Artificial intelligence (AI) is inextricably connected to big data, involving computers performing unimaginably fast calculations on astronomically large volumes of data. But big data also opens the door to greater possibilities for something going wrong, such as privacy breaches and hacking.

The rising incidence of hacking, cybercrime and industrial espionage is creating robust demand for corporate security. It’s the sort of unstoppable trend that confers long-term profits, but as I explain below, only for those who make the right investments.

One of the world’s top researchers in AI is Yann LeCun, an expert in facial recognition at Facebook parent Meta Platforms (NSDQ: META). He has commented that AI is a matter of engineering, not math. For instance, he said, AI experts don’t actually know why facial recognition works (to the degree that it does). They just develop a system through trial and error. There’s no mathematical formula that defines how AI comes up with an answer.

The implications are profound. That’s because in contrast to the purity of math, anything that is engineered, from computer algorithms to bridges, is almost certain to contain minuscule variations. These open the door for computers to be fooled, as indeed they can be.

Some examples come in the form of what’s called “adversarial examples.” In one case, for instance, after researchers subtly altered a few pixels that made up the pattern on the shell of a model turtle, an AI program misidentified the picture as a rifle. To a human eye, the changes weren’t even noticeable.

Finding Waldo: Not Child’s Play

In throwing AI off its game, a key word is “anomalies” — anything that is introduced into a computer that doesn’t belong and that the computer itself cannot detect as being out of place.

Almost every significant cybersecurity breach can be represented as an example of an anomaly, something out of the ordinary in the data generated by a particular computer, a set of computers, or a country’s worth of computers.

Anomalies are critical because computers operate according to particular engineered principles. If you can introduce an anomaly that escapes a computer, something it fails to recognize as not belonging, then you’re inside the computer and have free rein to threaten the entire system.

The efflorescence of big data, which forms the bedrock of AI, makes anomalies ever harder to spot. The greater the volume of data in any system of servers, the harder to notice any one piece that doesn’t belong that was snuck in for nefarious purposes and that can bring the network down.

It’s like the popular children’s books Where’s Waldo, featuring drawings of seas of tiny people among whom you have to find, hidden in plain sight, Waldo, who is distinguishable by his red and white striped garb. If Waldo is set among 10, or 100 other figures, he’s easy to spot. But if he were placed in a crowd of a million or billion or quadrillion people, good luck.

The Downside to Big Data

Not only does big data make dangerous anomalies more difficult to spot. What’s also true is that the more the world comes to rely on AI, the greater the damage that cyber attacks can wreak — potentially disarming the electric grid, sending false orders to military commanders, or scrambling financial markets, to name a few alarming but not farfetched possibilities.

Techniques for spotting anomalies are themselves complex, and no one solution works in all cases. Most cybersecurity systems are based on firewalls, which keep track of incoming and outgoing traffic on a computer system. The idea always is to be able to detect which information doesn’t belong.

Judging firewalls is inherently difficult since by their nature they have to be opaque. In assessing a cybersecurity company you need to rely on customer satisfaction surveys and on the company’s financial results.

As I’ve just explained, AI remains a huge investment opportunity. However, you need to be wary of the risks, which include high valuations and media hype.

Perhaps you’re eyeing the Big Tech stalwarts that are making massive investments in AI. Well, they’re obvious plays on the trend and they’ve been trading at lofty heights. A lot of cybersecurity companies are overpriced right now, too.

How can you safely invest in AI? Consider the advice of my colleague Robert Rapier.

Robert Rapier is an income investing legend. He’s the chief investment strategist of Utility Forecaster, Income Forecaster, and Rapier’s Income Accelerator.

Robert has found a better way to make money from the AI super-boom, thanks to a group of reasonably valued, under-the-radar tech plays. Robert is locked in on AI right now because of the incredible income opportunities it has created. To learn more, click here.