AI fever has been reaching unprecedented heights in the business world for several months, but will it last? Data complexity and ethical issues could easily lead to a new “winter” in the AI ​​sector.

The building Dangereux products Once a hazardous materials warehouse, it now serves as a small, cozy event location on the Tour & Taxis premises. At the Dynatrace BeLux Forum the room is filled to capacity: anyone who doesn’t make it to a meeting on time has no guarantee of being able to claim a place. “I couldn’t imagine a more appropriate place to talk about AI,” opens keynote speaker Geertrui Mieke De Ketelaere (Vlerick Business School). “You hear a lot of good things about AI, but also stories about it being a dangerous technology. Who or what should you believe?”

(Generative) AI will also be an inevitable topic of discussion during the Dynatrace meeting. The development of AI has been going through ups and downs for decades. Today we have undoubtedly reached a high point. “In the past year, AI has experienced explosive growth. Thanks to ChatGPT, I no longer have to explain to people what I actually do,” jokes De Ketelaere, who has more than 25 years of experience with the technology.

For all the enthusiasm and optimism surrounding AI, De Ketelaere also emphasizes the need for realism. “The development of AI has already had two ‘winters’ in the past in which enthusiasm waned. These previous dips were due to a lack of data and/or computing power. There is enough today, but an ethical debate could bring about a new winter.”

Complexity is an advantage

De Ketelaere’s explanation provides much fodder for further discussion. We meet Florian Ortner, Chief Product Officer at Dynatrace, during the event at Gare Maritime at Tour & Taxis. Although Dynatrace’s headquarters are in Boston, the company also has strong roots in Europe. “We left 95 percent of our research and development in Europe because we believe there is a lot of IT talent here,” says Ortner.

Transparency and overview are more important than ever for companies today. Last week, Dynatrace released the results of a survey of a thousand companies worldwide. One key word emerged clearly: complexity. Companies can no longer see the forest of data trees and solutions to manage that data, often spread across multiple cloud environments. “Customers process petabytes of data through our platform every day,” says Ortner.

He looks for the deeper cause. “This complexity is often a result of the desire to be both fast and successful. You can roll out or implement something faster than your competitors, but do you automatically have the best solution? The battle between speed and success takes place on many fronts.”

AI is never 100% right or wrong, but it helps people find answers in data sets too large to spend a lifetime looking at.

Florian Ortner, Chief Product Officer Dynatrace

3 + 1 lessons for AI

We’ve heard at many tech conferences that AI will solve every problem imaginable, but Ortner and Dynatrace don’t want to fall into that cliché. The research also found that while many companies have an AI strategy in place, they continue to face the same issues. “Companies don’t always understand what AI is,” explains Ortner. “You can’t expect AI to give you all the answers. AI isn’t right or wrong: it just explains the probability that something can happen. This is never 100% accurate and that is not the idea at all. AI is designed to make you more productive.”

Ortner shares his three lessons for building AI in your business. Lesson 1: Be efficient. “Measuring is one thing, but you also have to add value with the data you have, and do it as cost-effectively as possible. There is no point in solving a problem that costs a hundred euros with a solution that costs a thousand euros. You won’t buy a shovel if you don’t have a garden.”

Lesson 2: Look at your data. “The data includes both raw data and aggregated data. Additionally, keep track of the raw data and context. Queries can be very complicated, I wouldn’t know how to start them myself. With our Davis AI engine, we can make it more accessible and usable to gain deeper insights from data.” Ortner’s third and final lesson: “Always stay a little playful.”

De Ketelaere has now joined us. She adds a fourth lesson. “It doesn’t start with data, but with setting clear goals. Many companies view data as a value in itself, but it represents a cost that processes and policies must be aligned with. A “master plan” is created that spans several years, but the value of the data is variable. Start with baseline and run tests. If it doesn’t work, see what data you can use to gain more concrete insights. This process requires costs, but once the wheel is in motion, those costs are ultimately converted into value.”

Many companies view data as a value in itself, but it is a cost that you need to convert into value.

Geertrui Mieke De Ketelaere, Associate Professor of Artificial Intelligence at Vlerick Business School

The three P’s

Three keywords are central to any technological development, each beginning with a p in the English language: Profit, People, Planet. De Ketelaere illustrates this in her lecture using an object that we use every day: the car. “The invention of the car has ensured that we can move around much more efficiently: that is benefit. But the first cars were anything but safe, so innovations were gradually made to increase people’s safety (People), such as the belt. In the next phase, we have also started to look at the impact of cars on the planet.”

AI will have to follow a similar path, but for now we are still in the winning phase. “The impact on people must be at the forefront of the discussion. This often only happens once a technology is in use. With AI, it’s not that easy because not everyone understands the technology equally well. Autonomous systems are not dangerous in themselves, but the way we use them is. The context in which the systems are trained may not always match the context in which we use them. So there is a lot of power with those who train AI systems.”

In order to prevent the potentially dangerous use of AI, the European Union created the AI ​​Act after years of debate. De Ketelaere certainly has reservations about the draft law. “The AI ​​Act places great emphasis on human rights and I think that is absolutely a good thing. But I still feel that there are some problems. First, there has been an attempt to combine all types of AI into one proposal, with the United States making greater distinctions between different types of algorithms. “In addition, the European Union had to make some concessions regarding the open source models and the size of the models because some member states protested.”

“But what if open source models fall into the wrong hands? Who will carry out the checks on this?” Mieke De Ketelaere asks loudly. “Another problem is limited budgets. The AI ​​Act stipulates that member states must make their own resources available to carry out controls. Because of all these factors, I fear that the AI ​​Act could suffer the same fate as the GDPR, namely that the big players will find a way to shift their responsibility to smaller players. Looking for mistakes afterwards becomes far too complex: I’m afraid Europe has caught up.”

AI itself is not dangerous, but the way we apply it can be. The context in which the systems are trained may not always match the context in which we use them.

Geertrui Mieke De Ketelaere, Associate Professor of Artificial Intelligence at Vlerick Business School

Decentralized AI as a solution?

The third “P”, our planet, is also not yet at the forefront of AI development for De Ketelaere. “Current AI solutions are energy-intensive. Any request you make to ChatGPT must be sent to a data center, processed there and returned to you. Users are not sufficiently aware that this costs energy. This cannot be solved easily either, because new business models have to be found that are not based on computing power.”

De Ketelaere is a firm believer in decentralized AI. This type of AI does not run on LLMs with trillions of parameters located in power-hungry data centers, but rather works on smaller models that run locally. According to De Ketelaere, this offers benefits in terms of all three Ps. Decentralized models require less energy, but also offer privacy benefits as you do not have to share data with large tech giants. Because the models remain local, there are also fewer connectivity and latency issues. “I see this as the future of AI for numerous use cases,” she concludes.

So it’s still summer in the world of artificial intelligence and the companies in the industry are also convinced that the sun will shine forever for them. But after every summer comes another fall and winter. If the tech sector doesn’t find an answer to increasing data complexity and ethical questions, winter could approach sooner than expected.