Rise of AI Summit 2020 · Berlin · Society & Strategy Day | Future Space

The Global AI Landscape

Hans Uszkoreit · Scientific Director, German Research Center for Artificial Intelligence (DFKI); Founder and Chief Scientist of an AI company in Beijing

At the Rise of AI Summit 2020, Hans Uszkoreit, scientific director at DFKI in Berlin and founder and chief scientist of an AI company in Beijing, compares the AI strengths of the United States, China and Europe after nearly a year back in Europe following three years in China. Building on Kai-Fu Lee's comparison of the two AI giants, he argues Europe is not leading in AI for internet services, consumer electronics or finance, but holds strong chances in manufacturing, Industry 4.0 and enterprise AI, and in the next wave of AI that combines machine learning with explicit knowledge graphs. He describes the 'intelligent enterprise' as combining data, knowledge and business processes, while cautioning that AI cannot automate the most dynamic, high-level decisions. He argues Europe's persistent weakness is the last mile from basic research to scaled commercialization, and that German 'Angst' around data, illustrated by the country's Corona contact-tracing app, is holding AI adoption back. He proposes secured, trust-governed centralized data storage instead of current data protection norms, and closes discussing whether AI is dangerous and how enterprises can innovate despite AI's bias toward broad, data-rich topics.

Key takeaways

  • Uszkoreit argues Europe is not leading in AI for internet, consumer electronics or finance, but has strongholds in manufacturing and enterprise AI.
  • He frames the intelligent enterprise as combining data, knowledge and business processes, an idea he traces to 1992.
  • He argues AI cannot automate high-level enterprise decisions because markets and regulations change too fast and unpredictably.
  • He says Europe's persistent weakness is the last mile from basic research to scaled commercialization, citing GPS, Siri and Google Translate.
  • He attributes Germany's slow AI and data adoption to a widespread fear he calls German Angst, illustrated by the Corona app's low data-sharing rate.

Detailed notes

Where Europe stands between the two AI giants

Uszkoreit revisits Kai-Fu Lee's comparison of US and Chinese AI strength, which last year left Europe as a marginal 'safe' player. He offers a more fine-grained view: Europe is not leading and likely will not lead in AI for internet services, consumer electronics or finance, and automotive and public transport remain open questions, but manufacturing, Industry 4.0 and increasingly enterprise AI are strongholds where Europe has real chances.

Enterprise AI and the intelligent enterprise

He frames digital transformation as two waves: first digitizing data processing and resource planning, then adding intelligence through Industry 4.0, digital twins, robotics and AI applications, toward a vision of the 'intelligent enterprise' that he traces to 1992. He describes this as a complex organism of people, technology, knowledge and intelligent processes, resting on three ingredients: available and interoperable data, managed and increasingly machine-usable knowledge, and business processes, both operational and support, that must be adapted or redesigned and modeled digitally. He stresses limits: AI can automate narrow tasks and even some end-to-end processes, but not higher-level decision making, because markets, regulations and external conditions change too fast and combinatorially for any system to predict tomorrow's decisions from past data. He lists main enterprise AI applications: interpreting unstructured data, demand forecasting, anomaly and event detection, business process modeling, and human-machine interaction. He argues the central challenge for the next AI wave is combining the implicit knowledge learned by deep learning models with explicit knowledge held in resources like the Google and Baidu knowledge graphs and Wikidata, and describes his own company's approach of web crawling with partner Ubermetrics, extracting roughly 80 types of events into knowledge graphs used for supplier, competitor, portfolio and compliance monitoring.

Europe's basic research strength and its last-mile weakness

Uszkoreit says European basic research has long led in AI and is well positioned for hybrid AI and human-centered computing, but Europe repeatedly fails at the 'last mile' between applied research and scaled commercialization, including venture-capital risk-taking. He cites DARPA-linked technologies such as the internet, GPS, Siri and Google Translate as often rooted in European research but ultimately developed and commercialized in the United States.

