What Is AI Anyway?

It’s depressing, it really is. As Holden Caulfield would say.

Everyone on the vendor side of the marketing technology space says AI is table stakes. Almost everyone claims to be offering it, one way or another. And how many surveys have I read this year in which senior marketers say they’ve already allocated budget to AI, or they plan to do so in the immediate future.

Then I read articles online, purporting to offer authoritative overviews of the use of AI in marketing and marketing technology; but they describe AI as a “technique” or a “solution.” I admit I haven’t yet seen it referred to as a “platform.” I’ve even seen it suggested this week that if machine learning can serve a business objective, AI might be unnecessary. All of which suggests to me that many people commenting on AI don’t know what AI is. 

Stuck with the term

The AI tag isn’t going anywhere. It’s a pity, because its philosophical and computer science roots lay in the attempt to model human intelligence in a computing environment. Success would be achieved, said the pathfinder Alan Turing, if the difference between computing responses in a defined experiment, and human responses, were indistinguishable. Turing was wrong about that; or at least his perspective was too narrow. But at least it was somewhat clear.

But in recent years, and in this marketing tech space, AI has come to encompass everything from machine learning to basic automation. 

In the case of machine learning, this is arguably valid. Machine learning takes place when algorithms can correct themselves (hopefully in the direction of better performance) without human supervision. This falls way, way short of modeling human intelligence, but at least it features some independent action by machines which in the broadest sense might be described as rational.

On the other hand, let’s be clear that automation based on heuristics set by a human operator, and not revised by the machine, is not AI, no matter how valuable and useful it might be.

An easy example makes the difference clear. If I set up a crude, Amazon-style book recommendation engine, packed with rules like “If she buys Dickens, recommend Thackeray, but if she buys Tolkien recommend Rowling,” and set it running, I have not created an engine powered by AI. No matter how smart it seems to the end user. If, on the other hand, I build an engine which can independently change and improve those rules, without my supervision, I’ve created some really simplistic AI.

So it’s “intelligence”

AI isn’t a technique you apply to problems, it’s not a tool, a solution, or an application. It’s “intelligence” in the sense of being able to learn independently. In humans, “intelligence” is a much broader capability, so the historical view that AI is achieved only when machine intelligence is indistinguishable from human intelligence has fallen by the wayside. 

That’s not to say that cutting-edge AI hasn’t made enormous strides. Thanks to “deep learning,” explained here, AI has become extraordinarily proficient at detecting patterns. This had traditionally been a weak spot for computers, great at doing calculations at dizzying speed, bad at telling a dog from a cat. We’re now at the point where AI can be trusted to analyse and catalog vast libraries of images, interpret and summarize texts, and even compose texts without supervision.

These impressive capabilities have real, obvious use cases, especially for marketers wrestling with content creation and digital asset management at scale.

It can also, it should go without saying, ingest and analyse consumer data in such volume and at such velocity that the insights it surfaces will appear to be, and might reasonably be described as, “predictive.”

Note that, in addition to the development of deep learning algorithms, the rapid advances in AI are not so much due to the math as due to the fact that the quantity of available data, and the computing power to deal with it, have increased dramatically in recent years.

And Microsoft?

Big Tech is investing heavily in AI.  Google was leading the field last year, with almost $4 billion in play.  Amazon, Apple, Intel, and Microsoft were below the billion mark. Last week’s announcement that Microsoft, in one fell swoop, splurged $1 billion on OpenAI is a game-changer; or at least a game-changer for everyone other than Google.

But what is OpenAI?

Simply, it’s a small team of scientists in San Francisco, working on AI, but working on it ethically. It was strongly associated with its co-founder, AI panic-monger Elon Musk, who pulled out earlier this year. At the heart of the capped-profit company’s mission is the objective of developing “safe” AI, and distributing its benefits broadly. To its credit, OpenAI understands that being able to fulfill that mission demands high technical proficiency, and not just policy-making chops.

Is that where Microsoft’s money is going?

According to reports, the investment is likely to see the commercialization of some of OpenAI’s initiatives, and the exclusive provision of cloud services to OpenAI by Microsoft. Certainly Microsoft expects a boost to its own AI capabilities from the deal.

The gutter, the stars…

You remember the line from the old Pretenders’ song? “All of us are in the gutter, but some of us are looking at the stars.”

Most marketers, or marketing ops professionals, have a foot in the gutter and a foot on the pavement when it comes to AI. In other words, everyone (surely) is using basic automation processes (or sophisticated, but non-self-learning robotic automation), which aren’t really AI but are easily labeled as such. And many are using real AI, real machine learning, which can improve itself in limited ways with limited human supervision.

At the other end of the spectrum, the likes of Google and Microsoft are gazing into the deep space of deep learning based on vast cloud computing resources. The benefits of that will surely flow to enterprises and their marketing teams, but not immediately, and perhaps not with immediately clear use cases.

In the mean-time, understand what’s available; understand what is AI and what isn’t; and don’t worry about anyone creating a thorough model of human intelligence any time soon. Improving the subject lines in emails? Sure.

Related Posts