goals about how to actually uh drive business value with it meaning they have the AI product owners in AI stewards for example that are helping

with that and on top you still need some crossunctional guidance right like an AI review council that's actually looking at it from a technical point of view about architecture about policies and all these kind of things and a strategy board that of course prioritizes and decides over funding for AI right those that actually say yeah we are investing in this but not investing in this and we

want to focus on this and not on that and that should come again from a cross functional point of view and an agreement and this way we can balance autonomous innovation experimentation from the bottom up but also have alignment with core principles and standards from the top on right and again I think one thing about it here is that this is a nice target state and um

we are today right now in a state where we're just not as mature yet to reach that target state right often you have one central AI team that does everything for the organization and that is also normal right like every innovation starts with a center of excellence that services everyone and then over time to make it scalable so it's not a bottleneck anymore you kind of distribute

and decentralize more and more so you end up in this federated model so I'd rather take this as like a target state down the line but something that we should consider the more mature we get to move towards when the time and then let's talk about um yeah basically solving actual problems and what I mean by that is that I feel like we're playing a lot of

pilot theater nowadays and we're not actually following the pain what I mean by that is pilot theater happens in three acts mostly right someone builds a dazzling application everyone says wow Oh, very cool what you did and then no one knows what to use it for and then it slowly dies and no one knows actually what what was built for in the first place. That happens probably

everywhere in organizations and the real trap here is right you haven't actually started with a a property statement you started this with a solution that you don't know what problem you're solving with then what is the point of building that solution so what I would suggesting as the opposite is a following the pain approach right and it starts with identifying and acknowledging pain points first and foremost

right so if you're thinking about business processes or your day-to-day tasks somewhere there's a problem you want to solve first right and it might be for software developers of course that some of the task you're doing like reviewing or you're specifying all those takes too much time so you actually introduce AI to make that faster and make it more scalable that's great right but now we think

about bigger scale and day-to-day of business people they're not as reflected as developers are right they might not even think about they might think it's great what they're doing dayto-day and they haven't even considered that some of the things they're doing are actual problems that can be solved So helping them to understand that and identifying what is possible to uh fix basically is really key and once

you have those pain points right you can clarify the requirements and the urgency of things right because if someone acknowledges finally that there's a problem they know exactly how good looks like right so it's not only about I'm complaining about this but they think I wish that my day-to-day would look differently like this right so you have a target state you want to have and they might

have an urgency too right because they know exactly how much time they're wasting or how much money they're missing thing on making by not having this fixed. So you also have a business case behind it, right? And that leads to then defining and implementing the right AI solution and by having that you have a proof of value, right? And that leads to advocacy because people that have

been helped with solving the problem are now big fans of it and they want to spread the word and say look AI can actually solve those problems. lets everyone talk about pain points and then ideally you come into the cycle of following the pain and having one more painoint solved instead of just building solutions with no real problem statement right and all that leads actually into what

I call the AI purpose framework and that is actually um for those who know the eeky guy a little bit of a derivative of it right the eeky guy is basically finding purpose in your profession it's a Japanese framework with exactly also a ven diagram of four circles In this case, actually the four circles are more from an um business sense and from what the purpose on

enterprise scale could look like. And it starts with what is exciting to do, right? Because we still need to acknowledge people are really excited about AI. So let's not stop that. Let's get people excited and let them think about what is exciting to do. But that needs to be aligned with business strategy, right? So it doesn't have to cannot be only exciting. It needs to somehow help

with our business objectives at what we do as a business goal, right? That needs to be aligned with whether it's ethical to do, right? So, is it somehow hurting society? Is it somehow hurting our employees? Is it somehow a kind of um creating a risk for us about how we deal with this in the right way? And then you have lastly what is feasible to do, right?

