AI and Warfare
As the world has descended into a darker place, so governments still feel the need to use AI to assist them. Now Large Language Models and other AI systems are being used to help summarise intelligence reports, process satellite imagery and maps, and even suggest plans for battlefields. This is alarming for a technology that is still in its infancy - many of these AI models are only months old and they all still suffer from the same known problems: they are extremely fallible. Yes people make mistakes, but that's why we have checks so that others can confirm and corroborate before decisions are made and actions are taken. When an AI makes mistakes, it is so good at presenting its output as plausible and realistic that we may find it very hard to detect its error. Let me get this clear - there is nothing inherently evil about AI. It always does its best. But if you ask a 4 four-year-old child to determine who lives and dies, even if it does its best, there may be chaos. Trusting AI with the lives of our friends and families in making - or even helping to make - decisions like these is terrifying. The bottom line is that until we can know for sure when an AI is being accurate and when it is making sh*t up, these software tools are too dangerous to use for any safety-critical purpose. They are just not ready yet. Frankly it's only those who wish to make a profit that are pushing them so hard on us. Would you fly on a plane made six months ago that has a tendency to get a bit confused and tell pilots entirely the wrong thing? That is known to cause crashes all the time? Not me!
I recently voiced my concerns on this topic for an article. Quite a lot of my comments were printed - especially at the end of the story. I'm also contributing to other similar stories around the world.
The story was published here:
