Bots should be toasters, not knock-off humans
Since the creation of the concept of robots, they have usually been imagined as humanoid creatures, modeled after our own appearance: head, torso, arms, legs and, so on.
Even though there’s no good reason why robots should have those things, people tend to imagine the future by taking a facsimile of what we know today and projecting something very similar in the future, with an added twist of Future-ness.
There are some robots that aren’t simply knock-offs of people — R2-D2, the Daleks, and many others — but they usually have some human characteristics, like a “head” area with “eyes” and a “legs” area (wheels). They are more the exception than the rule.
In the 1950s and ’60s, a wave of optimism emerged around robots that would, it was hoped, do a lot of our drudge work in the future. They looked like humans and mirrored our conceptions of housework and tiring manual labor: The Jetsons robot, Rosie, wore a housemaid’s outfit.
At the same time, actual robots went into mass use in the US, but they weren’t called “robots” — they were “appliances.” Dishwashers, washing machines, toasters, and and the like became ubiquitous: single-purpose devices that did one thing well, and other things not at all. (If you used your dishwasher to make toast, the results would be bad.)
The “humanoid error”
With the bots and AI boom of the last couple of years, bot-makers have fallen into the same trap the Jetsons creators did. Instead of making simple, focused bots that work like appliances — strictly limited to specific tasks and handing the rest to other specialist bots or to humans — the bot-makers tried to fake it.
They presented bots as much more capable and general-purpose than they are, and they pitched bots as able to do “jobs” as opposed to tasks. The implication of Rosie’s housekeeper attire is that it’s multipurpose, like a human, so it can dust, wash the dishes, do laundry, and so on. These general-purpose bots do not exist in reality.
By overpromising and under delivering, consumers feel stung by failed bot interactions and less willing to use bots in the future. After enormous hype about “conversational commerce” and bots superseding the apps paradigm, there will soon be a bust, with brands and software companies growing more skeptical.
We need more toasters
The better approach for businesses adding bots is to deploy them only for simple, predictable tasks — where they have been designed specifically to accomplish this one thing and will, generally, not fail at it. Ring-fence the bot for just that target task, and put in an “escape hatch” to detect when the bot is outside of its comfort zone. You wouldn’t use a dishwasher to toast bread, so don’t use a bot for tasks it was not specifically designed to do.
Our customers — large banks, telcos, retailers, and others — have told us repeatedly that it’s not realistic to put generalist bots in front of customers: there is too large a variety of questions and communications styles across any customer base numbering millions of people. And when they have tried generalist bots that attempt to take on too wide a role, they have worked poorly.
One large European bank started with some “appliance”-type bots that address only 70 specific topics out of the many thousands customers ask about. The bots tackle those questions, while humans handle the rest. A wireless carrier we work with launched bots in January to automate common interactions and free up staff to answer more complex questions. For example, processes like “activate my SIM card” are handled by bots.
A kitchen full of appliances, not a toaster-fridge
Apple CEO Tim Cook mocked a combination laptop and tablet by saying that “you can converge a toaster and a refrigerator, but you know those things are probably not going to be pleasing to the user.”
With bots, the situation has been worse. Companies have launched toaster-fridge-dishwasher-blender-microwaves in the name of mimicking the “humanoid” experience: talking to a general-purpose bot that can approximate how humans speak without the need to pay a human to do it.
What works better are single-purpose specialist bots — explicitly labeled so consumers know what they are dealing with and what to expect — that do one thing well, rather than multiple things badly.
This post was originally published on BusinessInsider.com