thank you very much so the F the first question is of course the most important one because none of you can concentrate without knowing what he did wrong and so it was just a bike accident biking on the bike path and then in in orus which is a very dangerous city to be biking and and then a car turned right uh right in into me and I broke both of my arms which is not so good for a knowledge worker but it gave me we discussed in the beginning it gave me a chance to to try to explore some of the text tox technologies that are here so it's a very fortunate time to break both of your arms and be uh mobile with two fingers or something like that so I'll show you some of that the results of that and I'll get into hyper intelligence and we also talked about uh previously that I've been talking about hyper intelligence for for for half a decade now and and it's really now coming into the mainstream uh but everyone is asking it sounds really cool and it sounds like something we should do but what is it and I'll try to give you uh an answer uh throughout this talk here because it's not um not so trivial in particular because what we are trying to develop is a definition of hybrid intelligence which would allow us to say that something is not hybrid intelligent because there's so many things that you would always say there's no code or no application interface place where where there are no people in the loop somehow so I hope that we can take that Journey uh and I will try to tell you how well I mean you you you probably know about generative Ai and and and how generative AI is sort of a unique time for us to be creative and it's and and and how we can foster that creativity so in order to do that I just want to poll you first and just see how many of you are using generative AI in your in your code which would be co-pilot or chat gbt I I guess I should have asked about who's not okay so it's about it's about 50% uh of you and and then and then how many of you have been using some of the other text 2x so that is text to image for instance generating images yeah about a third and how many of you have tried to generate videos so text to video okay a couple so so I had to think a little bit about how to present generative AI in in in this audience and and of course I could have just presented co-pilot and and the ways in which we can autogenerate code um but really I'm trying to take a different narrative in in this coming 45 minutes and take you on a journey of seeing what is it that the people that you are developing applications for know and how they think so that we can have a more systematic way maybe of incorporating in their perceptions both their P posst their hopes of what such an interface could be if um if you are generating so let's say I'm specializing to Virtual assistance now so how do we Define define a virtual assistant and how do we design it in such a way that users say I love to get that that's going to be a a revolution for my workflow and they don't say I am afraid of that and I I might lose my job because of that so so it's those fears and hopes that we are navigating in that I'll give a little bit of a impression with so start with a couple of perceptions of you know where is this what are the boundaries what is what are the sort of the high-profile explorations of that how many of you have not heard about Auto GPT Auto GPT uh just a a few of you and I'll just show so so autog GPT as an example of an architecture where you can have agents that generate agents I'd like to order a pizza what type of pizza would you like to order 11in pepperoni pizza regular crust okay great I will first search to find a good pizza place in your area I have found a pizza place called zap Pizza I will now create a pizza ordering bot who will call this location for you and will order your pizza super amazing Pizza ordering bot reporting for Duty I have been instructed by the other AI to order you a pizza I will now call Z Pizza using the twilio voice API and order you a pizza so the point of it is that that there's an architecture now so in the creativity fields we said okay well there's always something for the human because either it's sort of defining something at the end or in the beginning or at the end of a process or some some sort of task management which is sort of that unique human contribution and autog GPT and those kinds of Technologies really uh the answer to that to say there's nothing which is uniquely human and there's going to be nothing which is uniquely human but there is a discussion now and in the coming decades about how what is the most efficient way to build these interfaces so just the fact that AOG G exists does not mean that those kinds of architectures are the best way to do it but I don't know how many of you have have tried to create something useful without how many of you have created something useful with AOG gbt so so there was a lot of people who have heard about it maybe also experimenting with it and we've also been experimenting with it and and and and so far you know a a $10,000 Pizza is is about the supremacy of of what you can achieve with that technology so it really shows in principle it is there in principle it works but or it could work but it doesn't and and we need to a military operator responsible for monitoring activity within Eastern Europe they've just received an alert that military equipment is amassed in a field 30 km from friendly forces AIP leverages large language models to allow operators to quickly ask questions show me more details they ask what enemy units are in the region and leverage AI to build out a likely unit formation what enemy military unit is in the region our operator requests additional imagery to build a more complete picture of the potential enemy equipment on the ground task new imagery for this location at a resolution of 1 M or higher AIP surfaces the option to deploy a nearby drone to collect video task the mq9 to capture video of this location the footage confirms a Potential Threat the Drone footage shows an enemy t80 main battle tank generate three courses of action to Target this enemy equipment so here we see here we see the so the the GPT sort of on steroid and saying okay you can control an entire War using just the chat interface and where it's heading and and where also the compan companies and corporations are looking at this and saying we will also probably very soon have to turn our strategy session and our execution into a war room like this and control it and of course at the end of this movie they say they say that they send it up the chain of the command they have all those three options and then one of them is chosen by after careful deliberation but of course if you are now I'm on this journey of saying not necessarily uh understanding our stakeholders or our Target audiences and saying the people that we're building for uh it's not um unlikely that they will be thinking yes some people are doing this chain of command and then others are probably also going to short circuiting and so this dangerous this technology is extremely dangerous uh to to to us and to humanity and to our Workforce and so that's what I'm be be be talking a little bit more about now one of the sort of exciting so those were two exciting trends that happened in the spring and and and one of the sort of thing that that that really gave a lot of Hope was this leaked document I don't know if you saw that from Google that essentially said uh we have no mode and neither does open AI so what they're saying is that in terms of these llms then where we thought it was a Winner Takes it all or the the sort of the two first Prime movers of it are going to be the move first movers and uh the winners of this because they have so much data and they have so much training capabilities uh they are saying open source is just as efficient as us as doing this and so that's really good news and it really tells us that this llm landscape is going to be much more pluralistic going forward and that these open source strategies are going to be very efficient and and so that's something that uh for instance if you have data then you training a your your separate GPT model on that like the Bloomberg GPT is going to be better than what Google or uh Microsoft can do because you have that detailed knowledge the uh the approach of MAA is as you know probably to to go entirely sort of Open Source on that and soberg has has been sort of okay how is he going to make money off of this and the investors are asking okay why are you not trying to monetize on this and and I'll give you sort of a on this this this football field over here I'll I'll show you sort of a little bit of a speculation what I'm saying what uh uh this is just the entertainment part what do I think that Mark Zuckerberg's master plan for World Dominion is uh even if he's giving it out as free open source because there's one other really interesting technology Trend that's happening at the moment so this was text tox and we've seen lots of this but what about the next phase of that which is going to be thought to X and so there's already sort of experiments where you can put an electrode in there uh and then you can start to do some machine learning and then you can start to read thoughts in this way but what's really exciting uh maybe also scaring is that some of these uh researchers medical researchers are trying to do that also um for what is called the fmri scanners so that means you can sit in a scanner now you have nothing in operated with you uh but still we can start to read your mind and now you can sort of think okay what is it that Mark cleberg once when he says that he wants to combine Ai and machine learning and then the metaverse well he wants to build uh these sort of miniaturizing fmri scanners into let's say a billion of these headphones and then he'll be reading our minds in that way so so that's sort of a cool way to get lots and lots of data and I don't know if you saw it but but apple is also so sort of putting this kinds of sensing into our airpod so so this not just text tox but but you can say x2x is is just about to come so now the question is generally speaking why is it that it is happening right now and and so this is the sort of the answer to that and and and you if you've been following the field over the years then you sort of know that that we have seen now over the last let's say mini decade here that a number of really important curves have crossed the line of human performance and so that means that we now have uh the gates are opening to multimodality uh applications as the ones that we're seeing now so what is the long-term perspective of that well here's one answer to that and that is uh three or 400 of the AI researchers that are uh ask the question when is the when is there a 50% chance that a super intelligence will be formed uh and and sort of the let's say good news for some people would be to say that there are some outliers out here who say that not going to happen the next couple of hundred years but but the really sobering news is that half of these experts say that within the few next few decades it is going to be there so so so that's a really serious sort of debate to to to be had when we finished with this with with my talk I'll have introduced a little bit of this hybrid intelligence notion and I would venture to say that uh the majority of these AI experts don't have let's say state-of-the-art knowledge about the human brain and about soci ology so that I think can infect our curve and I think that's a positive story that line that I would come to is that hybri intelligence is really sort of an alternative to these questions and I think hopefully that will become clear soon so first of all uh if you follow the IBM developers conference then then then the CEO of IBM says no one will lose their job uh to AI but they will lose their job to people who use Ai and so this this blog post I don't know if you've seen it it's one of my favorites is is how did I lose my job overnight to a program called mid Journey which is a im text to image generator and this guy he was working in a graphics company and he was the best of all of the people because he had the skill of finalizing sort of putting the finishing touches on all of the things that went out so in the S same sense that you may have someone sort of who in the code world who will s put the finishing touches and knows sort of how to uh let's say