Home of Innovation 2021 - AI Special

Page 1

Home of

Innovation Stories about innovation and societal impact June 2021

The future of work

Workers and robots improving each other

TU Delft AI Initiative

AI will have an impact on the whole university

EU sets the standard for ‘humancentred’ AI

Home of Innovation | Stories about innovation and societal impact



Home of

Innovation Stories about innovation and societal impact


Colophon Contents Production TU Delft | Innovation & Impact Centre, Malou Spruit (Concept & Art Direction) Jurjen Slump ( Editor in Chief) Monique Stuffers (Traffic) Contributing writers Agaath Diemel, Bennie Mols, Jochen Meischke, Karin Postelmans Translations UVA Talen Infographics and illustrations Iris Jönsthövel Photography Guus Schonewille

4

Lay-out Liesbeth van Dam Print Edauw en Johannissen © TU Delft | Innovation & Impact Centre June 2021

INTRODUCTION Tim van der Hagen

5

IMPACT ‘AI will have an impact on the whole university’

4

Geert-Jan Houben

FEATURED With AI and into AI: connecting AI experts with scientists who apply AI

10

Infographic

POINT OF VIEW EU sets the standard for ‘human-centred’ AI Dragoş Tudorache

12

EDUCATION 24 Delft AI labs give boost to education

18

VISION All TU Delft students to be educated in AI

20


Introduction

23

NETWORK Netherlands AI Coalition accelerates innovation Kees van der Klauw

25 26 28 30

Startup Smart software assistant knows what your colleages would do

Corporate partner AI is key to ING’s digital strategy

IMPACT ICAI: innovating with leading companies in the field of AI VISION The future of work Workers and robots improving each other

32

OPINION The missing link in the AI ecosystem

34

TREND Regional collaboration in AI, Data and Digitalisation

Paul Althuis

This edition of Home of Innovation is focused on a topic that is set to play a huge role in society in the coming years: artificial intelligence or AI. As a university of technology, we are at the forefront of this development. Not only are we conducting research on AI, much of our research already makes use of AI, and just about all of our specialist fields will be impacted by it. TU Delft is even one of the major players in the Netherlands when it comes to AI. And in the coming years we will be giving this a huge boost in our teaching and research. Our AI experts are working together in 24 special laboratories with scientists who use AI in their research. This combination of research into AI and with AI, means that we can directly implement AI in countless applications. We are also setting up a large-scale curriculum for students in every specialisation. At the end of 2020 we appointed professor Geert-Jan Houben as Pro Vice Rector Magnificus for Artificial Intelligence, Data and Digitalisation to put all of this into action. You can read more about him in this magazine. In addition, we are spotlighting five very promising areas of application for AI: peace, justice & security, port & maritime, energy & sustainability, health & care and technical industry. We are also examining developments in fundamental research and looking at examples of innovation among our own start-ups as well as in the collaboration with corporate partner ING. Of course our AI efforts do not stop at the boundary of the campus. We have joined forces with Erasmus University Rotterdam, Leiden University, Erasmus MC and the LUMC and we are also jointly participating in the recently established Holland AI, in which companies and knowledge institutions in Zuid-Holland are collaborating on AI. The international perspective is also covered. For more on this, read the interview with Dragoș Tudorache, Chair of the European Parliament’s Special Committee on Artificial Intelligence in a Digital Age. Professor Tim van der Hagen

Rector Magnificus/President of the Executive Board TU Delft

5


6

‘AI will have an impact on the whole university’ AI, data and digitalisation are increasingly essential to solving important scientific and social issues. This is why, with its TU Delft AI Initiative, the university is investing heavily in education, research and innovation in the field of AI. Prof. Geert-Jan Houben is coordinating the initiative. Text: Jurjen Slump


IMPACT

‘Our main interest is the engineering context and how we can innovate with AI in the various fields in which TU Delft already excels’

A

t the end of last year, the Executive Board appointed Houben to the position of Pro Vice Rector Magnificus AI, Data and Digitalisation, tasked with ensuring that TU Delft excels in these fields. He is also spearheading regional, national and international cooperation on this theme. TU Delft is making serious investments: the budget for AI, data & digitalisation is set to double to €70 million and 24 interdisciplinary AI labs are being established to strengthen AI education across the university by teaming up AI experts and scientists who apply AI in their research and education. There is also cooperation with the business community and TU Delft is working closely with its regional partners and knowledge institutions.

TU DELFT HAS ALWAYS HAD A STRONG REPUTATION IN COMPUTER SCIENCE. WHY THIS BOOST? “At TU Delft, we recognise that a great deal is happening in this field, especially when it comes to combining fundamental developments in AI and its application in specific contexts. We intend to actively invest in this, in order to enable cross-fertilisation. AI will have an impact on the whole university: almost all degree programmes, fields of research and areas of application will be affected.”

WHAT DISTINGUISHES TU DELFT WHEN IT COMES TO AI, DATA AND DIGITALISATION? “TU Delft is a university of technology with a strong focus on applications. That’s a good starting position for AI, because AI is always applied in a context. Our main interest is the engineering context and how we can innovate with AI in the various fields in which

TU DELFT AI INITIATIVE •

The AI, Data and Digitalisation budget for education and research is being doubled to reach €70 million a year

The establishment of 24 AI labs is leading to the addition of 120 academics, including 24 tenure track candidates and 96 additional PhD students

At TU Delft, approximately 700 academics are engaged in fundamental developments ‘into AI’, In addition, there is an approximately equally large and fast-growing group of academics who are engaged ‘with AI’.

TU Delft is the only Dutch university listed in the Nature 2020 Artificial Intelligence Top 100 Index

Home of Innovation | Stories about innovation and societal impact

7


ELLIS DELFT: GUARANTEEING ACADEMIC EXCELLENCE TU Delft has been selected as an ELLIS unit for research into artificial intelligence and machine learning with societal impact by the European Laboratory for Learning and Intelligent systems (ELLIS).

8

By providing funding to the tune of more than €200 million for more than 30 European units, ELLIS is reinforcing the position of academics on the frontline of AI research. This is a welcome move, because the boundaries between academic and industrial AI research are becoming blurred. There is a risk of fundamental research and valuesensitive design falling behind commercial AI, undermining its quality and social aspect. Talent can also be lost to the AI industry. The TU Delft ELLIS unit brings together leading researchers from different disciplines and connects them with European counterparts, thereby ensuring a significant acceleration of knowledge. More info: https://ellis.tudelft.nl

GEERT-JAN HOUBEN Houben is Pro Vice Rector Magnificus AI, Data & Digitalisation. He has been Professor of Web Information Systems at TU Delft since 2008. He was involved in the research programme on Data Science for Open & Online Education and the research on Urban Data Science as principal investigator within the AMS Institute. He is chair of the national Data Science Platform Netherlands and a member of the AI Knowledge Forum (Kennistafel AI).


TU Delft already excels. That distinguishes us from general universities. Of course, the technology itself also needs to be further developed and that’s what we’re good at here at TU Delft.”

SO, THE MAIN FOCUS IS ON CONNECTING AI EXPERTS WITH SCIENTISTS WHO APPLY AI IN THEIR FIELDS? “Yes, we call this the combination of ‘into AI’ and ‘with AI’. That’s where our strength lies. In this, we are focusing on five strong areas of application: peace, justice and security; ports and maritime; health and care; energy and sustainability, and the technology industry. The 24 labs being established aim to develop new academic education and research in areas where innovations ‘in AI’ come together with specific areas of application ‘with AI’.”

SUCH AS? “Take the process of designing a building or an aircraft wing – you can use AI to accelerate that. You no longer need to design and test a prototype if you can also simulate those tests using AI. That way, you can really speed up innovation.”

IS THE CROSS-FERTILISATION LIMITED TO TU DELFT? “No, we’re explicitly aiming to work with other knowledge institutions in the region. As part of what we call convergence, we’re working with Erasmus University Rotterdam, Leiden University, Erasmus MC and LUMC on our five themes. Through far-reaching cooperation and by combining ‘in AI’ and ‘with AI’, we aim to achieve pioneering solutions in the fields of for example energy transition, security and health.”

