Artificial intelligence “can be of immense importance in detecting things that are almost impossible to detect by manual work,” cybersecurity consultant Amit Meltzer said last week at a conference on AI in Tel Aviv.
The event, “AI for Human Language,” was organized by Basis Technology, a software company that provides AI solutions for the understanding of multilingual and unstructured texts, and hosted 100 to 150 attendees.
The conference focused on how technological innovations brought by AI have transformed natural language understanding (NLU), a branch of artificial intelligence dealing with machine reading comprehension and understanding data when it is in the form of text or speech.
One panel, moderated by CEO Amit Bohensky of Israeli startup Zoomd, dealt with how AI techniques of natural language processing are and will increasingly be of importance to government intelligence agencies.
Using such technologies will enable intelligence agencies to uncover data patterns and understand them, in order to better perform preventive and operative tasks.
Bob Flores, the CEO of Applicology Inc., who is also a former chief technology officer of CIA, said in the panel that in today’s world “everything revolves around data.” Gaining access to as much data as possible, and extracting strategic information, helps intelligence.
Flores explained that in the past, security analysts could only concentrate on certain sets of data. If they were interested finding out how much wheat the Russians were going to be able to sell, for example, they could only rely on wheat and agriculture databases, because they lacked the computing power and the memory capacity to process greater quantities of data.
“Today we try not to neglect any set of data,” he said, adding that AI techniques, for security as well as all other aims, are used to understand the meaning and correlations between disparate datasets.
But how do AI and natural language processing concretely help in the hunt for cybercriminals?
Meltzer, who worked in the past as CTO in the Israeli Prime Minister’s Office, explained that AI text analysis “improves our ability to monitor” and identify criminal activity. He added that AI’s use of mathematical supervised methods can help trace “indications of hidden activity patterns” that human intelligence analysts would not be able to extract from a huge volume of data.
He described to the audience a scenario in which an intelligence agency is monitoring the conversations in a foreign language of a criminal on a so-called “onion site” – an unindexed website used by people who want to stay anonymous and difficult to trace. Text analysis allows analysts to determine if the criminal is a native speaker of the language detected and also to glean details about the environment the lawbreaker comes from, Meltzer said. In the example he brought, the language was Russian and technology identified the person as coming from the former USSR.
“I want to know as much as I can about his identity attributes,” Meltzer said, because these attributes are what help intelligence agents when they want to match a virtual identity, or an avatar, with a real person, he said.
Nonetheless, asked about the way in which criminals themselves could or are using the same technology to their advantage, both Flores and Meltzer admitted that the risks of that are not negligible.
“Bad people have the same access to technology as good ones,” Flores said, adding that AI can be used to predict the best time to push malware into an enterprise system or to rob a bank.
So, how can AI provide a reliable form of security against such threats?
The two experts underlined that some AI solutions are better than others, and most times customers need to be wary of generic and ready-made solutions.
Flores said this technology needs to go through a sort of customization because not each AI solution fits every company. He also said that it takes time and money to do so and that many firms are not willing to undergo this process.
Flores, while reluctant to predict any future AI innovation, did say that it would be interesting to see a solution that would determine whether a person is lying, in real time.
According to Meltzer, textual communication and cybersecurity will increasingly use AI as a component, adding that developments will help uncover “more meaningful and more complex patterns that hide in today’s existing data.”
Staffan Truvé, the CEO of Recorded Future, a US internet company that collects and analyzes information from the open and dark web to detect and predict cyber threats in real time, said AI is starting to be used to do things that “are harder or impossible for humans,” like analyzing data when the volume and complexity are beyond the capability of humans without the support of deep learning machines.
“AI for the foreseeable future is not substituting for people, but helping them,” he said.