Can You Hack AI? Yes, and Deep Learning is Being Used to Both Attack and Defend

Being online these days is scary, but there are simple safety measures you should take

We’re all familiar with Siri, Echo, Google Maps and Translate – as well as IBM – using artificial intelligence, or AI, to answer questions, find locations, help us make sense of foreign languages, and predict the weather respectively.

But AI is getting smarter, which of course, is exactly what it’s designed to do. “Deep learning” means the software gets better at guessing as time goes by.

If you haven’t taken a ride in a recent model by Tesla you might be a tad surprised. A large “dashboard screen” shows any and all relevant journey information and the entire car is connected to your phone ecosystem.

While the screen shows you exactly what’s to your right (a truck), your left (a sedan), and in the rear (a motorcycle), you can dictate an email while being given directions as music from your iTunes collection plays softly in the background.

Its language recognition AI software is startling (we tested it in both English and Mandarin Chinese; it passed both with flying colours). And of course, there is the autopilot feature, which some argue isn’t 100 per cent bug-free yet – but Musk et al still plan to have self-driving cars in garages sooner rather than later.

The technology available today naturally gives us a glimpse of tomorrow – a “tomorrow” that takes Moore’s Law into account. The thought of leaps in AI machine learning leads to obvious questions: Considering all of the AI functions that are entering our daily lives – how vulnerable are we?

Few serious people are all that concerned about a movie-like Skynet takeover by suddenly semi-sentient robotic masters, but as people begin to understand how pervasive artificial intelligence is becoming and how invested it will be in our future, some are asking, “Can you hack AI?” The answer appears to be a clear “yes”.

Cybercrime is already a lucrative pastime for e-criminals and online threats to your personal data are becoming more sophisticated and more widespread.

An “ancient” 2015 article from Business Insider had a statistic claiming a hacker using ransomware – that doesn’t require a PhD to figure out – could make on average around $84,000 per month.

Not a bad paycheck and certainly sufficient motive for those with ill-intent. It’s definitely time you switched to a browser that doesn’t track or allow cookies and you can use a VPN, but adding a safety website checker app to those lines of defence is a good idea as well.

Such safety apps go a long way in stopping a leak from “browser fingerprinting”, and they have functions that scan Wi-Fi systems, let you set-up “blacklists” so your kids don’t see stuff they shouldn’t, and in short – answer the question: “Is this website safe?”

Online security begins at home and if you haven’t upgraded and updated your protection measures, get on it now. A safe browser app is easy to download and install – and many are free.

It’s spookily impressive how quickly hackers and attackers have learned to get around security systems – until you remember that the bad guys are using AI as well.

For decades, we’ve tried to outsmart computer bots by using CAPTCHA systems, hoping a non-human agent would be flummoxed by distorted text or be unable to surmount the challenge of picking out which squares look like a bicycle, etc.

But studies show that with optical character recognition (OCR) 99.8 percent of these challenges can be overcome. AI is also figuring out passwords much faster than back in 2017 when Science Magazine reported on the malicious hack that released the personal information of 143 million users of credit reporting agency Equifax.

The same article reported that scientists used AI (and remember, this was tech available almost four years ago) to figure out close to 30 per cent of passwords out of 43 million LinkedIn profiles.

They used a machine learning-based system known as a “generative adversarial network” or GAN, to make faster, more intelligent guesses – and the speed at which it learned how to crack passwords was phenomenal.

These computer scientists were doing the hacking in a bid to improve security, but you can bet your bottom dollar that others out there are using improved, 2021 tech for evil rather than good.

So yes, AI can be and will be hacked – but AI might also be the solution to stopping AI hacks and cracks.

The genie, to use an old cliché, has been let out of the bottle and AI isn’t going to disappear any more than any other new technology has. We will have to learn to live with machines that are getting smarter with every byte of information it absorbs.

But in the meantime, get a safe browser app, create genuinely strong passwords, and, while it might seem like a hassle, use two-step authentication. The same article reported that scientists used AI (and remember, this was tech available almost four years ago) to figure out close to 30 per cent of passwords out of 43 million LinkedIn profiles.

They used a machine learning-based system known as a “generative adversarial network” or GAN, to make faster, more intelligent guesses – and the speed at which it learned how to crack passwords was phenomenal.

These computer scientists were doing the hacking in a bid to improve security, but you can bet your bottom dollar that others out there are using improved, 2021 tech for evil rather than good.

So yes, AI can be and will be hacked – but AI might also be the solution to stopping AI hacks and cracks. The genie, to use an old cliche, has been let out of the bottle and AI isn’t going to disappear any more than any other new technology has.

We will have to learn to live with machines that are getting smarter with every byte of information it absorbs. But in the meantime, get a safe browser app, create genuinely strong passwords, and, while it might seem like a hassle, use two-step authentication.