Irish experience: Dublin Tech Summit 2024

Last week, I had the pleasure of meeting like-minded people at Dublin Tech Summit 2024, listening to influential speakers, and learning a bit about trends in AI and tech in the business community. This blog post sums up my learnings and hints at possible trends of AI and tech in the business community.

About the conference

Dublin Tech Summit is one of the largest Tech conferences, with prolific speakers like Jack McCauley (co-founder of Oculus), Jeremy King (CTO of Pinterest), and Dr. Omar Hatamleh (Chief Advisor at NASA). It started in 2017 and is visited yearly by more than 8,000 attendees, with 150+ startups presenting their revolutionary products (including us at ReasonField Lab by SoftwareMill, offering Science as Service).

Interesting presentations

Building a more Human (and Humane) Internet with AI by Jeremy King, CTO of Pinterest

Skin and hairstyle features of Pinterest (source: pinterest)

Pinterest is the platform to get inspiration, and as such, is driven by client engagement. In their quest to improve the system, they did multiple ML mini-projects, which drove their engagement up:

• starting with a skin colour detector (and the possibility of filtering based on that),
• advancing even further with hairstyle detection, finalising with body type, most probably using zero-shot classification due to the large variety of classes.

This presentation has shown Pinterest’s journey to increasing user engagement through the inclusivity of its audience. It has proved that you do not always have to follow the crowd but rather make small steps and listen to the client's needs.

What caught my eye was definitely how the presentation was well maintained, with a clean design and understandable to a less technical audience, without encapsulating everything in the word AI.

Race against the Machine: How to Ensure Compliance with Fundamental Rights in the EU AI Act by Dr Marta Lasek-Markey, PostDoc at Trinity College London

Pyramid of risks inspired by EU AI Act

This presentation described the hot topic of AI regulation (check our blog post related to the EU AI Act) with different risk levels. It also included an interesting way of reasoning which area is classified as high risk by incorporating the motto, “If something is affecting the fundamental rights, it is more controlled.”. Apparently, various countries (i.e., the Netherlands or Canada) have frameworks for assessing risks posed by AI algorithms, which give authorities better hints on how to regulate new inventions like GenAI. Proper regulations could also oversee certain underprivileged groups (like children, senior citizens, or physically disabled) as they are more vulnerable to recent technological advancements. For more information on the EU AI Act, you can explore artificialintelligenceact.eu.

The Power of the Niche: Building Europe’s First Sovereign Legal AI by Prof. Michael Huth, Cofounder of Xayn

This presentation was eye-opening as it showed how difficult it is to apply RAGs and LLMs to LegalTech. This comes from the main characteristic of LLMs, which operate on probabilities and consequently synonyms of the words being good enough. As this works in other domains, the Law domain is highly detail-oriented, and bad phrasing may result in a potential lawsuit. A good example is a comparison of the words “possession” and “ownership”. From an English perspective, these words are synonyms. From a legal point of view, the definitions of these words are different and have entirely different consequences, so a “close enough” approach will not be sufficient. They pointed to the current AI race between the countries, with the USA and China leading the AI race, with France (i.e., Hugging Face, Mistral) and Germany (DeepL or Aleph Alpha) worth mentioning. Regarding money, Europe is lagging in this race, mainly because of: less capital, more regulations and more considerable diversity (in languages, culture and regulations).

Unleashing the Power of Collaborative Intelligence: Harnessing the Power of Multi-Agent Systems by Paul Walsh, AI Innovation Director at Accenture

Example of multi-agent systems, from AutoGen, Microsoft framework to LLM agents

This presentation was close to my latest interest in LLM agents (check the blog post related to the topic), so I decided to join it. It was a gentle introduction to the area of why agentic workflows are superior to simple RAGs (improved factuality), including interesting findings like electing a leader increases the quality (related paper) or interesting new directions of the area: self-evolution (agent learns as it operated) and dynamic generation (new agents are created on the fly).

Conclusions

During the conference, it was possible to grasp what is currently the most popular theme among startups (via their stands) but also larger companies and their presentations (i.e., Pinterest or Accenture). Definitely the main themes among them all are Retrieval Augmented Generations, Language Agents and multimodal services (generating pictures, analyzing charts or creating related audio recordings, all in one tool). All these areas are the main focus of our R&D department, so we are glad that we are on top of things and can forefront the latest developments in the AI space.

Special thanks

Thanks to support from PAIH, we, as a Reasonfield Lab by SoftwareMill, managed to join the conference and even meet Arkady Rzegocki, the Polish ambassador for Ireland, over a relaxed meeting in an Irish pub.