Maybe you have heard the term “neural network” and possibly you have a basic understanding of what it means. It’s a kind of software that receives input and “learns” how to produce correct output. But, if you are anything like I was not too long ago, what it means for the neural network to “learn” is a complete black box. Input goes in, magic happens, good things come out.
That was the extent of my understanding until recently, when I decided to fill the gaps and develop a basic working knowledge of neural networks. I wanted to know what was happening in that “magic” step. How exactly did this software “learn?”
This article represents my first step in that pursuit. I focused on one concept in particular: the perceptron.
What is a perceptron? And why start there? A perceptron is a binary classifier, which means it can only return one of two possible outputs (“classifications”). A simple way to think about it is that a perceptron is useful for answering “yes” or “no” kinds of questions like, “Do you want a cup of coffee?”
But the really cool thing about perceptrons—and the reason it is a perfect place to start learning about neural networks—is that a perceptron is a basic implementation of a single neuron. A “neural network” is simply a group of connected neurons. If we can understand how a single neuron works, understanding how they all work together in a network is surprisingly intuitive.
Before we get into it, a disclaimer: I am no expert in the fields of data science or machine learning. I am writing this because I found that understanding the perceptron was that lightbulb moment I needed to peer into the magic black box of neural networks. Once I understood this basic building block, the rest started to fall into place. I still have so much to learn, but now I have a fundamental understanding from which to build. I hope by the end of this article you will, too!
The Anatomy of a Perceptron
First let’s define some terms and talk about the components of a perceptron. Use this diagram to orient yourself as we walk through each component:
The input of a perceptron is the numeric data fed into the perceptron for which we want a binary classification. For example, if you were using a perceptron to determine whether or not you’d like a cup of coffee, you might have the following numeric input:
Do you like coffee? (1 or 0)
Time of day in minutes (0 to 1440)
Your tolerance for caffeine (between 1 and 10)
How tired are you? (between 1 and 10)
How much effort obtaining a cup of coffee will require (between 1 and 10)
Now, given a dataset like above, a perceptron should be able to learn and then somewhat accurately predict whether or not you’d like a coffee. But how? You’ll just have to… “weight”… to find out!
It’s perhaps a little simplistic, but you could think of “weight” as the amount of impact a particular input has on the outcome. For example, if you really like coffee, you may be willing to drink it any time of day! In that case, the weight associated with “Do you like coffee” would be very high and the weight associated with “Time of day in minutes” would be low.
Each input has its own weight. When input is passed into the perceptron, each input is multiplied by its corresponding weight. The resulting value is then added together with all the other input/weight combinations, and that single number is what is passed to the “activation” function.
The “activation” function determines whether the neuron is active or inactive. It takes the sum of all the inputs multiplied by their corresponding weights (an aggregate weighted input) and converts it to the expected binary output. This can get complicated and involve other concepts like “bias,” but a simple example of an activation function would be something like: If our aggregate input is greater than 0, return 1. Otherwise, return -1.
The output of a perceptron is simply the result of the activation function. It can only ever be one of the two values representing either “active” or “inactive.” In this article I’ve used 1 for active and -1 for inactive.
Training a Perceptron
That’s all well and good, but how does a perceptron “learn?” A perceptron “learns” through a process called “supervised learning” or “training.” Here’s how it works: First, you need a large set of example inputs (training data) for which you already know the correct output (active/inactive). Here’s an example using the coffee dataset from before:
Do you like coffee? (1 or 0)
1 (I love coffee!)
Time of day in minutes (0 to 1440)
1200 (8 p.m.)
Your tolerance for caffeine (between 1 and 10)
8 (caffeine doesn’t affect me much)
How tired are you? (between 1 and 10)
3 (not very)
How much effort obtaining a cup of coffee will require (between 1 and 10)
7 (gotta wash the carafe and everything)
Output: 1. Yes, I’ll absolutely have a coffee at 8 p.m., please.
Once you’ve collected a good number of inputs with corresponding correct outputs as seen above, you’re ready to train your perceptron! You train it by running each set of inputs through the perceptron one at a time. If the perceptron gets it wrong, that information is used to make changes to the weights associated with the inputs. The idea is that after those adjustments, the perceptron is more likely to produce correct output when that dataset (or datasets similar to it) get passed into the perceptron in the future.
So, training is actually simple—it’s just adjusting weights. You could think of those weights as knobs. When you turn each one to the optimal number, the perceptron will be most likely to produce the correct output for any input.
But how do you know how to adjust the weights? Each weight is adjusted individually with a function that takes into account the individual input corresponding to the weight, the current weight value, and the output. There are some fancier things you can do, but the basic function looks something like this:
Are you the kind of person that has to see code and get your hands dirty to learn? Me too. While I was learning, I put together this example perceptron implementation with a visualization of its training process:
The problem I’m using the perceptron to solve is very simple: Given two inputs that represent the coordinates of a point on a two-dimensional plane (x and y), determine whether that point is in the upper right half of the plane (represented by blue) or the lower left half (represented by yellow). The black line is there to visualize the truth threshold between the two halves of the plane, while the purple line visualizes the perceptron’s current “understanding” of the threshold location.
