Rise of AI Conference 2024 · Berlin

Global AI Echoes

Björn Bringmann · Managing Director and Lead, AI Institute, Deloitte Germany

At Rise of AI 2024 in Berlin, Björn Bringmann, Managing Director and Lead at the AI Institute of Deloitte Germany, presents findings from Deloitte's quarterly AI surveys alongside outside data such as the Stanford AI Index. He argues that ChatGPT made AI tangible almost overnight despite remaining one of the most opaque models ever released, and reviews recent milestones including AlphaFold 3, open-source models like Llama reaching top-tier leaderboard scores, the EU AI Act, new Nvidia chips, and European generative AI players such as Mistral and Aleph Alpha. Using an AI ranking that scores countries on implementation, innovation and investment, he shows Germany performing reasonably on government strategy and operational metrics but slipping in research and commercialization, and highlights a large gap in R&D spending between the top US, Chinese and German companies. He walks through Deloitte survey data on German and global sentiment: perceived threat from generative AI to business models, high productivity expectations, shifting adoption priorities, talent and governance barriers, uneven employee access to GenAI tools, and views on whether AI will centralize or decentralize power. He closes urging German and European companies to move faster and pursue more ambitious "moonshot" projects.

Key takeaways

  • ChatGPT made AI broadly tangible almost overnight despite remaining, in Bringmann's view, one of the most opaque models ever released.
  • Germany ranks second globally on AI government strategy but fell in research (6th to 8th) and commercialization (8th to 11th) over two years.
  • The top three US companies spend about 31 billion euros a year on R&D versus roughly 11-12 billion for Germany and China's top three.
  • About 55 percent of companies give no more than 20 percent of employees access to a basic generative AI tool.
  • A majority of respondents in Germany, Europe and globally expect generative AI to centralize rather than decentralize power.

Detailed notes

From explainability efforts to ChatGPT's overnight impact

Bringmann opens by noting that Deloitte now runs its AI survey every three months instead of annually to keep pace with the field. He observes that years of work on making AI models explainable, aimed at overcoming distrust of "black box" systems, produced steady but unspectacular progress, whereas ChatGPT made AI tangible almost overnight roughly a year and a half before this talk, even though it remains, in his view, one of the most intransparent models to date, showing that once people can use a system, transparency concerns become less central to adoption.

Recent achievements and European players

Bringmann highlights AlphaFold 3, calling it especially notable because it delivers direct impact in pharma and medical research rather than being another chatbot or voice assistant. He notes that Meta's open-source Llama model now scores close to the top proprietary, API-only models on leading benchmarks, arguing open-source progress will likely accelerate further given the larger number of contributors outside private companies. He also cites the EU AI Act as a flawed but valuable first step, new Nvidia chip generations expanding available compute, and Mistral and Aleph Alpha as Europe's two significant generative AI language-model players. He references German politician Christian Lindner publicly warning that Germany cannot afford to miss the AI wave as it has missed other technology waves, and cites statements from figures including Eric Schmidt and Amazon CEO Andy Jassy describing AI as still underhyped in the long run.

Germany's position: implementation, innovation and investment

Citing an AI ranking that scores countries on implementation (talent, operations, infrastructure), innovation (research and development) and investment (commercialization and government strategy), Bringmann says Germany ranks second globally on government strategy, largely due to the EU AI Act, and 13th on operational excellence, ahead of the United States at 28th, though behind China overall. Comparing the current ranking to two years earlier, he notes Germany improved on most measures but fell in research, from 6th to 8th, and in commercialization, from 8th to 11th, even though absolute investment rose, which he reads as Germany moving too slowly relative to competitors. He also cites average annual R&D spending among each region's top three companies: about 31 billion euros for the top three US companies, 12 billion for China, and 11 billion for Germany, and notes that Germany's top spenders are concentrated in automotive rather than software, with Volkswagen's R&D spending at roughly half of Alphabet's alone.

