Artificial intelligence (AI) has been on everyone’s lips for the last couple of years and marketers are no different. It feels that at least a thousand articles are written every day about how AI is already transforming marketing and how every tiny agency SIMPLY HAS TO HAVE AN AI STRATEGY.
As someone who has been around the block a few times, someone with experience in tech and someone with a lot of friends and acquaintances that are smarter than me, I can tell you that the reality is a bit different.
I’d like to share a few things I learned about AI that all marketers should know.
“Modern” AI Is 99% a One-Trick Pony from the 1980s
Virtually all of AI revered as something futuristic today is based on theoretical work that was done in the 1980s by Mr Geoffrey Hinton. He figured out a method called backpropagation that can be used to train layered neural networks to start exhibiting a certain level of autonomy when analyzing certain sets of data.
The reason why we had to wait thirty years for his theories to be put in practice is that we just didn’t have the computing power to do it efficiently and cost-effectively.
Now, we do.
The only issue here is that the vast, vast majority of all the new breakthroughs in the field of AI is mostly engineering work that squeezes another iota of efficacy out of this same process.
In essence, for the most part, AI developments that we see today are not really that revolutionary in the great scheme of things.
Mr Hinton says this himself.
AI Is Not Really that Intelligent
The AI in the sense of the word that is used today is a spectacular thing. In simplest terms, after being trained, the layered neural networks start exhibiting behavior that shows a certain level of autonomy in finding ways to analyze data. They figure something out on their own.
That is a huge, huge thing and it should not be understated.
That being said, all of the AI models in use today are spectacularly limited. Most glaringly, they are extremely pigeon-holed. If an AI model is trained to, for example, recognize dogs in photos, it will not be able to recognize cats, let alone do anything else that might be considered intelligent.
In addition to this, AI models simply cannot deal with curveballs. If the input diverts even the slightest from what they were trained for, they will not be able to handle the challenge and the output will be useless.
AI Really, Really, Really Needs Humans
Despite its often staggering capabilities, AI is actually worthless without puny, squishy humans. For one, every AI model still needs to be trained by humans. In essence, a team of people will have to spend copious amounts of time feeding data to a model, showing it the ropes and pointing in it in the right direction. They will also spend huge amounts of time tinkering with the model, guiding it as it develops.
On the output side of things, you still need to have humans who will make the final decision. This is necessary for one very simple reason — AI cannot explain itself. Its decisions may be 100% correct, but you still cannot put absolute faith in it if you don’t know how the decisions are made. With AI, we simply don’t know.
Try explaining to a client that you wasted their money on a market segment that an AI wrongly recommended.
Many AI Companies Have Nothing To Do with AI
AI has become a veritable gold rush-type situation, with every other company slapping AI in front of its list of services or products, expecting a payout. A recent report showed that 40% of all AI startups in Europe have nothing to do with AI. Or at least, AI is inconsequential to their operation.
As a marketer who is thinking about maybe incorporating some kind of AI capabilities in their operations, you have to be very careful about this. There are a lot of swindlers out there trying to sell you basic statistics and visualizations of data sets as some kind of AI-based snake oil.
In other words, instead of buying some crowdsourced, cloud AI B2B lead generation solution from some two-person “company” that ends on ‘ix’ or ‘yx’ (as they so often do), maybe do your homework and learn how to use your website for lead generation.
AI Is Difficult and/or Expensive
As a marketer, there are two ways you can go about adopting AI. The first option is to build your AI capabilities the way that Unilever does it either as a way to help your clients (or company) or to sell them forward.
In order to do this, you need extensive resources, and I mean extensive. You need data scientists and AI experts which do not come cheaply and you need enormous amounts of usable data which is a huge expense as well. On top of all that, you need a lot of time to develop capabilities that will be useful to your clients and/or customers.
Alternatively, you can approach one of the companies that already provide AI-based services to marketers. Provided that you do your research well and avoid getting taken for a ride, you will still have to hope that the service/product you chose really answers your needs and solves the problem you have. It has to be a really tight fit to be effective as you will be buying a solution of the box (to a larger or smaller extent).
And don’t get fooled, these do not come cheap either. (Let’s just say you won’t find the pricing page on most these companies’ websites.)
Trying To Sum It All Up
AI is in a funny place where it is astounding and not-so-astounding at the same time, where the whole marketplace feels a bit like the Wild West and where cost-effectiveness feels more esoteric than in most other fields.
For marketers, it is important to keep a cool head, do plenty of research and never think of AI as some kind of a magic bullet.
It can be a nice (and expensive) tool, but only if used by an expert. And only then in certain specific situations.