SlideShare a Scribd company logo

Applying AI technologies on industrial needs - Caj Södergård 2023.pptx

The slides are from my keynote at the AI Forum in Helsinki 2023.

1 of 23
Download to read offline
Applying AI technologies on industrial
needs - challenging the AI productivity
paradox
Prof. Caj Södergård
NextAI
Caj.Sodergard@nextai.fi www.nextai.fi
Content
• Lagging productivity
• Generative AI to boost productivity
• Example: development of new materials
• Mitigating risks with ethical guidelines & regulation
Machine Learning
- Incl. Deep Learning
What is Artificial Intelligence ?
Sources:
- International Joint Conference on AI
www.ijcai-18.org/wp-content/uploads/2018/01/demos.txt
- A definition of AI, EU AI High Level Expert Group, 2019
AI enables machines,
devices, programs, systems
and services to act in a
sensible way depending on
goal and situation
Adapted from Russel and
Norvig, 2014 & HLEG 2019
Machine Reasoning
- Planning & Scheduling
- Knowledge representation
& reasoning
- Search & optimisation
- Multi-agent systems
- Perception (speech, vision)
- Conversational Interfaces
- Generative AI
- Natural Language Processing
- Robotics
Perception & Interaction
18/12/2023 3
AI fear…
Productivity lags in Finland
Productivity = GDP____
working hrs
Companies recruited few AI specialists
AI skills
Source: Asta Bäck, Caj Södergård: Tekoälytaitojen kysyntä
työmarkkinoilla työpaikkailmoitusten valossa. T&Y, 2021
AI skills

Recommended

20210325 jim spohrer future ai v11
20210325 jim spohrer future ai v1120210325 jim spohrer future ai v11
20210325 jim spohrer future ai v11ISSIP
 
[DSC Europe 23] Luciano Catani - AI in Diplomacy.PDF
[DSC Europe 23] Luciano Catani - AI in Diplomacy.PDF[DSC Europe 23] Luciano Catani - AI in Diplomacy.PDF
[DSC Europe 23] Luciano Catani - AI in Diplomacy.PDFDataScienceConferenc1
 
Applications of Computational Intelligence, Internet of Things and Cutting Ed...
Applications of Computational Intelligence, Internet of Things and Cutting Ed...Applications of Computational Intelligence, Internet of Things and Cutting Ed...
Applications of Computational Intelligence, Internet of Things and Cutting Ed...Christo Ananth
 
Looking Beyond the AI & IoT Research and Industrial Opportunities: How two Br...
Looking Beyond the AI & IoTResearch and Industrial Opportunities:How two Br...Looking Beyond the AI & IoTResearch and Industrial Opportunities:How two Br...
Looking Beyond the AI & IoT Research and Industrial Opportunities: How two Br...Haithem Afli
 
Ert 20200420 v11
Ert 20200420 v11Ert 20200420 v11
Ert 20200420 v11ISSIP
 
Robotics & Artificial (RAI) Intelligence webinar: Law & Regulation for RAI In...
Robotics & Artificial (RAI) Intelligence webinar: Law & Regulation for RAI In...Robotics & Artificial (RAI) Intelligence webinar: Law & Regulation for RAI In...
Robotics & Artificial (RAI) Intelligence webinar: Law & Regulation for RAI In...KTN
 
Greening Artificial Intelligence
Greening Artificial IntelligenceGreening Artificial Intelligence
Greening Artificial IntelligenceAlexandra Petruș
 
Webinar: Why risk managers should look at Artificial Intelligence now?
Webinar: Why risk managers should look at Artificial Intelligence now?Webinar: Why risk managers should look at Artificial Intelligence now?
Webinar: Why risk managers should look at Artificial Intelligence now?FERMA
 

More Related Content

Similar to Applying AI technologies on industrial needs - Caj Södergård 2023.pptx

Internet of Industrial Things Presentation - Sophie Peachey - IoT Midlands Me...
Internet of Industrial Things Presentation - Sophie Peachey - IoT Midlands Me...Internet of Industrial Things Presentation - Sophie Peachey - IoT Midlands Me...
Internet of Industrial Things Presentation - Sophie Peachey - IoT Midlands Me...WMG, University of Warwick
 
