AI developers think smart to boost cybersecurity

Today is Data Protection Day 2016, an educational global event aimed at raising awareness of data privacy and security protection and promoting best practice. The event will see countless debates and conversations around what it means to keep both personal and business data private and secure, but one of the key discussions will centre around the potential of artificial intelligence (AI) or machine learning techniques and their application in this space.

Cybercriminals have been using automation to take down networks for years. In what’s known as a distributed denial of service (DDoS) attack, servers can be transformed into “bots” to flood recipients with unwanted traffic. This gives perpetrators the ability to cause substantial damage to a big organisation. Sony and the BBC are examples of large groups that have recently been hit.

In response, businesses and cybersecurity professionals are tightening up their defences through machine learning techniques – the advantage being that they’re able to find patterns in malware behaviour more quickly, accurately and efficiently than humans (and at scale).

Among their uses, machine learning solutions can be applied to specific enterprise applications – in the financial sector, for example, where intelligent software is used to spot anomalies that might identify insider trading. More widely, machine learning can be used to scan incoming email for malicious programmes.

Turning problems around

AI is an approach that turns problem-solving around, says Duncan Hodges, lecturer at the Centre for Cyber Security and Information Systems at Cranfield University. “Rather than creating something to solve a problem, we create something that learns how to [solve a problem].”

Outside of the security space, AI is common in systems that try to get a rich understanding of their surroundings. Driverless cars are already using machine learning techniques, while deep learning models have the ability to label pictures and films with rich, detailed descriptions about their content.

But as the technology becomes more sophisticated, there is a move towards full AI: systems that are entirely self-learning. It’s still early days, but potential use cases are emerging. Some modern systems are already surprisingly intelligent. Take, for example, the security cameras using AI to identify and detect unusual behaviour. An automated camera called AISight by BRS Labs works by monitoring feeds in real-time, alerting authorities if it spots any abnormal activity.

Moving even closer to the realms of science fiction, manufacturer Hitachi has developed crime-predicting technology. The brand claims that the system is capable of accurately predicting crime by collating data such as weather patterns, public transit movements, social media activity and gun-shot sensors.

Devices learning for themselves

The increasing power of mobile devices is also enabling AI capabilities to be added to much smaller hardware. Qualcomm said its latest Snapdragon processor, the 820, can run machine learning algorithms that could previously only be used on computers with huge processing capability.

“The machine learning market is evolving towards deep learning, which is more sophisticated,” says Gary Brotman, the company’s director of product management. “The actions and datasets are far more robust and complicated. This is where you teach devices to learn for themselves over time.”

Developed using Qualcomm’s Zeroth machine learning platform on the 820, the company’s Snapdragon Smart Protect software is an AI-based approach to identify zero-day malware (previously unknown viruses that have no available antivirus software signatures) more quickly. It does this via around 350 different trigger points.

The AI challenge

These intelligent techniques have great potential, but there are challenges.
The huge amount of data needed to accurately perform AI poses privacy issues if businesses don’t comply with regulation – specifically, the update to EU data protection laws coming in over the next two years.

There’s also the question of how intelligent a machine is allowed to be. “Whenever a computer takes over from human decision-making, you need to question what values on which they’re based,” says Olaf Groth, global professor of strategy, innovation and economics at Hult International Business School.
“If they are human, then whose values are you using?”

At the same time, AI is continuing to evolve on the attack side. It’s being used more by organised criminals and nation states. The future could give way to systems able to produce tailored attack software, or ways of quickly identifying and determining vulnerabilities. There’s also the risk of criminals deploying intelligent botnets that are resistant to attempts to shut them down.

In the end, protecting data is all about keeping up with the attackers, so it makes sense that AI will feature heavily in the fight back. However, as Data Protection Day will surely highlight, businesses need to keep on top of protecting all information, including that which AI systems themselves draw in.

As David Ferbrache, technical director at KPMG’s cybersecurity practice, warns: “As these systems become more sophisticated, they will draw in increasing amounts of data in a faster way that’s less easy to scrutinise and monitor.”

To get weekly news analysis, job alerts and event notifications direct to your inbox, sign up free for Media & Tech Network membership.

All Guardian Media & Tech Network content is editorially independent except for pieces labelled “Paid for by” – find out more here.