How Design Ai will Transform Work

Jared Bloom
 min read
How Design Ai will Transform WorkHow Design Ai will Transform Work
Table of Contents

We believe that Ai can—and should—make it easier to do beautiful work.

With, our goal is to give everyone the experience of working with a world-class designer every time they build a presentation. This means we’ve had to build the “intelligence” of a designer directly into the tool.

The following post is an introduction to Design Ai, the expert design system that we are developing to visualize a user’s ideas and automatically enforce the rules of good design. As it evolves, we believe that this system will help revolutionize the role of design in everyday work.

But first… is this really Ai?

The Ai effect

It’s been an interesting decade or so for the tiny island nation of Anguilla.

Historically known as a great place to kick up your feet or hide your taxable income, this small country in the Caribbean is now a central player in one of the biggest technological transformations of our lifetimes—Artificial Intelligence. As the official administrators of the .ai top-level domain, the government of Anguilla began selling domain registrations to companies all around the world in 2009. And, at more than $100 per registration, business is good.

Last year, hundreds of new .ai domains were either purchased outright or transferred between owners. And in September 2017, 4 of the 15 most lucrative domain sales were .ai websites, with selling for a whopping $45,000. (Anguilla, unfortunately, does not get a cut.)

The explosion of .ai websites is just one of many signs that the hype surrounding Artificial Intelligence might be outpacing the reality. It seems like every tech company in the world is in the midst of developing—or, at very least, selling—its Ai strategy. And you can’t launch a startup these days without promising to transform your industry with natural language processing (NLP) or machine learning. In fact, a quick google search will point you to dozens of companies developing Ai for Robotics, Healthcare, Transportation, HR, Sales, Marketing, Customer Service, Security, Real Estate, Retail, Driving, Flight, Coding, Alcohol, Shopping, Education, Fishing, and Ai itself. And there are dozens of “Ai-powered assistants” with names like Olivia and Clara standing by to help you schedule meetings, recruit candidates to your company, or chat with your customers.

#1 Noise Illustration-1.jpg

The problem, of course, is that when everyone claims to be doing Ai, it can be incredibly difficult to separate the wheat from the chaff. In many cases, you don’t have to scratch too far below the surface to see that Ai is little more than a marketing strategy for many companies.

But there’s a flipside to all this skepticism. Quite often, new efforts to build “real” Ai are dismissed as fluff because, unlike sentient robots, they don’t feel like Ai. There’s actually a name for this phenomenon: The Ai Effect. As Pamela McCurdock explains in her 2004 book, Machines that Think, Ai technology very quickly gets “assimilated” and ceases to be considered Ai at all, even when it’s applied to a completely new domain.

Douglas Hofstadter summed this up nicely with his own cynical definition of Ai: "Artificial Intelligence," he wrote, "is what hasn't been done yet."

As we continue our work on Design Ai, we understand that we’re operating in a fuzzy domain where nearly nothing has been done yet. We’ve already made significant progress, but we’re still in the very early days in this effort, so it’s only natural that the very idea of Design Ai will be met with its own fair share of skepticism from both the Ai and design communities.

With that in mind, we’ve created this post to explain how we think about Ai, where Design Ai fits into the conversation, what it means for users, and where we hope it goes from here.

Superhuman Ai vs. Expert Ai

There are almost as many definitions of Artificial Intelligence as there are companies claiming to do it. But most of us probably picture the same thing when we hear the phrase: robots.

We may not be that far away from having our own robot butler, caretaker, or therapist. But before we do, we have to solve dozens of critical Ai challenges. These robots will need to understand human language, interpret human behavior, and have some sense of empathy; they will need to know how to navigate the world and avoid obstacles; and, most importantly, they will need the ability to learn new skills. In other words, they will need to understand how we do what we do, and then do it better and faster.

#2 Superhuman vs Expert Illustration-1.jpg

This type of Ai technology—which includes disciplines like machine vision and natural language—falls into a category that we call Superhuman Ai. The goal here is to help machines reason like humans and act autonomously, and the ultimate objective is to make it unnecessary for humans to perform these tasks at all. This is the type of Ai that we usually see in movies, but it isn’t the only kind of Ai.

There is also an equally important strain of Ai that doesn’t seek to replicate human intelligence; but, rather, to democratize it. We call this Expert Ai, and the goal of these technologies is to give users access to a specific kind of human expertise that they either don’t possess themselves or can’t afford to hire. And the word “expert” is critical here, because it helps explain what makes this so difficult.

Expertise, by its very nature, is rare—and that’s because it requires two kinds of knowledge:

Factual Knowledge: These are the rules and laws of a particular domain that all experts agree on. Much of this knowledge lives in books or online, often written by the experts themselves. For example, writing experts generally accept Webster’s ruling on whether a word is a noun, a verb, or both—along with the basic rules of grammar and style. But if you’re building artificial intelligence to help someone improve their writing, these laws typically aren’t enough.

Heuristic Knowledge: At some point, an expert needs to apply “rules of thumb,” frameworks, or best practices to solve problems within their domain. There is no textbook definition of good writing, for example, so we rely on experts to use their judgment and they apply their own rules, laws, and inferences. The key difference here is that, unlike factual knowledge, heuristics can vary between experts. One writing professor’s idea of good writing—and how to achieve it—can vary widely from another’s. So Expert Ai, like the experts they’re emulating, needs to be opinionated.

The trick in building Expert Ai systems is to combine a “knowledge base” with an “inference engine,” and then fine-tune that inference engine over time. This means that the way you evaluate the effectiveness of this kind of Ai is unique, as well. Instead of looking at the novelty or sophistication of the technology being used, the best way to measure Expert Ai is by the experience of the user. Because, in this case, perception is reality.

If the user is satisfied that the Ai is delivering a result that (a) is consistent with what an expert would deliver, (b) she couldn’t have generated on her own, and (c) she can trust, then the Ai is doing its job.

But to reach this point, we must create a reliable Expert Ai system. And this generally requires three phases of development:

1. Build the Expert System. This is where factual and heuristic knowledge are combined to create a set of commands for the software to follow when presented with a set of inputs. Laws are defined, heuristics are translated into rules, and basic algorithms are developed.

2. Train the Expert System. Once the system is designed, it must learn in the real world, and this is where true Ai methodologies are applied. For example, the system may use machine learning to test it’s own rules and update it’s heuristics based on the training data it receives.

3. Make the Expert System Self-Sufficient. The ultimate goal of any Expert Ai system is to turn it into a self-learning system. This is technology that continues to evolve over time. It unilaterally adds new laws to its knowledge base and creates new rules of thumb to deal with them. In other words, the system goes from mimicking an expert to being an expert.

This approach takes time—and it almost always moves in fits and starts—but it’s precisely the path we’re following with Design Ai. The challenge in our case, though, is that we’re essentially starting from scratch.

What is Design Ai?

The first question that most people have when they hear about Design Ai is a good one: “How can software guarantee ‘good’ design? Isn’t ‘good’ a subjective term?”

In a sense, it is. One person’s perception of “good” design can vary widely from another’s. But, to understand why we still believe it’s possible to build an expert Design AI system, it’s helpful to break down what it is that an expert designer actually does.

Graphic design is, in its simplest terms, visual problem-solving. The goal of any designer—regardless of his or her medium—is to communicate a message clearly and effectively; and to do so within the constraints they’re given. These constraints might be the amount of content they can fit in the space they’re given, the functionality of the tools they’re using, or even the limits of human perception. For example, we know that humans can only read text beyond a certain size, and that we can’t see an element when it’s set against a background of the same color. These basic constraints are obvious, but they set the foundation for what good design can and cannot do.

#3 Artist Robot Illustration-1.jpg

Beyond these basic limitations, there is also a broad set of design best practices—rooted in academic research and centuries of experience—that are widely accepted by the expert community at large. For example, designers will emphasize items to draw attention to them because our eyes tend to focus on outliers; they will use complementary colors because certain color combinations make us uncomfortable; and they will organize content in a grid because we’re more likely to comprehend well-organized information. These general principles of good design—everything from consistency to balance to contrast—may be scattered across textbooks, universities, individual thought leaders, corporations, and the internet at large, but they are certainly knowable.

This common set of rules and best practices is what allows us to begin building a knowledge base for good design. And these are rules with a clear if/then structure, which means that they can be translated into an algorithm and applied automatically.

That said, we realize that design is about more than simply following rules.

For one, many of these rules are, by their nature, ambiguous. “Contrast” may be the best way to communicate the difference between two items, but there are many different ways to express contrast, and so designers must use their own set of heuristics to fill in the gaps. Similarly, designers may decide that a given rule should be ignored, or even broken, in certain cases. If you’re trying to make someone uncomfortable, for example, you might actively decide to use dissonant color combinations or an unbalanced layout. This is what we traditionally call “creativity,” and it’s what separates good designers from the truly great ones.

