Ex-CIA, Israel government officials: How to get right AI for missions

Amit Meltzer, now a cyber security consultant, said a key new power for Israeli intelligence from AI was to detect “patterns which human analysis cannot find.

Cyber hackers [illustrative] (photo credit: REUTERS)
Cyber hackers [illustrative]
(photo credit: REUTERS)
An ex-CIA and a corresponding ex-Israeli government official said on Thursday that artificial intelligence (AI) cyber methods have empowered intelligence agencies to achieve their missions at a whole new level.
Speaking at a conference in Tel Aviv organized by Basis Technology and with limited media access, ex-CIA chief technology officer Bob Flores and ex-Israeli government agency chief technology officer Amit Meltzer discussed different aspects of the new powers their agencies had gained through AI.
Flores, now CEO of Applicology Inc., said that US intelligence agencies are now “trying to do everything, to get all of the data and analyze it all. In the old days, you had to pick which data to look at.
“You could look at wheat databases regarding Russia and wheat, and you needed to disregard everything else because there just was not the computing power to analyze everything. Today, you use it all,” he said.
Continuing, Flores explained that the next step was “to use AI and machine learning techniques to massage data and figure out what it tells you. Data in a disparate data set may be relevant to what you are doing and you may not realize it,” without the proper methodology.
MELTZER, now a cyber security consultant, said a key new power for Israeli intelligence from AI was to detect “patterns which human analysis cannot find... the ability to detect intelligence-related activities usually beyond the spectrum and capacity of an intelligence analyst.”
He also discussed using AI-based intelligence for “target-based activity... to improve the understanding of a subject – his background,” in trying to apprehend and attribute attacks to a specific cyber adversary.
Giving examples, Meltzer played out a scenario where Israeli or other Western intelligence agencies might be clandestinely monitoring a cyber adversary’s conversation online in a foreign language.
“I want to know about his background. Maybe he is talking in Russian, but is it native Russian or is he from a former USSR environment? This information could come from a different area, like from his passport,” he said.
Along with that, “sometimes you can use communications intelligence from his IP address and similar indicators or textual aspects” of what he writes, adding that: “I want to know as much as I can about identity attributes when trying to match an avatar [anonymous online name] with a real person.”
Using AI, all of these disparate points of data are much easier to triangulate and use solve the question of who the cyber adversary is, said Meltzer.
In addition, he commented that AI’s “split personality” can help intelligence agencies with both following the broad “landscape-level analysis in following popular trends,” as well as helping to delve deep into a specific issue by “prioritizing raw data.”
Flores said that AI is already being used to better determine when information technology breaches occur and to understand their nuances at a much deeper level which might otherwise require a thousand security analysts –  “and we do not have a thousand security analysts to do the job.”
DISCUSSING WORST fears of how adversaries might use AI against the CIA or the West more broadly, Flores said, “sure they are using it: this is going to happen. The bad guys will look at it and predict patterns – like when is the best time to rob a bank or to push malware into an enterprise system.”
On a similar theme, Meltzer said that adversaries could still beat systems defended solely by AI, as opposed to systems mixing AI and human-based technological defenses.
“It is a lot easier to cheat AI than humans because the algorithms are simplistic. You just need to figure out the attributes which go into the process,” pointing to beating AI facial recognition systems as an example.
“You just figure out the points and indicators which go into the AI algorithm for the unique facial pattern, and then you fool the actual process of getting the algorithm right... if we put the financial sector with AI as the first level of defense, it will be the first to go into Bitcoin because all of the money will be stolen,” he added dryly.
Offering a somewhat differing view, Flores said that he believed AI will “get to a point where it can police itself,” saying that algorithms can be written “to check whether facial recognition is being manipulated.”
Though a moment later he also dryly admitted that this would move the battle for cyber adversaries to whether they could crack the second-level algorithm which was policing the first level.
Basis Technology, which organized the conference, says that its “mission is to improve the process of extracting meaningful intelligence from unstructured multilingual text and digital devices by developing the industry’s best software.”