Server-side humans

On delivering distinctly human services at a superhuman pace

September 21, 2015
Jonathan Libov

The film Edge of Tomorrow takes place in a future where humans defend Earth against a plundering alien race. Although everything one would read in today's news suggests that wars of the future will be fought by drones, Edge of Tomorrow would still have beautiful humans like Tom Cruise and Emily Blunt donning exo-skeletons to fight the aliens on the ground, guerilla-style.

When one considers how difficult it is to maneuver in one of those suits, let alone to run away from a grounded, skidding, flaming helicopter... would have to assume that in Edge of Tomorrow's future, exoskeletally-enhanced humans are still better soldiers than drones. Perhaps in this future, artificial intelligence and robotics still haven't caught up to humans' sensory and communication faculties, nor their quality of judgment. Because if they did, why would you ever put Tom and Emily in harm's way like that?

Today's exoskeletons

One of the more interesting trends in tech right now is wrapping humans in technology and (albeit loose forms of) artificial intelligence to enable services that need humans at their core, but couldn't be handled efficiently enough by humans on their own.

The most prominent example is Facebook M, a digital assistant that lives inside of Messenger. M handles incoming requests via AI and natural language processing, and like Siri setting a reminder or performing a search of the web, it can even close the loop on basic tasks like booking an Uber. But when the messaging concierge tasks get tough, the humans get going:

"[Facebook M] is a virtual assistant powered by artificial intelligence as well as a band of Facebook employees, dubbed M trainers, who will make sure that every request is answered...The thing is: that’s a person on hold on your behalf. Facebook’s M trainers have customer service backgrounds. They make the trickier judgment calls, and perform other tasks that software can’t. If you ask M to plan a birthday dinner for your friend, the software might book the Uber and the restaurant, but a person might surprise your friend at the end of the night by sending over birthday cupcakes from her favorite bakery."

In healthcare, consider this bit from the About section of Icebreaker Health's website:

"We’re ashamed that too many people in America don’t have access to affordable healthcare...The solution isn’t more of the same, but technology that helps scale a doctor’s expertise logarithmically so that one doctor can help hundreds of patients per hour. This is the only way to improve access, increase quality and radically decrease cost."

In other words, take all those questionnaires you fill out when you visit the doctor and have software analyze it on behalf of the doctor. This level of diagnosis, what doctors call "differentials" , usually boils down to a rubric anyway. Why not spare the human doctor all the rubric-like work that software is really good at? That way the human doctor can focus only on the two minutes of the appointment that you really need human senses, judgment, and compassion for? Not only is that more efficient, but it frees doctors up to work on the more creative and challenging aspects of medicine that we need human doctors for.

We're also seeing humans deployed in surprising ways by the rash of digital-yet-there's-a-human-on-the-other-end assistants that have emerged: Magic, GoButler, Operator, Pana and other messaging-based concierge services. While there are obvious benefits of deploying humans to interact with consumers in a more conversational, human format than traditional user interfaces, it also defies everything about the "software is eating the world" mantra we've come to expect from new consumer services. The proposition: Instead of consumers taking the time to use a software-based user interface to procure a service or good, for a fee they can just ask another human on the other end of the chat to do it.

This is perfectly rational in a division-of-labor sense: A travel concierge service like Pana promises all the same upside—efficiency and expertise—as a travel agency. The question here is an economic one, one that again defies this zero marginal cost model that defies the Internet software revolution. It's unclear whether consumers are really willing to pay enough to support the non-zero marginal cost of human assistants. This is where many of the unit economics for on-demand services appear to break down.

There's only one way that human-capital-intensive would seem to work as a business model: Augment the human concierge providers with software, possibly even AI and NLP software, that enables these new-fangled "human operators" to process requests at exponential rates compared to traditional call center services. This is presumably the philosophy behind Facebook M and their M trainers. promise to offer these AI/NLP/dispatch services to startups, so they too can leverage human faculties at a superhuman pace.

Services as a service

The way in which this does align with trends is our ever-growing transformation to a service-driven economy. As software and machines take over jobs that were once filled by humans, those humans are increasingly moving into the service sector. As pointed out in an excellent episode of the EconTalk podcast, it's easy to forget that in the middle of the 20th century, massages were something that most Americans only witnessed in the movies because they were only accessible to the rich and famous. Today you easily find massage services in strip malls, research them all on Yelp, or even order them to your home via an app.

This seems to be happening all over again with on-demand services that bring just about everything to your door. Only the rich could afford room service, which is what many of these on-demand companies amount to. Returning to the question of unit economics, do all these companies have something up their sleeve beyond the dispatch layer to make the on-demand labor force work at a superhuman pace? Is the demand-side of the economy really just willing to have an entire supply of humans on the other side of their apps?

There is a long-term future where software (including robots, drones, and self-driving cars) really does eat everything, and there are services like financial advising which already seem ready to be eaten by software. So while it's true that software is eating the world, in the near- to mid-term there's a funny potential outcome: The more humans that are freed from manufacturing and administrative tasks because of software and robotics, the more humans we might see deployed "server-side", armed with software-based superpowers. fulfilling tasks that we thought we needed to wait for really advanced software to fill.