Generative AI has created a lot of hype, but artificial intelligence has been around longer than ChatGPT. In fact, classic AI algorithms are already doing a lot of useful work in mature implementations. Generative AI has enormous potential, but one technology is not necessarily better than another.

“AI is not new; It was first talked about in the 1950s,” notes Gregory Verlinden, Associate VP Analytics & AI at Cognizant. He sits at the table with four other specialists during a conversation organized by ITdaily. The topic is AI and the hype around generative AI has helped define this approach, but everyone at the table agrees that AI is more than GPTs.

Acceleration tsunami

“There are moments of acceleration where technology is an important factor. We are currently experiencing such a wave of acceleration around generative AI,” says Verlinden. Véronique Van Vlasselaer, Analytics & AI Lead South West & East Europe at SAS, also recognizes the importance of the current hype. “Our grandma now also knows what AI is. Generative AI makes the technology more tangible and opens up opportunities for everyone, including SMEs and local retailers. AI is no longer just for companies.”

The iPhone had a hundred million users after sixteen years. ChatGPT only took three months.

Frank Callewaert, CTO BeLux & EU Institutions Microsoft

Frank Callewaert, CTO BeLux & EU Institutions at Microsoft, illustrates the effects of generative AI with some numbers: “After 16 years, the iPhone had a hundred million users. Facebook reached this number after 4.5 years. ChatGPT only took three months.” Generative AI is a lot of hype, but it is built on a solid foundation of interest and tangible practical use.

Generative does not mean omnipotent

Koen De Maere, board member and director of marketing and communications, government relations and advocacy at Isaca, also sees a big impact from GenAI, but adds nuance. “Generative AI is particularly useful in departments with a lot of unstructured data. Classic AI focuses more on structured data. It is important to choose the right model for the right situation. Traditional models deliver better results than generative AI in certain contexts.”

Traditional models deliver better results than generative AI in certain contexts.

Koen De Maere, Board Member and Director of Marketing and Communications, Government Relations and Advocacy Isaca

There are other considerations too, such as budget, notes Antoine Smets, co-founder and CEO of Klassif.ai. “The license costs of SaaS solutions are constantly increasing. The development of these costs is clear. You should primarily use generative AI where it offers concrete added value so that you… Return on investment can stay in balance.”

What’s more, generative AI can make a surprising difference in the truest sense of the word. “Generative AI can do many things that it has neither been trained nor tested for,” notes Van Vlasselaer. “For example, ChatGPT was not originally trained to generate code, but it quickly became apparent that it was very powerful for this purpose. Users are often not explicitly prevented from using genAI for such unplanned applications. We don’t want to and aren’t allowed to work like that in this industry.”

Generative AI can do many things that it has not been trained or tested to do.

Véronique Van Vlasselaer, Analytics & AI Lead South West & East Europe SAS

1+1=3

“Most use cases today require a combination of traditional AI and generative AI,” says Verlinden. “The combination of both will achieve the best possible results.” Like the other participants, he sees a future in which various classic AI models are combined with one or more Large Language Models (LLMs), with specializations in tasks performed by generative AI benefit. “Even with generative AI, we will not work with one model, but with different models that compete with each other for the best answer.”

Most use cases today require a combination of traditional AI and generative AI.

Gregory Verlinden, Associate VP Analytics & AI Cognizant

Implementations of AI are not in the future. The general public has only just become aware of AI, but large organizations have been working on it long before ChatGPT. “Large companies such as banks, automobile manufacturers and pharmaceutical companies are already using classic AI,” says Verlinden. “This technology is really entrenched, but there is still massive experimentation with generative AI. Few actually put it into production, and when they do, it is on a small scale and the responsibility lies with the user.”

Strong together

De Maere considers this situation to be entirely logical. “If classic AI systems are configured correctly, they can make more objective decisions than humans. Generative AI will help you, but you still need to verify the results. This takes a lot of time. You should avoid using such AI in situations where the output is difficult to control.”

“It’s not for nothing that we talk about a co-pilot,” adds Callewaert. “The user remains the pilot and bears responsibility.” Verlinden says that generative AI primarily enables the expansion of human tasks. “Conversations between machines in which systems make decisions currently only take place in classic AI implementations.”

New solution to current problems

That’s not to say that generative AI doesn’t have an impact today. With Klassif.ai, Smets is actually introducing generative AI solutions in the production of large companies. “We decide on a controlled rollout in response to an urgent problem.” He points, among other things, to the increase in productivity in the back office, where generative AI helps employees to complete the work with the available manpower. “The demand for AI is driven by a business need where we are solving a problem that has no technological solution without generative AI.”

The demand for AI is driven by a business need where we are solving a problem that has no technological solution without generative AI.

Antoine Smets, co-founder and CEO of Klassif.ai

This approach is important, according to Van Vlasselaer. “Generative AI is an accelerator, but not the solution for everything. This impression is sometimes present. You should think carefully about whether genAI is the solution to your problem and whether it requires a more traditional AI approach. I know of cases where genAI is suggested as a solution when it is not. We still have work to do to raise awareness.”

Promising

The conversation rarely focuses on classic AI for long without one of the experts once again building a bridge to generative AI. Everyone agrees that classical AI, with its more transparent behavior and better handling of structured data, is still crucial, but the promise of genAI is too great to ignore.

There is now a strong focus on LLMs and limitations that make GenAI unsuitable for industry are quickly being removed. Callewaert: “Developers are already producing special models tailored to specific applications. We are moving from a single LLM to a combination of multiple LLMs and specially trained smaller SLMs (Small language models, ed).”

Beware of the hype

Generative AI has taken interest in AI to new levels and will continue to dominate the discussion for some time. Classic AI now operates more in the background, but often incognito. For companies considering AI, it remains important to first determine what problem they want to solve. Do you have large structured data sets of machines in a production environment? Then classic AI can probably help you optimize processes. Is your administration overloaded with valuable employees wasting time extracting data from PDFs? Then genAI may be able to offer a solution today.

The truly intelligent AI that can handle structured and unstructured data will also come, but this system will not be built with an LLM. The future belongs to multimodal systems that combine LLM, SLM models and traditional AI. “Gen AI allows you to communicate with your data in a human language, and therefore many companies are already using Gen AI applications where they chat with their own company data,” adds Callewaert. In any case, today is a good time to think about AI. But don’t get carried away by the hype: AI isn’t just generative.

This article is part of a series following the AI ​​roundtable organized by ITdaily. Read more here.