ESET CTO: AI Can Work With Correct Human Intervention

AI and machine learning technologies need training and human intervention to work as expected.

Speaking as part of ESET’s Virtual World event, CTO Juraj Malcho said there are perceptions that AI is evil or mysterious, but “it is not magic, not self-aware and it is invented and programmed by humans; it doesn’t have any obscure intentions” and it relies on inputs.

“There are some companies out there that claim they have magic solutions, but that is not the case,” he said. “I like to say it is advanced computer assisted automation.” He admitted whilst that made it sound uninteresting, “it is a beautiful thing if you look under the hood,” as we didn’t have computers and technology to utilize machine learning capabilities for many decades, but we have other ways to apply them now.

Citing an how automation can be used in malware detection, Malcho said unique clusters of malware samples are often classified by common traits. In one example, he referred to a case where 7.7 million Emotet attacks had been detected by ESET, and as the company was able to classify using machine learning of a single DNA detection, three million attacks were discovered “thanks to us seeing common traits of a family.”

Malcho admitted that machine learning is not accurate but it is a fast way to detect, and “accuracy is best when you have a human involved and work hand in hand.” However, machine learning also comes with challenges, he added, such as when you feed it with data “you may find you don’t have the capacity of your computing systems to process all of the data.”

This requires a hybrid approach, where you pre-select the samples and train your models. “The trick here is to have it balanced, as if the model is imbalanced and not representing the real world properly, you are basically getting junk in and the result is junk out,” he said.

Consideration also has to be made regarding the malicious use of automation too, he continued, and malicious usage can include generating and distributing spam and phishing, and Malcho said automation is also commonly used in language translation.

He also claimed that attackers can detect intruders in their infrastructure, identify patterns in generated content, create false flags and choose the best target and attack methods.

Concluding, Malcho said that AI “is far from 'Skynet' and trying to control us, it is just a tool that we have at our disposal” and it depends on how well you are able to use it.

He said: “AI without data is just beautiful math, and data without AI is basically just a bunch of ones and zeros. One doesn’t exist without the other. So when the perfect combination of these elements is achieved and properly validated data is fed into the properly-designed systems, a euphoric moment is created.”

What’s Hot on Infosecurity Magazine?