Survey findings: threat perception, priorities and barriers

Bringmann presents Deloitte survey results showing German business leaders initially feeling more relaxed than the global average about generative AI threatening their business models (about 50 percent seeing no threat versus a 36 percent global average at the end of the prior year), a gap that narrowed toward more realistic concern three months later. Across regions, roughly 90 percent of respondents expect a significant productivity increase from AI and about 80 percent expect substantial transformation within three years. Top adoption priorities shifted from efficiency and productivity, which fell from about 63 percent to roughly 50 percent in Germany, toward using AI for fraud and risk management, which rose from about 10-15 percent to nearly 30 percent, while using AI to accelerate software development fell from about 25 percent to 18 percent. The most cited barrier, in Germany and globally, is a shortage of skilled talent, followed by employee resistance and governance and regulatory concerns; German respondents' confidence in already knowing the right use cases dropped sharply between survey waves, which Bringmann reads as growing realism about how hard use-case selection actually is.

Uneven tool adoption and societal concerns

Bringmann reports that simple GenAI tool rollout inside companies remains highly uneven: about 55 percent of companies give no more than 20 percent of employees access to a basic generative AI tool, while only about 16 percent of companies give 80 percent or more of employees access. He also presents survey data on whether respondents expect generative AI to centralize or decentralize power: a large majority across global, European and German samples expect centralization, with the United Kingdom showing the least concern, though still leaning toward centralization; just over 30 percent expect decentralization. He notes public trust concerns are shifting from whether a model itself can be trusted to whether the companies using it can be trusted, illustrated by Stack Overflow users objecting, and being blocked from removing their answers, after learning their contributions were used to train a commercial large language model under terms they had clicked through without expecting that use.

Missed opportunities and recommendations

Bringmann recommends companies invest in people, set up governance properly, avoid fixating on a single use case, and adopt a beginner's mindset of strategic experimentation, adding that AI should not be siloed in one part of a company and that efforts should target broader transformation rather than isolated pilots. He argues companies should not wait for the technology to feel finished before starting. On missed opportunities, he notes that AI investment skews toward marketing and commercial applications with comparatively little directed at societal problems such as climate change, poverty or health, that German companies tend to move slowly because of internal pressure to justify investment upfront, unlike faster-moving US subsidiaries of the same firms, and that Germany lacks ambitious "moonshot" projects, citing early German self-driving car research in the 1980s that was not followed through. He closes by calling for Germany and Europe to move faster and pursue bigger bets.

Quotes

The future is here, it's just not very evenly distributed.

Björn Bringmann

It's really hard to shrink your way to success.

Björn Bringmann

If you want to be around in two years, you better start now.

Björn Bringmann

Questions answered in this talk

How does Germany rank internationally on AI compared to the US and China?

According to Björn Bringmann at Rise of AI 2024, Germany ranks second globally on AI government strategy and 13th on operational excellence, ahead of the US at 28th but behind China overall, though it has slipped in research and commercialization rankings over the past two years.

How widely have companies actually rolled out generative AI tools to employees?

According to Björn Bringmann at Rise of AI 2024, Deloitte survey data shows roughly 55 percent of companies give no more than 20 percent of employees access to a basic generative AI tool, while only about 16 percent of companies give 80 percent or more of employees access.

Do people expect AI to centralize or decentralize power?

According to Björn Bringmann at Rise of AI 2024, a large majority of survey respondents across Germany, Europe and globally expect generative AI to lead to further centralization of power, with just over 30 percent expecting it to lead to decentralization.