put the uh let's say the for instance the error finding and and what he's then discovered was that all of his skills the ones that he sort of made him leading went distinct extinct overnight because that was exactly what mid Journey could do and of course his reaction could have been to say okay this this technology comes and it it is going to enrich my life and I can do more things but he considered himself not as an employee but as an artist and if you consider yourself as an artist and then you see your artistic sort of sort of profession being taken away then he left his job and then his colleagues who said I think this is fantastic because I can now do some things that I couldn't be doing before H are sort of the winners of the game so those are the you know the the danger signs of losing your job is that if you have something that you really want to do and you don't so here's a a cool book I think about range and sort of being able to extract Knowledge from many many different knowledge domains so that you can generalize rather than just specialize on one thing so so now I'll just walk through in the coming 10 minutes or so this will be sort of a little bit of a tour of what people who are not like you are interested in in in generative AI in terms of how it is changing their workflow and the reason that I'm putting all of these sort of examples out there is that that shapes then their impressions of the technology and how we should uh be building it and so first of all no let me skip that one and say Okay so so so in the chat GPT then how many of you are using plugins uh in in chat GPT okay so so chat GPT so if you if you if you have the GPT plus and then you enable the plugins then you can pull in the PDF readers and you can say okay you can get the summaries of the PDFs you can also uh pull in the uh smart slides for instance and then you can have this uh sort of okay uh please create a slide deck summarizing these main points and then you can download the presentations St there so so those are the kinds of eye openening things in the sort of everyday work of of many of the of of of your colleagues in the different uh uh institutions the chatbots uh are really easy to train now and so so I I usually have this sort of eye openening video of it it takes one minute now to train a chatbot if you have a service like any it could be chatbase or any other service and you just have website and then it scripts the website and it generates that we and it generates that chat Bo so it's fetching the links and loading it in and what it's doing underneath is vectorizing all of that material and then it's good to go and you can embed it on your website two minutes later so that really shows Hands-On how you can do a lot of Technology with that and and how we can start to experiment with technology also before we start coding we can actually build a first prototype of this if you were building a chatbot then why not build sort of a a scotch taped chatbot like this one and then start attacking it and and sort of saying okay this is Den bank's I think uh one or one that I made for dankbank and and then we start to attack it and say okay well uh what is the capital of Denmark and then figuring out that this particular chatbot is is well trained in the sense that it doesn't want to talk about the distance to the Moon but it's not so well trained in terms of of of the capital of Denmark and then the main point is sort of a no code response to that is that that every one of these chatbots then has a base prompt here and then you can go in and then you can just add sort of into this Bas prom you can say okay uh please uh act as a document and also do not answer questions about cities for instance and then without any coding these people have now interrogated their technology product and they have found errors of it and they have implemented those error modes into a new product that they can then iterate on and then they can go to development staff and then say okay now I know exactly what I want in a product so I think that's a really great thing for for sort of rapid prototyping uh in this sense we went out for Nick christansen for instance and and then we I had a couple of guys here who had no IDE about how to code but they used all of this and then they did sort of iterative prompt base engine base prompt engineering and so of iteration by iteration they went out to the stakeholders tested it out both internally and externally and every time they only changed this in this this uh uh base prompt but they could increase the functionality of it step by step and and they could get sort of really positive responses but also they could start to investigate how would this technology product be used not just as a chat Bo which fulfilled its primary and first purpose which was to give the answers to the customers but allowed us to start thinking more holistically about that technology and what kind of value that could be bringing and one of the things that came out of these stakeholder discussions where everyone was around the table was to say actually it's really interesting that if we have a chatbot then maybe we could discover trends that were different in Oro from Copenhagen and which cars we really wanted to have on stock and then if we feed that back to our purchasing department and our Logistics Department then we are having what I would later on call a system solution of AI rather than what is called a put Point solution where we just slot in a technology for a very well- defined purpose so that shows a little bit I I also like to sort of flesh out Oto for instance because uh they sort of a little bit first movers in in terms of meetings and so I use it for my recordings and then for the meetings and then as soon as we start up this we invite the auto uh recorder into the meeting then it also SP Wars up its its spot and that means that when my people come to the meeting and if they are 5 minutes late then they will just ask okay what was mentioned so far and what were the tasks that were given to me and so all of these are sort of ways in which we can see how it's integrated into the workflow uh of of of the organizations so now uh I'll just advertise for uh if you want to get a quick