Ai tech in india
Ai tech in indiaAi tech in india
Ai tech in indiaGUNASAI2
 
Impact of Generative AI in Cybersecurity - How can ISO/IEC 27032 help?
Impact of Generative AI in Cybersecurity - How can ISO/IEC 27032 help?Impact of Generative AI in Cybersecurity - How can ISO/IEC 27032 help?
Impact of Generative AI in Cybersecurity - How can ISO/IEC 27032 help?PECB
 
Global Governance of Generative AI: The Right Way Forward
Global Governance of Generative AI: The Right Way ForwardGlobal Governance of Generative AI: The Right Way Forward
Global Governance of Generative AI: The Right Way ForwardLilian Edwards
 
AI Summary eng.pptx
AI Summary eng.pptxAI Summary eng.pptx
AI Summary eng.pptx진희 이
 
How Can AI and IoT Power the Chemical Industry?
How Can AI and IoT Power the Chemical Industry?How Can AI and IoT Power the Chemical Industry?
How Can AI and IoT Power the Chemical Industry?Xiaonan Wang
 
IT Cluster Skolkovo Presentation at FRUCT.org conference
IT Cluster Skolkovo Presentation at FRUCT.org conferenceIT Cluster Skolkovo Presentation at FRUCT.org conference
IT Cluster Skolkovo Presentation at FRUCT.org conferenceAlbert Yefimov
 
Applications instead of Capabilities: Innovating Return on Investment of Netw...
Applications instead of Capabilities: Innovating Return on Investment of Netw...Applications instead of Capabilities: Innovating Return on Investment of Netw...
Applications instead of Capabilities: Innovating Return on Investment of Netw...gkout
 
I4ADA 2019 Presentation Stepheni baraki
I4ADA 2019 Presentation  Stepheni barakiI4ADA 2019 Presentation  Stepheni baraki
I4ADA 2019 Presentation Stepheni barakiPaul van Heel
 
HPCC Systems Engineering Summit Presentation - Collaborative Research with FA...
HPCC Systems Engineering Summit Presentation - Collaborative Research with FA...HPCC Systems Engineering Summit Presentation - Collaborative Research with FA...
HPCC Systems Engineering Summit Presentation - Collaborative Research with FA...HPCC Systems
 
e-SIDES and Ethical AI
e-SIDES and Ethical AIe-SIDES and Ethical AI
e-SIDES and Ethical AIIDC4EU
 
20090327 Software Engineering -- What's in it for me?
20090327 Software Engineering -- What's in it for me?20090327 Software Engineering -- What's in it for me?
20090327 Software Engineering -- What's in it for me?Arian Zwegers
 
Generative AI - Responsible Path Forward.pdf
Generative AI - Responsible Path Forward.pdfGenerative AI - Responsible Path Forward.pdf
Generative AI - Responsible Path Forward.pdfSaeed Al Dhaheri
 
Connecting AI technologies with industry needs
Connecting AI technologies with industry needsConnecting AI technologies with industry needs
Connecting AI technologies with industry needsBig Data Value Association
 
Internet of things_by_economides_keynote_speech_at_ccit2014_final
Internet of things_by_economides_keynote_speech_at_ccit2014_finalInternet of things_by_economides_keynote_speech_at_ccit2014_final
Internet of things_by_economides_keynote_speech_at_ccit2014_finalAnastasios Economides
 
AIOTI presentation
AIOTI presentationAIOTI presentation
AIOTI presentationIoTUK
 
AIVS - AI, Industrial Data Space, and Innovation Transformation
AIVS - AI, Industrial Data Space, and Innovation TransformationAIVS - AI, Industrial Data Space, and Innovation Transformation
AIVS - AI, Industrial Data Space, and Innovation Transformationpantapong
 

Similar to Applying AI technologies on industrial needs - Caj Södergård 2023.pptx (20)

Internet of Industrial Things Presentation - Sophie Peachey - IoT Midlands Me...
Internet of Industrial Things Presentation - Sophie Peachey - IoT Midlands Me...Internet of Industrial Things Presentation - Sophie Peachey - IoT Midlands Me...
Internet of Industrial Things Presentation - Sophie Peachey - IoT Midlands Me...
 
Ai tech in india
Ai tech in indiaAi tech in india
Ai tech in india
 
Impact of Generative AI in Cybersecurity - How can ISO/IEC 27032 help?
Impact of Generative AI in Cybersecurity - How can ISO/IEC 27032 help?Impact of Generative AI in Cybersecurity - How can ISO/IEC 27032 help?
Impact of Generative AI in Cybersecurity - How can ISO/IEC 27032 help?
 
