Artificial Intelligence is all over the place. If you attended our post-summit AI symposium we held during Global Summit 2018 in San Antonio, you certainly got a taste of the varied use cases where AI can make a difference. But what does it take to build an AI-powered application? Do you start implementing tedious data-gathering processes for training your models? Or do you first scour the job market for a handful of those elusive data scientist unicorns, which itself may take years?
Meet Monique. As you can see from the picture, she uses a laptop and delivers presentations.
Not a big deal, at first.
I first saw Monique on television a month or two ago and learned a few things about her that might give you a different perspective. Monique is blind. She suffered from congenital glaucoma, which paralyzed her optic nerve and took away her eyesight when she was still a toddler.
But that hasn’t prevented her from being an optimistic, outgoing and entrepreneurial woman. Amongst other jobs, she worked as a receptionist, wrote a book, and lectures to various audiences. For the lectures, she carries around that laptop and operates it with a very impressive efficiency using a device that reads things out loud and leverages braille. Reportedly the only thing she really can’t do by herself is correctly aim the projector – her own, of course.
And by the way, Monique is a mother as well.
Monique also has a passion: photography. She maintains a blog on Facebook and loves the interaction with her audience when she writes an article or posts a picture taken with her mobile device, which she operates probably just as fast as you and I. Unfortunately, taking pictures isn’t particularly easy when you are blind, and that’s where the television program, and ultimately Artificial Intelligence, came in.
In the program, hosted by a comedian and scientist, a team of “makers” designs solutions for people with physical impairments who have trouble with everyday tasks. In this episode, they wanted to help Monique with her photography challenge.
I’ll fast-forward a bit through the different iterations, but the “maker” in this episode was a software architect who assembled an app from a variety of AI-powered services and other features ranging from simple to complex:
- First, she leveraged the gyroscopic sensors in the device to check if the phone was upright. The app provides an audio tone with varying pitch according to the tilt. This isn’t quite AI yet, but it certainly helped Monique a lot in steering clear of ill-aimed pictures, a primary concern for her before.
- The app also measured the overall light and contrast in the picture, to make sure the picture isn’t all black, all white or just has black-and-white silhouettes when facing the sun. Again, this is not really AI, but already a little more involved than a direct sensor read. As with the tilt check, an audio hint indicated when the picture risked getting a little imbalanced.
- Clearly the biggest win was being able to embed a piece of Artificial Intelligence that recognized objects. When pointing the camera at one of a few hundred types of objects it was trained on, it would tell her in real time “small bottle next to laptop” or “person in lower right”. A bit creepy at first, but invaluable for Monique.
- The final important capability they embedded in the app was face recognition. After taking a few pictures of a person and telling it the name of this person, the camera app would recognize him/her in pictures and, quite importantly for Monique, save the names of people in the shot in the picture’s metadata, so she could more easily look up the pictures and recollect her memories of the event afterwards.
All in all, a very smart app and good television. Now, what really happened, and what’s interesting for us solution providers with an interest or opportunity for AI? The fact that this got built entirely within the budget of the Belgian public broadcaster, with the help of a volunteer software architect sacrificing her spare time. That’s interesting. There was no Silicon Valley behemoth that built this from scratch as a single Machine Learning project, but rather a smart combination of existing tools, APIs and services. Yes, some of those services were hosted by a mega vendor from Seattle, but the interesting thing is that Monique’s app was assembled rather than developed.
This development paradigm, leveraging APIs and services rather than building everything from scratch, is not new. It’s the core idea of service-oriented architectures, of microservices, of low-code development. It’s actually a way of working that hinges on a platform’s interoperability capabilities. This development paradigm is becoming even more important in the context of AI, as they will tell you on interest groups such as PAPIs or large cloud vendors like Amazon, Google and Microsoft.
In summary, building amazing AI-powered applications doesn’t mean you need an army of data scientists to get started. By identifying the individual tasks within “amazing” you can often find APIs, SDKs and services that will get you a long way. Just good development skills, a solid interoperability platform and a drop of imagination are all you need to start delighting your customers with AI-powered apps.