sort of overview of the field of text to image uh then there's a a really cool website Fabian Mel he has sort of put together a timeline like this here and I'm just sort of very quickly flicking through it where you can see all of the different generations of technologies that are emerging what it is that is sort of the uh the link to it and what is the science behind it and so it moved as a field from so five years ago we were using these big Gans this generative adversarial networks in order to do our generative Ai and then came sort of the mid Journey d 2 people think di 2 is outdated now because they really spent all of their energy on on the chat uh um on the chat bot instead uh but then I'll go back to Firefly also so the the important thing about seeing that technology being rolled out is that it sort of goes through phases every one of these applications goes through failure modes that are sort of becoming a part of The Narrative of saying I don't know if you you have heard or seen so so the first one here is sort of okay challenge number one is to really understand what is it that a text is there and and so if it's a the man paints a picture of himself and then we get these sort of odd odd failure modes at the early stages of a technology development and then all of the providers say okay we have to fix this kind of Common Sense thing and then it gets into some some more subtle sort of error modes and and maybe you have sort of seen talk a few months ago just saying I'm never afraid of these kinds of deep fakes because I just look to the hands and then there are these telltale signs and of course then it spawns a lot of work on that and then this gets fixed and then the next one is to sort of say okay if we really want to make use to it then we would love to have something where we could produce some some some images but also combine some text and and this is how text looked a few weeks ago and then a new stability AI comes out with something that can actually also handle these things that means that you can start to do logos and all the kinds of Technologies so so it's really tough to be giving this kind of a talk uh uh because because the one I used a month ago is is completely outdated and so I'll show you I'll show you some more things that I added last night um so so so that means that what's happening now is that these deep FES are there and they sort of have a real effect and and and we have to start uh uh living in a world where there are some people who launch a deep fake uh of for instance the Pentagon and they do that with one purpose and that is that they know that the stock market is going to take sort of a millisecond dive in the second after in in sort of the the time after this is released and if you sell or buy stocks at exactly the right time then you can game the system so that is the new reality that we are living in um and that we have to sort of uh uh with here's uh one of my sort of favorite sort of these technology comp companies Adobe and I work with them uh on something that I'll show you later but but really just as an illustration of what we can now do and we can specify we don't want the surfboard we can specify that we always were curious about what was happening on the other side of the baby swimming there and these kinds of things so in painting all these Technologies are are are moving uh pretty rapidly voice cloning uh is is is another one and um um um emulate the taste of like like non-chicken nuggets yeah I went too far okay so these nuggets are made from chicken but they're made to emulate the taste of non-chicken nuggets dope oh that's much better then edit all the blather out of your videos because my time is very precious oh that's fire so so what this shows is just that these kinds of narratives are emerging where the workflow of people who are editing videos is going to be sort of really dramatically disrupted if if it's not that it maybe necessarily does something that you weren't able to do before you could always do video editing but the new thing is that you can do video editing just the way that you do text editing and just say look at all of this stuff and then you just cross out whichever words you don't want and then they all of the sort of integration you cut out that and then the the flow from frame to frame is seamlessly done using the generative AI so it paints really a completely new picture of how we can engage with that technology and and and one part is then also sort of okay the the voice dobbing now I've spoken for about 20 minutes so now you recognize all my my uh embarrassingly Danish accent that that I have and and so I trained uh my my voice also to do following you will hear some text that I have never jber I jber uploaded 30 minutes of my voice recording and let the script over up my voice I can now write anything and select my voice as the speaker or alternatively choose any other voice male or female and with any kind of jab or intonation so so so sorry about the words that was because I was on my on the low PID tier and then I couldn't get so much text and then they inserted some of those but you really can see at least hopefully you can recognize my voice that's that's about a month ago and then the field is such that the the script were the best ones I tried out many of them and then okay so so they all can do something similar let me say it's becoming more and more mainstream what was really difficult a month ago is becoming something that more and more products can do so it's really a quickly rapidly moving thing and and what I'm trying to say with all of this is you know this all shapes our expectations of what kind of products we have um uh coming out in the very near future video creation is the next one that happens uh again Runway was my favorite last month uh and I've been working with that we've been working with the municipality of orus around uh using generative AI where we got lots of pictures uh for instance the town hall and then they could use generate text to X and then they could make changes to that and and citizens around orus could then make changes to see okay which kinds of positive and negative visions of the future could they create but then uh but then now you could