Cheat Sheet: AI for business

The buzz around artificial intelligence is hard to ignore. Davey Winder explores how AI can make a real difference to the way we do business

Davey Winder explores how AI can make a real difference to the way we do business.

It seems like people have been hyping AI for years: is it ever going to actually help us make money?

First of all, I’d take issue with saying AI has been hyped – I would say it’s been over-hyped. Not least as most of the high-profile AI developments can be more precisely defined as machine learning (ML), which is a specific, relatively basic subset of AI. But, for the sake of convenience, let’s pretend that the two are one and the same thing – after all, that’s what the media seem to be doing.

Fine, so what is machine learning going to do for our company?

Well, how does the idea of AI-enabled chips grab you? Amazon, Google and Microsoft are all investing heavily in custom chips that are optimised to speed up AI-enabled applications in general, and ML training models in particular. You won’t be buying these chips yourself, of course. But they will be snapped up by the companies that are investing heavily in AI, and they will be using them to offer AI services that can streamline your processes and improve your productivity in ways we haven’t yet begun to think about.

Can we really put our business-critical processes in the hands of an AI algorithm? What if it makes a mistake?

AI does make mistakes – but computer algorithms learn from their mistakes, and can quickly become smarter and more reliable than any human. Look at the growth of facial recognition techniques: not many years ago, this seemed like science fiction. Today, we take for granted the way that Apple, Facebook and Google can automatically tag your friends and family in photos – and that’s just the tip of the iceberg. Subaru is creating cars that can predict when the driver is about to nod off, while medical imaging technologies use AI-driven facial analysis to detect diseases and even help manage pain.

So while it’s easy to imagine a badly programmed AI selling off all the company stock at rock-bottom prices, a real AI mistake is likely to be something vastly subtler than a human would never even notice in the first place – and which is subsequently used to improve accuracy and performance in the future.

What about security? How can we be confident that an AI is wholly trustworthy?

You can’t ever know that any solution is wholly trustworthy, but the security industry has given AI a big vote of confidence and ML is fast becoming de rigueur for incident detection monitoring.

That’s because it helps to solve two problems at once. The first arises when incident alerts are generated faster than your analysts can deal with. If a multi-pronged attack hits your network out of the blue, an AI can react thousands of times quicker than a human can.

The other issue is that battle-hardened security analysts are hard to come by, especially on a limited budget. So even if you have plenty of time to address a security issue, an inexperienced analyst may miss the warning signs. ML-trained machines can help by flagging up issues that look worthy of further investigation.

So AI solves out security problems? Great!

Unfortunately, there’s a flip-side to all of this: cybercriminals are also using ML techniques to make their attacks smarter and more insidious. For example, automated vulnerability reconnaissance can be used to find new, previously undiscovered holes through which an attacker can gain entry to an application or network. This sort of attack isn’t new, but AI techniques make it faster and more intelligent than ever before.

So AI basically means smarter apps?

That’s certainly part of its potential, but the implications go deeper than that. A new development model is emerging that we could call AIOps – AI/ML principles applied to the DevOps process.

The pairing might not be immediately obvious, but if you think about DevOps as being about agility and flexibility, you can see how AI slots in nicely. Your AI can function as a sort of Mystic Meg, anticipating from a very early stage how different approaches and processes will impact production, and making changes if that impact is found to be negative. It’s also likely to drive a benefit in terms of data organisation: AI works best when it has the largest possible dataset to work with, which will hopefully mean an end to siloed systems.

“If an attack hits your network out of the blue, an AI can react thousands of times quicker than a human”