We are entering a period where machine learning is going to dramatically change the nature of work across a broad swath of service and knowledge industries.
Extremists see a world where all human work evaporates rapidly, driving a conversation about human dignity and equality in a world of technological unemployment. A more sober group believes we’re on the brink of eliminating a class of repetitive white-collar knowledge work—not unlike transitions we have gone through in the past.
My view is that this debate is several steps ahead of reality. Those that concentrate just on the elimination of some specific job categories underestimate the impact of new technology.
What is worth instead discussing is how the fundamental nature of human work is going to change—for better and worse—based on the reality that everything you do professionally will be measured and processed and predicted at scale.
Here is how I see things playing out:
Everything Will Be Measured for Everyone
It is now possible to measure everything. Every message and keystroke will be stored, every conversation will be recorded, parsed and saved. Every meeting and interaction will be remembered and analyzed.
This already happens for certain categories of jobs, especially in the on-demand 1099 space. There, workers have flexibility and can make their own hours, but their output is measured closely. Cheap data-warehousing and processing, coupled with all sorts of new sensors and tracking mechanisms for the workplace, will bring similar insights to all sorts of sales, office work and engineering jobs shortly.
In a sense, this is just the acceleration and personalization and natural extension of the 100 year old-plus Taylor method. Rather than measuring overall output just in steps of the industrial process, we are moving to a world where we personally measure all steps in everything. The data will be used for three separate things:
- Anything that can be automated away from workflows will be either explicitly done by machines, or machines will make the right suggestions—leaving only approval to humans in the loop. This will make a huge number of activities dramatically more efficient.
- Where things can’t be automated, the data will be used to help you understand how you work and improve what you do.
- The same data will be used for performance management. This by no means suggests that if you have a bad week the machine will fire you, but it does mean that it will be very evident to managers and the company at large who is doing well and who is not in whatever job they are doing.
Human Work Will Become Intellectually More Difficult
In general, I have found that people don’t enjoy doing work that they know a machine could do for them (or could help them do more efficiently). Construction workers would—largely—be insulted at this point if you asked them to use hand saws and drills, knowing they could do more with power tools. Similarly, I predict that when self-driving cars eventually become a reality, it will be seen as pretty demeaning to drive a vehicle. By and large, I believe people derive pride from contributing things that machines can’t easily do.
As machines get faster and more able, the work that is left for humans should become more fulfilling. At the same time, it will become intellectually harder, on average. The challenge is that without the creation of new jobs, there is indeed less raw work to go around.
There is a question about how long workdays will be in this new world. While it might be possible to do rote work like turning a crank for 10 hours a day, I have found that more creative and intellectual work is harder to do for hours and hours on end each day. (That’s part of the reason why I find creative writing for long stretches of time to be harder than basic programming for hours. Programming has lots of rote tasks associated with simply creating the output, whereas writing is almost pure creative output).
Work Should Become More Flexible but More Engaged
In our work today, we suffer a tragedy of partial attention. People, especially Americans, spend long hours at work and yet aren’t as productive as they could be. A lot of the workday is funneled into browsing social media and the news, chatting with friends, etc. This partial attention time, I believe, makes us look far less productive per hour of “work” than technology actually allows us to be. Blame tabbed browsing, smartphones and notifications.
At the same time, because so little actual work is measured, teams and companies look at crude measurements—like how long people are spending in the office versus how productive people actually are toward meeting their goals.
Measurement and machined feedback should fix a lot of this. First, it will make clear to people just how much time they are spending in partial attention state, not really doing productive work. With the right feedback, people will rapidly figure out how to limit distractions in the office and spend less time in the office and more time at home or with friends.
Simultaneously, the more that measurement allows us to be outcome-oriented, work will become far more self-directed and flexible in many cases. No one will care if you are at your desk so long as your measurements look good.
Of course, quantifying what matters to success and then measuring on the individual level with enough granularity to get a valuable signal is extremely hard and data-intensive. At Fin, the company I am currently building, we have spent the better part of the last two years understanding how to build human-computer hybrid systems with a premium on leveraging data to coach people. It is very difficult, but I believe it is ultimately doable.
Some organizations, no doubt, will value and measure the wrong type of output. Others will not understand how to properly “chunk” the work they do so they can manage and optimize intermediate steps with new technology. They will have trouble surviving. But, for the companies that learn how to organize themselves to take advantage of granular performance data to automate and coach people, the advantages over competitors is going to be massive.
There Will Be Fewer Management Roles
There is an irony that in some ways, the role that will suffer the most as AI measurement grows in the workplace is going to be management. It’s far easier to automate most of the day-to-day manager’s job of giving performance feedback and guidance than it is to automate most types of actual work.
To be clear, managers will still exist. But rather than only managing a few people, they likely will be able to use technology to efficiently manage many more simultaneously—with automated insights and suggestions.
Done correctly, this likely will make a smaller number of managers better at managing, giving them more opportunities to coach different people and spend less time on Administravia. It also means that organizations will be more heavily weighted toward individual contributors versus layers of managers. This should make organizations more nimble and better places to work.
Work Should Become Fairer
I am walking into dangerous territory using the word “fair”—but it’s a point I want to make nonetheless. The more metric-oriented the work is, the less politics and other social dynamics come into play. It becomes very clear who is doing a great job and who isn’t in an objective way, and that, I believe, should make work more fair and erase politics from the office.
There are, of course, very complicated issues about exactly what you measure to create the right incentives and outcomes. But the reality is that the more you measure indisputable output—ranging from athletics to academics to sales—the harder the system is to game. The more the best people get hired, star performers become clearer and rise to the top.
Measurement means that advancement becomes far less about being buddies with the boss, and far more about demonstrating to your peers and the leaders at the company that you are doing excellent measurable work that contributes to the team.
Technology Work Will Turn Into Service Work
In the last 50 years, the primary role of technology companies has been to observe the world and encode the basic logic of how things can be done most efficiently. In effect, the job of technology companies has been to “skim” out of roles the most basic tasks that computers can address and leave for people the more complicated—if still rote—parts of their jobs.
A good example of this is scheduling, email and calendar programs that automated the most basic work of setting up and remembering meetings, offloading that work from assistants. As a result, companies around the world effectively cut the number of assistants they hired per executive and placed the work of now-more-efficient scheduling back on their executives.
The net effect is that the average mid-level executive at the average company almost certainly spends more personal time today on scheduling tasks than they did before, not less.
Machine learning and measurement, I believe, will help solve this by forcing technology companies to evolve into service companies—becoming experts in delivering end-to-end solutions.
This is a natural step forward. A generation ago in the IT era, “technology” meant simply giving companies the basic tools to customize and use as they saw fit. We have more recently gone through an era where “technology” meant programming parts of workflow on behalf of companies. The next logical step is to move toward a service definition where “technology” means managing whole functions end-to-end for others.
Where We Are Going
Everyone is going to struggle with the future of work, not just those in outsourced call centers and whose entire jobs can be automated sooner than later. Measuring every second of everything is going to have its challenges socially, as has been discussed in movies ranging from “Gattaca” to “Apollo 13.” The fact that there will be fewer management jobs and more meritocracy will be certainly better in some ways. But it also will likely make the patterns of careers and promotions more stressful and fraught for some.
Ultimately, I believe the combination of measurement and machine learning that is coming to the workplace will ideally make work more open, productive and fair—and fundamentally make human work more human. But it will not be a simple task to navigate.