also create videos that were sort of animations of what could be happening and so so it's really the new thing is not that this can be done this has been available for a long time but it's available to the masses now which really changes the way that this technology is is taking speed and so every one of these sort of has then different advantages there's one which is open source and so you can there's a very much of a plurality of of services that are being offered then the next cool thing is that we can also start to say okay we can if I have some video then if I want that video to be also available in other languages uh then hyen launched this very very uh recently so so they have sort of an avatar where you can do you have this Avatar and this Avatar can speak and and their speciality is lip movement and and being really sort of good at integrating lip movements and doing that across different languages and so here you can see uh dubing in different speaking eight languages fluently to translate a video into any language just follow these steps first click the link in the description and create a free account with h&i then upload your video file then select the language you want to translate your video into click on the submit button and boom you'll have a video translated into your chosen language and so an example of of how quickly this moves is that it's so so this technology moves so quickly that the the companies themselves can't even keep track of it because I was really frustrated when when I wanted to test out this technology and for free I could get one uh voice One recording uh uh translated and so I picked one of me speaking Danish and then and then I sent it up in the queue and it said you are now number 700 in the queue you could pay and then you could be number one in the queue and I waited and then I I researched a little bit more and then I saw that Danish was not not actually on one of the supported languages so I was really frustrated because I couldn't even pull it out of my queue so I had to to just uh wait for it to come out and be worthless and then the next morning it came out and it turned out that they had also added Danish uh since then because but it just wasn't on their website so so so I picked just a little bit of of of of footage from that's my previous computer oops just quantum physics is fascinating not only because we and everything around us are made up of atoms but also because it may enable us to build new computers more powerful than all the world's computers combined it requires that we and so this was a little bit of a technology tester because first of all it's in a new language Danish is not super supported and and I speak very very quickly compared to what is optimal for that kind of a technology but but you can see it really is starting to work uh U and and be at least moderately convincing and will become more and more convincing on this so then the question was these were sort of the individual uh Technologies and I just took that as a sort of a primer for for what is then to come which is building these sort of fullscale applications where we take the workflow of someone in some company and then we say okay how do we deal de develop sort of a portfolio uh offer that can really solve all of their tasks or at least help them in solving all of those tasks and and customer relationship management CRM systems are uh are sort of one of the places where there's a a lot of money to be had and so here's too for instance just a few of the ways that we see customer AI transforming your customer relationships customer AI connects the dots for your marketers pulling data from every prior interaction a customer has had with your company and then personalizing each and every message and finding the right people for every campaign with generative audiences automatically bringing together the right people for the right message so the point of it is really to say okay everything that the marketers were doing or that the support ticket responsible people were doing is something that can now be incorporated into a flow and and Salesforce has the same sort of convincing sales story and it's all contact them if you want to uh if you want to be sort of a part of it the point that I want to make here is that well I I'll show you in the next video this is Adobe uh that we are working with and what we're working with Adobe on is not generating this necessarily the core product itself but the virtual assistant around it because there's a lot of different kinds of uh interactions that can be pulled in and the more comprehensive it goes into the workflow of the people the more we have to think about what their perceptions are and and and I'll show you some examples of it but but this is the Omni channel to how much can Delight your customers people want more more more faster than ever and it's got to be personal but creating personal experiences across every customer Market Channel and device that's been more than any marketer or even a whole team could handle but with Adobe experience manager you can marketing has gone from manual to semi-automated in a matter of years but that's just not enough if you want to connect with every customer oneon-one thankfully technology is accelerating faster than ever and with Adobe you unlock the power of Automation and can do things that just weren't possible before like so what you can see is that these companies who have a lot of data and who have a customer base are now saying okay they're sort of fighting for the same territory but Adobe for instance comes from the creative side and says here's now integration into how you do all of the products and all of the material whereas twio comes from the opposite end and says here's the ticket and the handling and the data science around it that they have usually done and so they they try to generate these sort of 360° uh solutions that you have so so now for the remaining part uh so sorry I still haven't sort of told you about hyber intelligence yet but that's exactly what sort of will be uh will be the topic of the last parts and so so the Point of Departure of that is that um BCG and and McKenzie are also saying starting now to say that that this kind of hyperd