Accelerating the Adoption of AI – Our First Community Event

On the 11th October, we hosted our first community event to present the Alpha Hub innovation platform in our Cluj office in Romania titled “Accelerating the Adoption of AI”. This event was aimed at beginning to create a virtual community of start-ups, academics, investors and corporate partners. The specific purpose of the event was to promote discussion and knowledge around Artificial Intelligence (AI) initiatives while connecting different parts of the local community. We brought together experienced researchers and practitioners in the field, while also sharing what we do in our European division, Paddy Power Betfair.

We broke the evening into two parts. Firstly, a presentation on the Alpha Hub platform initiative by Flutter Technology to connect to the international start-up ecosystem and present the collaboration and growth opportunities we could offer. The second part was an in-depth panel discussion with specialists working with or researching AI technologies, especially Machine Learning/Deep Learning and Knowledge Representation.

Guest speakers

PPB Participants

  • Oana Iacoviță – Head of Data Delivery at Betfair Romania Development, focused on delivering Business Intelligence, Advanced Analytics, Data Platforms and Automation Implementations for Paddy Power Betfair.
  • Anca Lazăr, Sorin Circo – experienced Data Software Engineers, working on data processing, processes orchestration and productionizing ML Models.

The Panel Talk

Technical Overview

Venturing into the areas of expertise of the respective panel speakers, the discussion covered a wide range of topics in the realm of AI with a focus of some of the main technical areas, challenges and what accelerating the adoption of AI means to Europe.

Firstly Răzvan Florian explained the direction of AI research in the Romanian Institute for Science and Technology (RIST) institute. With 22 researchers working in Machine Learning (ML), with special focus on Deep Learning (DL), they are divided into two groups. The first group is automating the creation of web pages and studying curiosity-based learning (similar to how children learn exploring the world around them and discovering new learning models). The second one is studying the application of Non-Euclidian mathematics in DL to improve convergence methods. The latter can be applied to text, audio and video processing. The institute has also initiated two spin-offs in the past based on their research.

Although ML and DL are underpinning most of the practical applications of AI, Adrian balanced the view with a fresh perspective: there is not a single “party” in the AI world, but two main ones. One is based on statistics and extracting patterns out of data, including Bayesian networks, NLP, and machine learning. The star of the show is Deep Learning, something that has solved the classification problem based on CNN (Convolutional Neural Networks) and RNN (Recurrent Neural Networks). One of the disadvantages of DL he pointed out is the explainability issue and the different outputs that you can get by applying different algorithms on the same data. The other “party” is called Knowledge Representation – representing knowledge in a logical format and rationalizing based on that to solve complex problems. One of its branches, XAI (Explainable AI), is aimed at making the black box models of DL transparent. University research in AI at UTC-N covers both domains of research.

On the other hand, Răzvan confirmed that advanced DL algorithms have loads of applications in business environments and there is collaboration openness from the RIST institute. One of the main conditions of applying Dynamic Programming in a company is to have data available with good quality and in large quantity. As quality of the outputs depends of the quality and the quantity of the data that can be made available to algorithms – the more data the higher probability of accuracy. RIST offers two models of collaboration:

  • Consultancy – implementing DL models for business specific needs
  • Research contra equity – applied research directed mostly to start-ups

A concrete application of the RIST research can be found in the automation of interfaces creation that is being pioneered by the AI start-up teleportQH, headed by Paul Brie. TeleportQH was founded 2 and a half years ago, one of its objectives being the automated generation of user interfaces code. From that the next step was trying to generalize to other use cases, an approach that opened a lot of opportunities – evaluation metrics for websites, predictions of usability and website performance etc. Paul Brie has pointed out the challenge his team of software developers had in learning ML at the level of production readiness, as well as the heavy work of curating the data for the algorithms, that takes about 90% of the development time. He has also highlighted the importance of the tight collaboration between research centers and companies looking for innovative solutions. The teleportQH initiative is a clear example that ML  has opened a wide range of possible options that would not have been possible with deterministic programming.

On Challenges in AI

One of the main challenges of ML projects that Paul observed was the collaboration between software programmers and data scientists, or of people with a ruled-based mindset and people used to the word of probabilities that are more risk prone. There is an acceptability limit of the risk in each industry depending of the perception of performance for the specific use case.