German Angst, data and the Corona app

He argues Germany's deeper problem is a pervasive fear of digital data and algorithms, contrasting it with his experience in China, and illustrates it with examples like overflowing paper mailboxes, cash-only restaurants and public offices without email or IT infrastructure. He criticizes Germany's Corona contact-tracing app as an example of good intentions producing a nearly useless tool: about 25% penetration, with just over 10% of infected users sharing positive results, far below the roughly 60% penetration Oxford researchers estimated was needed for real impact, which he attributes to political decisions rather than the technology. His proposed fix is a highly secured, centralized data store, which he suggests could be called GAIA-X, governed independently like a supreme court or central bank, with strict legislation and harsh penalties for abuse, shifting focus from restricting data collection to protecting people from misuse.

Is AI dangerous, and Q&A highlights

Weighing a German researcher's warning that AI is more dangerous than the atomic bomb against Andrew Ng's view that fearing evil AI is like fearing Mars overpopulation, Uszkoreit sides with caution grounded in current limits: today's AI lacks self-consciousness and free will, so it cannot be evil, but critical deployments still warrant care and continued work on transparent, accountable AI. In the closing Q&A, he agrees that an intelligent enterprise increasingly acts as its own decision-making subject, since more than 90% of decisions are already collective rather than made by a single CEO, and, asked whether AI is inherently a conservative consultant favoring broad, data-rich topics, he says the fix is infusing systems with explicit, condensed knowledge alongside the implicit knowledge learned from few examples.

Quotes

Data are not harmful for health or environment, and data are good, nothing bad about them.

Hans Uszkoreit

We should be very cautious and we should continue to work on transparent, reliable and accountable AI.

Hans Uszkoreit

Questions answered in this talk

Where does Europe stand in the global AI race between the US and China?

According to Hans Uszkoreit at Rise of AI Summit 2020, Europe is not leading in AI for internet services, consumer electronics or finance, but has strong chances in manufacturing, Industry 4.0 and enterprise AI, and in combining machine learning with explicit knowledge for the next wave of AI.

Why does Uszkoreit think Germany's Corona contact-tracing app achieved little?

According to Hans Uszkoreit at Rise of AI Summit 2020, the app reached only about 25% penetration and just over 10% of infected users shared their positive test results, far below the roughly 60% penetration Oxford researchers estimated was needed to be effective, which he blames on political decisions rather than the technology itself.

What is Europe's biggest weakness in turning AI research into successful companies?

According to Hans Uszkoreit at Rise of AI Summit 2020, Europe is strong in basic research but weak at the last mile between research and scaled commercialization, so technologies with European roots, such as GPS, Siri and Google Translate, were ultimately developed and commercialized in the United States.

Mentioned

knowledge graph, intelligent enterprise, digital transformation, Corona-Warn-App, narrow AI, hybrid AI, Industry 4.0, GAIA-X

Keep exploring

More from Hans Uszkoreit

More from Rise of AI Summit 2020

More on Enterprise AI & Adoption

Transcript

Auto-generated transcript, not edited. · 4,486 words

Show the full transcript

i'm happy now to announce our next speaker who was already mentioned very often this morning he is the founder and chief scientist of giants technologies in beijing he's also scientific director at the german research center for artificial intelligence dfke in berlin where he is currently coordinating china activities and he's also honorary professor of our technical university here in berlin he has over 250 international publications under his belt he has worked in ai for over 30 years i will not list his numerous other awards and honors because that would significantly reduces time and he needs every minute of it please give it up for Hans Uszkoreit good morning and thank you for the nice introduction thank you for having me here today at last year's rise of ai i talked about my impressions my experience with ai and china after my first three years in china and today having been back to europe for almost a year now i want to talk about my impressions of europe as seen through the eyes with somebody who has

been in china for three years last year i was discussing this picture from the book by kaifu lee in which kaifu lee compares the strengths of the two giants in ai the united states and china i won't go through this here today but actually the comparison came out very favorably for china not surprisingly and today i want to say and back then already there was the question where's europe in this global race of the two giants and the picture that came to my mind after reading kaifu lee's book was this you may remember from last year there are two giants fighting but where's europe according to kaifuli it should be down there you see the green guy there that should be maybe europe europe is on the safe side still making ai more secure and more good before really starting so today i want to talk about the respective strengths of the global players in ai i want to uh suggest a two-pronged zweigley strategy for europe and germany i want to talk about enterprise

ai a huge chance for germany and europe the next wave of ai for which i think all bets are still very much open and i want to address the question why does europe and germany not have equal chances yet i think what are the handicaps especially for human-centered ai and i would finish with an outlook and answering the question what should we do next so this is a little more um this is a little more fine-grained than kai fully's findings and it also includes europe and europe as fabian already said is not all that bad in all fields yes it is true in in ai for internet and ai for consumer electronics europe is not really leading and won't be leading in the next few years also in the financial sector maybe not with automotive we don't know we see lots of future developments in china and the us but we don't know yet whether europe still has a chance when it comes to public transport yes maybe we can be we could come become

better but also united states is not very strong there as you know but our two strongholds are manufacturing industry 4.0 and also growingly our chances would be an enterprise ai and this is what i want to talk about next about enterprise ai so one of the two prongs is to exploit our excellent chances in business ai smart manufacturing but today since i've given many talks on uh industry 4.0 already today i want to talk about this new emerging field of enterprise ai so as you know many people talk about digital transformation intelligent transformation what does it really mean so the digital transformation comes in two big waves the first wave is the digitization of data processing resource planning and so on and the second wave of digitization is the one that brings in intelligence the sophisticated things industry 4.0 digital twin robots ai applications and the goal according to companies like sap but many other players is the vision that is already very old it comes from 1992 of the intelligent enterprise so the intelligent

transformation the second wave of digital transformation leads hopefully to the intelligent enterprise there is now a complex definition of intelligent enterprise but you don't really have to read it here actually i will abbreviate it in simpler words it is a complex organisms consisting of people technology resources of knowledge of learned experience also machine learned experience and intelligent processes that show signs of intelligence of collective intelligence that go beyond the intelligence of the individual players so i will highlight some of these ingredients in the following slides the first ingredient is of course data i do not need to say much about that so a prerequisite for applying ai and analytics deeper intelligence is the availability the fusion the interoperability of the data in the whole enterprise clear the second one is knowledge this is the old dream of knowledge management but now on a different level the management of the knowledge resources that are on paper hopefully then digitized that are partially already machine usable in structures such as digital process models i'll come back to

that and in many other forms but let's not talk about that that's knowledge management this is the important part that will make the difference the third part is the processes so business processes come in two flavors there are the operational processes the support processes and the management processes and parts of the management processes are the strategic planning the controlling the coordination on the top they overlap of course all of this yeah and on the core operational processes in a production company it's production assembly quality assurance these things and then there are support uh functions support processes like human resources marketing purchasing and so on so we all know that and now the the thing is what can we do with them then in the next step in the intelligent enterprise so we need to adapt existing processes to this higher form of digital transformation adapt them i will come back and and say how we need to design new processes that use the data and knowledge resources that were not there before yeah either adapt

the old or define new ones and we have to create and this is the core part digital process models people are already starting that but it's yet not really connected with knowledge resources like the corporate knowledge graphs and it is not connected yet with machine learning people are trying it already for instance in robotic process automation and trying to watch people watch processes and then model these processes and improvement and then we need facilities for monitoring measuring and optimizing the processes with the help of ai and finally it comes to the automation of individual steps in the processes and of sub-processes but now the question is can ai also automate entire processes yes in general yes there are some processes that and can be modeled uh not modeled automated entirely in in in in the business world but be careful now and this is really the essence then of understanding the limits and the challenges of enterprise ai there are some processes on the higher level of decision making that ai even if people tell

you the contrary cannot be automated by ai so what does it mean that means that even if you have the best machine learning systems and let the best managers let these ai systems watch the best managers and get all the data of their decision they cannot predict or make the decision of today and tomorrow why because we live in a constantly changing environment the world is very dynamic regulations the market situations are different today than they were last year and the system and also the combinatorics of all the external data the macroeconomic data the uh market uh the ever-changing market they are so complex it's impossible so one of the challenges of ai is now to amplify human intelligence and to make sure that the decision making that still exists is better decision making by having much better input by having much better decision for possibilities for decision making and so on and also because of language technology and visualization to bridge the barrier between the machine and the decision makers that are there so

this is a real challenge for ai so the main application areas for ai and the enterprise are the following one is the interpretation of unstructured data for instance for quality management yeah but also for written partially written business documents transform them into a digital machine-usable way learning from unstructured data for prediction demand forecasts and analytics optimization then the detection of anomalies event predictions of events and then two very very big areas one is business process modeling and management i already mentioned that and i would predict that as we have big language models today like gpt3 yeah or the bird based models we will have process models of huge models of processes that are trained and that are able to learn the the the the the processes and learn from the syntax and the semantics of processes and then the last one is the human machine interaction that i just mentioned that is absolutely necessary the seamless integration of human and machine intelligence now there is one challenge of course and we see this already in

the enterprise why didn't the enterprise why wasn't the enterprise before internet ai and electronics ai and uh and perception area the reason is very simple the the hurdle is very high so the ambition the demands are very high because in industry in unexplainable learning can only solve individual tasks this is a narrow ai here you can apply ai either for demand forecast or for for maybe quality assurance for any of these things maybe for speech output for something but every bigger thing like a task driven chatbot that talks to the customers or to the partners needs knowledge it cannot only be based on a model on data so there's ever-changing knowledge so the best systems of that type now in addition to having machine learning they are already using knowledge sources and they combine maybe speech and language understanding and language generation several things this is going from narrow ai to broader ai also in business process modeling going from written specification of processes and transform those to business models yeah this is requires quite

a number of components and also support for decision processes i'll come back to that so it is the and and the main big thing which will be the next uh revolution of ai that many people know is almost everybody in ai feels this is necessary but most of the best people are already working on it is how can we combine now the implicit knowledge learning of deep learning how can we combine that with explicit knowledge which can also be learned of course but it's a different way it's discrete learning yeah so how can these things how can the huge amounts of knowledge that we have in our knowledge graphs like the google knowledge graph the baidu knowledge graph with billions of facts here also wiki data how can they be combined with the experience learning yeah so this is the big challenge now and this is absolutely needed for success in the enterprise world it will come because only that will enable bringing the ai in the higher in the higher floors of the enterprise

so the next thing i will briefly mention is the combination of internal and external data exploitation usually starts with internal data but external are becoming more important so for instance an sap in sap people are talking about x and o data x are mainly experience data and then the ai brings those together for analytics but it's much more than customer experience it is your customers your suppliers your partners your competitors your employees or prospective employees your products relevant technologies regulations and so a big company already now has to follow all of these and since many companies are globally operating you have to do it all over the world in many languages and that you cannot do without ai especially without language understanding because most of these news are in language few of them unstructured form so actually this is the the external data are full of untapped potential and i don't know if you know that the 16 intelligence services of the united states mainly use open data 95 of their stuff does not come

from spies taking secret pictures but it comes actually from open data so what we need then is something that brings now the data from all over the world together with these internal data right now it is still separate so marketing gets their data sales customer supplier supply chain monitoring and so on competitor monitoring but of course in the next sorry in the next phase that needs to be combined to in be able to be fused with internal data so that's a big thing so we are doing this in giants we are doing powerful web crawling with our partner company uber metrics then we extract information and then we give it to the customer and we integrate it already in knowledge graphs so if we get knowledge then we get them into knowledge graphs knowledge graphs of that form so we ex oh the chinese show okay it doesn't matter so we we extract some messages yeah we extract some events and use and then we put them in knowledge graphs and we do that with

many many topics financial so about 70 no now around 80 different types of events we can detect already we use them for supplier monitoring competitor watch portfolio monitoring compliance watch customer watch and so on many applications and it combines now let's come to the second prong now european basic research has always been on the forefront of ai and european research is well equipped for the next phase for hybrid ai and european researchers and funding bodies have already embraced human centered computing wonderful topic human-centered computing ai for education healthcare personal assistance and all these things why can't we do it i am extremely pessimistic instead of building fantastic ai applications for our citizens european ai is asked to start with rules constrained standards even the praised white paper it's good that we have a white paper as was pointed out for instance by rasmus 20 pages and most of it is actually regulation and constraining and only a few pages maybe four pages are about doing something we block most kinds of human-centered ai collection and

we have to overcome our worst weakness our worst weaknesses are that in the in this innovation process you can see it here i won't have the time to explain it in detail so applied research then the startups commercialization then the scaling phase right self-explanatory our problem is here here and here it has been and it still is so it has not changed yet it is the last mile of research the one that darpa is doing and others long breath shared criteria competitive last mile until things are usable and then as fabian already said the venture capital in the very beginning risk-taking also risk-taking in big companies so if we see the darpa success stories darpa you know right the is the arpanet the internet the mouse the global positioning system conversational ai siri google translate all of these things many of these things were started in europe the base foundations were laid in europe but here the last mile was not done people went to united states many of these things were developed by europeans

in america with american money and then commercialized in america so then our problem is angst the german uns the fear of digital data fear of algorithms and it's i coming back from china you cannot realize that you are here all the time it's much deeper and much more complex than you were thinking of the fear of ai ai is too good it will steal our jobs no no ai is not it's too bad it's not good enough to make critical decisions it has biases it makes the wrong things and ai is evil it will soon turn against us so but data is the raw material for knowledge creation the fuel of science yeah the key to understanding the world for data are not harmful for health or environment and and data are good nothing bad about them but germany is the country of overflowing letterboxes in china in the three years i got two letters in my mailbox here it overflows after one week lots of paper that i have to deal with it is

we have cash only restaurants we have public administrations without email addresses schools without i.t infrastructure no data interoperability mass pixelated buildings from on on street view because every citizen can erase the facade every tenant of the house facade and we have the most feared data protection officers take as an example our corona app this is my a picture of mine voluntary this is a good thing every politician says it's great that this is voluntary 25 penetration 57 of infected users share the positive results a little more than 10 of positive tests are shared via the app and you get every day these things 10 encounters with low risk nobody knows what to do with it what do you do with that useless absolutely useless it's not the fault of telecom or sap it's the fault of our society of our politicians to come up with something totally useless university of oxford researchers estimated it takes at least 60 penetration and then it will really help quite a bit if you get close to 100 percent

all this fear that the gesundheits enter may break down and cannot track our trace anymore would go away it would simply go away so it's political decisions it should be voluntary decentralized because decentralized then only google and uh and apple get it yeah but not the german gesundheit's empty wonderful yes should be decentralized should be voluntary and no other data should be collected or they although they are badly needed for understanding the pandemics and no other uses are allowed although they are badly needed so this is terrible we have this privacy data of austerity so we just heard yesterday from our chancellor that the right to life integrity of the person has to be ranked higher than the right of freedom of movement or the right of assembly clear yeah we have to stay at home because we want to save lives there's only one thing that gets ranked even higher this is data privacy and data austerity that is rang tire because no the the app should not be made like is in china

mandatory and in china it really helps people quite a lot and people like to use it they feel safer with it so the solution is simple have a tech have a highly secured centralized data storage call it gaiax or whatever you like and it can be a public trust concern independent of government just like a supreme court or a central bank have strict legislation governing data utilization with harsh penalty for abuse and focus finally focus on protecting people from abuse and of their data instead of protecting data data protection in the old sense is outdated so there are lots of excessive systems that we trust existing systems we trust the air traffic control system it really helps we trust central banks and other large banks we give our money there we are not scared we trust the armies some european countries with some without nuclear weapons we trust them that they don't do this yeah we trust intelligence agencies but we wouldn't trust a simple controlled database it's ridiculous so now the question how dangerous