So, not only from a technology point of view, do we have the right tools, but also from a skill set point of view, right? Do we even have the right people to run this? And do we have the right expertise across our employees to make it work? The good thing is that if you basically check what you're working on across all four of those, you might need

to involve the right experts from different functions as well and talk to them about it before you just start doing it because otherwise if you don't talk to them up front, they might come in and stop it anyway, right? So if let's say legal comes in and they say this is not ethical to do and you have already invested like a month on working on it, then

they just stop it, right? because then what's the point? So having a little bit of a lean alignment up front actually helps you to prioritize because actually then you can see what the purpose is and you create a backlog of actually useful and valuable use cases rather than only things that are experiments that might not move forward anyway. Cool. And that moves us moves us to communication,

right? And communication really for me I think is about first how to communicate the value of AI. And I think nowadays because um AI can act autonomously um that productivity uplift or automation uplift and saving cost on things is a really strong business case right and we can now very straightforward calculate these kind of things nowadays as well. Um and that is a good thing that can

basically inform everyone why this is important and why we should do it. But I don't think that human beings individually are personally that excited about it across all of the employees, right? Especially thinking about those people that feel threatened in their own job immediately by it because they see oh if they're automating this then what am I doing with 80% of my time now right? So that

is kind of how many people are thinking but they might not say it out loud in organizations and those will become those people that are quietly sabotaging a efforts. So what we need to do is to not only frame impact on business level but also frame it on personal level by understanding who AI is actually impacting and what it means for them in a positive light. Right?

So for example, if the business impact is about information findability because now you suddenly have much quicker access to specific data and information and that is great. That means the speed to knowledge has increased. But for personal reward is basically more about higher confidence, right? like people that now can access the right information at the right time in a fast way. They can act much more confidently

because otherwise they might just feel like, oh, I only found these documents. I don't know if they're the correct or outdated ones. I hope that this is the right result, but I'm not actually sure. And to move that actually into confident acting, which saves some time as well. And then of course we have for example also marketing efficiency, right? Which is all about higher creative output, right?

So basically think about all of the creative assets that are being produced like banners and pictures and social media videos, right? And then through AI you can have a higher creative output because AI can assist to that. But what actually it means is not that all those people don't have to create assets anymore. They have more time for ideiation now, right? So they can spend more time

on having the better ideas to then guide what needs to be produced and not get rushed into producing things when they're not sure about how good the idea was in the first place. And also that leads to talent development, right? Um because then we think about how that can AI can also tailor learning to people, right? Not only replacing knowledge but actually teaching people in different ways

how to learn new things. That's all about learning and growth. And what it's actually about maybe is about career planning, right? So give people now alternatives about how they can advance. And it's not only about their jobs might being redundant but in what directions they could take their skills into a different direction that's relevant again to the point from before in this AI world AIdriven world and

having AI adjacent roles for example in the future and when we talk to people about what the value of AI is we should not only think about business cases right depending on who we're talking to let's frame it in a personal way and for that we need to understand what drives human beings too right People are motivated by different things. Some people just want to move up

the career ladder. Some people just want to learn more, right? Some people just want to get promoted. All these things are things that we don't publicly talk about, but the more we understand them, the better we can frame it. And then we we can position AI as helping with this rather than threatening people with it. Right? And now the other side of communication, not human to human

communication, but uh human to AI and AI to human communication. Right? the more we use LLMs now in communication as well like specifically I'm thinking about customer service where of course a lot of that innovation is happening right then you're using a lot of chat bots nowadays right to make that customer service happen but um we all know right that if you don't give it the right

instructions it's just another generic LLM that's going to use the same way of communicating like any other foundation model we see and that might not be good especially if you don't put any gatras around it and what I want to focus Is that exactly that voice and tonality of those L&Ms? Right? Because in the end it should represent your brand just like how customer service agents, human

agents have been in the past been trained of how they should communicate to customers and how they should react to critique or react to different uh customer inquiries. We should have the same guidance as principles for chat bots as well and LLMs, right? So basically reflect the core brand values into communication principles. for example, optimistic but never cheesy. But then at the same time, we still need

to provide some flexibility just like customer service agents in the past had different personalities and would know how to talk to different customers in a different way. Meaning to put like an equalizer in, for example, to let those people that are actually owning the customer relationships also have a way to be more flexible. And then also of course thinking about variety, meaning we should monitor and govern

the voice in a different way. Plus a community of practice never hurts, right? So if we think about voice stewards, for example, in the future that are actually responsible for tuning that voice in the right way, they could all come together from different functions and they could talk about best practices or how to recalibrate those guidelines, right? Um I think of course this is not really um