Mentioned

EU AI Act, Stanford AI Index, Generative AI adoption, AI talent gap, Centralization of power

Keep exploring

More from Björn Bringmann

More from Rise of AI Conference 2024

More on Enterprise AI & Adoption

Transcript

Auto-generated transcript, not edited. · 4,387 words

Show the full transcript

wonderful thanks for the nice introduction and a pleasure to see you again here um on stage as just mentioned and uh ah awesome um going to have a little bit of a look into Global area echos so what has been happening around the world and uh that in some numbers there a bunch of surveys we run a survey usually every year we decided that that's not fast enough so we run it every 3 months by now to kind of stay on top and stay at the pace uh the topic is moving and I brought some numbers with it to really have a look into um what we have achieved uh how it looks right now and what maybe we have missed or might miss so without further Ado um that one slide doesn't really fit the rest of the presentation um but it's still something that that struck me as as pretty impressive for years the community has been working on making AI explainable because the whole back push was hey um this is a

complex model I don't understand it so I can't use it so there was a huge effort into making them more transparent um more understandable all these kind of things which I think was a great effort and had certainly uh quite some good impact in moving forward but nothing really magical happened out of this and then overnight um there was one system which made AI incredibly tangible and suddenly a flurry of activity started um and you know what I'm talking about that's Christmas about one and a half years ago and what I think is striking is that that model is still despite being newer models available by now one of the most intransparent that there ever has been so it turns out that as soon as people kind of get a feeling for how to use it and what it can do all the transparency I don't want to say is out the window but apparently become quite a lot less relevant um so all of this basically brought us here and that's just a slide

that's basically out ofd by definition but obviously you know there's a lot of stuff going on in the media um that's hard to keep up with looking a little bit deeper um and that's from the uh from Sanford from the state of AI uh different topics that AI researchers working on and that used to be human only so I just put AI in one corner and human intelligence in the other one and just quick feeling of what you see there uh what's basically where the AI would win and where the human would win and I'm I'm afraid we are on the retreat here um which is obviously an achievement on the side of AI but also typically BS many questions as to okay what is the human then going to do right most of the time the answer is typically collaboration and it's not that easy looking a little bit different in terms of achievements and obviously that's a very very small selection right there's tons of things going on um one thing that has

been released I over the weekend last week was Alpha fold the third version um which I think is pretty impressive because it's slightly different from all the other uh typal media things where you have voice Bots and chat Bots and whatnot which are all very tangible and very interesting but that one um actually helps directly in the Pharma industry um medical health and so on uh moving things forward that also just like go or chess previously were thought to be unsolvable uh with a machine and now you know it's it's certain Fields forward that's the other topics that was Christian L actually um so the other topics on here bottom left might not be the most intuitive picture but that's one of the famous leaderboards in terms of models competing with each other and the one that's marked is the Lama model um which is almost on the same score as the top private models so for quite a while the um the statement was the private models so owned by companies built by companies

not really available except for an API are the ones which are really leading in the market but then the open source community so to speak with some support in this case obviously for meta um picked up and pushed forward and the open source models are now playing in the top tier as well which I think personally will further accelerate because there just many more people in the open source Community than within private companies um other things obviously the UI act I'm not going to say too much about this but it is an achievement many on the rest of the world were discussing here there was a lot of skepticism but we have it and it turns out it might not be perfect but uh it's better to have it than not to have it um down there uh a new chip you know Nvidia is crunching them out by the months in new versions kind of uh again quite a lot of capacity um available for this kind of technology and then more on the

European side I mean the UA obviously as well um two startups which are usually in the news Mr you will all have heard of I Alpha you will all have heard of um which are the two significant players um regarding uh language models in generative AI from Europe built here designed here and run here then we just had the mention Zer I just put him there as an example and I've heard it before from Peter and and others in the panel as well a lot of the research is actually coming from Europe um obious not all of it but we're quite good at this and there are certain achievements let's see how the uh X lstms are going to do not that easy to pronounce uh but good good uh interesting um train of thought here and finally um that's a guy you just heard speaking between chrisan Lindner uh a few weeks ago I believe on the omr stage U one of our top politicians obviously going out there and saying we cannot miss

the whole AI thing as we have missed so many others in Germany so I was quite happy to hear that there is somebody going on a big stage basically just saying this there obviously was not directly an approach on how to but awareness is a is a good first step and when you look and these are obviously um mostly us folks on how much relevance the whole topic will have I was quite surprised by these statements which are basically all say um we're all talking about quite a bit of hype on AI but if you listen to these people they say well that's not enough yet so there's more to come some say okay maybe now it's a little bit overhyped but in the long term it's underhyped Eric Schmidt just said this stuff is underhyped um you know there might be Personal Agenda still um Andy CEO of Amazon uh said recently you know if there's some big bet I would go with right now then it's actually geni so again three which basically

