Why an “AI Day” at Hager Forum?

Because there’s not a day that goes by without artificial intelligence being mentioned. AI is everywhere, be it in the media or in our lunch break conversations. Who hasn’t seen “Balenciaga Pope”, an eye-catching picture of Pope Francis wearing a puffer jacket which trended world-wide back in March?

What’s interesting isn’t so much why it went viral, but that this sudden craze around ChatGPT and generative AI-based tools has given visibility and substance to the invisible. In doing so, it allows us to question AI: what it is and what is it not? How might it change our lives and businesses?

These are the questions we tried to answer with our guest speakers on stage - Marion Moliner (Data Science & Data Engineering Team Leader at Hager), Rodolphe Gelin (AI Leader Expert at Renault) and Michel Lutz (Digital Factory Head of Data at TotalEnergies) during a dedicated “Unbox your mind” conference.

There was more: workshops with local experts and a poster and demo session for all Hager employees to learn more about the technical side of AI as many of us still struggle to fully grasp how it works and what it truly entails. 👇

What we learnt in a nutshell:

1. You were already using AI fifteen years ago 🕑

AI happening right now, and it isn’t even new. Why? Because everyone has used AI at least once in their lifetime. In fact, most of us use it daily, often even without being aware of it. And we're not talking about these last months trending tools or more classical examples such as video streaming services, shopping recommendations and cashier-less supermarkets.

Have you ever used your navigation system to look for a shortcut to get to your appointment on time? This is rule-based or ‘symbolic’ AI relying on data. Have you ever done a captcha for Google? That is supervised AI. And using it, you were teaching a machine to sort and label data. A final one? It might seem natural that spams have been automatically filtered in your mailbox for more than fifteen years now, but that’s also down to AI. AI is definitely here and everywhere. Ever increasing computing power makes it omnipresent.

2. ChatGPT is a well-crafted app, but isn't revolutionary ⚙️

Mention AI today and chances are people associate ‘OpenAI’ or ‘ChatGPT’. Looking past this trend, the real game changer is that the general public can now play with AI easily whereas for years, it had been seen as a technology only accessible to a closed circle of experts. But fundamentally, ChatGPT is an evolution, not a revolution.

It is successful marketing rather than a technological breakthrough which makes it so all-pervasive. The same applies to generative AI. It looks magic on the outside but it’s not. If you look behind the curtain, you’ll see that 'Deep Blue' defeated Garry Kasparov at chess in 1997 because we trained it to do so. Since then, computers beating professional players became a tradition.

However, even if AI is based on learning, AI cannot develop by itself. It will always need brilliant engineers to get better. Without human intelligence and structured data, there is no AI. Yes, AI can reproduce an artist’s style. But it won’t replace living, breathing artists: it’s not built to create art out of nothing. There are still many technical limits and biases ahead that needs to be overcome so that AI has a real chance at being able to solve complex use cases for us. That is why the industrial use of ChatGPT for such problem-solving isn’t yet a reality.

3. AI will continue empowering us, rather than overpowering us 💪

You shouldn’t be afraid of AI. It is here to simplify our daily jobs and support us in achieving certain tasks quicker or more effectively. It’s just another tool which will amplify human intelligence rather than replace it. How? Either by increasing our potential or by allowing us to bring our added value elsewhere. This happened before with the advent of the internet: job titles and missions evolved, and it will happen again. It’s only natural. Still, never forget that AI is a co-pilot, the steering wheel remains in our hands.

We are the ones accountable for the process to succeed, not the machine. Machines’ intelligence is still narrow. If we change one parameter, it won’t work as planned or not at all. Which is why an untrained autonomous car with the best vision system won’t get you anywhere. And even then, the best trained systems won’t be 100% correct. They will prevent some accidents but generate new ones along the way.

We are more agile, able to improvise and therefore able to face unpredictable situations. AI simply cannot. So why continue competing against it in its domain of excellence? Rather, we should learn from it, take advantage of what is has to offer and stop fearing what it can do for us. At Hager, for instance, our operators at the Relay vision station can now rely on AI to check for defects on components. It was a tedious task carried out manually and required hours of training before. Human resources can now be reallocated on more important missions on the production line.

4. Implementing AI in the industry is challenging 🛠️

Factories have existed long before AI was even a thing. But the technical limitations and challenges AI teams face today aren’t the same from one industry to the other and depend on a factory’s profile. There is a huge difference between challengers and regular players on the market, as well as old versus brand new factories.

Adding AI into already existing production processes is a big challenge of our day: it means transforming a running enterprise with machines and industrial processes in place without reinventing the wheel. It’s like playing Tetris instead of Legos. This is the difference between ‘brownfield’ projects (updating existing factories) and ‘greenfield’ project (industry 4.0 by design factory).

Our brown-field approach at Hager means anticipating possible problems: how to gather data? Where to put sensors first? Which kind of storage solution needs to be implemented? What kind of computing power should we rely on to run this AI? In our Relay vision station, AI turned out to represent less than 10% of our teams' work. Being able to show that there is a default on a square part was quite easy. Implementing it on the field was another story but we’re finally crossing the final line.

5. Working in an AI department isn’t always what it looks like 🔍

Without structured data, there is no AI. This is the less visible part, but the most paramount one: data is the fundamental ingredient in any AI. ChatGPT as well as any other previous AI based on deep neural network and reinforcement learning, are only part of the job and technologies specialists rely on daily.

Because AI can be done through a lot of different methods. That is why projects relying on deep learning only represent 20 to 25% of what the end-product or project required. Most of the time, experts use traditional learning methods, maybe not even data-driven ones. Other fields are therefore always involved, may it be 'operational research' or 'optimization' for instance. However, in the end, all projects have one thing in common: the best solution usually dereives from the combination of all methods.

6. Anticipating the carbon impact of AI projects is our responsability! 🌍

By trying to build systems that generate less CO2 emissions, we gained a lot of maturity regarding the environmental impact of AI during the last few years. However, since the beginning of 2023 and with the sudden craze around ChatGPT along with the rise of many other AI-based tools, we use growing computing power each day. Another reason to be even more careful and not forget the best practices we’ve learnt before!

Which is why, when our teams begin a new AI project, the environmental aspect is part of their thought process from the very start. We still have a long way to go with many challenging projects to come, especially within a ‘brownfield’ frame, but anticipating CO2 emissions is one of our main targets. And if you want to learn more about our Sustainability approach at group level, feel free to look at our brand new ‘Sustainability report” here.