Global Governance of Generative AI: The Right Way Forward
Global Governance of Generative AI: The Right Way ForwardGlobal Governance of Generative AI: The Right Way Forward
Global Governance of Generative AI: The Right Way Forward
 
AI Summary eng.pptx
AI Summary eng.pptxAI Summary eng.pptx
AI Summary eng.pptx
 
How Can AI and IoT Power the Chemical Industry?
How Can AI and IoT Power the Chemical Industry?How Can AI and IoT Power the Chemical Industry?
How Can AI and IoT Power the Chemical Industry?
 
IT Cluster Skolkovo Presentation at FRUCT.org conference
IT Cluster Skolkovo Presentation at FRUCT.org conferenceIT Cluster Skolkovo Presentation at FRUCT.org conference
IT Cluster Skolkovo Presentation at FRUCT.org conference
 
Applications instead of Capabilities: Innovating Return on Investment of Netw...
Applications instead of Capabilities: Innovating Return on Investment of Netw...Applications instead of Capabilities: Innovating Return on Investment of Netw...
Applications instead of Capabilities: Innovating Return on Investment of Netw...
 
I4ADA 2019 Presentation Stepheni baraki
I4ADA 2019 Presentation  Stepheni barakiI4ADA 2019 Presentation  Stepheni baraki
I4ADA 2019 Presentation Stepheni baraki
 
HPCC Systems Engineering Summit Presentation - Collaborative Research with FA...
HPCC Systems Engineering Summit Presentation - Collaborative Research with FA...HPCC Systems Engineering Summit Presentation - Collaborative Research with FA...
HPCC Systems Engineering Summit Presentation - Collaborative Research with FA...
 
e-SIDES and Ethical AI
e-SIDES and Ethical AIe-SIDES and Ethical AI
e-SIDES and Ethical AI
 
20090327 Software Engineering -- What's in it for me?
20090327 Software Engineering -- What's in it for me?20090327 Software Engineering -- What's in it for me?
20090327 Software Engineering -- What's in it for me?
 
Generative AI - Responsible Path Forward.pdf
Generative AI - Responsible Path Forward.pdfGenerative AI - Responsible Path Forward.pdf
Generative AI - Responsible Path Forward.pdf
 
Teollinen internet: mistä liiketoimintahyötyä? - Heikki Ailisto
Teollinen internet: mistä liiketoimintahyötyä? - Heikki AilistoTeollinen internet: mistä liiketoimintahyötyä? - Heikki Ailisto
Teollinen internet: mistä liiketoimintahyötyä? - Heikki Ailisto
 
Connecting AI technologies with industry needs
Connecting AI technologies with industry needsConnecting AI technologies with industry needs
Connecting AI technologies with industry needs
 
Internet of things_by_economides_keynote_speech_at_ccit2014_final
Internet of things_by_economides_keynote_speech_at_ccit2014_finalInternet of things_by_economides_keynote_speech_at_ccit2014_final
Internet of things_by_economides_keynote_speech_at_ccit2014_final
 
2015 imcrc
2015 imcrc2015 imcrc
2015 imcrc
 
Innovations using PowerAI
Innovations using PowerAIInnovations using PowerAI
Innovations using PowerAI
 
AIOTI presentation
AIOTI presentationAIOTI presentation
AIOTI presentation
 
AIVS - AI, Industrial Data Space, and Innovation Transformation
AIVS - AI, Industrial Data Space, and Innovation TransformationAIVS - AI, Industrial Data Space, and Innovation Transformation
AIVS - AI, Industrial Data Space, and Innovation Transformation
 

Recently uploaded

AWS Overview of AWS Clarify, Feature Store, Hyper parameter Tuning
AWS Overview of AWS  Clarify, Feature Store, Hyper parameter TuningAWS Overview of AWS  Clarify, Feature Store, Hyper parameter Tuning
AWS Overview of AWS Clarify, Feature Store, Hyper parameter TuningVarun Garg
 
Augmented and Mixed Reality Solutions for Frontline Medical Professionals
Augmented and Mixed Reality Solutions for Frontline Medical ProfessionalsAugmented and Mixed Reality Solutions for Frontline Medical Professionals
Augmented and Mixed Reality Solutions for Frontline Medical Professionalsthirdeyegen65
 
