Alexa, what’s AI?

Source: Deep Learning on Medium

Go to the profile of Frank Beirne

It seems like artificial intelligence (AI) is everywhere these days. All the products, services, and software you’ve used for years are new and improved. It promises to make you smarter, faster, more charming, and better looking. A lot of the time, “AI-powered” is just thrown on top like spice without much context at all. BAM!

BAM! A dash of AI. Ship it.

The team at Dig recently attended Collision in Toronto, North America’s biggest tech conference. Everyone with a tech product was there and a big chunk were claiming “AI”. When we asked them what the AI does, or how they created it, we got a lot of blank stares followed by…. “it’s AI!”

Don’t get me wrong, I want it. Take my money and all that, but what am I being sold here? Just hype and buzzwords or is there something to it?

First, let’s talk about what AI is without getting too nerdy about it (data scientists don’t @ me). As the name suggests, artificial intelligence means incorporating human intelligence into machines. Any time a computer completes a task that we’d classify as “intelligent”, that’s artificial intelligence. In the broadest sense, this ranges all the way from a calculator to HAL (context for Gen Z readers). But these days when we talk about AI we usually mean something more.

Enter the cool stuff- machine learning and deep learning. These are algorithms, trained by humans, that learn from inputs, process information, and give us some output. There are esoteric differences between the two. Deep learning uses neural networks, algorithms designed to mimic the human brain (although we don’t really know if they do). But the goal of both is the same. Alexa can now figure out what song you want and Google Maps can find you the best way home. The name of the game here is automating decisions what were previously the realm of the human brain.

The AI onion (simplified)

There’s no magic here. Humans can do all this stuff, usually better since we’re the ones training the models. We show them examples of what we want and they learn to repeat it. But machines can do it faster and at a bigger scale than our puny human brains. Like the old saying goes, “Give an AI a cat photo and nothing happens, teach an AI to recognize cat photos and it can keep doing it a billion times a minute with 93% cross-validated accuracy forever with no sleep”.

Impressive AI really is everywhere, like in your Tesla. This made it an attractive term for marketing folks. They realized you could slap a “now with AI!” sticker on just about any old tech, suggesting there’s some fresh magic under the hood- “This product uses the same tech as autonomous cars and Alexa”. After all, no one will really question you except some data scientists who don’t hold much purchasing leverage anyway. The CEO wants AI…. who cares what the techs say?

Even worse, many companies are making stretch claims about AI that they didn’t develop themselves, or offers no customer value. Like having a chatbot on your website, or some AI feature built into your CRM. This is backend stuff that offers me no value. It has no place in your marketing.

The AI catfish

If you get caught touting some AI that’s just a bunch of “if-this-then-that” statements, you have an admissible defence. That technically is AI in the broadest sense, the outer layer of our onion. You get to lump yourself in with something much more impressive. The price of entry is programming logic that’s been around since the computer itself. You’ve pulled off the classic AI catfish.

It’s not “wrong” to do this, but it’s misleading. Technically, a Tamagotchi and a MacBook are both computers, but which would you rather have? (keep in mind vintage Tamagotchi’s are very valuable).

The fact is, proper deep learning (but less so machine learning these days) is really hard and really expensive. The big tech companies and some niche players are the only ones doing it well. It is getting more accessible now that Google, Amazon, and Microsoft have been kind enough to open source their work. You can use their “pre-trained” neural nets and customize them for your use case. This takes a lot less training than starting from scratch. But even doing this is no small feat and requires major resources. I speak from experience.

We put some deep learning features in our Upsiide app. The goal was to automatically process our image and text data to tell our clients the “why” behind certain consumer decisions in addition to the “what”, which we could already do. Not unlike cat/no cat (or hotdog/no hotdog). Starting from scratch was not an option so we leveraged open source code and built on top. Even with a head start, and some of the smartest people we employ working on it, it took time and money. Lots of it.

Now we’ve got some neural nets doing stuff that we could just have a bunch of analysts doing. But this way we have speed and scale so the tradeoff was worth it for us. It’s faster and cheaper for our clients and that’s valuable. But it certainly wasn’t fast or cheap to develop, and we’re a consultancy with data scientists and math PhDs coming out of our ears.

None of this is to say that products without AI, or with “AI” but not the machine/deep learning kind, are less valuable. It’s just a toolkit to be used when appropriate. It encompasses a broad range of technologies, some valuable and some not. Depends what you need to do.

Next time someone tells you their product has AI, question it. What does the AI do? Is it machine learning? Deep learning? Normal logic? Do I care? After all, you probably wouldn’t ask what the programming language is, so why do you care about the algorithms? It’s the value you’re after.

Asking these questions can help cut through the noise and dodge the AI catfish. There’s a ton of hype here and a great deal of it is unwarranted. If you’re tempted to buy a calculator because Texas Instruments tells you it has AI, ask them what that means. If the answer is “it can add, subtract, multiply, and divide”, well, maybe pick up the cheaper model without AI.

We brushed over some big topics. Mainly machine vs. deep learning, there’s different kinds of AI too (narrow vs. general). You can find info online, but we have our own take on the business impact for marketing and research. Stay tuned…