um all bit on this topic which leaves us basically with this um it's a topic that we really should pick up on and push forward and we do this here which I think is great because otherwise we're going to be left in the dust to the numbers when you look at um and in this case it's from GEOS media the AI index and it's a ranking but the order on the left is not the correct one it's just that I wanted to highlight Germany here um they look at three different three different levels one is implementation um um another one is innovation and a third one is investment and you can see in implementation you know it's about Talent it's about um operations it's about infrastructure Innovation is about research and development and investment is uh pretty much about commercialization and government strategy we're number two in this ranking regarding government strategy which I believe is mostly due to the eui ACT and overall we're not too bad right if we look at operational excellence

number 13 the US is at number 28 so while they're leading overall and China being second um the numbers don't look that bad for Germany what's interesting I believe is how it changed over the last uh two years I believe so these are numbers from 2023 and the one before was from 2021 we went up on almost all of the topics except for two which is research we went down we went from six to two uh sorry from 6 to 8 um where I don't have an explanation for yet uh but sounds like research is lacking a bit probably Talent moving overseas and what's painful Discovery was that the commercialization actually went down when everybody is saying that that should go up so let me just highlight this this went down from we have been on number eight now we're on number 11 and the absolute numbers were actually going up which means we're just being too slow coming back to Fabian wordss from earlier um a bit of speed wouldn't be a bad idea

another one if you look at R&D spent from the top three companies average these are striking numbers that 31 billion euro per year from the top three us companies 12 billion in China and 11 billion in Germany now 11 to 12 is not so so much of a difference but if you just quickly consider how much you get for a billion in Germany and how much you get for a billion in China uh there might be quite a gap again the other striking party I believe is if you look at which these top three companies are you will see on the top six software companies and on the bottom three Automotive companies that might be an issue and if you sort them and the list is top 10 I shorten it to six on the bottom um is no other German company um there's HH in there which at least is European but you can see again um the top four actually from the US so Apple didn't make it in the top three um

and then there's UI from China and while Volkswagen is constantly spending a lot on R&D they half of what alphabet is spending on it and alphabet is just one of the companies doing R&D and I've heard this before uh today there's quite a lot uh that we probably need to catch up with and it kind of highlights this problem with the commercialization um because obviously these are commercial companies a question from the safier that we ran how threatened do Executives or decision makers feel in Germany in this case by regarding the business model U by generative Ai and what you can see what is quite striking if you add the none and the little up that's 50% 50% of the surveyed people um so all Business Leaders basically said I I don't see any threat that was end of last year we were in Germany leading with being you know relaxed about it 36 was the global average so globally the average about roughly 3,000 people said you know onethird was relaxed about the topic

of AI regarding the business model moving three months ahead the global number hardly changed but Germans got a little bit more realistic about it which I think is good news so there's a lot more attention uh quite apparently on that topic uh being somewhat threatening to more business models than previously thought also the extreme threat from 1% is gone but I mean 1% um fair enough if you look at what's being expected it's quite stunning that no matter how you cut the data set if it's Global if it's Germany if it's the financial services industry it's always roughly 90% who expect a significant increase in their productivity significant impact on the productivity and same count roughly 80% expect substantial transformation with in the next 3 years the question for me always a little bit there's at least 10% that want to have quite a lot of productivity increase but don't want to change I'm really curious to see how that will work um but again huge expectations uh about many changes coming up when we

