Corporate data remains a challenge for Artificial General Intelligence (AGI)
While research continues toward developing Artificial General Intelligence (AGI) – a system capable of handling any intellectual task at least as well as humans can – AI still faces significant barriers when it comes to being integrated into companies.
Many organizations hope that the use of AI will enable them to optimize their internal processes. However, corporate knowledge consists not only of documented information, but also of decades of know-how that often only exists in the minds of experienced employees. Even the most intelligent AI quickly reaches its limits here: the most powerful AI techniques, too, cannot automatically understand contradictory data and chaotic processes or categorize implicit knowledge.
Another major challenge for AI implementation in companies is data protection and regulatory requirements. AI models require large amounts of training data, but sensitive information such as personal data or company secrets cannot simply be fed into neural networks. Once learned, information cannot easily be “forgotten”, leading to potential conflicts with data protection regulations such as the GDPR or the EU AI Act.
Even if AI can analyze enterprise data, the question remains: how do organizations control what information is accessible to which users? Early implementations, such as Microsoft Copilot, demonstrated that inadequately protected data can lead to unwanted disclosure. In some cases, sensitive information, such as employee salary data, was inadvertently disclosed as a result of AI accessing unprotected Excel spreadsheets.
The solution lies in careful and modularized planning of AI implementation. Companies need to create clear structures to ensure that their data is consistent and protected. Rather than a sudden revolution, the integration of artificial intelligence into operational processes is more likely to be a gradual development. The success of this transition will be measured not by how quickly companies can deploy AI systems, but by how effectively they can integrate them into their existing operations while maintaining reliability and trust.
Efficient new AI models will reduce energy consumption
Another critical issue in AI productivity is the immense energy consumption required by modern AI models. Large AI models, such as those developed by leading tech firms, demand enormous computational power, significantly increasing energy consumption. This has led to a growing push for efficient AI models that require fewer resources.
New AI models such as Deepseek, developed in China, show that powerful language models can be trained with a fraction of the previous resources, but can still compete with ChatGPT or Claude in many tasks. This raises the question of whether the need for huge, energy-intensive AI systems could become obsolete in the future.
In fact, it seems that language models based on today’s digitally available knowledge can actually be created more cheaply and efficiently than originally assumed. However, new AI models are currently being trained that not only learn existing knowledge but are also designed to develop new insights and solve unsolved problems, such as in mathematics and physics. Similar to how a scientist spends years developing a new theory, such AI models require large amounts of time, resources and energy.
A promising solution could be to use advanced AI models to train smaller, more energy-efficient AI models. These more compact models could then be used for specific applications without the massive energy consumption of the current leading language models. The success of such low-cost AI models shows that it is not too late for Switzerland to successfully position itself in the race for intelligent AI models and that the barrier to entry for new developments is lower than originally assumed.
While it is not always easy to keep pace with the rapid developments in the field of AI, it makes sense, at least today, to understand AI as a tool that supports us humans in our creativity and decision-making, without trying to give up our authenticity and responsibility to it.
