Many results of common uses of popular artificial intelligence (AI), such as Google’s Gemini or OpenAI’s ChatGPT, are like a brother-in-law at a Christmas dinner: able to talk about everything with unreliable references. 20% of the responses of these models present hallucinations, false outputs with the appearance of truth. But it is the great bet of the technological giants, mainly from the United States, who aspire to turn that family member into a helpful assistant who, in addition to speaking and creating images, stories or texts, analyzes and proposes solutions (agents). The industry in Europe, where the auditors of the EU Court of Auditors see a risk of missing the AI train, distances itself from these developments. They are betting on business applications, smaller in size, proportional to the processes that are intended to be improved and that do not amaze. They are two parallel careers with different roots and different objectives. In this sense, Siemens, in collaboration with Microsoft, has just launchedin the German city of Munich, a Copilot development specifically aimed at industrial automation.
Mihails Kozlovs, member of the Court of Auditors who has led the latest audit on AI in Europe, warns that this technology will condition the economic growth of the EU in the coming years and that ignoring it can relegate the continent: “In the AI race, there is the risk that the winner takes everything. If the EU is to achieve its goals, the Commission and Member States must join forces more effectively, accelerate the pace and unleash the EU’s potential to succeed in this important ongoing technological revolution.”
However, for Tom Hurd, researcher and creator of the Zeki organization for identifying experts and trends in AI, the situation in Europe is not dramatic but different. He agrees that it is necessary to be there because, in his opinion, there are only two options: “either produce artificial intelligence or wait in line.” But he considers Europe, especially Germany, the Netherlands, the United Kingdom and Switzerland, to be in the first group, based on its ability to attract talent. “Spain, France and Italy are beginning to change,” he points out.
The difference, according to this researcher, is in the model. He believes that the United States still leads the field of equipment, programming and conversational applications, but he observes that experts are beginning to abandon these companies (he assures that half of them can only last in these 18 months) and opt for European companies, “more diverse.” and inclusive,” as he explains, and aimed at automation in sectors such as finance, industry, defense and, above all, health.
There are several reasons for this trend. One is security; According to Hurd, many abandonments in OpenAI occur because it is perceived as “high risk.” A group of 11 former employees and workers of this company have published a letter in which they warn that the company prioritizes commercial incentives over the dangers of increasingly advanced AI systems.
Another main reason is investment in a sector such as conversational applications that does not arouse European interest. “Each version of ChatGPT costs 10 times more than the previous one. We are not competitive there,” admits Hurd. A Stanford University report puts more specific figures: “The training costs of next-generation AI models have reached unprecedented levels. For example, OpenAI’s GPT-4 spent about $78 million on this task, while Google’s Gemini Ultra cost $191 million.
Michel May, an artificial intelligence researcher at Siemens, agrees with Hurd that the European path in this field is different. “The industry has a different tradition. He does not throw first and correct later, because confidence is a priority. And not just for ethical reasons,” he says.
“AI in industry cannot afford hallucinations,” adds Norbert Gaus, vice president of research at the same company. He means that an error in the response of a conversational robot in a school assignment, a recipe or a purchase is not irreparable, while in an industrial process it can be catastrophic.
In this way, he affirms that European developments are conditioned by safety (it is one of the main concerns of the continental industry due to regulations and cultural differences, according to the manager), reliability, trust, and the efficiency of the data. While a chatbot requires a large language model (LLM) with a context window of trillions of tokens (basic unit of information that can be understood as a word, number, symbol or any other individual element that constitutes a part of the input or output data of the program), an industrial application, when the training phase is passed, does not need an accelerator or an LLM, as Gaus explains.
This strategy does not exclude the use of platforms such as ChatGPT, but with limited use. In the robotics and industrial twins laboratory of the University of Munich and Siemens at the Garching research campus it is used to facilitate commands to operate the systems. Artificial intelligence creates the necessary codes for machines to respond to requirements and saves dozens of hours of programming for each specific task.
The European line of action is similar to the AI models proposed by IBM and which use a lower range of parameters than systems from other companies. “It is the trend of the industry: to obtain the performance that is needed for the use cases and at an affordable cost,” as summarized by Darío Gil, vice president of IBM in the presentation of its annual meeting. Think in United States.
In this sense, industrial artificial intelligence is powered only by its own and reliable data, those relevant to the process that is intended to be optimized, and at a lower cost. It aims to be flexible, open, helpful and interoperable, capable of exchanging data securely and automatically, regardless of geographic, programming or organizational boundaries.
For Gaus, European developments “are already everywhere, even if we don’t see them.” They are used in design, automation, maintenance, in services and as a diagnostic tool. For Michael May, he is a “launcher of productivity that does not replace the human because he always has to be at the end.”
With these criteria, Siemens, in collaboration with Microsoft, has just launched in Germany a version of Copilot (Microsoft’s AI) specifically oriented to the industry, a path begun just over a year ago with a difficult start. “The first developments seemed good, but they were insufficient,” admits Rainer Brehm, head of automation at the German multinational.
With the objective of “supporting humans in the industrial value chain”, the process has matured after consolidating proofs of concept (implementation to verify that the theory can be developed), adaptations, verifications and operational experiments. “It is not a product, it is the transformation of the entire organization,” highlights Brehm regarding a development that, according to him, will be available in July.
This orientation of the European industry greatly differs from the more effective developments of other giants focused on personal agents and coincides with the vision of other large companies, such as IBM. “The approach I see in Open AI and Google,” he explains Robert Thomas, commercial and programming head of this multinational, “it is to very different markets, to the interaction with the consumer. “This kind of thing is not our focus, but rather enterprise use cases, digital work, which is automating repetitive tasks in an organization that then extends to things like data and AI governance,”
You can follow The USA Print in Facebook and x or sign up here to receive our weekly newsletter.
Subscribe to continue reading
Read without limits
_