Each time you click “Train Again,” a new dot is added to the plane at random and the perceptron guesses whether the dot should be yellow or blue. You’ll notice in the image above the perceptron got it wrong a few times while it was learning! When the perceptron gets it wrong, we feed that back into the training function and update its weights. This results in the purple line moving to visualize the perceptron’s new “understanding.”
Eventually the perceptron’s weights are adjusted to the point that it essentially always gets it right. When we reach this state, the purple line will be in almost the exact same position as the black line.
Of course this is a silly example. You could solve this problem programmatically without any machine learning whatsoever in just one line of code. But shh! Our perceptron doesn’t know that. Let it learn! Try it out! Tweak the code and see what happens to the learning process!
Putting It Together: The Network!
A single perceptron on its own is a surprisingly effective tool, but it is limited to binary classification (“yes” or “no” kinds of questions). But what if we had multiple perceptrons? And what if the output of a set of perceptrons served as the input for another set of perceptrons? Now you have a network of neurons—you might even call it a “neural network!”
A network like this could be configured in a number of different ways to solve all sorts of complex problems. For example, you could change the number of layers and the number of neurons each layer contains. Training a neural network looks much the same as training an individual perceptron: Feed training data in, when the output is incorrect, adjust the weights within each individual perceptron.
Of course there is much more to neural networks than what we’ve covered here, but I want to demonstrate that the fundamental concept is quite simple. If you’ve understood the perceptron (a single neuron) then a neural network is essentially just layers of perceptrons feeding into one another.
If you’re interested in learning more about machine learning and neural networks, I recommend this video series by Daniel Shiffman—from which I learned much of what I’ve shared in this article. It’s very beginner-friendly and thorough.
Happy (machine) learning!
Last year, I wrote about some of the interesting content-related trends I foresee coming our way. In The Future of Content’s first episode, “A Glimpse of the Future,” I build on that forecast with eight predictions about content in the not-so-distant future.
“What we’re seeing now with tools like Paragraphs in Drupal and the Gutenberg editor in WordPress…there is an explosion of interest in creating better editorial experiences.”
New technologies like machine learning and augmented reality are changing the content landscape.
Content production is set to become more vertical-focused.
Content delivery will become increasingly contextual.
Some of my past predictions that missed the mark.
“Machine learning can elevate a basic CMS to the level of an enterprise digital asset management tool.”
Stream Episode 1 now, or subscribe on your favorite podcast platform below.
Note: This transcript may contain some minor wording and formatting errors. Apologies in advance!
[Voiceover] Welcome to The Future of Content, a podcast exploring how we create, manage and distribute content. Brought to you by Four Kitchens: We make BIG websites.
[Todd] Welcome to The Future of Content. I’m your host, Todd Nienkerk. Every episode, we invite a guest to explore an aspect of content and to make predictions about the future of that content. We’ll talk with people who create content, writers, marketers, filmmakers, performers, and people who build the tools that manage and distribute content. Our conversations will be equal parts creative inspiration and technical how-to. If you create, manage or publish content, welcome! This podcast is for you.
I’ve built websites most of my life, but I’ve also been a writer, editor, and publisher. I started my first ‘zine in middle school. I designed it in WordPerfect, made photocopies at my dad’s office, and mailed them to my closest friends. In high school, when GeoCities and the X-Files were at their peak, I ran a UFO conspiracy website, and later I started an e-zine to compete with my high school newspaper.
In college, I met my Four Kitchens co-founders at the Texas Travesty, the student humor publication at UT-Austin. After graduating, we launched an alt-weekly called That Other Paper. It was our first foray into making big websites and using modern content management systems like Drupal.
I love helping people create and share content and I love talking with people about the future of content. What’s next? What new devices will transform how we experience content? What new technologies or tools will help us manage and share these experiences with others? So in our very first episode, I’d like to kick things off by offering some predictions of my own. The first of these will be technical, focusing on content management systems and editing tools. If this isn’t your thing, hang in there. I’ll get to the creative, experiential predictions soon.
[Music] Prediction number one: I believe that content management systems will be content repositories, not website managers.
So to understand what I mean by a website manager, we need to go back in time a little bit to the early 1990s, the early web. In these days, content was tied to presentation. These were the same, stored in hard-coated files. You had content alongside HTML and later, CSS, and we used tools like desktop publishers, FrontPage, to manage these files and help us upload. Dreamweaver, and tools like these… And there was very little functionality contained in websites. Websites that did do something—like let you buy something or order something—were very complicated to build and maintain…
But then in the 2000s we saw a rise of content management systems which was a new kind of term. In 2000, Drupal. WordPress and DotNetNuke in 2003, and Joomla in 2005, just to name a few. These web-based CMSs were a single piece of software that divided the frontend and the backend, and the frontend was the display of content, and the backend was the management of content. It was the first time that those two things began to be separated.
So early on in the early days of the CMS, we could display content in a web browser and that was pretty much all that a CMS would do—just display the content when you go to a URL—but then came things like RSS and atom feeds which allowed us to syndicate our content to other sites, and then later came smartphones and tablets. And those ushered in a totally new way of designing for the mobile revolution, ways to interact with sites and expectations that your audience and users had visiting your website, but they really, strangely, did not fundamentally change our approach to content management.
On the backend side of things, CMS stored text, media, and user-generated content like profiles and comments and things like that. And as these CMSs became more popular, website administrators, webmaster started to demand more backend functionality not that there were people adding comments. Well, now you need the ability to additional logins and accounts. And that means permissions. You don’t want every user to be able to access all of the administrative functions of a site. So we started creating permission schemas. Later, website managers wanted tools like layout managers and integrations with third parties. And of course, all of this had to be configurable within a user interface. Nobody wanted to rely fully on developers to make a code change every time you wanted to alter a layout or user information.
So our CMSs, as they became popular, became website managers. They were really more about managing an entire website, the users and the functionality and layout and all of these things and not just content. That made content management systems really heavy and really complicated. And CMSs were really no longer fundamentally about content management. They were about managing your website or managing your commerce platform, fundraising campaigns, communities, all of the things that CMSs now try to do.
So around 2008, we saw a rise in new devices, apps, and channels because of the mobile revolution led mainly by the introduction of the iPhone. And this proliferation of devices and channels has led to a fragmentation on the web. It’s caused website managers and digital strategists to spin up new websites and new content management systems and new ways to have to manage each of these devices and experiences. And in the scramble to keep up with all these new devices, we keep creating more sites. And they pile on top of each other. And they build technical debt. And now, we have multiple instances of WordPress, or Drupal, or custom home-brewed things that we all have to maintain. So of course, this greatly increases the effort and cost of maintenance. But it also leads to really inconsistent practices, inconsistent user experiences, bare-bones user experiences, just bad stuff in general.
What we’re seeing now, though, looking ahead to the future, is people are fed up with this fragmented approach to the web where every app needs a different backend to manage the content there and sharing your content. And experiences between multiple sites is really hard to do. So we’re seeing a centralization of the content. And that’s leading to what’s called decoupled architecture.
[Music] Prediction number two: Content will be extensible and modular.
So a modern CMS shouldn’t treat the website as the primary experience. A modern CMS should be multi-device, but not designed for any specific device or context. And that means you have to think about your content first. Not the website, not devices—content. You should let your CMS deliver structured content, and then let the device’s software or application handle the rest of the work.
If your content is properly modeled—meaning, if you have reusable fields in really robust content types or post types—you can quickly support new devices and experiences as they are invented and as you want to adopt them. And one of the things that makes modern CMSs like Drupal and WordPress so great is their ability to quickly add new fields and content types. So let’s say, for example, that you wanted to add more iPhone support to your digital presence, either by creating a better mobile website or a more mobile-friendly website or app. One of the things that you may want to do is add a location field, because an iPhone, of course, is a location-aware device. But once you add that location field, for free in a way, you’re now getting support for all location-aware devices. But that only works if you have all of your content centralized and modular. So if you have multiple content management systems, one for your iPhone app, one for your Android app, then you don’t get that win.
Or let’s say you want to start adding more video assets to your site. Well, there’s all kinds of interesting metadata associated with video assets. It could be the length of the video. It could be certain tags that you want to use. It could be break points in the video to insert ads—whatever you want to do. If you wanted to, say, add Roku to your digital presence, you can add certain metadata fields that are reusable by YouTube, Vimeo, Apple TV, Samsung Smart TVs—that whole fleet of streaming video services and applications and devices.
We also see things like the proliferation of things like AMP, Accelerated Mobile Pages. These are really similar to Facebook Instant Articles and Apple News. So if you start to create something like an AMP feed, it’s very easy to translate a lot of that work into Facebook Instant Articles and Apple News if you use an extensible content model. So when content is modular, it can easily be published to entirely different experiences and not just device specific context.
[Music] Prediction number three: Content creators will finally get the tools they deserve.
One of the number-one complaints we hear from editors and maintainers and content creators is that their tools—their CMS, the editing interface, whatever they’re using—is just really complicated, cumbersome. It’s hard to maintain. It takes too long to publish content.
But we’re seeing now tools like Paragraphs in Drupal and the Gutenberg editor in WordPress. There is an explosion of interest in creating better editorial experiences. So coupled with the need for modular content, this has led to new ways of thinking about structuring and assembling content.
So as Paragraphs and Gutenberg both illustrate, there is a trend towards creating blocks of content, or chunks of content, that can be reordered, reassembled, remixed into different posts or for different experiences or different devices. And these editorial experiences, in making this more user-friendly, are actually further entrenching the block approach to creating content.
I suspect also that we are going to see a similar interest in standalone editing interfaces that are totally CMS-agnostic. And these could theoretically replace many content management systems’ out-of-the-box editing tools. So I envision that, in the not-so-distant future, there will be content creators and editors who carry with them, in a way, an editing interface that they prefer. So rather than having editors demand that a newsroom adopt a platform like Drupal or Arc or something else because they prefer the editing experience, instead, those editors can say, “Oh, I prefer ABC Content Editor.” And another editor can say, “I prefer XYZ.” And both can coexist in a newsroom because they’re agnostic, and they’re portable. So editors then get to kind of build their own tool or use their own tool regardless of the underlying content management system that is storing and distributing the content that they’re producing.
We’re also seeing a proliferation of hosted content services like GatherContent and Contentful. GatherContent is a hosted platform for managing content, and it’s focused on what they call “content operations.” They refer to themselves as a “content operations hub.” So what they’re leaning into is not just content management but content workflows, content governance, campaign management. They want to be the hub around which entire organizations structure their content strategy and their marketing campaigns. Then we have tools like Contentful. And Contentful was leaning more into the technical side of things. It’s positioning itself as an “API-first CMS,” focused on the editorial experience and content modeling.
Another common complaint that we hear from content creators is the inability to preview content before it’s published. So now we’re seeing a lot of tools like Gatsby Preview that allow you to view the changes to your content live or close to live in another window that you have open. And it’s actually using the live website or a development instance of the website, whichever you prefer, and it updates that live as you change the title or change the lead or update text move images around. This third-party solution to a common problem is something we’re going to see more of. We’re going to see more third-party systems like portable editorial interfaces and previewing tools that then get attached to existing CMSs rather than the CMS trying to be an all-in-one solution solving all problems for everybody.
We’re also seeing some really bold, new ways to make traditional CMSs more editor-friendly. So of course, there’s the Gutenberg initiative within WordPress. There’s also many examples in the Drupal space. An interesting one that’s rather timely is something called DX8, which was built by a group called Cohesion. It’s a version of Drupal that focuses on the digital experience that can be created within the Drupal CMS. And just this week, Cohesion announced that Acquia, a major player in the Drupal space, has acquired DX8.
And one last note on content creation, and the different tools that are starting to pop up left and right. We built something called Emulsify at Four Kitchens. Emulsify enables teams to create design systems, and then build CMS themes based on those design systems. Emulsify has gone through a few iterations since it was launched in Spring 2017. But after only two years, it has almost 80,000 installations. And this is just one of many examples of how the demand for tools that enable content creators and designers is enormous.
[Music] Prediction number four: CMSs will focus on specific verticals and use-cases.
We’re already seeing a lot of specialization in the content management space. There are marketing automation platforms, newsletter and email management tools, constituents and campaign management platforms. All of these are especially strong in some verticals, like enterprise marketing, publishing, in the case of constituent management, nonprofits and associations. So a lot of these tools just kind of naturally aligned with the needs of some verticals. But what’s really interesting is the sudden rise of CMSs that are specifically targeted to media, entertainment and publishing.
Here are a few examples. There’s Thunder, which is a distribution of Drupal. It’s a flavor of Drupal. It’s sponsored, maintained and used by Hubert Burda Media, which is one of Germany’s largest publishers. And Thunder is intended to be a mostly out-of-the-box, though still highly configurable, Drupal-based publishing platform for media companies out of the box.
We also see tools like Arc Publishing. Arc is Washington Post‘s CMS that has now been productized and is being licensed to other publishers—The Los Angeles Times, for example. So there are lots of newsrooms that are starting to adopt Arc Publishing, and that’s because Arc really focuses on newsroom management. It’s not just a content management system, it’s a newsroom management system. Plus, it leans heavily into content monetization. So Arc has a lot of tools out-of-the-box that allow you to monetize your content through throttling of paywalls and registration walls—and all kinds of other tools. They realize that one of the biggest concerns and biggest challenges that the publishing industry faces is: How do they continue to make money when a lot of their content is available or given away for free? And these tools, like Arc, are trying to help address that by packaging best-in-class tools around content monetization.
And we also have Chorus. Chorus is Vox’s proprietary CMS that they have now opened up for licensing. And the way Chorus positions itself is: “Chorus is the only all-in-one publishing, audience, and revenue platform built for modern media companies operating at scale.” I think Arc might have something to say about that because it’s positioning itself in a very similar way, but it’s yet another example of media companies that are turning their home-brewed or customized content management systems into something that’s marketable. And that’s actually a really interesting subtrend to highlight.
So all three examples that I’ve given here—Thunder, Arc, and Chorus—were all created by large media companies and then made available to the public either through open-source (in the case of Thunder) or proprietary licensing means. These publishing companies then—and here’s the interesting trend—publishing companies are also becoming software companies. And they’re trying to reduce or recoup their costs by releasing their internal products to the public and having them support or help pay for those.
So we’re going to take a quick break. When we get back, I’m going to talk about the more creative and experiential side of what I see happening in the Future of Content.
[Voiceover] The Future of Content is brought to you by Four Kitchens. Our team creates digital experiences that delight, scale, and deliver measurable results. Whether you need an accessibility audit, a dedicated support team, or a world-class digital experience platform, the Web Chefs have you covered. Four Kitchens: We make BIG websites.
[Todd] Welcome back. So we’re going through eight predictions about the future of content, and we’re up to prediction number five: Machine learning will help us manage and create content.
Modern CMSs can read, see, and hear your content. This is a paradigm shift, but it’s going to happen in the background without people really being aware of it. For example, if you use any kind of Google product, particularly Gmail, you may have noticed that they’re introducing a lot of auto-complete. And these auto-complete tools, at least in my experience, have been extremely, unnervingly accurate. So as I’m writing an email in Gmail, it often knows how I talk. And as I see the suggested text, it’s, I’d say, about 80% of the time, it’s really accurate. It’s how I speak. Well, of course, they’re using machine learning to look through the years and years and years of emails that they have on file, so some machine somewhere has learned how I write and it’s just simply repeating that back to me in a way. That’s an example of a content management system that’s reading, seeing, hearing content and it’s feeding back to the content production workflow. So that’s a way in which content management systems and machine learning can help manage and create content.
Machine learning is also going to drastically simplify media management. So in the past, you had to tell your CMS where your content belonged on your site, but with the help of machine learning, your CMS can add metadata to that content and perhaps file it away more quickly and accurately than you. And this is especially true of media management: images, videos, sound files. Machine learning can elevate a basic CMS to the level of an enterprise digital asset management tool simply by being able to employ image recognition or speech-to-text or clever tagging techniques.
CMSs are also going to create content using artificial intelligence. In September 2017, Digiday reported that the Washington Post had published 850 AI-generated articles in its first year of operations. This included 500 articles about the 2016 US elections. And those 500 articles generated more than half a million clicks, “Not a ton in the scheme of things, but most of these were stories that the Post wasn’t going to dedicate staff to anyway.” So what that means is rather than replacing the reporters and journalists that they have on the ground these tools, at least in this instance, are augmenting them by being able to provide more content that otherwise would not be reported on at all. Since then, the New York Times, Associated Press, Reuters, Yahoo Sports, to name a few, are also using AI to write stories. And in March , the Press Association, which is UK’s equivalent to the Associated Press, claimed that they can publish 30,000 local news stories per month using AI.
And another fun thing to consider about how machine learning and content creation can interact: CMSs can also interpret the emotional tone of the story, not just the facts, and react accordingly. So some publishers are already employing sentiment analysis in ad delivery systems to avoid embarrassing ad placements. So, for example, if an algorithm detects that an article is critical of Dow Chemical, Dow products will not be advertised on that page. So if you want to see an example of machine learning-powered content creation I encourage you to visit something we built called HappyGram. It’s at HappyGram.ai. It’s a proof of concept we built to show how editors can use machine learning to help them tell emotionally impactful stories and to speed up the content creation process.
[Music] All right. Prediction number six: Reality will be augmented.
If you’ve seen me speak at an event in the last couple of years you probably heard me talk about how big VR is going to be. Well, I got a little too excited about VR. VR will definitely have a strong adoption in gaming, training, and industrial applications, but I have to admit to the average person, it’s really just going to be a novelty and it’s going to be used occasionally, if at all. But it’s still important. For example, VR is being used in combination with cognitive behavioral therapy to treat combat PTSD and phobias.
AR, however, is another story. So VR, in case you’re not aware, VR is a fully occluded, fully immersive experience where you shut off all of your awareness of the outside world and you enter a virtual world. AR, augmented reality, is augmenting what you already see around you in some way. And it’s quietly infiltrating our everyday lives. Its potential is huge precisely because it’s so useful but subtle. So a really good example is Google Lens. You should take a look for it. Google Lens, according to CNET, “feels like a pair of smart glasses without the glasses.”
There’s an example that they give in this article that CNET wrote where Google Lens is being held— the device is being held over a menu in a restaurant, and it’s highlighting the menu’s most popular choices based on reviews that it’s pulling off the internet. And then it links those popular choices in the highlights to photos of the dishes. And all of this is being handled through a combination of machine learning and augmented reality.