Obstructive jaundice is a medical condition characterized by the yellowing of...
Obstructive jaundice is a medical condition characterized by the yellowing of...Obstructive jaundice is a medical condition characterized by the yellowing of...
Obstructive jaundice is a medical condition characterized by the yellowing of...ssuser7b7f4e
 
Regulation is Coming - Trusted Media Summit 2023
Regulation is Coming - Trusted Media Summit 2023Regulation is Coming - Trusted Media Summit 2023
Regulation is Coming - Trusted Media Summit 2023Damar Juniarto
 
Model Jaringan network jaringan komputer.pdf
Model Jaringan network jaringan komputer.pdfModel Jaringan network jaringan komputer.pdf
Model Jaringan network jaringan komputer.pdfgalfinprihardiputra0
 
Red shadows ringing in Japan's Cyberspace
Red shadows ringing in Japan's CyberspaceRed shadows ringing in Japan's Cyberspace
Red shadows ringing in Japan's Cyberspacesttyk
 
UGB INTERNETBANKING FACILITY LAUNCHED.pptx
UGB INTERNETBANKING FACILITY LAUNCHED.pptxUGB INTERNETBANKING FACILITY LAUNCHED.pptx
UGB INTERNETBANKING FACILITY LAUNCHED.pptxRitesh Sahu
 
Augmented and Mixed Reality Solutions for Aerospace & Defense
Augmented and Mixed Reality Solutions for Aerospace & DefenseAugmented and Mixed Reality Solutions for Aerospace & Defense
Augmented and Mixed Reality Solutions for Aerospace & Defensethirdeyegen65
 

Recently uploaded (8)

AWS Overview of AWS Clarify, Feature Store, Hyper parameter Tuning
AWS Overview of AWS  Clarify, Feature Store, Hyper parameter TuningAWS Overview of AWS  Clarify, Feature Store, Hyper parameter Tuning
AWS Overview of AWS Clarify, Feature Store, Hyper parameter Tuning
 
Augmented and Mixed Reality Solutions for Frontline Medical Professionals
Augmented and Mixed Reality Solutions for Frontline Medical ProfessionalsAugmented and Mixed Reality Solutions for Frontline Medical Professionals
Augmented and Mixed Reality Solutions for Frontline Medical Professionals
 
Obstructive jaundice is a medical condition characterized by the yellowing of...
Obstructive jaundice is a medical condition characterized by the yellowing of...Obstructive jaundice is a medical condition characterized by the yellowing of...
Obstructive jaundice is a medical condition characterized by the yellowing of...
 
Regulation is Coming - Trusted Media Summit 2023
Regulation is Coming - Trusted Media Summit 2023Regulation is Coming - Trusted Media Summit 2023
Regulation is Coming - Trusted Media Summit 2023
 
Model Jaringan network jaringan komputer.pdf
Model Jaringan network jaringan komputer.pdfModel Jaringan network jaringan komputer.pdf
Model Jaringan network jaringan komputer.pdf
 
Red shadows ringing in Japan's Cyberspace
Red shadows ringing in Japan's CyberspaceRed shadows ringing in Japan's Cyberspace
Red shadows ringing in Japan's Cyberspace
 
UGB INTERNETBANKING FACILITY LAUNCHED.pptx
UGB INTERNETBANKING FACILITY LAUNCHED.pptxUGB INTERNETBANKING FACILITY LAUNCHED.pptx
UGB INTERNETBANKING FACILITY LAUNCHED.pptx
 
Augmented and Mixed Reality Solutions for Aerospace & Defense
Augmented and Mixed Reality Solutions for Aerospace & DefenseAugmented and Mixed Reality Solutions for Aerospace & Defense
Augmented and Mixed Reality Solutions for Aerospace & Defense
 