I would say common nowadays, but the more we're letting actually AIS talk to people, right? the more I think we have to think about how to make it still tailored to different human experiences. Um, I think most of us maybe feel like that if you had before always like a human customer service and that was really good and suddenly you're talking to a bot, right? That it

feels a little bit like being let down, right, in the beginning, especially when they're not doing what you want and then you're just very frustrated and you end up escalating to a human customer service agent anyway, right? So, in a way, I think if we want to avoid that scenario, we need to just spend more time on the communication itself as well. And that leads me to

creativity, right? So creativity um again what I think is the main tension field is what is human creativity in the age of geni creativity. And I think what let's go down to what foundationally actually geni how it works because when we think about LMS and all these generative outputs they are being trained by human creative output right they're trained by everything we created like books and pictures

and music and everything that's out there that is our output but they are not being trained with what happens in our heads when we have new ideas right so the process of how we come up with new ideas and original ideas is not being represented. Meaning in a way it's kind of just being trained on the surface with what came out of us. But it's not actually

being trained by understanding how our human brain works. Even us human beings don't understand how our brains work completely, right? And that is also I think that generative AI will always be incrementally creative, right? It can mix and match different outputs from human beings and basically build new things. But it cannot be disruptively creative, right? Because this is where human beings have truly original ideas. like imagine

coming up with a whole new um programming language or coming up with a whole new style of art right those are things that are true mainly human and those are things also if we don't apply them then everything that's created only by AI it would become training data for the next uh AI training again and we all know that recursive training is also not a good thing

so in a way I think we should differentiate creativity right and also put it there where it belongs that we as human beings right we have original disruption we can then work together with AI to collaboratively iterate based on the original ideas we have and we can let AI automatically refine things for us as well and basically incrementally create it but this is where our creativities come

together right so I don't think that geni will replace human creativity it will actually help us but we should also not unlearn how to be creative ourselves and how to have original thoughts which is generally a great idea still to have in the age of AI and I mentioned that for a little bit, right? Because creative work and creative execution actually changes because of what productivity increase

AI gives us, right? So when we think about traditional creativity work, it roughly broke down into a small time period where we ideulate and then we have to reserve all that time to execute, right? For example, let a few graphic designers manually create all the assets, let an agency do all of the work, give them like four weeks in advance so they create all the assets. Now

with NAI of course we can make that faster and more efficient meaning we can actually reduce that time of execution significantly but instead of now basically saying okay then it replaces all of us people we actually have more time to idiate right and when before we actually didn't have that much time we have because we had the time pressure of execution now we actually have more time

to idate it should actually increase the quality of the output again and thereby make our creative outputs especially in creative jobs also much more impactful and valuable as well. And that leads me then to the last C, conscience, right? So I think there are so many things that we can do wrong with AI and so many things um I think that we can learn from about what

can go wrong and what can go right that it's more about combining internal practices with external frameworks and defining our organizational specific principles of practice. Right? So when we usually in organizations already have as internal practices for example our brand values right so what is our brand what is our mission our vision our core values how do we position ourselves so let that flow into our internal

practices then often we have a code of conduct right so we have dos and don'ts of human behavior but we can actually um uh represate that exactly also for AI and agents and then we have of course policies already right so enforce guidelines with actual consequences and things that we can translate there well and then we have also a lot of external frameworks that give a very

good guidance on how to actually deal with AI risk in the right way. So for example in the EU AI act there's a risk classification of systems that already I feel like is very actionable and can be very much used to based on different risk tiers define the right processes. Then the OECD has also AI principles more about human- centered values uh transparency and accountability and the