And of course, because augmented and virtual reality experiences often involve three-dimensional models, it’s no surprise to see that we’re seeing an explosion of 3D assets on the web, which is really just another form of content: 3D models. So an example of this would be Sketchfab. Sketchfab is a website that is a massive marketplace for 3D assets. It’s like a Shutterstock for 3D assets. There are also sites like Pinshape. Pinshape is like Sketchpad for Shutterstock, but for printable object models. So if you have a 3D printer, you can download one of these models, print out whatever you need and use it. All of this stuff is content, and all of this stuff has to be managed in some kind of system.
[Music] Prediction number seven: Content delivery will be context-specific. By that I mean: Content will be delivered to you based on where you are, what you’re doing, and what you’re trying to achieve.
A really good example of this is the death of iTunes. So earlier this year , Apple announced that iTunes is going away. It was launched in January 2001. It was a total game-changer. It arguably, maybe, ruined the music industry, but it was extended over time to add other media: movies, TV shows, podcasts. 18 years later, iTunes is a relic. So now iTunes is being fractured into multiple new apps. The apps are called Podcast TV and Movies. If you’ve been using an iOS device, you’ve already seen this happen. But now this is going to happen on the desktop.
The takeaway here is that iTunes has realized that it’s trying to do too much, and that each experience has a certain context associated with it. There’s a context in which you want to experience a podcast, TV show, or a movie, and they’re all different, and they don’t all belong on one application. We’re going to see this happen on the web everywhere.
So if you want a really quick demo of context-specific delivery in action, go to your desktop or laptop computer—not your smartphone—and google “dog.” The search results are really typical. You’ll see a Wikipedia page about dogs and photos of dogs. Now try this on a smartphone or tablet. In your Google search results, you’re going to see an animated 3D model of a dog. And if you tap on it, you can drop that 3D model of a dog into your room. It appears to scale, you can walk up to it, you can walk around it, and sort of interact with it. And this is an example of that content delivery knowing that you’re on a smartphone, knowing that you’re on AR-enabled device, and knowing to give you better search results that are related to what your device can do.
[Music] That brings us to our last prediction. Prediction number eight: Distribution channels will be restricted and monetized.
Streaming services are going to be built around exclusive content. This has been happening for a little while. Of course, Netflix has been producing, famously, original content for years. Hulu’s doing the same. But now we’re seeing individual networks start to do the same thing. CBS, for example, launched All Access in—I think it was 2016. And All Access provides exclusive access to streaming-only content like The Twilight Zone, Star Trek Discovery. By August 2018, it had 2.5 million subscribers. That’s less than two years after its launch. And in January of this year, NBC Comcast announced that they intend to enter the streaming wars, and this means that they’re going to be ending their deals with Netflix and Hulu. So if you want to watch Friends or Seinfeld or The Office or Saturday Night Live, you can no longer go to Netflix or Hulu. You need to go to NBC directly.
One thing I want to add—since, hey, we’re on a podcast—I want to talk about the changing face of the podcast industry and how seriously people are now taking it. So don’t let the DIY roots of podcasts, especially this one, fool you. Podcasts are serious business. So just last month, the Interactive Advertising Bureau [IAB] reported that podcast ad revenue in the United States increased 53% in 2018, totaling almost 480 million dollars. The IAB expects that ad revenue to grow another 42% in 2019 to 680 million dollars. Podcasts, which used to be platform-agnostic and just delivered through an RSS feed, are now consolidating on paid platforms with exclusive content deals. A really good example of this is Spotify acquiring Gimlet Media, a podcast network, and Anchor, a podcast creation and distribution platform. Spotify continues, as part of their digital strategy—their content acquisition strategy—to acquire more podcast networks and related tools.
And one final word before we wrap up, give you some food for thought about advertising and how advanced it is these days. You may not be aware, but when you download podcasts from some of these major networks, ads are injected into the MP3 file based on whatever information they know about you. So if you’re using an RSS feed to listen to podcasts, they may not know much. But if you’re using a closed platform like Stitcher or Spotify, they probably know quite a bit. They know your listening habits, they know a bit about your personal data, and they’re going to start delivering content to you that is specific to your location to the other podcasts that you listen to. So that’s why, frequently, those ads that say, “Hey, you should check out this other podcast,” tend to be really accurate. Or you start to get ads that are local for some reason—a local mattress company or whatever. That’s because podcasts have ads injected into them at the moment you download.
One last thing to think about advertising: When you visit a site, the ad space is auctioned off based on everything they know about you. The highest bidder gets their ad shown, and this all happens within milliseconds. Every time you open a page, there is a live auction that happens where your information is being traded, rated, and sold, and then you are showed an ad paid for by the highest bidder.
These are just my points of view and my predictions. I’m probably going to be wrong about some of them like I was VR a few years ago. In the future, we’re going to invite guests to provide a diverse array of perspectives on the future of content. Because after all, it’s something we all have a stake in. Thank you very much. I hope you keep listening. Have a wonderful day. Enjoy your content.