use about which topic the expectation is that there will be an impact and you see the slightly shaded bars um are again from end of last year the ones you can see a little bit better are from the first quarter of this of this year so improving efficiency and productivity went from 63% being by far the most important topic for Germany again globally there was also leading but about 10 percentage point less went down to just about 50% um same with reducing costs went down just a little bit um the topic of encouraging Innovation growth uh is roughly the same as before and you see on the bottom end there are another 10 that I did list here which are all somewhere in the in the 20s um interestingly I think is that fraud and risk went up from about 10 15% to nearly 30% so using jna for these topics uh apparently folks have discovered that there's a lot that they can do while at the same point as surprising to me using it

in software development to accelerate it went down quite a bit from you know 25% to 18 or something um but still is basically the same message that companies tend to focus on rather technical topics at the moment right short term cost and efficiency improvements and while it has gotten better strategic and long-term bats that might be a little bit harder to develop um are still kind of on the hindsight seeing all of this um the big question is usually is what's the problem what's holding us back and what you can see here is really um again the percentage numbers from now outstanding globally and in Germany is Talent right skilled employees if it's the people that develop it the people that use it doesn't matter there needs to be a better understanding more people that can actually handle AI um at the same point there's quite a bit of employee resistance and these are again German numbers of pushing uh the usage and the implementation of AI bank so two culture topics in one way

or the other then there's another Bunch uh regarding risks governance regulatory topics and so on which are reasonably high and lastly and topics more implementation the gray dots are again the ones from end of last year skill shortages and talent issues have apparently gone down quite a bit from 41% to 35 36 now um kind of alliance with the rest lots of the part in the middle has gone up to me the story I see here is that people have understood more and have realized that they're more challenges than they expected to be and use cases was the most surprising to me everybody was like oh that's easy that's 15% like Germans were absolutely convinced that they know what the use cases are we we were the best in this globally 3 months later it turns out Maybe not maybe maybe that's actually trickier than we thought um brings us basically to three topics of what to do right obviously invest in people make sure you get this governance set up right don't hunt a

single use case um and lastly this whole idea of experimenting strategically having a beginner's mindset is something that that should be encouraged and we see successful companies Doo and that brings a whole list of subtopics with it I'm not going to go through those um but I think the most important ones uh not having it siloed into one particular part of a company is quite important thinking about not just single use case but a little bit bigger call it Target operating model call it transformation something um and lastly what's probably the most simple statement and the most important one at the same point is don't wait we've had so many companies so many uh sea levels that told me like okay so it's not really completely finished yet this gen stuff or this AI of when should I start well you know if you want to be arounded two years you better start now um even if it's not finished because it's mature enough and even if what you do right now doesn't work out

in the end you will learn a lot um the other question was about talent and again end of last year what was quite stunning is that Germany was leading in terms of saying Talent is our main issue and the gry Bars were what Germany was doing to of Germany that we were pretty much on the last position so that's uh somebody saying like we really have a big problem here but we're not going to do anything about it and I was really surprised I have the numbers for like you know 10 days or something I was really surprised globally everything went to you know 30 to 40% all averag 36% I made the calculation five times it is actually the same for all of three um and for Germany look at that I mean apart from educating Workforce you know where we're not so I mean it's about what seven percentage points it's not too bad the other two we're kind of overshooting right at least if you compare to Global averages so apparently recruiting

um that's something that many people really took to heart so if you're looking for a job might be a good opportunity not that that was bad before um and in terms of reskilling workers that's that's like a 14% plus that's a lot so there's a lot of Education and Training apparently going on right now or at least being planned the other one is about adoption which a most simple way to adopt I believe is introducing some simple gen tool you know white label bar Gemini jbt whatever uh into your company so your employees can actually work and play with it and these are the now global numbers German numbers look roughly the same and you can see 25% of the companies say that less than 20% of the employees have access to such tools right then you have and it's luckily quarters roughly 50% let say up to 40% have access to these tools and so on and on the top um there only 16% of companies where 80 or more employees 80% or more

