Designing Invisible Products "Product people - Product managers, product designers, UX designers, UX researchers, Business analysts, developers, makers & entrepreneurs 16 November 2016 True Artificial Intelligence, Artificial intelligence (AI), Humanised Products, Invisible Products, User Interface Design, Mind the Product Mind the Product Ltd 583 Dennis Mortensen talks about designing AI products at ProductTank NYC Product Management 2.332
· 2 minute read

Designing Invisible Products

As we created our AI assistants – Amy & Andrew – to help people book meetings painlessly, we encountered quite a few product design challenges which were, quite frankly, completely novel. When designing an artificially intelligent agent that humans would enjoy interactive with via conversation, the factors you need to consider and the hoops you have to jump through can be quite unexpected.

Defining the Universe in Which Your Intelligent Agent (Product) Exists’s AI assistant, Amy Ingram, does one thing only – schedule meetings. We wanted to design an agent that could take over the entire task of scheduling the meeting start to finish. Once you define the universe in which your agent exists, you have to go about figuring out what data you need to collect and annotate. had to teach AI to understand a phrase like “Do you have time to meet up later today, tomorrow or perhaps early next week? I am free most days after 1PM.” It took the team over a year just to define “time” as a concept in the language of meeting scheduling!

Making the Choice To Humanize Your Technology (Or Not)

When designing an agent that exists only over dialogue, you have to make the decision as to whether or not to humanize your agent – there is no in-between. Companies like Amazon (Alexa), Apple (Siri), and (Amy and Andrew) have chosen to do this; whereas companies like Google (Google Assistant) have not. If and when you do decide to humanize your agent, you really have to try to make that work, by naming your agent, having them show empathy, and making them believably human. However, we are very clear and open about the fact that our agent is AI, and I don’t advise trying to fool anyone! wanted to humanize its agent to encourage people to speak to it as if it were human. This is both better user experience and better for data science.

What It Means To Build Invisible Software

You don’t operate’s agent via any visual interface. There’s no app, no password, and nothing to download. The product exists entirely within email, over dialogue. Despite the lack of visual interface, there’s a lot of work that goes into not only being able to read the incoming information, but also into how the system will respond to that information.

Instead of hiring a UI/UX designer, we created the entirely new role of AI Interaction Designer to own the “voice” of our AI assistants by crafting dialogue models, scenarios, and working in appropriate emotions such as empathy, politeness, and professionalism.

Wrapping Up

I’ll boil this down to a few key things to consider if you’re involved in designing artificially intelligent agents. First of all, you have to fully understand the domain you want your agent to operate in, and be able to completely define that domain – and the human foibles within it – in software. Secondly, you have to decide exactly how you want your customers to use this agent, and from there decide whether or not you want to humanize – bearing in mind that the kinds of interactions it can and can’t have will vary depending on this decision. And given that your product will likely be ‘invisible’, more rests on this question of interaction than you might at first suspect – be prepared to invest a ton of effort into getting that right!

Comments 0

Join the community

Sign up for free to share your thoughts