[Voiceover] You’ve been listening to The Future of Content, a podcast from the Web Chefs at Four Kitchens. Hosted by Todd Nienkerk. Produced by PJ Hagerty. Theme song is PAFRATY by DJ Listo. Find us on Twitter at @FoCpodcast and get in touch by email firstname.lastname@example.org.
Now that we’re back in the kitchen, we’re parsing a lot of amazing information. And we’re talking about the products and processes that are most likely to benefit or influence web design and content creation over the next year.
Front and center: Personalization and Machine Learning
We weren’t surprised to see that many attendees were as interested in ML as we are. The “Personalizing News for Readers” panel, which included Four Kitchens’ Mike Minecki and Grace Stoeckle, was standing room only. The conversations afterward reiterated the excitement around the topic. We talked with many niche news organizations, most focused on local news or even one news niche (e.g., crime) within a local community, who are severely understaffed and under-resourced. These teams naturally want to know how can AI and ML can help them, and whether the technology is something they can access.
One of the most interesting revelations from these conversations: Organizations are seeing the potential of using ML to improve the personalization of email newsletters.
“We spoke to one publisher who publishes an email newsletter focused on local crime in LA,” Grace said. “Newsletters like this offer powerful ways to reach readers — and are an excellent playground for experimenting with personalization and localization.”
Mike noted that many publishers are just becoming aware of ML’s potential. “Most publishers are unaware of what’s possible today with machine learning and natural language processing, for fractions of a penny per query. Many have a metadata gap that prevents them from optimizing personalization — and ML can help solve that problem.”
We also saw several other sessions and demos that illustrated the ways that publishers are using ML. One that caught our attention: Toronto’s The Globe and Mail demonstrated Sophi, its homegrown ML-powered global optimization platform. Sophi enables The Globe to maximize both ad and subscriber revenue, addresses opportunity cost, and helps to remove promotional bias from both native and wire content. This is the kind of tool that we’re excited about helping clients build for their own content hubs!
A matter of trust
Although personalization and its intersection with AI and ML were dominant technology topics at the conference, we also heard a lot of people talking about trustbetween the media and readers. Specifically, content creators want to know: How do we build — and repair — reader trust in the information we publish online? And once we gain that trust, how do we maintain it?
Grace noted that GDPR is forcing organizations to handle data differently, but that this challenge is also an opportunity to establish trust. “GDPR provides an opportunity for publishers to be transparent with how they handle user data, and to build trust with readers,” she said.
From analysis to understanding
Many media outlets are also struggling to figure out the best pricing and subscription models for online news. Trying to view and understand their sales funnel and conversions in an effort to get this equation right is creating a huge strain. Most simply don’t have all the tools they need to translate site analytics and metrics into a thorough understanding of visitor behavior. (The desire to use that data to make informed decisions to improve the online experience was one impetus for The Globe to build Sophi.)
The connecting thread between all these topics? Publishers want and need help understanding readers’ needs and decision processes. The importance of the user experience is something we focus on at Four Kitchens, so we weren’t surprised to hear that it’s a priority across the online publishing industry, as well.
Are you as interested as we are in the potential of ML to help create a powerful UX? Contact Four Kitchens. We’d love to talk with you about the possibilities.
In our most recent blog post, Four Kitchens Director of Technology Mike Minecki told you a bit about how Machine Learning — as Mike put it — is “the scientific study of algorithms and statistical models that computer systems use to perform a specific task effectively without using explicit instructions, instead relying on patterns and inference.” Mike discussed some of the technical details of how Machine Learning (aka ML) can help content creators with time-consuming tasks like building out metadata and content tagging, creating summaries, and even moderating comments.
Today, I’d like to focus on the other part of that equation: the human experience that ML algorithms enable.
It’s a people thing
At Four Kitchens, we like to say that we build BIG websites. Well, a big part of those big sites is the effect they have on both users and the people in our partners’ organizations. That’s why we make the user experience (UX) central to everything we do. It’s why we help our partners consider communications strategies to engage users for both external and internal projects. And it’s why human emotions played a big role in this ML experiment.
All content, no matter how technical, tells a story. It’s how we humans experience the world. Understandably, images are a vital part of that storytelling. The right graphics can boost reader engagement and evoke the kind of emotion that makes a story, event, product, or brand memorable. But finding the right photos, especially for organizations that maintain massive image collections, can eat up a huge amount of time and bog down even the most efficient workflows.
When we began experimenting with how ML and artificial intelligence (AI) could benefit our partners, faster and more effective content creation seemed a natural place to start. But we also wanted to see whether we could use ML to generate content that helped people connect with one another.
Come on, get happy
At DrupalCon Seattle 2019 this past April, we introduced the results of our experiment: HappyGram.ai. This humble little Drupal site demonstrates how ML can streamline content-creation workflows by helping you sort through your image collection to find the right graphics for a given story. For organizations with massive banks of photos and other graphics, this capability alone can make ML worthwhile.
