UN panel warns that AI systems could “break down” into unpredictable behavior

• The UN panel warns about the soundness of the time for ensuring that the problem of the end-to-end functioning of AI systems is properly addressed.
• It also highlights that AI systems can degrade in diversity, with unpredictability and the possibility of harmful effects, as well as the risk of data drift.
• It warns that, at the point of deployment, the AI system may behave differently from what was expected, which can lead to dangerous outcomes.

During a meeting, the UN panel stated in its report that the development of AI can have unintended consequences, which can affect the reliability of the global language model.

It also notes a key problem: AI systems can be “controlled” by training on large amounts of data, but China has a different approach, with a 15% share of the global market. The report says that by 90% of the time, these systems are used for training and testing on local data, which can lead to significant differences in their performance.

The panel argues that the deployment of AI systems in Ukraine can be particularly risky. In particular, it points out that the infrastructure needed for such systems is not yet fully developed, and that their practical control and monitoring could become a challenge for local authorities.

It also emphasizes that the risk of AI systems changing over time is significant. The report notes that the main issue is “data poisoning,” where malicious inputs can be inserted into the training process, leading to a deterioration in the system’s performance.

To reduce these risks, the panel recommends that governments require strict compliance with standards and ensure that the systems are regularly audited and updated. It also calls for greater transparency in the use of AI systems, particularly in critical sectors.

The panel also warns that AI systems can be trained to produce harmful outputs, and that their behavior can become unpredictable. It stresses that the most important thing is to ensure that AI systems are used responsibly and that their risks are properly managed.

In conclusion, the panel says that the main challenge is to ensure the safe and reliable operation of AI systems, particularly in the context of Ukraine’s ongoing digital transformation.

Tags: Technology/Research/AI