Anything but easy: in AI, preparation is key. Takeaways from the Beltug N-sight: 30 November 2021

Artificial Intelligence (AI) empowers organisations in many ways, especially when it comes to unleashing the potential of all the data a company owns. With the help of AI, processes can be automated, smart decisions made, repetitive tasks managed, and complex problems solved in little time.


But while AI promises to make certain processes easy, the success of the technology and of your AI project is anything but! Proper preparation is key:  from assessing the readiness of your IT infrastructure, through defining the right goals and realistic expectations, to making sure the right skill sets are available, and ensuring collaboration from the start with all company stakeholders.


We had a look at good practices for working with AI, whether for (big) data analytics or other purposes. In this session, we started with guidance on creating and implementing machine learning workloads. Then we dove into the legal framework of AI, including the upcoming AI regulation. Finally, we heard a real-life story of using AI to turn the wealth of data held by the notarial profession into insight for data-driven tools – while complying with security and privacy requirements.


The presentations and the recording from the event are available to Beltug members below (after login).


A ‘well-architected’ framework helps you understand the benefits and risks of your decisions when building workloads for machine learning (ML), started Nicolas Metallo from AWS (see slide 4). There are 9 key design principles for machine learning, he continued: from assigning ownership, providing protection and ensuring resilience, to reducing cost and enabling continuous improvement.


Machine learning is very much about a cyclic, iterative process with guidance for developing the ML workloads. Nicolas illustrated that process and lifecycle, as well as the importance of the various components and different feedback loops in your learning process (see slides 9-11).


A real-life use case brought the theory to life, in a music recommendation module (from minute 25 of the recording).


Elon Musk is known for pushing back against regulation, but not in the case of Artificial Intelligence (AI). Peter Van Dyck, from Allen & Overy, shared a quote by Musk illustrating why AI regulation is needed: "AI is the rare case where I think we need to be proactive in regulation instead of reactive. Because I think by the time we are reactive in AI regulation, it'll be too late, [-] AI is a fundamental risk to the existence of human civilisation."


Europe is working on a proactive regulation, triggered from a safety point of view, as well as competitiveness and prosperity. But the current draft AI regulation (which is definitely not final or adopted) has a long road ahead (see slide 6). The text should be finalised by early-2024, and will only come into force 2 years after. This gives companies time to prepare for potential new elements in regulation.


There are 4 key features of the draft AI regulation. It offers a:

(from minute 45 of the recording).


Peter concluded with 4 steps to compliance with the upcoming regulation:



After theory, we turned to practice with Erik Klewais and Ellen De Munck from notary organisation Fednot. When a notary takes a decision in a file, the goal is to do so in an informed, objective way. Fednot's AI project aims to support that objective, as well as help optimise the collective intelligence available within the notarial profession.


Fednot faced several challenges that may seem familiar to those who have engaged in their own AI projects:

 (see slide 9 for the building blocks of Fednot's AI project).


Pseudonymisation was chosen over anonymisation because there is a lower risk of missing data. In reporting, pseudonymised data can also be used in statistics and analysis. Finally, the end-user has a more pleasant experience when reading a pseudonymised text.


In theory (see slide 11) this is a good plan; in practice, difficulties pop up. Human intervention can help to improve the algorithm.


To conclude, Ellen and Erik listed their key learnings so far:


Presentations and recording (after login):


Anything but easy: in AI, preparation is key

Ann Guinée, Communication Manager, Beltug (English)


Best practices when building end-to-end machine learning solutions

Nicolas Metallo, AI/ML Business Development Manager, AWS (English)


Regulating AI – the legal framework and practical recommendations

Peter Van Dyck, Partner, Allen & Overy (English)


User story: How AI helps the notarial profession to unlock sensitive data

Erik Klewais, Data & AI Manager, and Ellen De Munck, Business manager AI, Fednot (Dutch)



Dear visitor,

Access to more information about this topic and/or to download the paper is easy and fast, but exclusively for Beltug members (just login to get access).

Beltug gathers a lot of information. Here you find the advantages of Beltug membership

The Beltug Team

Click here to login