Applying AI technologies on industrial needs - Caj Södergård 2023.pptx

  • 1. Applying AI technologies on industrial needs - challenging the AI productivity paradox Prof. Caj Södergård NextAI Caj.Sodergard@nextai.fi www.nextai.fi
  • 2. Content • Lagging productivity • Generative AI to boost productivity • Example: development of new materials • Mitigating risks with ethical guidelines & regulation
  • 3. Machine Learning - Incl. Deep Learning What is Artificial Intelligence ? Sources: - International Joint Conference on AI www.ijcai-18.org/wp-content/uploads/2018/01/demos.txt - A definition of AI, EU AI High Level Expert Group, 2019 AI enables machines, devices, programs, systems and services to act in a sensible way depending on goal and situation Adapted from Russel and Norvig, 2014 & HLEG 2019 Machine Reasoning - Planning & Scheduling - Knowledge representation & reasoning - Search & optimisation - Multi-agent systems - Perception (speech, vision) - Conversational Interfaces - Generative AI - Natural Language Processing - Robotics Perception & Interaction 18/12/2023 3
  • 5. Productivity lags in Finland Productivity = GDP____ working hrs
  • 6. Companies recruited few AI specialists AI skills Source: Asta Bäck, Caj Södergård: Tekoälytaitojen kysyntä työmarkkinoilla työpaikkailmoitusten valossa. T&Y, 2021 AI skills
  • 7. Generative AI & Foundation Models • Create content with a quality close to humans. • Are based on deep neural networks & vast amounts of training data. • Are able to perform a variety of tasks (general purpose). • Create text, images, video audio, program code, DNA sequences, …
  • 8. Why will Generative AI boost productivity? • Promises: Generative AI adds 1,5 % units to productivity* • More efficient output (linear) & acceleration of innovation (exponential). • Supports mainly knowledge workers (~2/3 of all). • Writing and programming can be much quicker. • Captures organisational best practicies. • Is fastly rolled out (incl. SMEs) & interaction easy. @YLE, 2023 * Goldman Sachs 2023
  • 9. AI Empowered Material Scientist (AIMS) Designing superior materials in minutes instead of years Source: VTT iBEX 2021 Final Event 22.02.2022
  • 10. The problem AIMS Unsustainable fossil- based materials Need for extreme performance Need for renewable materials Weak materials Help from Nature
  • 11. 18.12.2023 VTT – beyond the obvious 11  Trial and error  Years to develop  Massive costs  Complexity beyond human reach Challenges
  • 12. . Our Approach AIMS Synthetic Biology Radically Accelerated Materials Design Artificial Intelligence
  • 13. New proteins & materials DNA Cells AIMS Material variant designs Synbio Lab Data on properties Expert
  • 14. New proteins & materials DNA Cells AIMS AI speeds up invention AI Deep learning, Generative algorithms Material variant designs Finalists Synbio Lab Data on properties Protein databases Generate Generate novel candidates Predict properties of candidates Expert
  • 15. Click to add text AI Predicts properties from DNA sequence
  • 16. AI invents never-seen proteins Seed proteins Variant sequences Predicted properties Selected best Complexity Challenge: An amino acid sequence of length 10: 2010 =10 240 000 000 000 combinations combinations.
  • 17. VTT – beyond the obvious • 1800 digital proteins proposed by AIMS AI • 144 novel & stable proteins filtered by AI • 27 diverse protein sequences handpicked by Synbio expert & synthetized • 10 novel protein materials validated in Lab with good results Result: 10 novel validated protein materials AI Human
  • 18. Our AI-invented proteins have superior stability 190 200 210 220 230 240 250 -100 -80 -60 -40 -20 0 20 40 CD (mdeg) Wavelength (nm) 20 °C 25 °C 30 °C 35 °C 40 °C 45 °C 50 °C 45 °C 40 °C 35 °C 30 °C 25 °C 20 °C Human-designed ELP Irreversible 190 200 210 220 230 240 250 -60 -40 -20 0 20 CD (mdeg) Wavelength (nm) 20 °C 25 °C 30 °C 35 °C 40 °C 45 °C 50 °C 45 °C 40 °C 35 °C 30 °C 25 °C 20 °C AI-designed ELP* Reversible = stable *ELP = Elastine Like Polypeptides Source: Pezhman Mohammadi
  • 19. Risks with Generative AI • Significant changes in job content and even vanishing tasks. • Impact on high-paying occupations - risk of significant drops in income. • Will the AI systems compliment or replace labour? • AI systems are developing fast – do complimentary skills match up? • Economists: Labour demand, supply and wage levels will adjust granting close to full employment in an aging society. • Need close attention from the society to manage adoption so that all can benefit.
  • 20. Ethical AI for mitigating the risk  Ensuring that AI is safe, trustworthy, accountable and transparent for users.  Avoiding bias in AI models that can discriminate against certain groups of people.  Protecting users’ privacy and their data.  Reducing the environmental risks of AI (e.g. energy consumption, e-waste).  Promoting human values and rights in AI design and deployment, e.g. avoiding disinformation.
  • 21. EU AI Act – first AI Law ever • Foundation model requirements: Transparency, legal content, registration. • Companies must demonstrate compliance before and monitor after AI deployment. • Non-compliance fines up to €30 million or 6% of global income. EU AI Office monitors. • European Parliament votes 12-15 June 2023. Thereafter, negotiations between Parliament, Council and Commission several months, even years. Unacceptable Risk Social scoring, manipulation, exploiting vulnerabilities High Risk Health, safety, rights, environment, transport, jobs, infra, biometrics, influencing Minimal Risk Spam filters, video games
  • 22. Conclusion • Productivity has lagged in Finland and elsewhere for over 10 years despite breakthroughs in AI. This AI paradox is visible in job advertisements. • Generative AI has the potential to radically raise the productivity of knowledge work and thus solve the AI paradox. • New materials can be developed in much shorter time with generative AI. • Generative AI will make jobs obsolete, but new occupations will compensate it. • AI has to be regulated to benefit all. AI Act is a good tool for this.
  • 23. Thank you for your attention! Caj.Sodergard@nextai.fi www.nextai.fi

