Artificial intelligence and ‘smart things’ are the biggest topics in tech today, and will remain so over at least the next year. If you’re working in software or hardware development, you’re probably asking yourself how you will find ways to improve your user experience by adding a layer of AI. But where to begin?
The good news is that you can start doing a lot of things that will get you along the way, long before you start coding the first smart feature into your app. I made a list of four things that you can start working on today:
Step 1: Learn the difference between smart and artificial intelligence.
First, to build a smart app, you don’t actually need artificial intelligence, and it’s important to make the distinction.
Artificial intelligence is when a machine is able to learn and predict based on historical data. As of last fall, for example, all Tesla vehicles have the ability to drive themselves. These self-driving cars will consistently improve their ability to protect drivers by consolidating data across all drivers and learning how they behave in certain locations.
Smartphone programs that are using AI include Google Assistant, which mines millions of Google search queries from years of searches to better help users answer their questions.
Smart on the other hand is not inextricably linked to artificial intelligence, as it can exist without AI. An everyday example of this is a pizza delivery app that suggest an option for users to place the same order they did the last time. This app does not rely on AI. It just needs to know that pizza consumers almost always place repetitive orders, so simply storing the last order of the user will make it ‘smart’.
Smart functions do not need to be integrated with machine learning capabilities or other forms of AI to help the user. They just need to proactively offer the right service at the right moment.
Step 2: Start thinking about the balance between proactive and intrusive
Smart or artificially intelligent applications are all about predictions – what will the user need next? While getting that prediction right can create a wonderful experience, getting it wrong can be devastating for the user experience, and thus for your app. You want to be at exactly the right intersection of simplicity, interconnectivity, and proactivity. In short, your app should be wonderfully intuitive, without being intrusive or creepy.
Microsoft’s Clippy is probably the best example of an application that was trying too hard to be helpful when people didn’t need it. (Also, research before the launch showed that women thought it was leering at them, a fact that Microsoft chose to ignore).
The problem with Clippy was that it kept interrupting us to tell us how to do stuff we already knew how to do. It didn’t add any value, so in our collective rage, we killed it.
Once you’ve ascertained that a smart feature will add value, you need to think about how to balance that value with the level of intrusion. Let’s take Facebook’s weather integration as an example. Suppose it’s been raining for two weeks straight. Do you need Facebook to remind you of that every day?
Contrast that to summer. The weather has been brilliant for two weeks. But this afternoon, during your weekly bike trip with your buddies, an afternoon thunderstorm is predicted. Here, a push notification does make sense.
A rule of thumb to consider about notifications and in-app suggestions. Ask yourself: would you send a text message to your spouse reminding them of what the app is notifying you about? If the answer is no – if sending that text feels weird or like a needless interruption — you might want to rethink the notification.
Step 3: Set up a big data platform
If you do just one thing in 2017 to future-proof your app it should be this: start collecting data – way more data than you’re collecting today. Data is the raw material for smart applications as well as predictive AI. Even if you only start developing actual smart features next year, you will be thankful for a ready-to-use, proprietary collection of actual user data.
The more data you collect, the more you’ll be able to tap into past patterns, as well as the environmental context of app usage, to improve the user’s experience. Apps can use this data to make a contextual decision based on where the user is headed, what their daily routine looks like, and how they usually move around.
Surface level analytics, such as number of clicks per hour, won’t be enough. Think broader than that: useful data can come from numerous sources and in multiple formats: online and offline, photo and video, clicks and contextual data, etc.
To collect raw, big data, set up a big data platform. Oracle published a particularly useful guide to help you with this.
Step 4: Look for plug and play AI (but don’t expect too much)
Before you start building anything, make sure you do a thorough search of the tools available to help you. Software development kits and AI-as-a-service are making it even easier to take raw data and apply AI on it.
Neura takes contextual data that your app collects and maximizes its potential for your marketing team. It can trigger your app to notify a user at a certain moment (“just after going for a run”) or in a certain location (“retail stores”). A health app could recognize when a user has just finished exercising and suggest they stretch or offer up a smoothie recipe to replenish nutrients, for example.
The biggest names in technology are engaged in a turf war to offer AI-as-a-service. Azure, Microsoft’s public cloud, offers one of the most advanced Machine Learning-as-a-service service with Azure ML, as well as numerous AI APIs under their Microsoft Cognitive Services APIs. Amazon Web Services, Amazon’s cloud computing arm, once exclusively focused on IaaS is now heavily investing in AI-as-a-service APIs, with the recent launches of Amazon Rekognition for deep-learning facial recognition and Amazon Lex for Natural Language Understanding.
However, if AI becomes the core of your business, know that you will hit a wall. Plug and play solutions will only get you so far – eventually you will need a custom solution that addresses your needs. Hopefully, you’ll still be able to count on these cloud giants, as they’ve open-sourced some of their AI frameworks, just have a look at Google’s TensorFlow, or the Microsoft Cognitive Toolkit.
Feel free to share your experiences and learnings with me on Twitter. I’m @ffeytons
This post is part of our contributor series. It is written and published independently of TNW.