Today’s post is from a brilliant friend of mine, Paul Golding. I first met him over a decade ago when he was a technology wizard for 02, a large UK mobile provider. Since then, he has continued to stay on the leading edge of a variety of technologies, including Artificial Intelligence (AI).
In this article, Paul outlines five of the biggest misconceptions that are preventing many businesses from unleashing the power of AI – and why you should double-down on AI during an economic downturn.
I know you’ll enjoy this!
Recent advances in AI have been staggering. Yet, business leaders in all sectors, even with strong IT competencies, continue to misinterpret the uses and benefits of AI. They do so at their peril, especially in the current downturn.
Misconception #1 – “AI is Exotic and Futuristic”
Folks see the crazy world of machines that make art, write essays, or drive cars, and proceed to position AI in their mind as some kind of exotic technology done by brainiacs working at Google. AI is simply a method to find predictive patterns in data, patterns so subtle that humans cannot see them — patterns with economic value for your business.
AI can find patterns in manufacturing data, sales data, user data, marketing data, pricing data, logistics data, fitness data, health data, inventory data — your data. If you have data, you can use AI right now to boost your business, including your bottom line. The true economic power of AI lies in the mundane, not the exotic. For the curious, read my post about why AI can find patterns (warning — it’s long).
Misconception #2 – “AI is for Experts”
Whilst it’s true that impressive demonstrations take plenty of brainiac talent, the bulk of AI applications are within your reach. Notably, the Fast AI course, specifically aimed at AI newbies, has generated many students who, despite having never seen an AI algorithm or read an AI scientific paper, went on to achieve performance breakthroughs in their domains. Many of those achievers had no technical background, except for some basic coding skills.
Sure, if your plan is to innovate in the latest algorithmic development to fly rockets to Mars, then maybe you need a brainiac, or two. But if your plan is to apply AI to your business to get results, then there’s really nothing stopping you. AI courses start for as little as $0 and online resources are in an embarrassing abundance.
Misconception #3 — “During Economic Downturns, AI is a Luxury”
I see it everywhere — potential AI programs being cast aside as “expensive science projects” high on the list of programs to cut during a downturn. The opposite is true. The powerful pattern-finding capacity I referred to in #1 is exactly what you need during a downturn. The most critical ingredient for AI is data — and you already have it.
Leaders everywhere always want the paradoxical: to do more with less. If ever there was a technology for achieving this, then AI is it. Why? Because it finds answers “for free”, as in answers already sitting dormant in the data. But it requires a firm commitment to the belief that there are answers in the data, which leads to the next misconception.
Misconception #4 — “AI needs a ‘Data-driven Org’”
Everybody knows that AI needs data, apparently lots of it. After all, those impressive essay-writing machines consumed the entire Wikipedia corpus, and more.
This is where AI’s secret sauce comes into play, called Fine-Tuning. The information AI learns about language from reading all of Wikipedia gets built into the models for use in your project. Let’s say you want to build an AI to read manufacturing reports, you don’t have to start from scratch. You can start from where the Wikipedia model left off and focus on adapting to your use case. This method is shockingly effective and means more results with far less data — data you already have.
AI only needs enough data, and often a surprisingly small amount. State-of-the-art image recognition of hand-written digits now only requires thousands of samples, not millions.
Perhaps the greatest misconception is the meaning of data-driven. Leaders have taken it to mean the art of making decisions via data, such as analytics. For AI, data-driven means that you let the AI decide which pieces of data to pay attention to, dispensing with having analysts poring over tons of dashboards.
Misconception #5 — “Build vs. Buy? AI is a Buy.”
This misconception is driven by the marketing of many vendors who claim that their tool is “powered via AI”. Leaders are slowly trained to think that AI adoption is about buying tools with embedded AI. This is often the weakest option. AI is powered via data, as in the data you already have and understand. The dirty secret is that many of these vendors are taking your data and, behind the curtain, using fine-tuning without any other magic sauce. In fact, there is no other sauce. Getting results from AI often requires lots of trial and error in setting parameters. It is often better to learn this skill in-house as a fungible skill applicable to a range of projects beyond what a single vendor’s tool has to offer.
The Fast AI course has already demonstrated that the power of fine-tuning is available to ordinary folks prepared to learn the techniques. Right now, there are folks in your organization who could pick these skills up.
One of the greatest missteps in the digital transformation revolution was handing too much power to data science. Many techniques are within the grasp of folks who can write Excel macros, yet many such analysts remain with Excel instead of migrating, as they should, to usable AI tools. The levels of automation in these tools has made them accessible to a far wider audience than most business leaders might suppose.
About the Author:
Paul Golding has been building AI systems since the 90s and has over 30 patents. He has worked for the likes of telco giants, like Telefonica and O2, for formula one racing companies like McLaren and stealth start-ups out of UC Berkeley. Not only has he invented AI techniques, but he regularly advises business leaders on how to roll out AI in their organization, free of mystique.