intelligence or everyone is talking about human AI teams and how human AI teams are a very precise number six times more likely to amplify AI success for instance um so so so that's that's what uh they are talking about but as I talked about in the beginning it really is not very discriminative to say what is then a hyperd intelligence or what is a human- centered solution and what is not what is human in the loop and what doesn't count as human in the loop and I've been somewhat uh uh competitive with some of the the human- centered AI researchers because I have am of the opinion that we can actually formulate a set of principles that can ensure that an application let's call it hybrid intelligence application is used for good purposes and not necessarily as many sort of computer scientists of my colleagues say that that the algorithm is just a hammer and we have absolutely no control about what the hammer is being used for and the price that you have to pay for that is that you have to treat it as more than a code and so for me AI is a piece of code and there's a risk of failure to deliver value there's a very high risk of employee dis dis Skilling um and also if you sort of use Ai and you are a minor company you say we want to be data driven then there's also a risk that someone else like Google or something like that has more data than you and then they can train a datadriven you can say value mechanism more efficiently that you can and that's why more and more companies are coming and saying okay how can we generate value in a new way that is not as uh uh let's say um focused on just uh on on just the uh Reliance on AI but is something that we can integrate all stakeholders around technology development something that guarantees upskilling of of employees and also very importantly I'll show you one example a couple of examples of it how well actually I showed you the the chatbot how that technology development if the technology development happened simultaneously where all of the stakeholders could see the product and uh in its making then you could also start to imagine how that could generate new value in completely new ways how we could transform our organization while we are creating the technology and not let's say after to adapt to it so what we're trying to do is that we're trying to formulate a field of hybrid intelligence that lies at the intersection between the algorithmics the interface the human computer interaction and then all of the management principles that sort of have also been researched for a long time and we're trying to develop a uh a set of principles and and here's sort of a set of principles that that we uh are uh marketing out um so I when for my TED talk we had this small we had this small red circle that we had to stay in so it was really hard for me I was all the way always almost stepping out of that I'm actually the only one I think in the world who has given a TED talk with only one shoe um and you can so you have to see it if you want to see why uh so so here's here's an example of how hybrid intelligence is different than Ai and that is that we are introducing something that we call a hyper intelligence PCT between management and the employees which means we're saying the purpose the foundational purpose of this technology development uh and how its success is valuated is whether it creates employee upskilling it's not something that we are lucky and we're trying to do it's actually something that we are aiming to do and what happens when we do that is that then the employees have this psychological safe space which means that when we start to engage with them uh we did a study with Autodesk uh and some some Architects I'll talk about later then they open up much much more and you get these sort of foundational uh both the pain points and the desires and the wishes of of the end users that you get in so so there are sort of three different uh I I sort of think of of AI and human- centered AI as the human exoskeleton so so we can use AI to make every one of us better but the uh hyperintelligence includes both a human exoskeleton and then an organizational exoskeleton so that is there are some principles on how an organization should be reforming and refining itself I say disrupting itself so so so just very briefly uh one of the things that two understand is really what are these new emergent job roles in the age of generative AI that we should be designing for and designing support for and one of the uh roles that is not there is the London cabby because it takes it took two years to become a London cabie because you had to study for a test in which you had to be able to answer the question what is the fastest route from a to be at any given time in London and so this is what uh in this framework here we would characterize as an entirely prediction task and if you rely on an entirely prediction task then you are up for an algorithmic disruption uh whereas normal tasks are sort of a mixture between prediction and judgment and we really don't are not accustomed to thinking about it and disentangling those two but if we want to make optimal use of AI Solutions then we have to disentangle those two and say okay for that part of the workflow we can use prediction and we can slot in and then we have to spend increasingly amount more of time on value generation in the Judgment part and so here's an example of uh Radiologists and Radiologists were by Jeffrey Hinton he said in 2015 we don't need to train any more Radiologists and of course he was completely wrong and the reason that he was wrong was yes we have crossed that curve where we can now start to at least under some circumstances do the image detection here at human level but if you take a look at what the Radiology does then there's a list of 30 different tasks that that person is doing and only one of them was something that had something to do with uh something that could be algorithmically replaced another example of why algorithmic job replacement is not as sort of evident as possible is what do you think would happen if we had self-driving cars and we put them in a school bus so what would we have to do the minute after we had fired the bus driver call the ambulance yes