Machine Learning for Web Development

Machine learning, artificial intelligence (AI), and deep learning – we bandy these terms about in conversation concerning modern technology, and for good reason: this new frontier governs a growing portion of our present-day world. As a web developer, I’m constantly looking for new tech and trends to integrate into the software we build. So naturally, machine learning is very appealing. But it’s not intuitive as to how a website can immediately benefit from it.

A Quick Primer

If you haven’t heard about machine learning, then first of all, welcome to the world of the future. Cars still don’t fly, hoverboards only kinda work, and our robot butlers live in our phones. Anyway, here’s a quick overview of machine learning.

As computer programs grow in complexity, it becomes harder for humans to understand and maintain them. That means more mistakes, bugs, and glitches occur. Also, some things are just too complicated for a human to program.

Enter machine learning — or more specifically, deep learning. Deep learning is modeled after the way neurons work in our brains by passing around small bits of information… But that’s a complicated description. Let me try again.

It’s like evolution, where things start off simple and random but gradually grow to complexity… Then natural selection takes over and… You know what? This video does a much better job of explaining it.

Now that Grey has properly informed you about how machine learning works, we can start to talk about how it’s used on the web.

Tailored Content (AKA that Pesky Internet Privacy Thing)

Love it or hate it, targeted content is not going away. In traditional advertising, marketers either had to appeal to a broad audience or choose a specific but large demographic. For example, maybe you aren’t interested in purchasing baby diapers or asking your doctor about Celebrex, but you’re going to see those TV ads anyway.

The Internet is a different beast, though. Websites know a lot about you — where you are, the device you’re using, even if you still write those Harry Potter fan fictions. OK, that last one might be a stretch — but only by a little. Using the information about you to present you with highly specific advertisements is a much better return on investment for marketers.

CMS Vendors and Machine Learning

Web content management systems are providing the ability to change the website content based on who is looking at the site, and they’re using machine learning to do it. In Sitecore’s case, they’ve created a simple robot dubbed “Robbie” that is similar to a virtual assistant, only not so virtual.

A robot with a ball for a head sitting on red shoes

Robbie, otherwise known as Wilson 2.0 (Cast Away 2120)

The people at Sitecore claim that in the near future, you may be able to visit a physical store with a robot similar to Robbie (but much more advanced, hopefully). You could ask it questions and get recommendations and maybe even pay it when you’re ready to check out. Then, when you’re back at home, you could visit the store’s website, and the website would be aware of your earlier interactions with the store bot. The website could prompt you to review the products you purchased and offer deals on accessories for an incentive. The in-person knowledge gained is also useful to the website (gender, race, attire, even personality) may be difficult to ascertain from only mouse and keyboard interactions.

While Episerver hasn’t created a robot (as far as I know), they also are banking on AI to sell their CMS. In fact, if you take a look at their website, a little chat bot called EpiBot will appear after a few seconds in the bottom corner of your browser. It’s not that amazing as far as chat bots go, but it shows that vendors are exploring new avenues to drive users to their website.

Like Sitecore, Episerver also offers a tailored experience based on user data. A few months ago, Episerver did a presentation showing how a visitor can browse a website and later receive a customized email newsletter based on their time on the site.

Full disclosure, Sitecore and Episerver are both Experis partners, and while I enjoy promoting them, that doesn’t mean that other CMS vendors aren’t just as involved in AI. Even WordPress has a number of plugins that benefit from our soon-to-be bot overlords.

Accessibility and Localization

Another way that machine learning can be beneficial to websites is by making them more accessible. Believe it or not, many websites use images without providing alternate text. For shame. (Please don’t check our images.)

Missing alt text may not be a problem in the near future, however. Some publicly available APIs now make it possible to generate alt text without human involvement. Here’s an example of the Azure Computer Vision API in action. Go ahead and try it out below.

See the Pen Dynamically Generated Alt Text with Azure’s Computer Vision API by Sarah Drasner (@sdras) on CodePen.

It’s not perfect, though, and often gives text that doesn’t quite match. However, as things improve, these APIs will undoubtedly progress and even offer better accuracy than most people could give. For example, if you were asked to describe an image of a boy using a slicker brush on his chinook, you probably wouldn’t describe it that way. However, if you knew that a chinook was a breed of dog and that a slicker brush is a common pet grooming tool, you’d see that it makes perfect sense to use that description on a website for dog breeders. An AI can have far greater domain knowledge than an average person.

It’s not just images that need alternate text. Videos need captions and descriptions to be considered accessible. AI has you covered there, too. YouTube already gives you the option to automatically create captions. Speech-to-text recognition is commonplace now.

Automatic Translation

Accessibility doesn’t stop with media. If you have a multilingual website, you might already be using some automated translation feature powered by machine learning. Google Translate is one of the more popular tools, and it does a decent job. Of course, if you require professional quality translation, we’re not quite there yet. Still, it’s an example of how common machine learning has become; it’s often hard to tell when it’s being used.

Art and Design

Think your job as a graphic designer is safe? Sorry to disappoint, but computers can even create unique artwork now. Sure, it’s rudimentary at present, but it will certainly improve. There may even come a time when computers can create specific art and photos on demand.

Imagine you need a black and white photo of a woman sitting on a wall in an egg costume (hey, it happens). You could simply say, “Hey, Siri, make me a photo from the 1920’s of a female Humpty Dumpty,” and presto! You get something like this:

A woman from the 1920's sitting on a wall in an egg costume

An AI did not actually create this photo.

It’s not just visual design where bots will one day excel. Machine learning can be used for writing as well. The news articles of the future may be completely automated. As a matter of fact, you may have already heard about the technologist who built a machine to write screenplays. The short film, Sunspring, is the result. Unfortunately, not even the star of HBO’s “Silicon Valley” can do the script much justice. However, the fact that machines are writing scripts at all is an achievement many would never have thought possible.

It may be some time before websites are able to benefit from machine-assisted design and writing, but you can bet that there are developers working on it at this moment.

We’re Safe for Now…

All right, this article is starting to sound a little too much like a Doomsday, bots-are-coming-for-your-job, brave-new-world future, so let’s sum up. There are a lot of ways a website can benefit from machine learning: making content accessible to users with disabilities, providing more relevant content to everyone, and maybe even designing entire websites one day.

I believe we’ll see a lot of advancement and improvement as machine learning algorithms grow, but we developers and designers are most likely safe — for at least a few more years.

If you’ve seen or considered applying machine learning to any of your web projects, let me know in the comments.

Leave a Reply

Your email address will not be published. Required fields are marked *