NIST AI risk framework also is basically defining risk across AI system life cycles as well that we can all use to do that. So instead of just reinventing these AI practices from scratch and how we do it, I think we can all just basically um reference all these existing things and at least have a starting point. And the more we learn about it, the more we mature

in our a efforts, the more we can just iterate and improve on those as And once we do that, I think then it comes to the question, who is accountable for all of this, right? And the real complexity I think is that every AI application actually has these three layers with each of them having their own responsibilities already. Right? So very high level speaking you have a

data layer, you have a model layer and you have an interface layer. Right? Now with the data layer of course who is responsible for the accuracy and valility of the data that's being fed to AI models right and that means for example is mislabelled training data or there's biased historical patterns or data poisoning. Then you have the model layer like who's ensuring explanability and interpretability of the

models right? So it's about the inference method. It's about the purpose of the model. It's about the contextual knowledge base. And then you have the interface layer, right? Who's responsible for the good and bad human experience with the AI model or agent, right? Meaning that you have users actually making bad decisions. So if a user use AI and makes a bad decision, whose fault is that then,

right? So when you think about it, each of them have an impact on what goes wrong with AI. But um if you don't figure out how they can actually work together and how they have accountability that they're all going to fingerpoint to each other again about whose fault it is if something is not working. So defining that process for a governance is so complicated because it has

so many layers and such a complicated landscape that it's really about finding an initial agreement and then basically from there uh saying that there's one accountable person multiple responsible person but we take it seriously and we just move forward with this. So we are not stuck on the problems and fingerpointing all this but we move from problem to solution as quickly as possible. And um lastly right

um even with everything there are some risks that we need to look out for when it comes to AI and again with a very human- centered lens across all of this right so the first one is about spotting and reducing bias I think with that right it's all about usually training data that we use um and what's really the paradox about it is that when we think

about human behavior the further we go back in the past the worse it gets right to according to today's standards We didn't always behave in the nice way in the past. But um famously for AI to basically be trained the right way, we need a lot of historical data and a lot of granular data. But what if the further we go back, the more biased we are,

right? And the more we were bad to other people and made bad decisions about human lives in the past. So how do you find that balance, right? about not going too far back in the past where actually I would then learn the bad behavior of human beings versus then having it basically enough of training data so it's accurate and can actually act in the right way and

the same is with preventing disinformation and that is um intentional and unintentional right so unintentional means someone who is not a subject matter expert just gets hallucinations by um LMS and they cannot fact check or they don't know how to fact check and they directly just take it and that was unintentional wrong information and you unfortunately made a bad choice with it. Right? But you also have

the intentional disinformation of creating deep fakes and creating air generated pictures that look real. So you're on purpose basically confusing people and not even think about what it does. And if we think about that not only on societal scale but also within organizations, we need to somehow prevent that, right? How do we make sure that we can avoid the disinformation of things coming intentionally or unintentionally and

basically creating malicious um kind of impact on organizations and lastly ecologically conscious choices right we all know that AI actually consumes a lot of energy and we don't actually need the most advanced ways of doing things always so how can we balance a little bit simpler solutions that cost less energy and are nicer to our planet with actually still reaching the goals and finishing ing our task

in the right way because otherwise right we're thinking very short term and this is really cool but it does burn a lot of energy water etc etc as well. So um in the end it comes down also to this that when we think our own moral values they have changed over time too right generation by generation time over time we are all growing more mature as human

beings and what we did in the past we might not actually condone anymore today so if we as human beings are changing it's our responsibility that AI learns from that too and changes with it according to always the most current standards and it's our responsibility to be the role model as a moral compass for AI and not the other way around and hoping for the best that

it might just change on its own. So some closing questions right so what values will remain stable in an unstable future there's so many uncertainties about AI now right so what are the ways that we want to stick to which decisions must stay human no matter how advanced AI becomes right so think about that for a day-to-day but also in any business context and when machines become

more intelligent are we becoming wiser or brackets more stupid right if they're more intelligent that is the So the future of AI lies in all of our human hands. Good luck to all of you. Before we go into Q&A, actually I want to play one music video. Maybe there to the Hello. Are they listen to me? Yeah. Yeah. >> I'm going to play the music video now.

>> Okay, perfect. So um just one uh little thing up front. So I make music about data. it's a bit more um approachable, right? But also this one particularly is because I think when we think about the mistakes we make with AI, we can learn a lot from the past, right? Not all of the mistakes I knew. We just kind of forgotten all of the best practice

from the past. So in that spirit, I made a parody of 80s songs that are popular and parodyied them for the age of AI. So please enjoy if 80s songs were about AI. Yeah. Heat. Heat. Heat. Oh. All right. Thank you very much.