exactly or hire uh a kindergarten teacher to to because there was you know these screamy children cannot be left alone and the school bus dri was actually maintaining another role that we were not so explicitly aware of but we will really be aware of it when it falls out and that's going to happen in many of these different if we don't uh consider them explicitly another I think more forth forward goinging looking uh sort of example is the job of a meteorologist which used to be a computational person so you get a computational person in who does all the calculations they can and then they go on the news and then they present the best predictions that they can make but now the job of a meteorologist is much much different because state-of-the-art algor algorithms can do these predictions and they can even do them so that we can start to put percentage on the risk of a tornado which means that it's not no longer just calculating those 5% or whether seven or or or 10 but the role of a meteorologist is now to be communicating with all the different stakeholders and knowing how would the citizens of this American city respond if we communicated in various different ways and how many people would actually die if we communicated in the wrong way and it was a 5% risk happening so so this shows how uh when we start to disentangle this prediction and judgment in our job roles that we can completely new uh workflows that are arising from that and in general this is what we will have to start getting good at is for instance creative writing we shouldn't just say Okay chat GPT is is now there so that is the death of writing but we should really be able to recognize what are all the different phases that we go through for instance in writing and more generally speaking in any creative or even non-creative task then there sort of here is for instance eight different tasks that we go through and we need to build systems to support all of them but not to a to augment various different ones and and not to take over uh all of them so here is is is finally sort of one of the examples that that where they have really M where they have managed to create what I would call the hybrid intelligence corporate solution and the way that they did it was that they had a a mail order company which worked like Amazon but not like Amazon so that you had a subscription model and they just sent you five clothing offers uh every month for instance and of course if you were data driven then what would you do you would sort of do the calculations and then you would know that the the items that are most likely to be have retention are socks and underwear and then white t-shirts right and so you would send those and they tried to disrupt that model and instead say that they hired a a thousand personal stylist so these are now the experts and their aim was to create a product offering for every individual and the only way they could do that is by having these personal stylists but then also taking everything else in their operations and making it completely sort of data uh datafied and you can if if you want to you can sort of search for uh they have a website where they show the whole data FL data uh process and how their Logistics is even more efficient than than Amazon's is because Amazon tried to do this they took a patent on something similar to this and they failed to do it because when they get their uh Parcels back then they often throw them out and so of course if you try to do this new kind of uh mail order service and you throw out the parel parcels when they come back then you have a lot of value loss in that process so so so uh well let me just skip the the the prompt Engineering in the interest of time and then and then just talk about this development and how I think that this development should be happening and I'll just talk about a little bit about how there is now a conflict in computer science uh between approaches that are sort of uh moving from human in the loop to human on the loop to human out of the loop and then the response of that uh which is this field that is called human- centered Ai and what the people so how many of you are aware about uh human centered AI as the principles so they tried to get a a billion euros from the flagship and become a become a u EU Flagship but they didn't manage to do that completely so instead they became just a network and there something called a Humane AI Network and and one of the founding principles of this human centered AI approach is that uh there's a dilemma in design or usually a dilemma in design which is when you want to do something that automates very efficiently then you have to choose whether you have high degree of automization or high degree of user control and what this entire sort of sub field of AI inspired by HCI is saying is that you can actually strive to achieve both of them in your applications and you should actually when you have an application that has let's say generative AI inside of it you have to design generative AI in such a way that the user always feels in control and so here's a here's an example of sort of a a rubric of how you can try to move more more and more towards in this case here patient guided and clinician monitored system so in every application you can really think about what is it that gives the user a sense of control of the process or of the output that is happening in that way and we sort of place hyper intelligence somewhat in the middle because uh uh it is human centered as such but then uh one of the basic foundational principles of this approach is also just to take Ai and consider it a tool so that means which is good so you don't sort of exaggerate the importance of it but you also fail to sometimes reimagine the workflows that could be changed compared to a purely human one that is just augmented and then a human AI workflow doesn't necessarily need to be the same as it was before and that's where there's potential of disruption and uh if you want you can have the slides but but there are some very very specific design uh recommendations related to this golden rules of this HCI pattern languages that you could be following but the algorithms underlying it is sort of okay something like uh