WHAT SIGNIFICANCE DOES THE AI INITIATIVE HAVE FOR EDUCATION? “Quite a lot. It’s giving education a huge boost and it’s wonderful to see so much enthusiasm, among scientists and students alike, to be involved in the new areas of knowledge emerging from the 24 AI labs. The same applies to the new talent that we’re attracting for the labs. These are people, many of them from abroad, who are demonstrating a lot of curiosity and enthusiasm in their work on the application of AI and data in their fields. That’s invaluable.”

IS SOCIETY READY FOR THE LARGE-SCALE APPLICATION OF AI? “The most important thing is that society deals responsibly with the opportunities offered by AI, data and digitalisation. Context is important here too. Society needs to take time to discover all the applications that are possible. Government can play an important role in this by enabling experimentation which, if properly supervised, will allow us to determine what works and what’s less effective.”

TU DELFT HAS A STRONG REPUTATION FOR RESPONSIBLE INNOVATION. HOW’S THAT REFLECTED IN AI? “It’s not only the development, but also how it’s used that needs to be responsible. We must ensure that end-users can use AI-driven systems responsibly and that key principles of responsible AI are embedded in the design. We’re devoting a lot of attention to this: it comes up in many of our courses and this approach is a key priority for our researchers.”

HOW IS THE BUSINESS COMMUNITY INVOLVED? “In addition to the 24 AI labs, we have also set up several ICAI labs, in which we’re working with major companies – DSM, Ahold Delhaize, ING – on the development and implementation of AI in specific sectors. The advantage of these labs is that they enable us to work on very concrete applications and to access data sets that are not otherwise available.”

‘The most important thing is that society deals responsibly with the opportunities offered by AI, data and digitalisation’

9


FEATURED

TU Delft: with AI and into AI

At TU Delft, we believe that the continued and successful application of an AI-driven approach demands a convergence of research into AI, advancing the field of AI, and research with AI, advancing the state of the art of a specific field of research. The following pages provide an overview of the focus themes of our research in AI, data & digitalisation. Text: Karin Postelmans

Smart systems provide support and also protect against prejudices hA

en

k dij

I

Mi c

Innovation in artificial intelligence is being funded to the tune of €300 million from the National Growth Fund. “That’s good,” says Prof. Inald Lagendijk, professor of computing-based society. ‘We need to focus on fundamental issues around human-oriented AI on a larger scale.”

Prejudices

wit

Inald Lag

Human needs and rights are strongly expressed in justice and security. But how do you implement that in AI?

he

l va

n Eeten

Smart systems help humans to make decisions and improve processes that involve a lot of data. The use of AI for government decisions or cybersecurity must not only meet functional but also social and ethical requirements, such as integrity and human control. Lagendijk: “Preferences and prejudices exist. We are researching how to design systems that protect us against that while also supporting justice and security.”

AI in the Peace, Justice and Security domain

Larger system

Provide direction

Prof. Michel van Eeten is professor in governance of cybersecurity. He researches what humans need to improve AI innovations around cybersecurity. “Plenty of smart systems have been developed that did not do what users thought. The result: a false sense of security.” The cause is a gap in knowledge between specialists in AI and organisational and behavioural experts. The socio-technological environment at TU Delft brings various disciplines together. “That enables us to discover what larger systems need, rather than users literally ask for.”

Research projects are currently small in scale. The Growth Agenda opens up opportunities for bigger partnerships with players involved in the introduction and use of AI. Van Eeten: “That matters, because the effect on human lives can also be big.” Lagendijk is eager to see TU Delft continue to do fundamental as well as applied research, and provide direction to government and businesses. “We intend to demonstrate that it’s possible to implement human needs and rights in AI systems.” <


Looking ahead for smart logistics TU Delft AI research offers solutions for future transport needs. Collaboration will be key to this. Thanks to TU Delft’s AI research, maritime infrastructure and equipment are becoming cheaper, more sustainable and more efficient. To achieve this, TU Delft collaborates with different universities and companies. “AI is not just technology,” says Prof. Rudy Negenborn, professor of multimachine operations & logistics. “Economics, law and ethics also matter.”

Different decision-makers

eg

en b

om

Matthijs

Sp n

Rud y

aa

N

Negenborn is working on smart shipping and smart logistics. Much of the information about port transport is spread widely across IT systems and

AI Theme Port and Maritime

organisations. This means that a decision about which cargo ship needs to travel how fast and where to is made by different decision-makers. His research into algorithms aims to optimise goods transport, especially maritime transport, with autonomous ships. Dr Matthijs Spaan, associate professor of computer science, is working on algorithms that make predictions and use them to anticipate. With current data alone, the system can merely learn through trial and error. That can prove dangerous. Spaan: “Systems that anticipate their own predictions can reduce risks. That makes a difference when port-based factories want to save energy, for example.”

Human control Interaction between humans and systems guarantees human control. Imagine multiple ships sailing too closely together because of a variable wind. The result is a clash of interests: safety, efficiency, cost-effectiveness. Spaan: “This is why we are teaching computer systems when they can decide for themselves and when humans decide.”

Sharing data

wit hA

Ale

I

One key challenge is the lack of real-life data for algorithms to learn from. Companies are reluctant to share data. But this is not deterring the researchers. Negenborn: “If necessary, we create databases from real data, but omit any sensitive company information.” Thanks to TU Delft’s network, students also do graduation projects at companies. Spaan: “That provides us with insight into real-life data and helps companies to appreciate AI more.”

dro Bozzon

11

wit hA

AI brings out the best in humanity and in medicine

I Catho lijn

J

on

ker

Human dignity, the inspiration behind our healthcare system, is what AI technology is missing. But thanks to the research of Prof. Alessandro Bozzon and Prof. Catholijn Jonker, that is set to change.

n ssa

AI Theme Health and Care

Healthcare faces urgent problems. They include staff shortages, health disparities and pandemics. AI opens up new potential, but fails to take account of something that is central within care: human dignity.

People first TU Delft’s system approach to AI also offers promise for application in healthcare. AI must strengthen the care ecosystem in which it operates, including the values that apply. People always come first, both as the beneficiary and controller of AI systems. That’s why you need to teach people how AI works; and teach AI how people think and what values they adhere to. That’s exactly what Jonker and Bozzon are doing.

Predictions Alessandro Bozzon, professor of human-centered AI, is attempting to teach AI which values matter. Take a smart system that predicts patients’ likelihood of post-operative complications. The system has been trained (see box) using medical data, such as risk factors. Bozzon: “But are the data really representative of the population? We’re looking for criteria and methods to find this out in interaction with AI.”

AI can also help care providers to improve their medicine. Catholijn Jonker, professor of interactive intelligence, is exploring how to do that. She’s developing mediating AI technology: AI that supports decisionmaking by means of mediation. Think smart mediators that support doctors consultations between colleagues. Jonker always takes people as a starting point: what aspects improve the outcome of that kind of consultation? What unspoken rules and values apply? “But also: how do you implement these kinds of things in a way that ensures doctors find the system to be trustworthy?”

Care as a touchstone Jonker believes that the care ecosystem offers TU Delft an effective touchstone for AI technology. “Application in the real world gives us direction: where are the real issues and pitfalls? And also helps us to ensure AI is in line with reality.”

Home of Innovation | Stories about innovation and societal impact


A more secure, smarter and faster route to the energy transition

He’s exploring what happens if you add large wind farms to it. Palensky: “TU Delft is a pioneer in this. That’s partly thanks to its collaboration with grid operators. Their real-time data improves our AI; in turn they benefit from our knowledge.”

Our electricity grid is not yet prepared for the energy transition that must have been completed by 2050. Prof. Peter Palensky and Dr Mathijs de Weerdt are helping to make a faster transition possible.

In the past, a map of the infrastructure was sufficient for grid management. But demand for electricity is increasing and supply depends on the weather. In addition, consumers are changing into producers (solar panels). Palensky: “With AI, we can more effectively, quickly and securely find out the effect of the energy transition in this context. And also discover what the solutions are.”

TU Delft is researching how we need to adapt the electricity grid for the energy transition. Peter Palensky, professor of intelligent electric power grids, is developing a digital twin (digital version) of our electricity grid.

AI Theme Energy and Sustainability sk y

h wit

Peter Palen

Flexible use

AI

Dr Mathijs de Weerdt, associate professor and member of the Dutch AI Coalition, is developing the algorithms needed for that. “Traditionally, we’ve relied on flexible production. But flexible use is more efficient.” An example of this is to schedule the charging of electric vehicles so that it’s spread smartly. But this shift in thinking raises new questions. How do you organise that kind of schedule? How do you prevent disruption for consumers? De Weerdt: “I start by using existing flexibility that has minimal impact. If the system works, we go further.”

Worrying

d js

Mat thi

De Weerdt: “Our standard of living makes us addicted to electricity. Smart systems can help the energy transition and keep society happy.” Palensky: “This calls for investments and regulation from government. But innovation is outpacing decision-making. That’s worrying.”

eerdt eW Arn old

He nk em i

wit

TU Delft knowledge enhances the potential of industrial AI

hA

I Robert Bab

ka

Combining the latest AI insights with the laws of physics or robotics, for example, creates new opportunities. That’s what Prof. Arnold Heemink and Prof. Robert Babuska are working on. AI is already helping industries to improve production processes. But not in areas where the laws of physics prevail, such as in electricity or water management. This is because AI cannot integrate the laws of physics. Arnold Heemink, professor of mathematical analysis and large-scale models, aims to change that. “I’m trying to bridge the gap between AI and calculation models, that can actually take physics into account.”

Essential Heemink is developing physics-aware AI. He is doing this for digital twins (digital copies of complex systems). Like a twin of the Dutch Rhine-Maas Delta, for example. With data about groundwater levels, water movement and salinisation, you can use it to do simulations and predictions. “Physics-aware AI is essential for that.” Theoretically, Heemink is already capable of making AI physics aware. But the algorithms require too much computing capacity. Heemink is now researching how to simplify the algorithms without undermining the results.

us

AI Technology Industry

Robots Robert Babuska is applying AI technology for industrial robots. He combines AI algorithms with machines. The result is robots that do not need to be programmed. “They use AI to work out the best way to complete a task. And they can also learn from their own mistakes and from humans.” Babuska collaborates a lot with others, for example as part of the NWO FlexCraft programme. In it, companies and universities are collaborating on AI research on greenhouse cultivation, food processing and food packaging. Babuska: “This is both applied and fundamental research. That strengthens innovation.”

Time to learn Babuska envisages challenges because of high expectations, based on trendy robot videos. “They do not work 24/7 without making mistakes. In a complex environment, robots need time to learn.” He hopes to reduce that time by having robots learn from each other “for example in robot crèches.”


The key to AI innovation: human-AI interaction

to combine both of these. “Think of care providers using a voice system to learn at work.”

Dr Claudia Hauff and Dr Frans Oliehoek engage in different research, but they have a single goal: to support interaction between humans and smart systems. How can a smart system help in searching for and learning from information, in order to support humans? That’s what Dr Claudia Hauff, leader of the Lambda-Lab in Delft, is researching. For example, she’s working on systems that improve search results by learning from human-AI interaction. “Think Google Voice or Siri, but better: smart systems that learn from the conversation what the user really needs.” Hauff is also working to optimise learning processes: what AI interventions can promote learning? Hauff’s ambition for the future is Claudia H

o int

0100100 0010011 1010101 0010011

au ff

Machine Learning AI

AI Fr

One key problem is the limited amount of data available for training smart systems. Fortunately, TU Delft can provide some of this itself, through online courses in which two million people participate. Hauff: “For example, we’re testing how behaviour changes if we present questions or other participants’ results. Data like that serves as training input.” Dr Frans Oliehoek, who is in charge of the INFLUENCE research project and involved in ELLIS Delft (see box), is also focusing on learning from interactions, but, in this case, between humans and AI and between AI systems themselves. He is also interested in decision-making in uncertain situations. “Hauff and I have a shared goal: to support human-system interaction.” Self-driving cars are one area where interaction and complex decisionmaking come together. They are already able to make decisions based on data about a road junction. Oliehoek is working to scale up the process. “AI systems that can manage thousands of variables. Traffic data about a whole city, for example.” This involves specific challenges. Oliehoek: “We need to teach systems to consider, for example, longterm and short-term results.”

an

sO liehoek

Oliehoek is using reinforcement learning to teach a specific type of smart system to make series of decisions in order to think in more abstract terms. Oliehoek: “We are testing the fundamental principles in traffic simulations. These can ultimately be applied in real-life situations. That’s where TU Delft’s strength really lies.” Jero

e n va n d e

n

ven Ho

o int

AI

More humanity thanks to AI TU Delft is putting people and morality first in its AI innovation in order to protect equality and European values. Fundamental research is contributing to this.

System-wide approach The cornerstone is the system-wide approach. “That involves taking on board the values that will apply to AI operations as early as possible in the design phase”, explains Van den Hoven, professor of ethics and technology. “And the preconditions, such as legislation and human-AI interaction.”

Deliberation Catholijn Jonker (professor of interactive intelligence) is using NWO funding to research human-AI interaction. Her field of work ranges from robotics to decision-making. “I look for methods and technologies that facilitate the sharing of information and learning from dialogues.” Jonker aims to develop AI that encourages people, organisations and AI to ‘deliberate’. “Knowledge of everyone’s

Jo er nk

The difference between AI and humans is that we have awareness, morality and an ability to reflect. AI will also need morality if it is to be usable, because people must be able to protest against the results of AI. They also need to be able to debate online across social bubbles created by commercial AI. This is why Prof. Jeroen van den Hoven and Prof. Catholijn Jonker previously set up the Delft Design for Values Institute, now innovation leader in the field.

Catholijn

Human-centered Ai

motives and standpoints creates space for others, and therefore also for new ideas and solutions.” Her ultimate aim is to use human-AI interaction as a way of providing insight and an overview in online debates. “If you can use that to run a simulation, you can find out how to ensure everyone who wants to be can be heard.”

High bar The results of TU Delft fundamental research are far-reaching. Van den Hoven is co-author of a publication about safeguarding human dignity in AI systems. This is helping to roll out European legislation. “Europe is setting the bar high. This is annoying countries with little understanding of ethics. But they need to cooperate, or risk losing the European market.” However, the Netherlands will need to invest seriously in this. “It will also deliver innovations that the world wants to see, because all people share the same needs: safety, freedom, sustainability.”


POINT OF VIEW

EU sets the standard for ‘human-centred’ AI


‘Inevitably, it will become part of everything we do and have a deep impact on society’

In April of this year, the European Commission became the first in the world to introduce ambitious regulation in the field of artificial intelligence (AI). This strategy, which prioritises citizens’ fundamental rights, is set to have a major influence on how other countries deal with AI regulation, according to Dragoş Tudorache, Member of the European Parliament. However, he also calls for unambiguous regulation, in order to prevent any barriers to innovation.

Dragoş Tudorache

Text: Jurjen Slump

T

udorache, a Romanian, chairs the Special Committee on Artificial Intelligence in a Digital Age. The committee was established to give the European Parliament greater control over AI. Previously, the subject was fragmented and discussed in various different forums. “There was a need to bring all of those different lines together, in order to gain a better overview”, says Tudorache. The European Parliament acknowledges the significant role that AI is set to play. Tudorache: “Inevitably, it will become part of everything we do and have a deep impact on society.” AI has the potential to make all human activities “more efficient and sustainable”, from healthcare to public services, the electricity grid, public transport, the business community and industry.

Guiding AI in the right direction But that potential needs to be exploited and accelerated, according to Tudorache, who is a member of the liberal Renew Europe Group. The political vision is there – the EU has invested heavily in digitalisation and innovation in recent years. And the legislation that has been announced also provides a legal framework (see “Pyramid of risk for the use of AI”). “We now need to actively pursue technological advances, help businesses to grow and ensure that the member states follow suit.” “We simply can’t afford to do nothing”, he warns. “As the European Union, it is our duty to assert ourselves on the world stage in order to ensure that we guide the social changes brought about by AI in the right direction. We have lost ground compared to the United States and China

Home of Innovation | Stories about innovation and societal impact

13


and must therefore ensure quickly that we have clear policy, clear legislation and the right incentives to enable AI to thrive in Europe.”

