Read This Before You Build That Chatbot.
A four-step guide to using artificial intelligence to solve real problems.
Tweet “AI is the future of branding” and watch the follows roll in. And to be fair, that statement is true. The hurdle to getting there isn’t that these new technologies aren’t available, it’s that we’re not sure how to deploy them, how we should think about them, and most important, where we even start with AI.
Most of the AI talk has been focused on chatbots, and while it’s fine to build a chatbot, the AI conversation is about so much more than text. So here’s where I’m going to ask you to put the brakes on that chatbot brainstorm you’re moments from kicking off and ask that we take a moment to think about how to AI like a pro.
AI is everywhere. Every week, there’s a new technology touting a new solution followed by a slew of articles on companies that are shoving artificial intelligence into gadgets to make them more useful and powerful. Amazon Echo is open for just about anyone to develop skills for Alexa, the voice-activated assistant in its smart-home device. Uber’s predicting where you’re going to be before you do. Google’s “Show and Tell” algorithm can be trained to recognize objects in photos with up to 93.9% accuracy, and even write captions for them. Facebook Messenger’s AI engine, DeepText, will be able to understand the textual content of messages to detect a user’s meaning, whether she’s trying to call a cab or order pizza.
And yet, with all the times we encounter AI over the course of a day, if I were to ask you right now, “What is AI?” you would probably answer with one of two things—it’s a chatbot that hardly ever gets things right the first time, or it’s a near-sentient robot-being bent on undoing the human race. But here’s the thing about AI: It isn’t a thing. It’s not your smartphone, an app, or a killer robot. And guess what else, it’s not the way you interact with it—whether through a text, a voice, or smoke signals. At its most basic, AI is the science of bringing a collection of invisible technologies together to do things that usually require human intelligence, such as learning, understanding, and problem solving. (If you want to hear it from Alan Turing himself, go here.)
Great, now you know something. The clouds are beginning to part, "But wait," you might say to yourself. “If I’m not making a conversational interface (read: chatbot), what am I making? Where do I start!?”
Like any other service or product, the first thing you’re going to do is identify the problem you’re trying to solve. Figure that out, set your goal, and you will make something worthwhile…or at least not total garbage. Now, let me take you down a typical path to building a solution with AI technologies.
The path to AI.
Find the data.
So how do the Googles, Facebooks, and Amazons of the world build AI? They start with data, some that they collect from their users, some that they find in other places. Here are some common data categories, and how to mine them.
Public: Data scraped from publically available sources, such as government surveys, Twitter, and Facebook.
Internal: Information a company has gathered from its users, from their purchasing patterns to their Web searches.
Purchased: Relevant data bought from third-party providers. Major data vendors focus on consumer interests and spending, finance, and demographics.
Derived: By applying a calculation to different raw data sets, an entirely new data set can be generated. The price-per-earnings ratio and the 200-day moving average are two examples used in financial applications.
Remember the great promise of Big Data? Its time has come! As computer processing power and algorithmic know-how have matured, we’re finally seeing that stockpile of data bear fruit. I strongly believe that so-called dark data, the stuff that companies and organizations collect but don’t use, are merely waiting for a purpose, and it’s up to technologists and marketers to figure out what to do with it. IBM Watson IoT, for instance, takes unstructured data and combines it with other information sources, such as weather and news events, to drive decisions on things like predictive maintenance and worker safety.
Add the smarts.
Data isn’t meaningful until it has structure. Let’s say you have a string of words: Missy Kelley, product design director at Huge, all-around neat person. In order to get value from that statement—Missy equals a person who works at a 1,200-person digital agency headquartered in Brooklyn—you need to run some algorithms against it.
Here’s where you have a choice about what services to use to process your data. If you’re looking for natural language processing (NLP)—a platform that can understand natural language—you can either go with a machine-learning platform, like IBM Watson, which has an NLP component (among other algorithms) that can derive emotional value from unstructured data. Or maybe you’re just looking for something simple like a service specializing in NLP, such as Google’s latest acquisition, API.ai.
For instance, here at Huge, we’re building an internal API to make our lives better. One of our solutions is running topic, tone, and sentiment algorithms against email correspondence to alert project leaders to changes in a client’s outlook.
Choose your interface.
After being improved by computer smarts, the data can be delivered through different interactions. If we’re monitoring the tone of email exchanges, we might choose to alert internal employees via email or push notifications. Or we could decide against alerts and simply feed this information into a visualization of the overall health of the relationship with the client.
A chatbot is just one type of interface. Some amazing services have been built on chat, including WeChat, Facebook’s M, and the forthcoming Siri/Cortana-killer and recent Samsung purchase Viv, which promises to understand the intent of, and have an answer for, any question you might throw at it. But chatbots and NLP needn’t be the default interface. Think about your iPhone. When you receive a call or text message from a person that isn’t a contact, the operating system searches available data to guess who it might be. This is automated without any interaction and displayed back to you visually. Or take Google Trips, a new mobile app that acts like a travel agent to help book everything from hotels to sightseeing based on your personal tastes and time restrictions.
Knowing all this now, how do you figure out what technologies are right for you? Like I said, before investing in any AI solution, figure out your problem. If users aren’t engaging with your brand, ask why. Could it be that they can’t find you through search? Is your PR machine doing its job? If either of those is the case, a chatbot will only make a problem sexy, not solve it.
So go forth and create great solutions with AI technologies. But as you do, ask yourself, “Am I making people’s lives better?” And keep close to your heart the words of my mother, “Don’t use AI as a solution to a problem that doesn’t exist.”