interactive reinforcement learning interactive machine learning uh inverse reinforcement learning these kinds of algorithms that were also used to train the GPT models that have a human in the loop in the training of of of that and so now here is my uh sorry take-home message uh technical delivery of my take-home message which is when you want to train these virtual assistants there are a number of success criteria that were not there before so basically us we used to say that we design a system uh this is a this is coming out of Information Systems management where you are designing a system and then what you're designing for is the willingness to adopt the product that you had and this is just completely outdated if you are designing a system that is going to be disruptive in terms of a knowledge workflow that you have then we've been working with Autodesk uh around creating a generative AI for engineering and for uh for Architects and and what we're showing over here is that that there's actually uh three different design criteria that you should be including in your approach and that is the willingness to adopt but also the willingness to co-develop which means that you have to consider what are the settings that I can have around my development process that makes my let's say focus group as eager to bring up in uh ideas about what the end product could look like and one of the examples um I'll show in in just a second now and then the last one is also to say that these kinds of virtual assistance they only gain the sort of the true potential if you as a user invest enough time in order to personalize it and in order to train it so we really need to sort of think about what are the conditions that we can have for that and so we've been running some some some sort of technical research uh uh AB testing where we have an product that is developed this is business intelligence with a AI narrative and with a hybrid intelligence narrative and what we show is that that the the engagement that we get just by phrasing the entire process as a hyper intelligence process creates this psychological safety and the willingness to engage in the process and open up and reimagine your workflow so I think that's something completely uh new and and really exciting finally I just want to touch upon how we could be designing for upskilling and what would happen when we do upskill our employees and so I'll show just one one example here where we had a project together with a company called esoft and and with esoft we let me just double check yeah with esoft uh they have uh 50% of the Danish uh market for putting houses on sale they send a photographer out the photographer takes three of these pictures here and then these pictures exposures are being sent to Vietnam and then they have a processing unit uh where they uh where they uh uh do blue sky correction and all kinds of things that the algorithms couldn't do so this company had a project with some computer scientists from DTU and and they and they very quickly sort of harvested a 40% efficiency enhancement of of this process by just including Ai and then they asked us to say okay what would it look like if we now put hybrid intelligence me methodologies into this and the way that we work with them is then that uh well first of all we built something called a benefit dependency Network which is either from the it enablers or the investment objectives you go back and the really interesting thing about this forget about it it's really looks impressive when it's on a big screen that's cool uh but here you can see that uh that out of it emerges sort of what we see as a low road which is use AI to automate standard edits and increase editors time on this and and so that's the low road of just enhancing your efficiency and then there's a high road here which is about increasing the skills of the editors and then the question is what could we do if we increased the skill of the editors and then we gave them more time for non-trivial tasks and the way that we find out that is that we do interviews then with both the editors themselves and we say what is it that's most difficult in your workflow what is it that is most challenging what is it that's most fun uh what would you like to spend more time on and then we also talk to the marketing Division and then we asked them what is it that you had always loved to be able to sell but you never had the capabilities sort of organizationally to do it and what came out of this year was that we could build what we call a mass customization of these Services now instead of saying that this is a production line where we do the same for every customer then we do what what the uh Stitch fix case did and we say what if we tell the customers that they can write down five words about what characterizes your house and then these five words are translated to Vietnamese and then they are sent to the editors and now the editors are not production line doing the same thing repetitive task again and again every task that they do involves a human Judgment of what do they actually mean when they say that my house is warm and then they do something which was crazy before they overexpose instead of underexpose their images and then they get this glowing feeling out and so that's one of the illustrations of how we could sort of go through and think of a technology development process in a holistic perspective and think of it with all of the stakeholders included not just about how we create better product but how we create better and more productive people but also how we change organizationally the way that we deliver value in companies so that's what we are uh trying to push and we're trying to uh put together sort of a a PCT or something like I'm trying to do a kind of thing in Denmark where it's where you can get a brand of hyper intelligence if you follow a certain set of principles and we would like to develop those principles together with you H and then maybe in the coming year or two we could have sort of like an export success uh of taking those principles and then going global with them and that would be an amazing journey to be with you uh so so reach out to me if you're interested in something like that thank you