0 votes
1 view
in AI and Deep Learning by (19.1k points)

Seems that requirements on safety do not seem to like systems that use AI for safety-related requirements (particularly where large potential risks of destruction/death are involved). Can anyone suggest why? I always thought that, provided you program your logic properly, the more intelligence you put in an algorithm, the more likely this algorithm is capable of preventing a dangerous situation. Are things different in practice?

1 Answer

0 votes
by (42.4k points)

When something goes wrong with AI, and the root cause of the problem comes to light, there is often a great deal of head shaking. With the benefit of hindsight, it seems unimaginable that no one saw it coming. But if you take a poll of well-placed executives about the next AI risk likely to appear, you’re unlikely to get any sort of a consensus. 

  1. Data difficulties. Ingesting, sorting, linking, and properly using data has become increasingly difficult as the amount of unstructured data being ingested from sources such as the web, social media, mobile devices, sensors, and the Internet of Things has increased. As a result, it’s easy to fall prey to traps such as inadvertently using or revealing sensitive information hidden among anonymized data. For example, while a patient’s name might be redacted from one section of a medical record that is used by an AI system, it could be present in the doctor’s notes section of the record.

  2. Technology troubles. Technology and process issues across the entire operating landscape can negatively impact the performance of AI systems. For example, one major financial institution ran into trouble after its compliance software failed to spot trading issues because the data feeds no longer included all customer trades.

...