But HappyGram does more than sort photos. It also demonstrates how ML can create a genuine connection between people.
HappyGram.ai uses Google Natural Language API, AutoML Model, a free stock photography database, and Drupal to show how easy it can be to combine storytelling and visual elements that automatically fit the content and mood of any piece. HappyGram analyzes text, searches a specified graphic collection for appropriate images, and applies filters to reflect sentiment and enhance mood.
The result is a highly shareable entity — we call them “HappyGrams” — that are constructed from real-life human experience and emotions. The experiment is fairly rudimentary and somewhat limited by our use of all free tools, but it shows the potential of ML where content creation is concerned.
See it for yourself
Mike and I, along with several of our Web Chefs, are getting ready to head to New Orleans for ONA19, the annual Online News Association conference. We’ll be demonstrating HappyGram in the ONA Midway on Friday, September 13. If you’ll be at the conference, stop by! You’ll be surprised by how quickly this simple ML app can handle a chore that is typically time intensive—and even make it fun.
Imagine being able to automate the initial selection of photos for any piece of content — even recorded interviews that need to be transcribed. Then come see it in action! If you won’t be at ONA19 (or can’t wait until Friday), you can read more about how we created HappyGram — and see some examples of past demonstrations — at https://www.happygram.ai/about. Have questions about how ML can simplify or speed your content process? Contact Four Kitchens. We’d love to talk with you about the possibilities.
Without a doubt, Machine Learning is an ongoing component of our understanding of technology and the world around us. From the ability to spot speeding cars in traffic to facial-recognition software that can unlock smartphones, Machine Learning is at the heart of many cutting-edge technologies.
But before I get into what Machine Learning can do for content creators, let me define exactly what I mean by Machine Learning. Machine Learning is the scientific study of algorithms and statistical models that computer systems use to perform a specific task effectively without using explicit instructions, instead relying on patterns and inference. Many people conflate Machine Learning with Artificial Intelligence (AI); rather, Machine Learning is a subset of the puzzle that is true AI.
Another term to define is Natural Language Processing. NLP applies Machine Learning to understand how grammar and speech work together in an attempt to gain a better understanding of content’s meaning.
Machine Learning meets content creation
At Four Kitchens, we’re naturally interested in the crossover between content creation and the world of Machine Learning. How can we teach machines to find a brand’s voice? How can we build content for humans through algorithms and prediction mathematics?
Not that we plan to set up a room filled with 100 machines, just hoping one will randomly churn out a work by Shakespeare. But we’d love to use Machine Learning to alleviate some of the repetitive tasks in your content-creation workflow, so that you have more time to focus on creating great content.
Building Machine Learning into publishing and content management system (CMS) platforms can help with various aspects of the content-creation process:
Auto-tagging parts of content in a page or article
Correlating images with text to better represent content
Automatic building of semantic metadata
Building a voice within your organization’s articles
Generating article summaries or bylines based on content text
Moderating user-generated content
These key aspects of creating content can often be a drag. Instead of spending time on “housekeeping” tasks like these, most content creators would rather focus on creating better articles, blog posts, and pages to engage visitors. Taking care of logistics that can stifle the creation process is an important part of producing richer, more vibrant content.
With the integration of Machine Learning into the content-creation lifecycle, your teams can delegate those tedious tasks. You get more time to zero in on the messaging they’re creating and on connecting with users and communities. We frequently hear that content producers are under tighter and tighter deadlines and shrinking budgets. Machine Learning offers a way to give yourself some breathing room.
More time for creativity
Machine Learning won’t write the next great American novel. But it can enable you and your team to spend more time doing the creative stuff that makes your content unique.
Your CMS can now read your text, look at your uploaded pictures and videos, and understand the content and context of both. We’re still early days, but I believe that in the long run, Machine Learning will profoundly affect how we create content.
I recently spoke about the benefits of using Machine Learning in the content-creation pipeline at Yale Digital. You can see that presentation here.
What about bias?
Much concern has been expressed around implicit bias in Machine Learning—and rightfully so. Any project that uses Machine Learning must include careful consideration of bias and diversity. At Four Kitchens, we treat this need as an integral part of every Machine Learning effort.
If you’re interested in learning about our off-the-shelf modules or working with us on custom Machine Learning workflows that might help your content process, contact Four Kitchens. We’d love to tell you more.
Starting your own design system can understandably seem overwhelming. But I’m here to tell you that a small design system is not only achievable … in many ways it might be the preferable way for plucky teams to start.
Four Kitchens clients have big websites that include many pages, images, PDFs, videos, and other files. These sites have so much content, in fact, that many of our clients are overwhelmed just thinking about going through it prior to a CMS migration.
For more smart content like this, subscribe to our newsletter: