Supporting AI with design — or how to teach AI a lesson or two
There are tons of articles out there around how much AI will disrupt everything in our lives and how it will change everything and make it better. And mostly I agree with all of them (well maybe except the creepy ads that pop up every now and then). There are other technologies we put a lot of effort in and I don’t think it’s going to lead to anything, but that’s a different story.
What a lot of people don’t realise though is how much effort it takes to make a good model and a good graph neural network. Not only it takes time, but it also takes our knowledge and perceptions of life which are highly limited to the bubbles that we designers and engineers live in.
I don’t want to talk about how things will be, but how we can make our technology better and more inclusive quicker. And I think the answer to that is design and people. The best way to teach a kid anything is by giving him tons of lessons by tons of different people. Think about it… the more we learn and the more people we interact with, the smarter we get. We also get a wider perspective of the world and the cultural differences. What’s acceptable in the UK might not be acceptable in China. And that’s perfectly fine. It’s situational and we need to know how to do things in certain places or situations.
I think in order to have a better experience, we need machine learning that interacts with people and learn constantly, rather than using a limited data set that might not be as accurate or inclusive as we might want it to be.
The thing that’s going to make artificial intelligence so powerful is its ability to learn, and the way AI learns is to look at human culture. — Dan Brown
So how do we create an experience where AI should be the super smart thing that can tell us anything? Well we don’t. What we should be doing instead is using the knowledge of the AI to its’ limits, but use human help to teach them and make them better. Google call it Federated Learning and I think it’s the better way of creating experiences.
One thing that Ben Sauer mentions quite often is how we need to make sure that experiences fail gracefully. And he’s absolutely right. But what that also gives you is an opportunity. An opportunity to use your users to train the model. But how do we do that so it’s not annoying?
Well rather than talking about it, I’ll give you few examples of how we can use design to make AI and machine learning better.
One good example is Maslo which is something like a personal notebook. You can see how much and what you talked about and it gives you a good idea about your thoughts over time. But how Maslo uses us is by connecting our thoughts with feelings and I think it’s a great way to start bridging the massive gap between computers, words and feelings.
One of the obvious ones is Spotify. The more music you like the better results you’ll get for your discover weekly podcast. It’s not only looking at your historical data. It looks at your recent plays which gives you relevant results depending on what you listen to.
Chat suggestions are also great example and are currently used by Faceboook Messenger, LinkedIn, Gmail and many more. By selecting the more relevant ones we help make the experience better with each interaction.
I’ve also worked on a tool called Qubie that gives recommendations on how to improve your team where we used people to add sentiment to their weekly suggestions by which the model could learn as well as rating suggestions to make better ones in the future.
At Attest we’ll be using similar techniques to filter out bad responses from surveys and eventually make better sense of the data we have.
I’m sure you can think of many other examples which is great. I think before we try to catch the starship to where we get the exact information we need in the fastest possible way, we need to invest time in teaching. I encourage you think of the opportunities you have and make use of them to help build a more inclusive tech. The more people we involve in it, the better. And what better way of using design than that.