employees have access to these tools that looks great right except that's the top 5% other side which is now the bottom 65 so I kind of cut out the middle part looks like this 55% of companies say that no more than 20% of the employees have access to a simple tool of whatever form that's not how adoption looks like when you ask me and that's really really easy to fix um so I hope this will change uh in the next story which is basically running right now so what we can see is the future is here uh it's just not very evenly distributed which brings me to one other point if we look at this from a social perspective and that was a stunning question to me um and I hope the ground gets it across because it took me a while to put these numbers in some image we asked the same people again and these are um well you can see you know Global numbers Europe uh Germany and so on uh end

of last year again will J I rather lead to a further centralization of power or to further decentralization of power the blue ones are all the people that said it would rather lead to a centralization of power and the green ones rather to a distribution of power you can see a vast majority believes that this will lead to centralization of power which I believe is a rather negative view for most um you know unless you're one of those few which actually get that power then and very few I mean still more than 30% but comparatively few think that is going to lead to a decentralization of power so democratization if you like and the interesting part is the UK is the one where least people think it's centralization and most people think it's decentralization and still it's more uh tilted towards centralization that's a bit scary to me to be honest so um I think that's something also to be kind of kept in sight when we discuss this topic to make sure this whole

thing doesn't go the wrong way because there's a lot of trust questions around it there have always been Trust questions usually they were about can I trust the model now it seems to shift more to can I trust the companies that use those models right there seems to be this shift there's just been recently I'm pretty sure most of you have seen this um St overflow providing their data and while in the terms and conditions it said that if you provide an answer to stake overflow it's theirs people when they signed you know with the typical click this one button thing uh disagreement didn't consider that their answers would be used into one large model that's you know being commercialized in some other way and now they're trying to basically some of them trying to basically take down their answers because they don't like their answers to be included in one of these models and are being blocked from doing this so there's quite an interesting fight going on um now you can say well

they signed this but then again you know how many terms and conditions of 40 pages do you click okay on every day so it's a bit of a tricky situation so uh social topics quite quite challenging and and quite interesting and I think here it's also to me pretty obvious that um maybe we should not get everything from outside of our European borders at least but make sure that um if we care about this from a societal Point uh make sure that we also support the people that we have here to make sure that you know we can actually have a say in this and lastly um missed opportunity is really really hard to put down because there it's very unclear and very many at the same time I discussed with uh my Global counterparts and what they believe are the missed opportunities one is there's a lot going into right now and it's understandable for the business point of view in terms of using this for marketing you know selling more kind of things

there's not so much on actual Society problems like using it for climate change or against rather um seeing if if we can do something about poverty or or health they do progress but a lot less than there might be for enough these might also slightly more challenging at the same point what we see certainly in Germany in in many discussions is that companies tend to do small things and reasonably slow there are not too many companies that say let's do this quickly and it's funny enough if I speak to German companies headquartered here in Germany but have subsidiaries in you know the US and where they say our us colleagues um are doing a lot more than we here because we have to prove so many things up front um that the investment is worth it while the US colleagues can just get the money and start with it so there's a lot on on the culture side I believe in the in the that we were too slow and Fabian mentioned this at the

beginning there's a dire need for speed um and the other one is a small part there's really lots of I wouldn't say tiny but rather small use cases being discussed they're not so many moonshots and Peter said this before as well um we certainly lack a number of moonshots and that's it's funny I talked to my colleagues globally and they say hey Germany you're not really a bad position why are you not doing any moonshots and the example which doesn't work for this topic so nicely but makes sense not it's like we would have expected self-driving cars being pushed out of Germany why didn't that happen and we had one of the first in the 80s right but that kind of then went in the Shelf um and now there are opportunities to do these moonshots it seems we have to get our egg together and it doesn't matter if I talk globally or in Europe what I basically hear a lot is like Germany please get your act together because for Europe we're looking

on you so that's quite some responsibility and I think it's not too hard to get there and certainly it will be beneficial so final word more moonshots it's really hard to shrink your way to success only you know cut costs and optimize these things final word number of employees from the survey supposed to go up so some good news here thank you so much

Rise of AI 2027 · Berlin · May 2027

Be in the room for the next one.

Ten years of talks started as conversations in one curated room in Berlin. The next edition brings 300 selected leaders together in May 2027.