How Kai Fu Lee captured the essence of Misty

Consumer expectations are sky high and undeliverable for “home” robots.

Kai Fu Lee knows a thing or two about artificial intelligence and, therefor, the likely trajectory of robots enabled with it. He is a well-known expert on artificial intelligence having received a Ph.D. at Carnegie Melon and subsequently holding AI jobs at Microsoft, Google and elsewhere before going over to venture capital.

In an article by noted tech journalist John Markoff in the New York Times last month titled “What Comes After the Roomba”, Kai Fu has a parting line that accentuates the entirety of the article:

“The problem,” said Kai-Fu Lee, a leading Chinese artificial intelligence researcher who is now a venture investor, “is that low cost plus high expectations plus no patience makes it difficult to make a great product.”

As other posts from yours truly have spelled out this is exactly the case and it’s why companies such as Kuri and Jibo have folded shop and why its our expectation that home robots such as Aibo, Sanbot and other “home” or “consumer” robots are likely to fail.

Robots are Expensive

Indeed they are. Robots that do much of anything beyond the toy category must have laptop-scale (or higher) processors within them in order to begin performing rudimentary machine learning (one of the disciplines of AI), handle a variety of simultaneous tasks including listening, speaking, motoring around and possibly lifting/manipulating. Processors, despite Moore’s tremendous prophecy and progress, still aren’t a few dollars (not the ones that can handle a variety of challenging computational tasks). This leaves a robot maker with a decision to place all the computation in the cloud — which may be workable to mass-market types of machine learning tasks and/or ones that don’t require instantaneous, onsite learning and adaptation (like we expect from our future consumer robots) — or “at the edge”, in the robot.

Misty has two quad-core Qualcomm Snapdragon processors inside — devoted to different components of her work: one component for processing audio and camera inputs, including autonomous navigation (SLAM) and the other component for processing all of the interfaces our software developer customers use including access to all motors and sensors (aside from the microphones and cameras), local storage of media and programs, etc… In fact, future generations of Misty may have several processors, each devoted to specific components or functions of the robot.

Robots are also hard to power. Consumer expect the moon these days. They expect their consumer robot will likely be independently operating, without being at a charging station, for half a day or maybe a full day. Meanwhile, they’re expecting that robot will move around (motors consume power); interpret their surroundings (interpretation requires computation which requires power); network with the outside and more. Meanwhile, batteries are expensive, heavy, dangerous and just not all that long-lasting when it comes to the tasks a consumer robot will likely have to fulfill for their owners. We’re still a long ways away from solid power plants that are affordable while also powering the robot itself. Sure, maybe an outdoor robot can harvest power from the sun (when it’s not cloudy) but an indoor home robot doesn’t have much of a “scavenging source” from which to operate. And, it’s not like it can constantly drag, extend and retract its power cord as it cruises around a house.

All of which makes robots expensive. When this expense gets applied to one task (like vacuuming) it’s very rare that expense is justified in the consumer’s mind. Something like vacuuming that’s incredibly mundane stands a chance but many other single-task (or few-task) robots that are expensive to build can’t overcome that value proposition from the consumer.

Consumer Robots can’t deliver on expectations

For as long as we’ve had television and moving pictures we’ve imagined science fiction. We’ve imagined “smart entities” that weren’t human but were capable of most / all of what humanity can do. Yes, the basics like walking, seeing, speaking but also more complex tasks such as identifying, interpreting, creating, manipulating complex objects, conversing fluidly in the chosen language of the speaker.

We’ve seen them — the Rosies, the C3POs, the iRobots of the movies. We all want them. We’ve been trained “that’s what a robot is”.

So, when a consumer hears “robot” and “home” they immediately go to those images and imaginary entities that have mass awareness. It’s hard for the human brain not to go there. So, go there we do and with it come all of those lofty expectations that, from a piecemeal perspective might exist (million dollar robots that can grasp complex objects (only); affordable “robots” like Alexa or Google Home that can speak some languages some of the time; robots that can walk up stairs (but that’s about it)or see enough to build a car.

