This is a cross-post from the Stitch Fix tech Blog: Multithreaded.stitchfix.com.

More Human Humans: One way in which our lives can be made better by ceeding tasks to machines.

Machines are going to take over the world and leave us humans without jobs. This is the meme going around in mainstream business books on the topic of Artificial Intelligence (AI). This is understandable as the number of things that machines [2] can do better than humans is increasing: diagnosing medical conditions, analyzing legal documents, making parole decisions, to name a few. But doing something better doesn’t necessarily make machines an alternative to humans. If machines and humans each contribute differently to a capability, then there is opportunity to combine their unique talents to produce an outcome that is better than either one could achieve on their own. This has real potential to change not only how we work, but also how we understand our experience of being human.

Different Talents

Machines and humans have very different talents when it comes to processing information. This shouldn’t be surprising given their very different DNA makeup. The abilities of each needs to be appreciated in order for us to evolve the way we work together.

What is it that machines are good at? Machines are exceptional at performing rote calculations — adding or multiplying sets of numbers, for example. This is precisely what they were designed to do with their switches, binary code, logic gates, etc.

What are humans good at? Relative to machines, we seem to excel at several things — improvising or doing non-routine work, leveraging ambient information, and relating to other humans. This is precisely what we were designed to do. These innate abilities are likely the result of evolution. In our ancient lineage individuals with a penchant for these things were more likely to survive and therefore pass on their genes (e.g. being better at absorbing ambient information increases the odds of spotting both predator and prey — each an advantage for survival).

Where is the tension?

Machines and humans seem to have very different talents. For us humans, ceding the domain of rote calculations over to machines is hardly controversial since most of us are uninterested in performing such tasks. We are capable — who do you think taught the machines? It’s just that we can’t do it at the scale or accuracy of the machines, and even if we could, most of us would not enjoy performing such tasks for 8+ hours a day. Most of us prefer to do the things that we are uniquely suited to do, such as relating to other humans. From our point-of-view the division of labor is welcomed; we ditch the unwanted chore of rote calculations and retain the tasks that are innate to us.

However, contention has emerged because it turns out that those rote calculations can be sequenced and organized into algorithms and then embedded in applications to do some interesting things. They enable automation of medical diagnosis, legal research, parole decisions, etc. Those are things that constitute jobs. To the degree that these jobs can be distilled down to a series of rote calculations, machines are going to be better at them (and almost surely cheaper).

Is that all there is to these jobs — a series of rote calculations? That is certainly not how we humans perform them- not explicitly anyway. We use our powers to improvise, leverage ambient information, and relate to other humans. These abilities provide value that is different from the rote calculations and that means there is opportunity for additive contributions — perhaps even synergies. This is the central idea of this post: it is possible for humans and machines to combine their unique talents to produce an outcome that is better than either one can produce on their own.

There is precedent for this.

The industrial revolution taught us that machines and humans can combine to produce new levels of efficiency. The introduction of machinery into processes like manufacturing, farming, and construction greatly increased output. We welcomed the division of labor that led to these efficiencies: machines reduced the amount of physical labor required by us humans, shifting our contributions to more enjoyable tasks. Granted, these are examples of physical tasks. What about intellectual tasks? Can the Information Revolution bring a similar phenomenon? Again, there is precedent for this.

Consider the game of chess. It’s long held as the pinnacle of human intellectual ability [3]. Yet, in 1997, Chess champion Garry Kasparov was beat by the IBM Supercomputer, Deep Blue. It was just a matter of time before a human would fall to the fast calculation abilities and brute force methods of a supercomputer. But, later a new style of chess emerged. “Freestyle chess” is where the player is allowed to use the assistance of a device of his/her own (typically a much more modest device than a supercomputer). This sort of teaming up — humans along with a machine — has been able to produce an outcome that is better than either can do alone. The human-machine teams can now beat a supercomputer. Kasparov himself engaged in freestyle chess and noted that, by ceding the calculations to a machine, he felt freed up to focus on the more creative aspects of the game [4].

Commercial Applications

It turns out that there are also commercial applications for combining the intellectual talents of machines and humans. Stitch Fix is a personal shopping service for apparel (for women and soon men!). Matching clothing to an individual’s preferences benefits from both the talents of machines and humans. When we shop online we benefit from machine-generated recommendations for discovery and navigation. Machines use their abilities to perform rote calculations to make recommendations based on our past purchases or the habits of similar customers. Yet, when we go to a physical store we get a different experience. We get a knowledgeable store associate who helps us in a very different way. She leverages ambient information — our complexion, how we present ourselves, how we respond to questions, etc. — to help us find relevant things. She also uses her powers to improvise, perhaps finding creative ways to stay within our budget. She is able to relate to us, understanding that we sometimes don’t see ourselves the way others see us. Such human-to-human experiences provide us with a sense of validation — that our particular preferences are being addressed.

The experiences provided by machines and humans are vastly different and not alternatives for one another; they contribute very differently. Just like in freestyle chess — they can be combined to produce an outcome better than either one can do on their own. This is the goal at Stitch Fix — our styling algorithm runs on both human and machine processors in order to combine their unique talents. Machines do their rote calculations and provide recommendations to our human stylists. The stylist will then leverage her additional skills — to improvise, to leverage ambient information, to relate to the customer, etc. — in order to curate and select the final five items that will compose the Fix (our word for a shipment). Each processor focuses on their unique strengths and their contributions are additive.

I imagine there are similar opportunities for other jobs. Medical diagnosis is likely more accurate when done using hard data and rote calculations vs. human judgment alone. However, outsourcing the calculations doesn’t mean we don’t need human doctors. The news of the diagnosis needs to be conveyed to the patient. While a machine could deliver the news through a siri-like voice, it will lack the ability to relate to humans to ensure the patient comprehends. I further suspect a machine would fail to leverage ambient information, such as a child in the room, and adjust its tone accordingly. These things are as valuable as the diagnosis itself and shouldn’t be sacrificed.

A New Division of Labor

Beyond producing a better outcome, teaming up with machines has implications on human job satisfaction. The Division of Labor has long been cited as the primary driver of economic wealth. In this case we are dividing the labor with machines — so that worries us. However, what is it we are losing? What is it we are doing more of? Where do we humans specialize? In ceding the task of rote calculations to machines it frees us to contribute in a different way. Just like Kasparov’s experience, when paired up with machines, our human stylists find that they are freed up to focus on the more creative aspects of the game. Absent the rote calculations to do, stylists tend to spend more time relating to the customers — perhaps something long overlooked by ecommerce where technology has historically insulated the retailer from the customer. And, relating to customers tends to be more fulfilling work. Our customers often share personal details of their lives such as:

“My husband is returning home from a 6-month tour in Iraq. He is disabled. I’d love something for what will be a very special date-night.” [see other examples here]

Our stylists are very much real human beings and they can’t help but be moved by such notes. They sympathise with the woman and the sacrifices she’s made. They imagine the husband and the sacrifices he’s made. And, they are compelled — through an innate need to connect with other human beings — to relate, to empathize, to help. It’s not in their job description but sometimes our stylists provide more than their styling services. They write back. In a case such as this they may thank the client for her husband’s services by sending flowers or a little gift. They want to impact the customer in some positive way. This is where we humans thrive. This is where we are at our best and doing the things that only we can do. This is fulfilling. And, it’s so rare that we are able to focus on this in our jobs. But through the delegation of work to machines we’ve found new freedom to do what makes us … more human.

Postscript

As an addendum I thought I would add an interesting statistic about satisfaction in the workplace. At Stitch Fix we use a service called Tiny Pulse to get feedback from employees on how they are feeling about their jobs (think: simple surveys like “on a scale of 1–10 how happy are you?”). The team that is consistently self-rated as the happiest is Styling — the more than 2800 people that curate the fixes. What is remarkable about this is that this group is remote (most work from home) and paid hourly. Such groups are typically the lowest in satisfaction at other companies. The contentment at Stitch Fix may have some obvious explanation — the relatively high pay, the flexible hours, etc. But I suspect that, because they don’t need to do the rote calculations, they are able to spend more time doing the things that they were designed to do — such as relating to other humans. Perhaps this is more fulfilling — satisfying an innate need.

[1] The title of this post was inspired by Brian Christian’s book, The Most Human Human: What Talking with Computers Teaches Us about What It Means to Be Alive (New York: Doubleday, 2011). It’s a wonderful book covering a historical accounting of the Turing Tests.

[2] I use the term ‘machines’ interchangeably with ‘computers’.

[3] Arguably, the game of Go is a better example of human thinking abilities as it relies less on brute force methods and fast computation. In this post, chess serves as a great example as much of it can be distilled down to rote calculations.

[4] Thompson, Clive. “The Rise of the Centaurs.” Smarter than You Think: How Technology Is Changing Our Minds for the Better. Print.