Our goal with Design Ai is to enforce the basic rules of design automatically, and then apply generally-accepted heuristics so that our users don’t have to create their own. In other words, our expert system has an informed opinion of good design, can enforce that opinion, and adapts over time as rules and heuristics evolve. We may never be as “creative” as an elite human designer—and we may never get to a point where we can break our own rules—but we are confident that we can make good design a given for anyone who uses our tools.

The next logical question, then, is this: “If it’s possible to build this kind of system, then what should it do?” Put simply, we believe that the goal of Design Ai should be to democratize the experience of working with a professional designer. It should understand the idea you are trying to communicate and suggest the best ways to visualize it. It should interpret the intent you have and the content you are creating and translate it into “good” design. It should update its heuristics to adapt to new design patterns as they evolve. And ultimately, it should learn your tastes and preferences and help you define a personal style that is truly your own.

We will use Ai technology like machine learning and NLP to achieve these goals, but only if we believe that they will help our users achieve their goals. After all, we are not building a superhuman; we are giving humans their own superpowers.

Why start with presentations?

We believe that there is enough consensus to begin building a set of evidence-based rules that nearly all designers can agree on. But we also recognize that the design community is large and diverse, so we made a decision early on to limit our scope to a specific domain of design where we can credibly—and quickly—build an expert system.

#4 Conveyor Belt Illustration-1.jpg

At Beautiful.Ai, we are presentation people. Our team has spent decades working on other people’s presentations—writing them, designing them, and building the software to make them. We’ve studied and worked side-by-side with the foremost experts in the field of presentation design. And we are confident that the heuristics we are putting into place are consistent with the widely-accepted principles of effective presentation design.

The other reason why we selected presentations as our beachhead is that they are, by far, the most common “design” deliverable created by non-designers. As a result, they are fertile ground for an expert system that can help users transform their ideas into well-conceived, well-designed visuals. And because there are billions of presentations created each year, they provide an excellent data set for training and improving the Ai.

The tool we have built, Beautiful.Ai, is the world’s first automated expert design system for presentations; and it already handles a significant portion of the design work that has traditionally been left to users. This is the first iteration of Presentation Design Ai.

That said, we have a long way to go before Beautiful.Ai can learn over time and instantly design you a presentation that is both objectively and subjectively beautiful. There is still plenty of “fuzzy logic” built into our system that needs to be tested in the real world, and there are some design decisions that we are not yet confident enough to take fully out of the hands of our users—everything from color selection to image placement. But we truly believe that we are pioneering the development of Design Ai, and we hope that we won’t be doing it alone.

The past, present and future of Design Ai

Our goal is not, and will never be, to replace designers with software. Rather, we want to free designers from repetitive work and give users who don’t have access to a designer the benefits of working with one. And, in that sense, we’re not alone.

Helping people to visualize their ideas through software is not a new concept. To date, though, software has solved this problem by “pre-designing” assets and giving users the flexibility to customize them slightly. This is the approach that Microsoft has taken with SmartArt and Canva has applied to everything from posters to presentations. These tools assume that you might want to communicate an idea visually at some point, and tries to make sure that you can find that visual when the time comes.

#5 Building Robot Illustration-1.jpg

We believe that our core technology Design Ai is the next logical step in this evolution. Instead of offering a library of visualizations, it seeks to interpret your intent in real-time and automatically create the best possible visual for you. And we have good company in this effort: Google recently introduced Autodraw, a really fun tool that attempts—with some success—to turn your rough sketches into professional-looking icons. (Seriously, you should try it.) Similarly, Getty has released a pen that designers can use to sketch images as a way of searching for stock images. Although they might seem trite, these are the kinds of technologies that will lay the groundwork—and gather the data—that will make Design Ai possible.Our goal at is to advance Design Ai by applying it to real-world challenges that our users face every day. We will start with presentations with our first product, but we believe that our technology will be equally valuable in the creation of everything from documents to dashboards to websites. And we also believe that we will not, and should not, be the only company committed to advancing Design Ai. We are looking for experts in both design and artificial intelligence to work hand-in-hand with us to build a system that brings good design to everything it touches.

We believe in a world where good design is a given, and where designers continue to push us even further. And we hope you’ll join us on the journey.

Want to try for yourself? Sign up for your free account today!  

Jared Bloom

Jared Bloom