I recently saw an article in the Wall Street Journal about researchers at Harvard University who designed a team of fist-sized robots, operating under limited instruction sets, that will assemble blocks into a specified shapes like a wall, pyramid, or even more complicated structures. None of the individual robots is very smart: it can move forward and back, climb up or down one step on little pinwheel-shaped rollers, pick up a brick or put it down, and see where the next brick is needed in the developing shape. Together, four or five of the robots swarm about on the growing structure, apparently fiddling around, occasionally putting down a brick, or moving it from one place to another … and eventually what they’re building takes shape. The researchers call it “swarm intelligence” and liken the way these robots work to termites building a nest.1
Indeed, it took the researchers four years to design these mechanical termites. And, yes, they are wholly impractical in the real world, unless you want a toy that plays with blocks while a child watches. But I found the image of these simple, mechanical swarms going about their tiny tasks riveting. They are the first physical generation of the cellular automata postulated by John von Neumann and others. And yeah, these toys took four years to bring to life, but we are only at the start of this learning curve, which is always the steep end. Future versions doing different and more complex tasks will come along faster. Scaling up to jobsite size will be no difficulty, either. And once future engineers know the principles of reduced instruction sets, they will be able to step easily from robots that pick and place blocks to robots that dig holes to specification, construct forms and pour cement, position steel beams and weld them in place, add floor and wall panels, and lay out piping and wiring circuits.
In the future, an architectural engineer will only need to decide on a pleasing design, consult a civil engineer for the technical details, and discuss a few of the job’s unique aspects with a computer assisted software engineer.2 All that will remain is paying out the bucks for renting and reprogramming a team of constructions ’bots and buying the raw materials they’ll use. Then everyone will sit back and let the termites get to work.
Erecting a building this way might take a little longer than having human workers with their vastly superior—even excessive—brain power show up each morning, study the drawings in the construction office, and decide how much concrete they will pour today or which floors they will finish off with drywall. And all those humans are working under a gang boss, a construction supervisor, who fills in progress on a Gantt chart3 and holds the keys to the tool room. Termite ’bots stumble around because each of them holds just a fragment of the design in mind, and the master plan is built into their tasks. But when you’re not paying the crew actual, time-based wages, plus overtime when anything unexpected happens, you can afford to take the extra week or two that using the ’bots will require.
Of course, the mechanical termites will at first need some watching, at least until human programmers get the hang of this new technology. Even the simplest programming error, replicated through a swarm of busy ’bots, can lead to a remarkable jumble and waste of materials. If you doubt this, go watch The Sorcerer’s Apprentice—the one with Mickey Mouse—again. But in time, we’ll perfect this way of working as we’ve perfected others.
We already have factories filled with relatively mindless robots, as anyone knows who has watched the Science Channel’s How It’s Made series. Stationary machines pick up a blank piece of sheet steel from a pallet, run it through a hydro-forming press to give it a particular shape, trim the excess with a cutting laser, and dump the finished automobile fender or frame part into a bin. Later, maybe in another factory or at the assembly plant, other machines pick these parts off pallets, position them inside a jig, and smack them with a couple of hundred precisely placed spot welds. Out comes the basic body of a sedan or truck, missing only the paint job, engine, critical systems components, and four wheels—and all of those except for the paint dip are already being made by machines somewhere else. The only human hands involved seem to be moving the bins and pallets of semi-finished parts around on the factory floor.
All of these factories manage their inventories and their just-in-time deliveries with computers. Yes, human beings look up facts on their computer screens, then lift the phone to call the suppliers, place orders, and arrange payments. But the banking behind the verbal authorization to pay is all computerized. Above a certain scale of business, no one writes checks, puts them in an envelope, and mails them anymore. All the backroom functions are managed by computer with human beings keeping track, reporting trends, and taking responsibility for the results.
How soon will it be before these factories start “eliminating the middleman”? Eventually, they will replace the human who picks a part out of a bin—either the bin attached to the maker-machine’s output slot or the bin storing the part in the warehouse—takes it across the factory floor, and puts it within reach of the robot performing the next operation. Instead, conveyor belts will link the two operations, and small robots with optical scanners and a preference for seeing a certain image will turn the parts just so, ready for intake by the next machine. That conveyor will be managed by the same software that now accounts for inventory, using just an add-on module that actually moves the inventory from place to place. And that same piece of inventory software will be given authority to perceive when stocks are running low and make the call to the supplier—another computer program, of course—to replenish them.
Humans used to make things as master craftsmen. One gunsmith poured the metal to cast his own gun barrels, shaped wood for his own stocks, cut his own screws, and assembled and finished all the parts by hand. No two of his weapons were quite alike, and none could match parts or interchange them with the guns made by another craftsman. If a part broke on a gun held by a soldier in the field, he was simply out of the action. Unified designs, modular assemblies, interchangeable parts made to micron tolerances—all were necessary for the soldier to be able to take the broken trigger mechanism from his own weapon and swap it for the trigger on a comrade’s weapon whose barrel was plugged or bent. Modular assembly of precision pieces made to specification moved us into the first stage of the Industrial Revolution.
Now—or very shortly in my estimation—we will move into the second stage. Human hands are too slow, too inefficient, require too many coffee breaks, rest periods, and five o’clock closing whistles for really efficient manufacturing. And our products are becoming too complex, the parts too small or delicate, the assembly steps too intricate to turn them over to big, clumsy human hands in the first place.
It won’t take an artificial intelligence on the scale of IBM’s “Watson,” able to play and win at Jeopardy, to run a factory, coordinate the individual robots and the conveyors between them, keep track of inventory, order parts, and pay the bills. A couple of smart accounting systems, an articulated barcode scanner, and the authority to make payments is about all it would take.
Traditional economics makes a vast distinction between capital and labor. One is the cost of the tools, equipment, and materials; the other is the people who use them. We are fast coming upon a time when factories will be all capital and no labor—not even a night watchman or a janitor, because a security camera can do the work of the one and a glorified Roomba the other. And given that good mechanical designs are easily replicated by other machines, and good software is easily copied, the second, third, and fourth generations of these factories will cost much less than the first. Machine labor will push down not only the price of products but also the capital cost of their production.
For a while, these automated factories will make us all rich in goods and services. Running around the clock they will easily make more products than any human population can buy. Eventually, the machines will have to be throttled back to just-in-time manufacturing, on order and to specification. But then, it won’t take much of a leap in machine intelligence before automated factories can handle personal preferences and made-to-fit sizing. Those are just choices, involving a bit more inventory and a few adjustments along the production pathway. So the factories will not necessarily turn out boring, one-size-fits-all sameness but be responsive to customer needs, demands, and dreams.4
The ’bots will make us all wealthy beyond the dreams of kings and corporate titans of even a generation ago. They will also free us from the drudge jobs, the boring jobs. If you expect to go to work and spend your day picking up parts and bolting them onto something, you will be out of work. If you spend your day looking at columns of numbers and making changes or writing reports about them, you’re out of work, too. Work—defined as “some boring thing I need to do for a paycheck”—will simply go away.
What will people do then? They will be surrounded by a wealth of products, made at virtually zero cost,5 but which they cannot buy because they have no paychecks. Some people—maybe a great many, maybe just a lucky few—will find those jobs that robots can only do badly or not at all: sing, dance, paint a picture, tell a story, cradle a child, train a dog, comfort the dying. Some will find niches making beautiful objects that other people will value for their craftsmanship, uniqueness, and handmade qualities. But the mass of people, those who have not the talent to be great artists, poets, and craftsmen, nor the compassion to work with people on a personal level, will not fit into the new, machine-made paradigm.
The dystopian view says that a few extremely wealthy people will end up owning all the factories, making all the products, and laughing at the rest of the human population in its poverty and misery. But that’s the stupid view. Economics doesn’t work that way. Factories without customers are a bad investment, not worth the land they’re built on.
Another view might be that we should simply pass laws to stop development of the new machines, throw our wooden shoes into the gearboxes, and celebrate the glories of boring drudge work. We can all spend our lives doing the 21st century equivalent of bolting steam boilers onto cart frames to make entry-level horseless carriages. Then there will be plenty of work making a small selection of old-fashioned, clunky products.
The better view—although not so simple to frame as an answer—is that our society will have to redefine the concepts of work, ownership, participation, and the social contract itself. If we play our cards right, it can be a golden age of leisure and ease for everyone, with time to love, explore, learn new wisdom and talents, and become whatever we might dream of being. Of course, we’ll then need a new “science of contentment” to give the people who don’t already have goals or ambitions their own sense of purpose.6
Teaching and counseling—offered as personalized, hands-on service—just might be the growth industries of the future.
1. If you’re a WSJ subscriber, see the online article, including a fascinating video, at “Termites Teach Robots a Thing or Two”, from February 13, 2014. A more detailed article and simulations can also be found at “Designing Collective Behavior in a Termite-Inspired Robot Construction Team” from Science magazine for February 14, 2014.
2. Computer assisted software engineering (or CASE) is the principle that a programmer no longer needs to write out software line-by-line in longhand, being mindful of where the commas and the semicolons go, and checking every program call and its referent. So much code has been written over the years that most of the definable tasks for a computer to do have already been formulated and perfected as preferred modules in each language. The CASE expert figures out what the overall objective of the project should be, determines what language it should operate in, then picks and arranges modules to achieve its goals. Essentially, he or she is playing with groups of code as if they were building blocks.
3. The Gantt chart, named for its inventor Henry Gantt a century ago, is a bar chart that tracks job tasks down the side and dates across the top, showing the duration and dependency of each step in a project. It’s a core tool of any project manager. Figuring out the dependencies between the steps—e.g., “we can’t pour concrete until we finish laying out and tying the rebar, and we can’t lay rebar until we finish digging the hole for the foundation”—is the key difference between a human project that gets done in a predictable amount of time and one that finishes up “whenever.”
4. Go online sometime with GM, Ford, or Chrysler, select the model you like, and click on “Build Your Own.” You’ll find that every basic car or truck model comes in about fifteen colors, four different engine variations, six transmission and drive wheel combinations, three levels of interior, three kinds of sound system, and a hundred other packages and options. The dealer gets a range of the most popular choices assigned from the factory to offer on his lot, but if you want something special and are prepared to wait while the factory programs it into the production line, it can be yours. And this is not because some human will hand-build the car especially for you; someone is just punching choices into a computer.
5. Not even the raw materials will enter into the cost equation, because those busy little ’bots will be swarming out across the land to mine metals, drill for oil and gas, harvest trees—if materials science hasn’t already given us a biotech alternative to woods and fibers—and till the soil. The rest is must machine processing.
6. This is not a new theme for me. See also The Coming Robotics Age from January 8, 2012, and Automation, Work, and Personal Meaning from February 27, 2011.