AI is one of tech’s most over-used terms. In 2019, an MMC survey concluded that 40% of EU ‘AI’ start-ups did not actually use AI. The term is appealing to investors, customers and analysts, leading to regular misclassification which businesses can be very slow to correct.
In truth, real AI doesn’t exist yet. Instead, we have machine learning: an algorithm that processes data very quickly and issues a pre-programmed response. While these pieces of code do enable technology to ‘learn’ we’re a long way from Turing’s AI test, where a machine emulates our intelligence so convincingly that a human will mistake it for another human.
Machine Learning vs AI
Machine learning might not sound as sexy as AI, but these algorithms are helping to protect businesses, prevent fraud and save lives.
AI conjures images of slick robots and futuristic worlds. Conversely, its appeal dates back to ancient Greece. Silver coins depicting a mythical automaton called Talos demonstrate as much.
Machine learning as a term was coined in 1959 by Arthur Samuel from IBM. It is the engine that powers today’s most advanced ‘intelligent’ systems and without it, AI will never exist.
Technically, the difference between machine learning, and AI is formal reasoning. Whereas machine learning uses advanced heuristics to adjust its response based on input AI can ‘think’ independently. Anyone who’s used an online chatbot or automated phone system knows how broad a gap we still have to bridge.
But is it wrong to use the term ‘AI’? While we can’t have an authentic conversation with an automated system, it is still analysing a shed load of data extremely quickly, demonstrating a sort of intelligence. The problems begin if you believe AI, as it currently exists, delivers more than it can.
When ‘AI’ as a term is a problem
Using AI as a marketing term is fine, providing nobody is being misled. If you believe that AI means you’re getting a superior product guaranteed to deliver a superior outcome, you can find yourself anything from disappointed to dangerously exposed.
You can overpay for something that will never deliver the results you hope for, purely because you’ve been seduced by the term ‘AI’. Then there is the question of risk. Allowing unproven AI access to your precious, perhaps regulated data can be a huge error.
A badly written algorithm will deliver bad results. It could be missing critical signals amid all the data noise. In the security world, it could be detracting from your overall security posture, as opposed to improving it, leaving you open to a serious breach.
We’ve recently worked with two clients, both of whom believed their endpoint security solution was superior because it ‘used AI’. Once we introduced them to SentinelOne, with its extensive proof of AI defeating malware, they realised their mistake.
What makes good AI good?
Good AI is defined by specificity and speed. In other words, it requires excellent data and lots of processing power. The more specific the dataset, the better informed the responses. The greater the processing power, the faster it will learn.
To select AI well, be clear about what you want to achieve, and the kind of results you can realistically expect. Ask to see unbiased proof. SentinelOne broke records during independent tests, identifying and isolating more malware more quickly than any other similar endpoint security solution.
Show me AI that is designed for a single, focused purpose and I’m interested. Promise me a ‘silver bullet’ and I am instantly wary.
Where AI is working very well
AI is being used to protect businesses from highly advanced threats. Algorithms can detect and isolate malware much faster than a human could.
Doctors are using AI to tackle interoperable data, meet increasing demand for personalised medicine, develop intelligent applications and accelerate areas like image analysis and life science research.
Accenture estimates that AI has the potential to generate $2.2 trillion of value to retailers by 2035, by boosting growth and profitability. Improved asset protection, in-store analytics and streamlined operations are predicted benefits.
AI is helping telecoms businesses tap into the power of GPUs and the 5G network. As our communities become ever more connected, AI can harvest and analyse data to generate real-time insights into congestion, pedestrian safety, parking or pollution.
£1.6 billion of fraud was prevented last year, according to Mastercard, using AI. When I was in Brighton recently at the T20 Vitality Blast cricket match, we stopped at a small shop way outside my usual spending locale. My contactless card payment was stopped, and my PIN requested. This is AI in action, checking it was indeed me using my card and preventing fraud.
Industrial and manufacturing
AI is empowering smart factories to improve quality, operational efficiency, reduce costs and build safer working conditions.
AI can deliver seriously impressive outcomes if you approach it in the right way. First, be clear on the results you want to see. Then seek proof, not promises. Remove trust, sweep away the hype and see good AI for the clever code it is.