- Michael Wooldridge argues that the AI risk can be driven by the so-called black box of AI in the UK in which it might trigger a catastrophe in the UK.
- The material also points out that it is not possible to determine the exact cause of the risk, and that there is no certainty.
- Forbes discusses the problem of hallucinations in “black box” AI, as well as AI interpretability for more accurate model reasoning.
According to the expert, the AI risk can be caused by the so-called black box of AI in the UK, which may lead to a catastrophe in the UK. As stated by Michael Wooldridge, a professor at Oxford University, the material uses The Guardian.
Wooldridge points to the unpredictability of complex systems, which can result in new AI instruments for the UK to go beyond what they were designed to do. He also notes that the more complex the systems become, the harder it is to predict the outcomes in advance.
In an interview with The Guardian, Wooldridge said that “classical scenarios for technology” include hallucinations that “momentarily make the UK” and “turn right into wrong,” which companies are trying to address with new instruments. He also notes that the UK’s ability to track and interpret systems can be affected by the “jagged” nature of the system’s behavior: it may be possible for them to have unexpected effects in certain situations, and at the same time the results may be unpredictable and hard to verify.
According to The Guardian, the AI that powers large language models can generate misleading information based on unclear wording. However, as Wooldridge explains, such systems may have “teeth” (jagged) capabilities: they can have unpredictable effects in certain situations, while at the same time being difficult to verify and explain.
Forbes also argues that the new systems are not necessarily able to explain why their hallucinations occur. Forbes also notes that OpenAI’s internal research shows the problem of hallucinations.
Forbes also describes the “black box” problem: hallucinations can occur within AI systems even when they are not supposed to, and it is difficult to explain why. That is why the material emphasizes that AI interpretability is crucial for understanding how models work and for preventing errors.
Forbes adds that the reason AI in critical systems can fail is due to the inability to implement guardrails that “what’s not that” and “what’s not that” can override the system’s concrete project.