Four ways to make AI products that people will love

In the 1890s, bicycles were thought to be dangerous contraptions that could cause diseases including appendicitis and something called “bicycle face.” Today, many people are making similar claims about AI.

After three years of excitement, the novelty has worn off and we’re starting to see articles suggesting that AI is making people dumber, that it’s ruining society, or that it’s causing mass delusion.

As the founder of three artificial intelligence startups over the past 13 years, I’m unapologetically bullish about the “Cambrian explosion” of AI inventions. I believe that AI and robotics — AI’s ultimate physical manifestation — have the potential to make enormous, positive differences in our lives in ways we can barely imagine today. I also recognize that many people are increasingly nervous about it.

This is an encouraging sign: It means people recognize AI’s power. Technology leaders have a responsibility to respond to that awareness productively, not by arguing, but by building products that are so useful, beneficial, and even charming that people cherish the opportunity to interact with them.

We are more than capable of addressing the risks in order to unlock the benefits of AI, which will far outweigh the downsides. Here are four steps to building AI products that people love.

1. Start with what people need

First, the classic design principle: Start by focusing on what the user needs, not what technology can do. It’s all too easy to end up with a solution in search of a problem.

At a previous startup, we were testing a competitor’s AI product that analyzed home power usage to spot costly issues. A week after installing it, a colleague got an alert: His pool pump was broken. The problem? He didn’t own a pool!

Our product was different. When we onboarded new customers, we simply asked them to pick the appliances they owned from a list. One of my engineers at the time protested: “That’s cheating!” As if using a checklist, instead of millions of AI parameters, was somehow beneath the dignity of an AI-powered startup.

Sometimes, as engineers, we get carried away with the thrill of solving a hard problem or using a shiny new technology. Focusing on the user’s needs often leads to simple changes that substantially reduce complexity for everyone.

2. Understand what AI is good at

With any new technology, understanding how to use it well starts with knowing its limitations: How and when will this technology fail, and what do we do when it does?

For AI, we can often measure failures in two dimensions:

• False positives: A system alerts you about a pool where one doesn’t exist or stops an autonomous vehicle for an imaginary obstacle.
• False negatives: A system can fail to detect a real pool’s wasteful power use, or a self-driving car might not stop for a real obstacle.

“Precision” is a measure of false positives, and “recall” measures false negatives.

Here’s the key insight: AI can be optimized for either precision (fewer false positives) or recall (fewer false negatives). But optimizing for both is extremely expensive and time-consuming.

There are a few applications, like robotaxis, where optimizing for both is so important, due to safety, that it’s worth investing tens of billions of dollars in research and development. For the rest of us, the key to making useful AI products lies in making smarter design decisions. And to do that, we must first decide: Do we optimize for precision or recall?

We built our home power product to catch every time something wasted power. In other words, it was good at recall. But we knew that dumb mistakes such as identifying a non-existent appliance (poor precision) would destroy customers’ trust.

Instead of trying to increase precision at great cost, we just asked the customer what appliances they owned. Problem solved.

3. Empower people to assist AI

Think about how robots can complement human effort and free people of mundane, dangerous, or difficult tasks. Too often, the discussion of AI and robotics focuses on whether they will replace humans. This overlooks the opportunity for humans and AI to work together. Humans can assist AI with the inevitable tradeoff between precision and recall.

With a well-designed product, each side complements the other. We can build AI to detect what humans find difficult to notice, like wasteful electricity usage patterns, and achieve great results by focusing on either precision or recall.

Meanwhile, humans can be responsible for the other dimension, such as knowing what appliances they own, which is difficult for the AI. By freeing people from difficult, tedious or time-wasting tasks that they don’t want to do, like analyzing data for anomalies or scanning text for typos, AI can enable them to engage in more fulfilling and enjoyable work.

If your product does all this, congratulations: You’ve gone further than many products ever get.

4. Exceed expectations

However, a crucial final step is required to truly win people’s hearts: You need to go beyond the basics and add something surprisingly wonderful.

It’s hard to predict what this will be, but you’ll know it when you find it. For example, with smart speakers, the core function is playing music. The unexpected aspect is that they can tell jokes and play games, making them endlessly entertaining for children.

For the friendly delivery robots that my company, Serve Robotics, makes, adding blinking “eyes” and individualized names helped people see them as cute creatures rolling down the sidewalk. It has nothing to do with delivering burritos, but the names and the eyes humanize them.

Kids go out of their way to talk to the robots, and adults cross the street to take photos or even give them hugs. This is especially important because the people who interact with our robots the most often aren’t our customers at all. They’re just regular passersby.

Like the bud vase in a VW bug, it’s the charming detail that takes a product from merely good to delightful.

Delight will make the difference for AI and robotics

AI presents limitless possibilities to rethink and reshape the way we do things in nearly every field. Over the next few years, there will be disruptions and unanticipated consequences, as with every technology revolution. But there will also be incredible advances that make our lives better in so many ways.

While we can’t predict every breakthrough, we can shape how they unfold by ensuring AI development serves human flourishing rather than mere technological advancement.

It might sound like an optional “extra,” but today’s AI-powered products need delight just like bicycles in the 1890s needed some tassels on the handlebars: They’re the key to making people love them, leading to widespread adoption, success, and better living for all of us.

About the author

Ali Kashani co-founded Serve Robotics in January 2021 and has served as its CEO and a member of its board since then. Prior to that, he was vice president at Postmates Inc., an on-demand food delivery platform.

Prior to Postmates, Dr. Kashani was the co-founder and chief technology officer at Neurio Technology Inc., a smart home technology company acquired by Generac Power Systems Inc. He is an inventor with 15 granted or pending patents.

Kashani received both his Bachelor of Science in computer engineering and his doctorate in robotics from the University of British Columbia and was awarded Natural Sciences and Engineering Research Council of Canada’s Alexander Graham Bell Canada Graduate Scholarship. He was a guest on The Robot Report Podcast in March.

The post Four ways to make AI products that people will love appeared first on The Robot Report.