Editor's Notes

  1. 1. Hello to everyone, both to you here physically and you participating remotely!   2, It’s exciting to be here today discuss with you a critical topic – how Artificial Intelligence can enhance industrial productivity,. SMILE   3. Personally, I have had almost a lifelong relation to AI – ever since the old days in Otaniemi more than 40 years ago, when I studied for the famous AI pioneer prof Teuvo Kohonen & made my diploma work about machine vision for robots, which took me to develop machine vision for sawmills in a company, before joining VTT, projects I have worked on, maybe under different names, but anyhow. I have also been active in several European Big Data and AI Associations. 4. Currently, I work for the European AI, Data and Robotics Association –Adra - through my company Next AI.
  2. Today, I will share with you my views on the currently lagging work productivity on how the latest wave of generative AI is expected to turn productivity into growth. I will give a recent example from my own AI research of how development of new materials can be radically boosted with AI. Finally, I will discuss some risks of Generative AI and how they can be mitigated with ethical guidelines and regulation.
  3. 1.Let us first define what we are talking about. Somebody jokingly defined AI as something that you cannot do with computers today, but something you can do tomorrow or after a year . I will propose a more precise definition drawing on sources mentioned on the slide, 2.I define AI as technologies that allow machines, programs, etc to act sensibly depending on goal and situation.”. 3.Three disciplines are essential in Artificial intelligence research and development 1) Machine Learning 2) Machine Reasoning and 3) Perception and Interaction , where I am myself been most active. If you go to a broad AI conference, you will find these tracks and topics represented. 4.Machine Learning, especially data-driven deep learning, is now very much driving the development, especially together with breakthroughs in Natural Language Processing enabling Generative AI, which I will mostly dwell on today. 5.However, to create something even resembling human intelligence, we need major breakthroughs also in the third discipline Machine Reasoning and world modelling and we don´t have those yet.
  4. 1.Although we're a long way from achieving what could be called as Artificial General Intelligence (AGI), recent hype around ChatGPT has led to an exaggerated opinions of what of current AI can do. This has sparked widespread fears that AI could even pose a threat to humankind like nuclear weapons. 2.In fact, most Americans seem to believe so as seen here. And leading AI scientist and industry leaders have also expressed their concerns, most recently this week in an Open Statement. 3.It is totally justified to be concerned of the long-range risks, but this should not hinder us to see AI´s short- term benefits to help solve our urgent challenges and raise the wellbeing for all od us. So let us explore how AI can elevate work productivity.
  5. 1..   2. From this graph, we can see a strong growth in the first part of the period up to the finance crisis 2008, Productivity is a measure of how labour create value. For a country, it can be calculated as Gross Domestic Product / amount of working hours. If you cannot increase the amount of work in a country like in an aging population, the only way for us to get higher standard of living is to raise productivity probably partly because of strong uptake of PC’s, Internet and web in companies during these years. This is in line with Prof Matti Pohjola´s findings that Information and Communication Technology (ICT) has contributed to about half of Finland's productivity growth in the last 25 years. Yet ICT is accounting for only around 10 % of the investments . 3.But even if there have been huge breakthroughs in AI during the last 10 years, why do the breakthroughs not show up in the productivity growth in Finland. That is the AI paradox. And the same paradox exists in other European countries and in USA. Waving….   4.One probable reason is the time lag between the technological innovations and their deployment in companies. New technologies deliver productivity gains only after a period of investment in new skills.  
  6. 1. We got supporting evidence for this time lag with my colleague Asta Bäck, when we studied ICT job advertisements in Finland 2013 – 2019. We could see that Finnish companies recruited AI skilled persons to fairly modest degree. Instead, the companies mostly looked for traditional ICT skills, like ICT infra and databases. 2. That was my first point I wanted to make for you today – that productivity lags despite the AI breakthroughs. SMILE 3.The question now is, with the current hype around Generative AI like GPT, can we expect the situation to change and the AI productivity paradox to finally be resolved.
  7. 1. So what can Generative AI do? 2. Generative AI, which is based on Foundation models ( fairly new term), refer to AI systems capable of generating content such as text, software code, images, video and audio, even DNA codes like in my example later. And this with a quality close to what humans can produce. § 3. How many of you have used some or several of these Generative AI programs like Chat GPT, Bing – please raise your hands ? SMILE ?Almost all, that is what I expected.
  8. 1. There are big expectations that generative AI will increase economic growth significantly. 2. The economic benefits of generative AI come from two (finger) – firstly more efficient output production. This is linear growth, you can produce, say double as much as much, with these tools. The second, even more important source, is acceleration of innovation through discoveries and inventions.. This grows productivity continuously and exponentially like interest on interest.   3. The main reason for the optimistic expectations is that Generative AI tools mainly support knowledge workers like we certainly are all here today. The service sector and therefore knowledge work represents over 2/3 of the Finnish economy, so the impact of generative AI will be great, especially as the systems can be very rapidly rolled out through Internet., like chat GPT, that went from zero to 100 million users in 2 months. The impact will be big, even if the system makes mistakes as long as the average performance is OK and is easy to use. 4. E.g., writing text , like press releases have been shown to be even twice as fast. Some years ago we at VTT and Univ. of Helsinki designed the bot Valtteri that wrote 2 million election news in Finnish, Swedish and English in some minutes. Also software code has been found to be even twice or more faster by using GPT based tools like Tommi Korikivi recently told on TV. However, it might be that testing the code can be more extensive. In fact, undetected programming errors can be fatal and therefore LLM use in programming should be considered with caution especially in critical applications (we all know Sarastia troubles in the city Helsinki).
  9. To further underline my point on Generative AI boosting productivity in research , let me give you an example of how Generative AI can be used in Research to radically speed up the creation of new materials. While being at VTT, I initiated and lead this project together with a team of VTT scientists from AI, material science and synthetic biology. Our AI empowered Material Scientist project aimed at designing new and superior biomaterials – and fast - in minutes instead of years!  
  10. 1. However, it takes more and more time – currently a year or more – and massive costs to discover and develop new materials and in many cases the breakthroughs are beyond reach of scientists using traditional manual methods. 2. Therefore, are needed to help material scnew methods ientists speed up the development.
  11. 1. SMILE We chose to combine two exponentially developing technologies – Synthetic Biology and Artificial Intelligence – to radically accelerate the design of new biomaterials
  12. Here is the key idea t for today. As said, the current design and production of biomaterials is very much trial-and-error and therefore slow and cumbersome.  3. The scientist does not really know which molecules make good materials. 4. She comes up with experimental DNA sequences  very much by guessing, then purchases the genes, synthetizes them in the cells and ends up with biomaterials, that often prove to be inferior and then all is repeated again and again
  13. 1. In AIMS, we radically speeded this up by putting the computer to do the repetitive trial and error guessing work. 2. The AI software generates a huge number of digital DNA candidates and predicts with our ML scheme which properties the biomaterials will have, when synthetized from that DNA. 3.We then select a few  DNA sequences with the best properties as finalists and send only those to the Synbio lab with good chances of success even at the first try.
  14. .
  15. 1. We put our AI system to create variations of well know flexible protein materials leading to totally un-seen proteins. 2.To avoid the  combinatorial explosion, we start with a couple of seed proteins, around which our Generative AI algorithm (a bit variational autoencoder) creates a huge number of resembling, yet sufficiently different, variants of the underlying seed DNA codes. We then predict the properties of the corresponding proteins, and select the best ones to bring to the lab
  16. 1. Our AI system came up with a little less than 2000 candidates that it filtered down to 144. 2.After this, our biology expert stepped in and handpicked the most diverse set of 27 genes to be purchased and then synthetized by the microbes in the Lab. 3.After extensive lab measurements of these 27, we ended up with 10 very promising protein materials
  17. 1.In the spectroscopic measurements with cycling temperatures, the AI-designed Elastine-like proteins showed superior stability compared to the manually designed ones. This is a very important property in many applications like dental implants that Pezhamn 2. This was an example of how generative AI can help us to discover new materials and medicines, like antibiotics, much faster and better than ever before, thus raising the productivity of work.The principle is applicable to many kinds of experimental research with many iteration of trial and error. . 4. But now some words about the downside, the risks of Generative AI.
  18. 1. Generative AI will to lead to significant changes in job content and even job losses in some cases 2. It is also clear that this time, this transformation will impact also high-paying occupations and bring the risk of drops in income. 3.One central factor is whether the AI system compliments or replaces labour. Or if it creates entirely new tasks. Currently, it seems to more compliment , we saw in the previous example that we need the scientist, he cannot be repalcesd. The same message is in this graph, where radiologist job advertisements from US. 4.The current wave of generative AI will not be the last, new techs will come. Here it is essential to develop skills that are complementary to AI services - could they be creativity, emotional intelligence and critical thinking.   5.Over the long run, most economists expect the labour demand, supply and wage levels to adjust granting close to full employment. Historically, Unemployment has remained low in western societies after WW2, despite the constant destruction of jobs through automation. 6. Nevertheless, Job losses are a big concern, and the society need to pay close attention so that all benefit from the development. How will this be done?  
  19. 2. It is commonly agreed that we need ethical AI, ethical guidelines and discussions around issues like: How do we ensure that AI systems safe, trustworthy, accountable and transparent for users? How do we avoid bias in AI models that can discriminate against certain groups of people? How do we protect users’ privacy and their data?. Essential How to reduce the environmental risks of AI systems, such as energy consumption and e-waste Owerall How to promote human values and rights in AI design and deployment, e.g. relating to disinformation
  20. 1. In addition to guidelines and discussion, we need regulation like th AI Act, which EU is preparing and to first law on AI ever. 2. AI systems are in the AI Act classified into 3-level risk pyramid. The uppermost level covers systems that create an unacceptable risk that are banned/strictly forbidden. These are applications that 1. Manipulate people, 2. exploit people’s vulnerabilities (young, the elderly, or persons with disabilities), or 3. Score people , like examples from China. 2. The middle level, high-risk applications covers applications causing potential harms to people´s health, safety, rights and environment, jobs, transport and infra and Biometric systems (face rec) and influencing voters and recommender systems are also included. E.g one system the I and others built at VTT and published about recently, is sensing your mental concentration from biosignals from wearables – that would have to be certified if deployed. 3. Lastly, most AI applications represents minimal risks , like spam filters and video games. These are left unregulated.   4. Especially interesting for us today, Providers of Foundation models, it is generative AI, must be a) transparent, such as disclosing which content is generated by AI. Further, b)the models must not generate illegal content and c)they must provide summaries of copyrighted material used for training (tough demand) 5. How does it work? Companies and organisations deploying AI applications are obliged to demonstrate that their systems comply with the laws before deployment and monitor the systems after deployment. 6. It can be costly for the organisation The AI Act proposes considerable penalties, if the rules are not followed. Fines can reach up to €30 million or 6% of global income..   7.M The law proposal is expected to pass the upcoming vote in the European Parliament possibly during the 12-15 June session. Thereafter, the final Act will depend on the negotiations (trilogues) between the European Parliament, Council and the Commission, which can take several months or even years.
  21. 1. I hope that I got communicated to you that firstly BIG SMILE!, productivity has lagged in Finland and elsewhere for over 10 years despite breakthroughs in AI. We could see this AI paradox in Finnish job advertisements. 2. Secondlt, Generative AI, has the potential to radically raise the productivity of work and wealth, and thus solve the AI paradox, because it accelerates all kinds of knowledge work. 3. In research, e.g., new materials and medicines can be developed in radically shorter time with generative AI. 4. More, Generative AI will make tasks and jobs obsolete, but new tasks and occupations are expected to compensate the losses. 5. And finally, that AI has to be regulated to secure that it benefits the whole society, and the EU Act is a good tool for this.
  22. Thank you for your attention, and I'm happy to take any questions you might have