Consumers have little patience

Lastly, we’ve been well-trained by Apple to expect “world class” products.

(Tangent: companies like Apple, Samsung, Comcast, Misty Robotics and others use a metric known as “Net Promoter Score” — where Apple and Samsung’s NPS scores are >75 which, according to NPS experts, is “world class.)

Companies like Apple produce virtually flawless products. Those companies have so much capital that the amount of resource they can deploy is virtually limitless and they typically sell tens to hundreds of millions of units of the product they invest in which creates an incentive to throw thousands of engineers at the product to make it virtually flawless. Televisions and computers are now decades into their maturation cycles that every minute imperfection has long ago been squeezed out of the manufacturing process. Chinese electronics manufacturers, equally, are decades into their own maturity cycle that their internal quality control processes and the sophistication of the supporting equipment is unparalleled.

Which leaves almost the entire universe of products inside your average Best Buy (i.e. more than, what, 95% (by volume) of all consumer electronic stock keeping units (SKUs))? I.e walk into any Best Buy and you’re not likely to see a product that has a high return rate for warranty claims nor a product that can survive if its NPS is “average”.

The other aspect of impatience comes from “edge cases”. Consumers, by definition, are highly unique in their personal preferences, their domiciles, their family construction, their selection of pet/no-pet, etc… Consumer product companies go through, often, years of user testing and in-field testing to find and iron out all “edge cases” so that consumers will continue to rate the product very highly. Anyone who has had Alexa or Google Home say “sorry; I don’t understand” knows all about edge cases. With physical hardware products, the edge cases get harder because there’s virtually no way for a manufacturer to test all of those edge cases unless either they’re well-known (TVs and computers are decades old) or wealthy (Apple).

This impatience, transferred to the incredibly nascent domain of “consumer robots” completes the trifecta of insurmountable obstacles to being able to sell a consumer robot at sufficient volumes, with sufficient profits to stay in business, with sufficient quality to overcome impatience and able to deliver anywhere near the functional features and benefits consumers have been taught to expect from the movies.

For Developers Only

Which brings us to Misty Robotics and how Kai Fu has captured the essence of what Misty is all about. Misty is for software developers only; consumers need not apply.

Because software developers have myriad tasks within their businesses, or businesses they’d like to start, fun tasks they’d like to explore or, as beginning developers, stuff they’d like to learn about they can afford a robot that packs a lot of punch. They’re not looking to scrimp and squeeze and save every last ounce of penny out of a robot, they’re more interested in whether a robot could possibly solve a business problem they have or put the joy back into education. It’s well worth paying for the additional processing capability and stomaching the relatively small battery power duty cycle — especially if a business or market problem comprising tens of hundreds of thousands of dollars is at stake.

Because software developers, themselves, know the ins and outs of software development they understand all there is to understand about “bugs”. They get that complex machinery and complex algorithms don’t always work the first time out the gate. They also understand the notion of “edge cases” and that “edge cases” take months, if not years, of perfection to iron out. In fact, they expect bugs and edge cases. In fact there’s almost a joy and pride that comes from a software developer when they help a supplier track down a particularly pernicious bug or edge case. What software developers expect is responsive and technical support to help them sort through the environment of the platform — separating any bugs they might have introduced from bugs in the core platform.

Lastly, because software developers, today, have absolutely, positively no prayer of being able to build their own robot to try to solve their business or market problem, or to learn about the complexities of robots, the expectations of what a robot platform can do for them are incredibly different from a consumer’s expectations of what a robot can do in the home. There are no (AFAIK) characters from movies that have been a “blank canvas robot” upon which the proverbial Gene Wilder can bring their particular Frankenstein to life.

Misty is the Kai Fu Lee anti-trifecta — affordable (for their purposes), high quality for experimentation and learning and filled with the kind of developer support, software updates and community supporting it to enable sufficient patience.

As software developers invent uses for the office and home it is they who will cover the edge cases, eliminate the bugs, offer hundreds or thousands of uses to make the robot useful and, over time, create the optimal consumer robot of the future. It’s just not ready for consumer yet; just developers.