The spying and cyberattack wars of the future may all be fought by computers battling other computers, says a report commissioned by GCHQ, so for that reason alone our tech battlefield needs to invest in developing an AI that can better spot the work of rogue state AIs.
That is about the best we can parse the information in the report by RUSI at least, which says an Official UK AI is needed for national security purposes. Apparently even today's narrow ability AIs are better at scanning datasets than humans, plus it may save money by enabling more automation of admin. An AI may also "proactively identify" incoming cyber attacks by keeping an eye on the nation's network traffic in real time, sort through bulk-gathered data and help transcribe audio/video content.
Which all sounds good, until you get to the part that warns: "None of the AI use cases identified in the research could replace human judgement," and adds: "Systems that attempt to 'predict' human behaviour at the individual level are likely to be of limited value for threat assessment purposes."
The need for an AI is limited to being used to "counter AI-enabled threats" including recognising deepfake material posted by our enemies, while also potentially reducing the degree of personal intrusion in data monitoring. Because, if it's an AI monitoring us, we apparently won't or perhaps shouldn't mind as much? [RUSI via BBC]