“it will certainly influence how other countries set parameters”. He also praises the “human-centred approach”: citizens and their fundamental rights are central to this.

AI diplomacy As chair of the special AI committee, Tudorache has played a role in what he calls “AI diplomacy”, engaging the US and other like-minded partners in recent months. Since President Biden took power in the US, there has been some rapprochement towards the EU. Although big tech companies in Silicon Valley are hesitant about overly strict regulation, it is possible to speak of “shared, fundamental values and democratic understanding”, he says.

No brake on innovation

This means that the EU and the US are virtually reliant on each other to halt the rise of China and ensure that a “shared EU-US model” becomes the leading standard. The European Commission has already made the first move: a social credit system like the one built in China is being banned.

It is also important that the different member states do not start implementing EU legislation in different ways. “That causes fragmentation, which would weaken a strong internal market”, he says. Scale is particularly important in the field of data and AI. Only if businesses can access the whole internal market will they be able to compete with big tech companies from the US and China, according to Tudorache.

US and EU’s shared responsibility

14 14

“We have very little influence over what China does within its own national borders”, says Tudorache. “But the EU and the US need to ensure that China does not overstep the mark. Chinese views on the use of AI must not become the international standard.” That is why it is important that the EU has been the first to introduce legislation. Tudorache does not go so far as to say that it will become the worldwide standard, as has happened with GDPR privacy legislation, but

However, it is important that the new rules do not put a brake on innovation. “Legally, there are still many unanswered questions and it’s important for there to be clear technical standards to provide a framework for start-ups and businesses.” When the European Parliament discusses the legislation, he intends to advocate striking the right balance between regulation and room for entrepreneurship.

Key role for universities Universities play a key role in innovation. “They’re often instrumental in setting up alliances with major industrial players and other stakeholders”, Tudorache says. “We need to support that, if we are to remain competitive and attractive for research as a continent. AI needs the brightest minds. We don’t want to see them go elsewhere.”

Pyramid of risk for the use of AI The AI legislation is based on a so-called “pyramid of risk”, which means that the higher Unacceptable

the risk of a certain application is, the stricter the rules are.

Unacceptable – AI systems that constitute a danger to EU citizens will be rejected, such as the social credit system being used in China. High risk

High risk – this concerns systems in which AI is used to issue mortgages, for example, or the use of AI in job application procedures. In such cases, AI applications will be carefully assessed before being allowed to enter the market.

Limited risk

Limited risk – this applies to AI systems such as chatbots. These must meet minimum transparency requirements, ensuring that users know that they are dealing with AI and can make an informed decision not to use it. Minimal risk – the most-frequently used AI applications for which no additional rules

Minimal risk

apply. Think video games based on AI or spam filters.


POINT OF VIEW

Kim van Sparrentak

‘Environmental labels for AI and digitalisation’ AI is often regarded as the silver bullet for solving the climate problem, but this overlooks the fact that all these algorithms also consume large amounts of power, which is often not sustainably generated. Kim van Sparrentak, who represents GroenLinks (Green-Left) in the European Parliament, is calling for environmental labels for AI and digitalisation. “The ICT sector has a huge impact on the environment and we need to move towards a circular economy in this sector as well”, she says. In her view, it is important not only to consider the energy consumption of the technologies themselves, but the entire infrastructure that is used, including production materials, networks and “energy-guzzling” data centres.

world leader in this area, “based on our European values”, van Sparrentak expects. In her view, “ethical AI, gender equality, anti-racism and the prevention of bias” have a central role to play. “Algorithms need to be safe and transparent and work for us and not against us.” For high-risk applications, the European Parliament is calling for a compulsory ethical check, which van Sparrentak is very much in favour of.

Guaranteeing fundamental rights Like Tudorache, van Sparrentak is a member of the Special Committee on Artificial Intelligence in a Digital Age. She is impressed by the European Commission’s proposals. “There are some good things in there, such as first steps towards a ban on biometric surveillance and social credit systems.” But she can still see a few legal loopholes. “If we want to guarantee fundamental rights in the realm of privacy, the last thing you want is have broad exceptions, as we see now for the ban on biometric surveillance.” The EU has everything needed to ultimately become

AI for green innovations and healthcare When it comes to AI applications, van Sparrentak advocates targeted innovation. “Invest in applications to achieve environmental targets, use data to reduce energy consumption, or consider the development of systems that use AI for improving treatment plans in healthcare.” In the latter case, AI has the potential to make it easier to identify differences between patients, for example. “Things still tend to be based on a standard 40-year-old white male – AI allows us to differentiate and provide better treatment.”

Home of Innovation | Stories about innovation and societal impact

15


A

Marcel Reinders

16 16

An increasing number of people are developing Alzheimer’s disease. It is expected that within twenty years’ time, 500,000 people in the Netherlands will be suffering from this incurable brain disease. Every reason to invest now in research to unravel Alzheimer’s disease, including with the help of AI. lzheimer’s disease is caused by the accumulation of various proteins in the brain, which

without getting the disease: centenarians without

causes brain cells to die.

Alzheimer’s.

Forgetfulness and failing short-

The emphasis in Amsterdam is on the biology,

term memory are often the initial symptoms. The

patient-related issues and data collection.

most important risk factor for getting Alzheimer’s

TU Delft is involved in the data analysis.

is age. “It is age­related disease number

Algorithms are being developed to discover

one,” explains Marcel Reinders, professor

patterns in the data. “Most of the research

Bioinformatics. “Nearly a third of people above

questions and the variables have never before

the age of 85 suffer from a form of dementia, with

been examined. That means that we constantly

Alzheimer’s disease being the most prevalent.”

have to think about how we can arrive at an

Centenarians

answer and how the algorithm should look. As a research team, it also helps if you understand

Even though there are treatments to slow

biology and of the material you are working with.

down Alzheimer’s, no drug has been found

The strength of Delft is that we are not only

that can halt or cure the disease. For this we

exceptional mathematicians, we also understand

need to know more about the cause. Reinders

what we are looking at.”

is working on this together with the team of dr. Henne Holstege from the Amsterdam UMC’s

Sound progress progress

Alzheimercentrum Amsterdam. “To discover what

The research has already produced important

exactly is going wrong, we compare the DNA

results. In 2017, TU Delft contributed to the

and protein data of patients with and without

discovery of specific deviations in a gene that

Alzheimer. We are looking in particular at people

could be a strong indication that an individual

who appear to have reached extreme old age

will develop Alzheimer’s. “We are making sound


Tech for Health

Algorithms in the fight against Alzheimer’s disease progress, but I would like to speed up. We can achieve

diseases, such as cancer or hereditary disorders.

much more with additional computers, people and

“Clinical geneticists can use it to better map hereditary

particularly the use of the latest data analysis techniques.

diseases. This dataset has many possible applications.”

But none of this comes for free. While Alzheimer’s is affecting an increasing number of people, there is still

COVID

relatively little funding available for research. If we really

AI can also play an important role in the fight against

want to do something about this disease, that needs to

corona. A student from Reinders, together with Erasmus

change!”

MC, is in the process of making a so-called risk predictor.

The importance of AI and data is increasing

Based on all available data, measurements and lab data, it can be predicted whether or not a COVID-19 patient will go to the ICU. Another model can predict whether a patient will

AI and data are playing an increasingly important role

die within 24 hours or not.

in healthcare. Reinders is now also involved in a large

“These tools are primarily intended as a warning system for

Parkinson’s project, in which the same research method

doctors and nurses,” Reinders emphasises. “This allows

is used to compare the DNA of Parkinson’s patients

them to give more attention in time to patients whose

with the help of AI. The data generated by the

situation may deteriorate.”

Alzheimer’s project is also valuable for fighting other

Will you help? Delft University Fund supports TU Delft by contributing to research, education and talent development. This year, the Tech for Health campaign focuses on providing financial support to excellent scientists at TU Delft who are committed to improving healthcare in the Netherlands. You can also contribute to this important work. With your support, we keep our healthcare of high standard, future-proof and accessible and affordable for everyone. More information can be found at tudelft.nl/techforhealth. Join us and help with a donation. Transfer a contribution to IBAN NL48 ABNA 0441 4822 95 to the attention of Delft University Fund, citing “Tech for Health”. For more information about the Delft University Fund, please contact one of our employees or visit our website: www.universiteitsfondsdelft.nl | ufonds@tudelft.nl | +31 (0) 15 278 64 09

The Delft University Fund has ANBI status (organisation serving the public interest). Your donation may therefore be tax deductible.


EDUCATION

THE FIRST SIXTEEN OUT OF TWENTY-FOUR TU DELFT AI-LABS

3DUU

TU Delft AI Labs

18

An important part of the TU Delft AI Initiative is the establishment of a total of 24 labs in which experts in the field of AI foundations work together with experts who work on societal and scientific challenges with the help of AI. The TU Delft AI labs are an embodiment of the bridges that are built between science in and with AI, Data and Digitalisation. This approach gives a significant boost to education and increases the impact of Delft AI on science, technology and society. The first 16 labs have already been established, the next eight will follow this year.

3D Urban Understanding. New methods to automatically recognise and model objects in the built environment in 3D.

AI*MAN Lab Transparent & Traceable AI in Human-AI Teamwork. Developing optimal and transparent decision making in human-AI teamwork.

DeTail Lab Training & innovation in tensor-based AI methods for biomedical signals. Fundamental development of tensor computing and its applications.

IRIS Lab AI for quantitative bioimaging. AI-based technology that improves microscopy methods for biomedical use.


AiDAPT

AidroLab

AIFluids Lab

AI for Design, Analysis and Optimisation in Architecture & the Built Enviroment. AI for a sustainable and resilient built environment.

AI for sustainable water management. Enhancing the adaptability and resilience of urban water systems.

Artificial Intelligence in Fluid Mechanics. Using AI and fluid mechanics to build better planes and wind farms.

CITYAI

DAI energy Lab

Design @ Scale Lab

A place where data, AI and behavioral theory come together. Translating data into insights into the structure of cities and its impact on human behaviour.

AI for sustainable, reliable and effective energy systems. New AI-based methods that contribute to managing (dynamic) energy systems.

Humans and AI tackle problems together. Integrating Participatory Design, Crowd Computing and AI to better address complex social problems.

DI_Lab

HERALD Lab

Hippo DAI Lab

Humans and AI designing the future in dialogue. Using AI in the development of new design methods.

Human-aware robust AI for automated driving. New AI methods to enable automated driving systems to interact with the humans around them.

AI for fair, efficient and interpretable policy analysis. Optimisation methods for the fair and efficient design and analysis of public policies.

KDAI Lab

MACHINA Lab

SLIMM Lab

From data-driven to knowledge-driven AI. Strengthening data-driven AI by integrating fundamental knowledge from applied natural sciences.

Machine Intelligence Advances for Materials. Creating a new route for designing novel materials and AI algorithms.

AI for smart materials modelling. Developing Bayesian inference tools specifically for application in material models.

Home of Innovation | Stories about innovation and societal impact


VISION

Minor Engineering with AI If you have an interest in creating AI-enabled solutions themselves, this minor is set up for you. It will require your technical understanding of both the underlying data fed into the AI system and the algorithm running the AI system. You will also be acquainted with the limitations and ethical considerations of AI. You will get to know all the ins and outs of AI, will be able to tune settings or implement specific AI algorithms in software, and will learn what is ‘under the hood’ of the AI toolkit.

Online courses

Elements of AI

Big Data to Insights

AI in practice

{} </>

Automated Software Testing

Mind of the Universe Robots in Society - blessing or curse?

Unix Tools

Hello Real World with ROS

Data, software and production engineering

(Robot Operating System)


All TU Delft students to be educated in AI Susanne van Aardenne and Willem-Paul Brinkman

make this strong focus on technical applications part of TU Delft’s AI Initiative is due to have a major, our AI education too.” and unique, impact on education. All students Van Aardenne: “As far as this is concerned, we are busy at TU Delft will have the opportunity to get developing a unique proposition. No other university in educated in AI. The Taskforce Education AI, Data the Netherlands is doing this.” & Digitalisation has been set up to manage implementation. Willem-Paul Brinkman and This sounds like a monumental operation. How Susanne van Aardenne, both members of the will you manage it? working group, tell us more about this transition. Van Aardenne: “We have the advantage of being a Text: Jurjen Slump

V

an Aardenne’s current position is that of Education Advisor Innovation at the Faculty of Electrical Engineering, Mathematics and Computer Science (EEMCS). Brinkman is associate professor in Interactive Intelligence and Director of Studies at the faculty. Their experience will be most useful in the working towards the TU Delft AI Initiative’s objective of offering every student at TU Delft education in AI.

Why are we putting such a focus on AI in our education? Brinkman: “Everyone at TU Delft is confronted by AI in one way or another, so it is important that every student is taught about AI, not incidentally, in a single course but, where relevant, as an integral part of the curriculum. As a university of technology we distinguish ourselves by our ‘engineering approach’, and we wish to

university of technology - mathematics forms parts of almost every student’s programme. We already offer a lot of AI education, yet accessible basic knowledge in AI is new to students in certain degree programmes. The minor Engineering with AI is the initial solution we will be offering them.” Brinkman: “We have a short as well as long-term view of the matter. In the short term we will be initiating the minor, the pilot of which will be starting in the new academic year. In the longer term, we want to help degree programmes integrate AI education in their curricula. This involves considerable organisation. We are talking to the faculties, degree programmes and individual lecturers about how this is to be done.” Van Aardenne: “We are also discussing this within our convergence with Erasmus University Rotterdam and Leiden University. This collaboration, of which the minor pilot is a first step, is structured to allow students from varying backgrounds to learn about AI together, and thus learn from one another while applying AI in practice.”

Home of Innovation | Stories about innovation and societal impact

21


22

How will this focus on technical applications be reflected?

Do lecturers who haven’t had much experience of AI also need to brush up their knowledge?

Brinkman: “The twenty-four Delft AI labs that are being set up will play a major role here. TU Delft has consciously chosen to develop knowledge and education in the field of AI within the various faculties in ways that are closely applicable to specific domains, as they are researched within these labs.” Van Aardenne: “An additional advantage is the wonderful cross-pollination between faculties, and hence the strengthening of the transfer of knowledge among themselves. And that is not limited to Delft.”

Van Aardenne: “That varies from one lecturer to the next, but we want to give all our lecturers the opportunity to get further training. That project is known as ‘teach the teacher’. In this way we will make it possible for them eventually to lecture on AI themselves. We also plan to do this for PhD candidates. Brinkman: “We want as many people as possible to receive the AI education that is relevant to them, from students to lecturers to PhD candidates. We want to see how we can address everyone’s needs.”

How is the ethical aspect of AI safeguarded in the teaching?

Will AI also be given centre stage in TU Delft’s online education, the MOOCS?

Brinkman: “We train responsible engineers. Particularly where AI is concerned, students must understand the possible implications of the algorithms they are using or developing. They need to understand that a certain bias can be introduced when algorithms are trained using incorrect data. We are teaching students how to work responsibly with AI techniques in a social context.” Van Aardenne: “Here, too, the AI labs have an important role to play in embedding these key values, since the labs are linked to specific domains in which theory and practice come together. It is our ambition to link student projects to these labs so that students can experience first-hand the consequences of applying the theory in practice, and can learn from this.”

Brinkman: “Yes, and that is not limited to TU Delft. Together with the Netherlands AI Coalition, we recently launched the Dutch version of the Elements of AI course. This is a free online course that gives participants insight into the basics of AI. This course is also interesting for business or public-sector professionals who encounter AI.” Van Aardenne: “We offer various online courses in the field of AI, Data & Digitalisation. The Elements of AI course is also worthwhile for secondary school pupils who are deciding on their further education. Together with the Innovation Center for Artificial Intelligence, we have also developed the AI in Practice programme for business professionals. AI will affect society as a whole and we as a university want to ensure that our curriculum addresses that fact.”

AIED – TU Delft in AI and Education For example, we can assess the quality of essays with AI tools or build a tutoring system for complex skills. These are hybrid intelligent systems, which form an interplay of teacher, student and AI. Thanks to the automated feedback and the use of AI for coaching, the support of students can be expanded and teachers are given the opportunity to use their time more efficiently. We are currently investigating various initiatives to give these hybrid intelligent systems a place in education. The challenges lie, among other things, in making these models suitable for

M ar cus

Sp

ec ht

Not only education in AI will increase significantly, the use of AI as a tool for education is also on the rise. TU Delft accelerates educational innovation with advances from AI, data and digialisation, such as data-driven analytics and large-scale automated feedback. This is done within the AIED programme (TU Delft in AI and Education), led by Professor for Digital Education Marcus Specht. Specht: “The goal of AIED is to support and facilitate students and teachers with AI. Educational tools based on data-driven methods and machine learning offer many new possibilities.

various application areas. We work closely with teachers and researchers to further develop the use of AI, both at a theoretical level and its application.”


Kees van der Klauw

NETWORK

Netherlands AI Coalition accelerates innovation In the years ahead, artificial intelligence is set to have a major impact on all parts of society. But the development of innovative AI applications in the business world is still proving too slow. If the business community does not want to become dependent on “big tech”, it will need to reinvent itself with AI, according to Kees van der Klauw from the Netherlands AI Coalition (NL AIC). Text: Jurjen Slump

A

s the coalition manager, van der Klauw is in charge of the NL AIC. It was established 18 months ago, based on the realisation that AI “is set to play a decisive role in our future prosperity and well-being”. In response to this, the Netherlands must build itself a “strong and distinctive position” in both the development and application of AI. “It is only through the proactive deployment of AI that the Netherlands will be able to determine how AI is used in dealing with societal and economic challenges and reap the benefits of that”, he says.

Accelerating AI developments and connecting AI initiatives The NL AIC is helping to facilitate this process. The coalition is a public-private partnership in which government, the business community, knowledge institutions and civil-society organisations are working together to accelerate AI developments and connect AI initiatives in the Netherlands with each other. The aim is to achieve a leading position for the Netherlands in socially responsible AI applications. Thanks to support from the National Growth Fund, it is now well placed to achieve the necessary acceleration and increase in scale.

Home of Innovation | Stories about innovation and societal impact

23


There is some urgency. Innovation involving AI applications in the Netherlands is currently too slow. “We believe that many businesses still underestimate it”, van der Klauw says. Many companies do not have their data in order or are only just starting to automate. “SMEs are also preoccupied with surviving the pandemic”, he points out. “Of course, we totally understand that, but the problem is that AI moves so fast that there is a ‘winner-takes-all’ effect. If you get on board too late, you basically miss the boat.”

Changing value chains

24

Even major companies often underestimate the fact that the value chain “is shifting”. “I’m not saying that everyone should implement AI tomorrow, but they need to start thinking about it now.” Van der Klauw cites home delivery platforms as an example. When they started out, they were clearly adding value for the participating restaurants. As a result of their turbulent growth, these platforms have now become more than that and adopted a dominant position: they have all the relevant data about customers, which means that the restaurants risk becoming little more than subordinate suppliers. “That has the potential to happen everywhere”, van der Klauw says. “In transport, in the manufacturing industry – technological advances are very rapidly changing the way businesses earn their money.” Other countries have “had this on their radar for a long time”. The Dutch business community needs to respond in order to avoid drawing the short straw and becoming dependent on foreign big tech companies.

New propositions with AI This response will partly be about improving their own corporate processes by using AI and improving their own products. But van der Klauw would prefer to see them seize the opportunity “to develop a totally new proposition with AI”. He acknowledges that this is easier for large companies than for SMEs. But if smaller companies join forces, they can also use AI to innovate. Take haulage companies, for example. It is in their interest to join forces in applying AI for fleet management and route optimisation in order to prevent being left behind by big companies that are capable of doing this themselves.

Growth capital for start-ups and scale-ups Van der Klauw is a strong advocate of support for startups and scale-ups in the field of AI. There is a lack of growth capital and these companies often have too little access to data. They also need partners. “Imagine you are active as an AI start-up in agriculture and horticulture. You couldn’t survive without partnerships with companies and organisations in that sector.” The Netherlands is well-equipped to gain a place at the forefront of AI, but for that to happen, other players also need to start making progress as well as the business community. Government bodies need to improve their level of knowledge, properly anticipate new regulations and citizens – the end users – need to be involved in the design process for AI systems. The starting position is not the problem, van der Klauw emphasises. The scientific research around AI, data and digitalisation is “outstanding”. The Netherlands has an excellent digital infrastructure and several very strong sectors, including the high-tech industry and agriculture/horticulture.

The polder model as trump card The well-known Dutch “polder model” could ultimately prove decisive in achieving the NL AIC’s ambitions. “We have unparalleled knowledge of how public and private parties collaborate. The polder model may sometimes have its disadvantages, but compared to other countries, we are very effective at collaborating across supply chains.” That’s an essential precondition with AI, because of the influence it will have on all parts of society. “If the Netherlands can demonstrate how to shape this in practice, introducing AI applications that already take prior account of social, legal and ethical constraints and applying all the knowledge available in the different sectors, we will be well positioned as a country to play serious catch-up”, van der Klauw insists. Ultimately, the result will be better AI. “Humancentred AI has the potential to become a great export product.”

Further information: https://nlaic.com/en/ and https://nlaic.com/en/ained-programme/


NETWORK

Caspar Chorus

Smart software assistant knows what your colleagues would do

Insure or not? Operate or not? Fraud or not? Many organisations have to make a lot of fast decisions that can have considerable impact. Spin-off Councyl helps them to digitalise their expertise in order to support this decision-making. Text: Agaath Diemel

D

ecision-support systems are nothing new, but Councyl finds the happy medium between existing rigid rule-based systems and unfathomable AI. Rather than big data, Councyl works with expert data. “Our principle is that people find it difficult to put their expertise into words; they are unconsciously competent, as it is known. We make this knowledge explicit,” says Caspar Chorus, Professor of Choice Behaviour. For this Councyl uses techniques from econometric market research. “We are not predicting what type of iPhone someone will buy, but what percentage of colleagues would make a certain decision and why,” says Chorus. Once in use, the software assistant continues to learn. “Interaction comes about between the digital assistant and the human expert, each of which inputs their own choices. This feeds new knowledge into the system, not via extra decision rule input, but via the heads of experts who keep feeding themselves with new

knowledge. This is really a technological innovation in my field,” says Chorus. There is a real need for this. “Take the Dutch childcare benefits scandal, for instance. People were sometimes mercilessly let down by the system if they earned just a few euros too much, because a stringent rule-based system was used. The human measure was lacking.” And the opaque System Risk Indication (SyRI) algorithm, that the government used to track down benefits fraud, was banned by the court in 2020 because it violated fundamental human rights. It was precisely his irritation about scandals like this that motivated him towards innovation. “In recent years we’ve been hearing a lot about hoe AI is going to take over the world, but there is also growing dissatisfaction with black-box systems. So much that organisations are even turning away from them. Before you know it, the baby will be thrown out with the bath water. A shame, because there is so much potential in AI that would then never get a chance. I realised that my methods, which I was using for a completely different goal, should be able to obviate the drawbacks of AI.”

‘It is a big misunderstanding that the black box will become white if we just continue to do enough research’

25


AI is key to ING’s digital strategy

26

2020 saw the launch of a special project by ING and TU Delft: the AI for Fintech Research Lab. This marks a new chapter and further intensification of the collaborative relationship the two organisations, who previously worked together on software development in the ING Tech Research & Development department. Elvan Kula, who works for ING and is the manager of the new lab, tells us about the role AI will play in the bank’s core tasks and how TU Delft and ING are collaborating to implement innovation in the fintech sector. Text: Jochen Meischke

I

t is ING’s ambition to be a “tech company with a banking licence”. It’s no wonder the bank approached TU Delft to jointly investigate the possibilities for the broader application of AI. The lab, which operates under the ICAI flag (see page x), was set up in the same year Kula graduated at TU Delft in Data Science and Technology and then went to work for the bank. Given Kula’s interest in AI, it was not surprising that the new lab appealed to her. Kula secured one of the ten PhD spots and became lab manager six months after its launch. The job is challenging and the environment inspirational, she explained. “In every ICAI lab there is collaboration between one or more companies and a university, thus bringing academic knowledge and real data – insights from practice – together. For ING, it is very valuable to validate research in this way with TU Delft as its scientific partner.”

Forecasting delays, incidents and interactions Within the lab, Kula focuses on three topics: recognising and predicting incidents, predicting delays, and the further development of chatbots. She laughs: most people have an idea what a chatbot is, but she thinks her other fields of interest are a little more out of the ordinary. She explains: “Forecasting delays is a form of data analysis. AI can help us predict the likelihood of meeting a deadline for developing new software, for example. For ING it is important to know when, for example, new functionality on the website or app can be launched. The sooner we know whether a delay is likely, the better we can take anticipatory measures. Our current model already predicts delays with 80 per cent accuracy, as well as how long any delays to a project may last. Another important topic within financial services is recognising and predicting incidents. Kula: “At ING alone, more than 50 million transactions take place per day. We are now building AI that can independently recognise whether there’s anything irregular about a transaction, such as money laundering or other forms of financial

‘Forecasting delays is a form of data analysis. AI can help us predict the likelihood of meeting a deadline for developing new software’


NETWORK

or economic crime. At the moment, we primarily look at anomalies (as they are called), non-standard transactions. The next step is the alarm function: our AI lets ING’s staff know that it has noticed something irregular among the millions of transactions.”

More reliable prediction with AI At present, Kula’s work focuses on predicting delays, which she does with the help of enormous amounts of data from ING. “ING introduced agile working in 2011, and since 2015 almost all of their teams work in this fashion, in which self-organising teams chop large projects into shorter periods in order to be able to deliver something useful sooner. Thanks to the data this yields, you can really get to work on the technical and social aspects of software development. Humans are quite bad at judging what it is that leads to delays. AI is much better at this, because it uses years’ worth of data. One of the first insights emanating from this research is on the subject of what are known as ‘rapid releases’: companies would like to develop software more quickly and release it early in the form of new apps, website or other financial services. The question is whether increased speed compromises the quality of the software and the service. Our research shows that this is a concern for many software developers, but AI detected no proof of a correlation between poorer quality and shorter projects.”

Living Lab for innovation There are two ICAI lab locations, one in Delft and one in Amsterdam. Each of these is actually a kind of living lab in which ING and TU Delft get together, says Kula. “Within our lab, researchers, students and employees of the bank and of the -university apply state-of-the art academic knowledge to ING’s existing knowledge

Elvan Kula

Chatbots, too, hold a lot of promise, says Kula. “The current generations of chatbots are still relatively simple. They can be improved, for example, to be able to classify reasons for customer contact: predicting why and about what a customer would seek contact. Say, for example, that a customer’s account has been blocked. If the customer then contacts you, you will have a reasonable idea what their question will be about if they have already been informed about this by the chatbot.”

and processes. The question really is: how can we, together, use the latest academic insights in real-world applications? The lab also gives birth to new concepts that have not previously been researched. If someone from TU Delft or ING has a good idea, we are in a position to elaborate and further test it here first.” It is important to be diligent when working with (customer) data and ING sees great importance in doing so in close collaboration with TU Delft. Kula mentions the solid reputation of the university: “TU Delft really looks at the practical applications of AI. That makes the collaboration valuable for ING. It is also one of the reasons why ING is looking into developing the AI for Fintech Research Lab into a fully-fledged department of the company.” ING has been using AI as the key to its fintech strategy for quite a while and the ICAI lab is part of this. Kula: “In the first year we worked hard at building awareness of what we do within ING. More and more departments are approaching us now to help address any complex challenges and problems they have, which was not the case this time last year. Of course, we can dream up all kinds of innovations, but it’s when colleagues within ING contribute input that we can make a real difference as lab.”

Home of Innovation | Stories about innovation and societal impact

27


IMPACT

ICAI labs with industrial partners

28 28

TU Delft innovates together with a number of leading companies in the field of AI. This is done within the Innovation Center for Artificial Intelligence (ICAI), a national partnership that aims to keep the Netherlands at the forefront of knowledge and talent development in AI. The collaborations consists of multiyear strategic partnerships between the university and an industrial partner. TU Delft is currently working together in three innovation labs with respectively Ahold Delhaize (AI for Retail Lab), ING (AI for Fintech Research) and DSM (AI for Biosciences Lab).

AIRLab (AI for Retail), with Ahold Delhaize The AI for Retail (AIR) Lab Delft is a joint TU Delft-Ahold Delhaize industry lab consisting of a robotics research programme and test site focused on developing state-of-the-art innovations in the retail industry. By expanding its focus to robotics, AIRLab Delft will further drive innovations for daily business while building more knowledge of the intersection between retail, AI and robotics. The expansion comprises a robotics research programme and test site for developing state-of-the-art innovations in the retail industry.

‘The rapid advancements in AI and robotics provide us with significant opportunities to make everyday shopping even easier for our customers and develop new solutions for our warehouses and last-mile delivery. Working together with academic partners such as TU Delft will enable Ahold Delhaize and AIRLab to shape a technology-driven world in a responsible way. It helps us become a frontrunner in AI research and development for retail and ultimately build capabilities that are scalable for the group’ Frans Muller President and CEO of Ahold Delhaize


AFR (AI for FinTech Research), with ING

AI4B.io Lab (AI for Biosciences Lab), with DSM

Mercury Machine Learning Lab, with Booking.com and UvA

AI for Fintech Research (AFR) is a collaboration between ING and TU Delft. The mission of AFR is to perform worldclass research at the intersection of Artificial Intelligence, Data Analytics, and Software Analytics in the context of FinTech. Ten PhD researchers will work in the lab on projects that will focus, among other things, on autonomous software engineering, data integration, analytics delivery, and continuous experimentation.

AI for Biosciences Lab (AI4B.io Lab) is a collaboration between TU Delft and DSM. The lab focuses on improving production technologies and developing bio-based products using AI. AI4B.io Lab is the first of its kind in Europe to apply artificial intelligence to full-scale biomanufacturing, from microbial strain development to process optimisation and scheduling. AI4B.io Lab is aimed at long-term innovation in the domain of AI for developing biobased products and optimising biobased production technologies.

The Mercury Machine Learning Lab is a collaboration between TU Delft, University of Amsterdam (UvA) and Booking.com. The lab focuses on the development and applications of artificial intelligence to the specific domain of online travel booking and recommendation service systems. The collaboration within the lab combines expertise of scientists from UvA (e.g. information retrieval), TU Delft (e.g. reinforcement learning) with the unique expertise, experience and availability of big data at Booking.

‘The Mercury Machine Learning Lab’s focus on developing better algorithms for recommender systems is highly relevant to our society as these systems guide many of our digital interactions, not only how we plan our travel but also the movies we watch, the news we read and the products we might buy. By addressing fundamental AI challenges, the results of the lab will also be valuable in other domains’ Geert-Jan Houben Pro Vice Rector Magnificus AI, Data & Digitalisation

29


TOEKOMST

FUTURE

Jaimy Siebel (left) and David Abbink

The future of work

Workers and robots improving each other The new FRAIM institute is set to focus on the future of work. It is based on the idea that higher productivity should go hand-in-hand with more meaningful work. FRAIM aims to achieve this by enabling workers and robots to learn from each other. Text: Bennie Mols

I

n half a century, robots have evolved from static, dangerous and extremely heavy machines only capable of helping to manufacture cars from behind fencing into mobile, safe, flexible and communicating devices that can sometimes work side-by-side with humans across various sectors. However, despite this impressive development, their practical application in the world of work has yet to meet the expectations of society and industry. Working processes are often so complex, varied and dynamic that robots are still too limited for them. “In order to significantly accelerate robot innovation and

enable humans and robots to collaborate smoothly at work, we’re setting up a new institute on the TU Delft Campus, called FRAIM”, explains Jaimy Siebel, currently managing director of RoboValley and soon-to-be operational director of FRAIM.

Meaningful work FRAIM will focus on the future of work. As the population in the Netherlands continues to age, fewer working people will need to produce and care for larger numbers of people who are not working. In order to make that possible, the use of robots for parts of the work will be essential. Equally, people also want to continue to do meaningful work. “FRAIM will therefore aim to


provide an answer to the question of how humans and robots can collaborate symbiotically at work”, says David Abbink, professor of haptic human-robot interaction at TU Delft and the future director of research at FRAIM. “We’re familiar with symbiosis from biology as two organisms evolving interactively. At FRAIM, we aim to achieve something similar between workers and robots: they need to improve each other.” (See box). In around three years’ time, FRAIM will have its own building on the TU Delft Campus, alongside YES!Delft. The building will have its own working labs: unique laboratories in which both technological developments and social processes can be studied simultaneously in a work setting. Five overarching themes have already been agreed on: production processes, retail trade, manufacturing industry, care, and

agriculture & nutrition. Ultimately, FRAIM aims to focus its research on ten specific themes.

New way of research The first three labs are to be established in September of this year, based for the time being on the site of the current RoboHouse field lab. The brand name FRAIM will be launched in the course of this year. Siebel: “Our current robotics ecosystem, started in 2015, attempts to keep robot innovation reasonably close to the market. However, what we still lack is a holistic understanding of the impact a robot has on processes at work.” “FRAIM’s scientific ambition”, explains Prof. David Abbink, “will be to involve both the social and human sciences in the technical design and development processes. That calls for a new type of research lab and a new way of conducting

research that transcends the traditional boundaries between disciplines.” How do Abbink and Siebel envisage FRAIM in five years’ time? Abbink: “By then, I hope we will have created a vibrant scientific community conducting pioneering research that links in seamlessly with people’s everyday reality at work. Someone from a company will then be able to walk into one of our labs and physically experience what the future of work will look like.”

Leading institute Siebel: “I hope that, in five years, FRAIM will have become a leading European, and hopefully international, institute in the field of the work of the future – a place where companies, organisations, researchers, talented scientists and start-ups are working on the future of our working processes.”

What constitutes a symbiotic relationship between worker and robot? For decades, robots had no brains and artificial intelligence had no body. However, robots and AI have been coming closer together in recent years. Thanks to new AI technologies, robot brains are improving all the time. They can see, move, manipulate and learn better. This kind of robot is known as a cognitive robot. Currently, they are mainly being developed in labs, and their future applications include robots that can support nursing staff with physical tasks in healthcare and robots that help people in supermarkets or with logistical processes. If cognitive robots are to collaborate effectively with humans at work, they

need to meet five basic conditions. These basic conditions, devised by Prof. David Abbink, will also be central to FRAIM as an institute. 1. Robots must be safe (not inflict damage on humans), they must be secure (against cyberattacks) and they need to be sustainable (energy- and material-efficient). 2. Robots must be responsive: a robot needs to understand what a human is capable of and wants to do, and what not; it must be possible for the robot to be adjusted intuitively by different people. 3. Robots must make transparent decisions: workers need to

understand what a robot is capable of and wants to do and why. Only if these conditions are met, will workers and organisations also be able to take responsibility for the work done jointly with the robot. 4. R obots need to be versatile and not solely capable of completing one specialist task in a specialist context. They also need to be able to collaborate with a variety of workers. 5. R obots need to challenge humans: workers must not be robots’ slaves who only do mind-numbing tasks that robots cannot do, but must be challenged to improve their own skills.

Home of Innovation | Stories about innovation and societal impact

31


OPINION

32

AI building set to be a hotspot for regional collaboration


‘There is a genuine need out there and the House of AI will help to unleash this demand much more easily, which we can then link to what we have to offer. It will become the hotspot for regional collaboration in AI’

In three years’ time, the TU Delft campus will have another new building in which all innovation activities in the field of AI converge. This will be a vibrant place with international appeal, where students, academics, start-ups, SMEs and large companies will work together on new AI applications. This new initiative is the missing link in the AI ecosystem, writes Paul Althuis, Director of the TU Delft Innovation & Impact Centre. Column by Paul Althuis

T

he world of AI is virtual and almost everything takes place beyond our sight, on powerful computers, in the cloud and in high-speed networks. However, these digital developments will have a major impact on the physical world around us. AI is important for the future of the industry in our region. Think of smart applications in the port area in Rotterdam, of mobility and healthcare. At TU Delft, we work together with public partners, companies and regional knowledge institutions (Erasmus University Rotterdam and Leiden University) on digital innovations that can contribute to solutions for the major challenges of our time. Because of this collaboration, it is important that there is a physical place where all developments in the field of AI, data and digitalisation come together.

Co creation

true in complex deep tech, such as AI. This cross-fertilisation between science, the business community and government does not happen automatically – it needs to be organised. In a building that acts as enabler of innovation. With regard to AI, TU Delft serves as a focal point for the region. TU Delft has the largest com munity of AI scientists, is in the process of setting up 24 AI labs (of which 16 are already established) and is working to ensure that AI becomes part of every degree programme. This is a superb springboard position for TU Delft to make an impact and stand out: as a university of technology, we can use the content of our engineering areas of application in AI.

Missing link This ‘House of AI’ forms the missing link in our innovation strategy. In order to develop spin-offs, generate contacts with the business community and secure public-private investments, a place where people can meet up and collaborate on new innovations is essential. We already know from existing field labs that they are excellent places for businesses to become acquainted with new technology. There is a genuine need out there and the House of AI will help to unleash this demand much more easily, which we can then link to what we have to offer. It will become the hotspot for regional collaboration in AI.”

It’s quite simple: in 2021, co-creation plays a crucial role in innovation. This is particularly

Home of Innovation | Stories about innovation and societal impact

33


TREND

Regional collaboration in AI, Data and Digitalisation

34

The Zuid-Holland region is home to one of the largest AI ecosystems in the Netherlands. To unlock its potential, both in scientific research and innovation, the various knowledge institutions, governments and companies in the province are working closely together. In this way they contribute to the major regional, national and international societal challenges.

M

any AI techniques have shown a strong development over the past ten years, both from the theory of algorithms, machine learning and human-machine interaction and from the practice where large-scale utilisation has become possible, for example because the amount of data has increased enormously.

AI Convergence

This gives us important tools to tackle societal challenges. This starts in the region: Zuid-Holland, with the Port of Rotterdam, two large cities and major challenges in the field of energy and sustainability, mobility, safety and health, embodies the major challenges that we also face at a national and international level.

• • • • •

If we tackle these challenges together and from different disciplines, then we can take large steps. TU Delft is involved in two platforms from which this is shaped. Within the so-called AI Convergence, the Zuid-Holland universities and the university medical centers are working together. This collaboration forms the scientific backbone of Holland AI, in which governments and the business community are also involved.

We jointly develop minor programs in AI, data and digitalisation, and prepare modules and materials that will be included in the curricula of all 85,000 university students in Zuid-Holland. In research and innovation we link up with the themes that have a strong regional profile. It is no coincidence that these are the same subjects as the focus themes within TU Delft: Energy and sustainability Port and maritime Peace, justice and security Health and care Technological industry

Holland AI Holland AI is a network organisation of knowledge institutions, governments and companies in South Holland that work with AI. The AI Hub Zuid-Holland unlocks the regional AI ecosystem, one of the largest in the country. The focus is on strengthening this regional ecosystem and on accelerating innovation around AI. Here, too, the focus is on the above five themes. Holland AI is affiliated with the Dutch AI Coalition and is a showcase for knowledge institutions, governments and companies that want to work with AI.

‘Societal problems are not either technological or social in nature: they are always both, and they require an integrated approach. They call for new forms of knowledge and knowledge development that will only be possible if scientific disciplines grow towards each other, if they converge. Only with this convergence can we achieve the necessary impact on our society’



36

TU Delft | Innovation & Impact Centre P.O. Box 5 2600 AA Delft


Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.