AI and information overload

 

It looks as if Israel's pioneering use of artificial intelligence (AI) to process, sort and apply intelligence gathered from multiple sources to produce a recommendation for action, is hitting a snag.  Strategy Page reports:

Now the IDF finds that it has often collected too much data on potential targets. This makes it difficult to find the current targets it needs to hit. Efforts to use AI to find the right targets in time to attack them have stumbled over the inability to sort out the available target data to find current targets that need to be hit. Another problem is that using AI to analyze the data and find the right targets concentrates on speed rather than accuracy. As a result the IDF is often hitting the wrong targets and doing so faster than before. Before AI, human analysts were used. This process was slower and often missed targets buried in a massive amount of target data. With human analysts you had better accuracy but the process was often so slow that the target had moved out of view by the time IDF air, artillery or ground attacks were launched.

Current criticisms of inaccurate targeting and civilian casualties are the result of Israeli target planners choosing speed over accuracy. Greater civilian casualties are also caused by the frequent use of the wrong weapon. Often a half-ton aerial bomb is used when a ground based anti-tank missile would have done the job. Another problem was that Israeli target planners had stopped keeping track of how many civilians were killed during attacks. When recent data on the actual number of civilians killed in recent Gaza attacks became known, it was an embarrassment for Israeli target planners and a tragedy for the victims.

There's more at the link.

It's the military intelligence version of the old builder's conundrum:  Fast, cheap, high quality.  Pick any two.  You can't have three.  If it's to be a fast, cheap job, it won't be high quality.  If it's to be fast and high quality, it won't be cheap.  If it's to be cheap and high quality, it can't be fast.  The same applies to actionable military intelligence (i.e. intelligence that leads to action at once or very soon, rather than being filed for future reference).  If you want a fast intelligence analysis and recommendation, it won't be as high-quality as one where you take more time to make up your mind.  If it's high-quality, it'll be slower - and your target may no longer be there when you arrive at a decision.  Effectively, Israel has been trying to use AI to speed up the process;  but AI is being overloaded with too much information, and is therefore making the same mistakes as we might expect from human analysts in that situation.  Unfortunately, it's making those mistakes faster - and people are dying faster as a result.

This is, of course, a tragedy for those civilians caught in the crossfire.  They're "collateral damage", if one uses the official (?) military term.  Trouble is, that term hides the reality of burned, maimed, wounded and killed men, women and children who were doing nothing wrong.  There have been thousands of them in Gaza already, and more will come.

There's no easy answer to this.  I can feel immense sympathy for the innocent on both sides . . . but as long as terrorists are prepared to slaughter innocent people in Israel, they are effectively bringing this down on their own innocent people's heads in Gaza, because Israel cannot - dare not - not respond to terrorism.

Peter