The access to talent was identified as a general challenge, with PPB also having challenges in terms of finding senior machine learning engineers. Part of the solution could be found in the collaboration between companies and universities to run together projects at the doctoral level that are specified by a company which also provides part of the funding.

For researchers training, Răzvan Florian has organized a Summer School of Machine Learning in 2018. He has brought to Cluj some of the top 50 researchers in the world, and the program had over 160 students from all over the word. Teachers were from Google Deep Mind, AI Research etc.

Also, in the panel he announced for the first time publicly an initiative to help grow skills in ML: a local study group for developers that can collaborate to solve ML problems (e.g. Kaggle competitions or community projects). With significant learning resources available, he noticed that the challenge is getting yourself motivated to learn and to get over moments where you are stuck with a problem. These are times when a ML community is valuable and can help grow the AI community.

On the Future of AI

Paul showed an optimistic perspective on the future of AI, considering that it will be able to fully replace human work so that people are free to make decisions that cannot be yet made by machines.

With a lot of transformation in the AI space and various online resources available (e.g. Amazon offering on their platform the training resources used for their internal staff), a lot of young people are approaching the domain successfully . This is what Paul Brie found out while searching for talent to hire for teleportHQ.

Paul Brie and the group agreed that AI is one of the systemic and strategic technologies of the future and in Europe we are already behind US and China in the space. We need to catch up if we want to be creators, not only consumers of value. In his view: “the stake is not to make more money, it’s to exist or not in a quite near future”. In 10 years, he says, may be already late. There is a deficit of ML knowledge at European level.

On Funding and the Adoption of AI in Europe

Collaboration was a topic that emerged as a way to advance AI at a local and international (European) level. Answering a question from the audience related to availability of funding for AI start-ups, Paul said that the ecosystem and understanding in Romania and Eastern Europe is scarce. Also, he noted that the duration of ML project release is longer than traditional software development, so it needs stakeholders to understand how the technology works. Răzvan pointed out that the European Commission has an initiative to invest publicly into AI research that has had innovative results. But, as Adrian underlined, the AI budget from the EU is 2% of China’s barring each country’s own investments in AI. And, as he showed, innovations in the US are immediately implemented in China in a race where other countries lag behind.

Therefore, a collective European effort should be directed to AI, Paul says. especially to develop the competencies in people. Through Spherick he will be lobbying for that at the EU level to influence the technical strategy in the next years. By the way the automation power grows due to better algorithms and more data being available the choice is simple: participate or be a spectator of the future.

One of the things we could do better to promote the technology, pointed out by Oana Iacoviță, is for technical people to find an even better way to explain to business the value of ML to stakeholders and convince them to invest in it. Other things worth doing more for the AI ecosystems are sharing knowledge, connecting people in the AI community to drive common initiatives and support each other.

Adrian Groza (UTC-N) has given a good example where other countries than China or US do significant work in AI. For example, the Arab Emirates that have started in 2017 an AI ministry and started an education program with the guidance of Oxford University and engagement from the public and private sectors around the adoption of AI. They have created summer schools for students and further introduced from the 6th grade Python classes to teach children ML. Adrian warned that 40% of people in Romania and generally in Eastern Europe are at risk of their jobs being replaced by AI automation. On a similar note, due to the financial advantages of IT-based jobs compared to other industries, especially in Romania, a lot of young people are attracted to the AI word. As Paul Brie observed, there is in Romania a great pool of talent, and “a lot of minds to be discovered and grown”.

Conclusion

The future is promising if we understand the technology direction and if we educate young people in AI technologies, while empowering them to use them in a smart way. We would again like to thank everyone who attended and gave so much great insight into the current and future of the field and their advice on what it takes for Europe to accelerate the adoption of AI.

Our aspiration is to continue this series as talks, meet-ups and general discussion amongst our growing Alpha Hub community. If you are interest in joining, please submit your details here and we will invite you to the next event.

As always, we have an interest in speaking with companies with capabilities in any of this future mobile technology. At Alpha Hub, part of our mission is to accelerate the adoption of emerging technologies across Flutter Entertainment. If you’re working with innovative technologies that you think could support our global brands such as Paddy Power, Betfair, Sportsbet or FanDuel, we’d love to hear from you.

Comments are closed.