is ai two opinions here one comes from christopher martzbook a german ai researcher from deep learning the threat posed by artificial intelligence is worse than the atomic bomb this is wonderful thank you colleague from the mars book our politicians and the public will love it they will be even more scared of ai andrew eng on the other hand well-known silicon valley and china ai person he says worrying about ai turning evil is a little bit like worrying about overpopulation of mars why does he say that because we are still far far away from understanding self-consciousness and modeling general ai with intentions desires and free will in order to be evil you need to free will yeah you need in order to be develop your own evil our ai is much too dumb to become evil today but we should remain cautious in deploying ai for critical tasks we should be very cautious and we should continue to work on transparent reliable and accountable ai but it's these people doing it they are already sabotaged by

this fear that is there and people are not taking serious yeah so this is a decision the situation is severe the attitude of our opinion leaders towards ai puts our countries in jeopardy there's actualism committees white papers regulations another commission another utterance another promised funding but politics cannot overcome anti-technophobia there is no impactful funding you could see now in the corona times here or when we had the refugee situation it is easy to spend huge amounts of money if they are needed our society is able to be decisive yeah how much money is spent but did you see the money even promised on ai was never really spent and allocated fully here so in that cases why are they so hesitant they are not bad people they are all governed by this german fear so this the clothing was platin the not must burst it's not a guardian knot that can be eliminated by a sword it's a paralyzing collective deadlock caused by spreading psychosocial disorder we have to be patient with the people who

have it because there are some friends there are some people who i went to school with they are scared yeah they are scared and technophobe we should be patient with them but we shouldn't be patient with the situation and maybe humanities and social sciences that should be oriented within shaft and that should give us orientation maybe they need orientation themselves because these people are totally disoriented in today's world yeah they are they are not there yet and they give orientation to politics it's the wrong way so the knowledge platform conclusion and outlook europe is still behind in many areas of ai but we have huge opportunities especially in industrial ai and enterprise ai and we have excellent chances in the next ai revolution going from narrow to broader ai by bringing together huge knowledge explicit knowledge and machine learning especially for human-centered applications for which this is needed and but europe and especially germany has a big handicap a fear-driven hesitation and the quest for being better having better ai in the ethical sense before

we ever have good ai we want to be better before becoming good and if we can overcome the german angst and european arrogance telling the whole rest of the world what is good and ethical we have the power to make our continent even better our wonderful continent thank you for your attention much thank you Hans Uszkoreit we have very few minutes left but there is a question by uh daniel again thank you daniel for your question can we think of an intelligent enterprise from the perspective of a company becoming a subject itself instead of being comprised of many subjects companies today already seem to be an organism yes compared to like an organism of people and logarithms or maybe algorithms i think he means they just need to understand themselves do you think that's true yes like with any complex social system that makes decisions and has information processing we need to understand it it is more tighter than let's say the market of an economy that's also a complex system and we need to

understand it through ai but in this case there is still much more centralized decision making but already today 90 of the decisions or even more are not made by the ceo anymore but they are collective and it's true the integration of these intelligent processes will make it an an own subject right very very interesting carson has a question while innovations are agile lightweights broad data rich topics are given high weight by ai systems ai is therefore a quite conservative consultant so how do we innovate ai sorry again so see all right all right i'll i'll i thought you would understand it right away i'm not sure if i understand while innovations are lightweights broad and data rich topics are given high weight by ai systems so i think like broad topics a lot of data yeah is like preferred by ai or maybe maybe ai lends itself to analyze topics where there is a lot of data and custom means that therefore ai is a quite conservative consultant uh only going for where there is

a lot of knowledge maybe not being inventive enough that is very much related to the prediction fusion of knowledge so how can we innovate this yeah because because there is now you have realized right that big self-trained models develop lots of implicit knowledge from general general knowledge and then they are trained by very few examples yeah one shot few shot learning but sometimes this is not enough sometimes you have to infuse the knowledge as explicit knowledge into the system you can do that already today actually you can teach the system and give it uh definitions and so and this is exactly the thing you know this is exactly the thing because we we need many many we have many applications for which we don't have these huge amounts of data but they are also crucial and very important and we have to infuse the combine general implicit knowledge with very explicit with very condensed few data but also with explicit knowledge

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.