The company that sells you a cheap robot could end up owning the expertise of everyone who works through it.
Everyone expects
Robots get an app store, like phones.
What I found
A profession becomes an app. Then the app becomes a feature.
RoboStore finances cheap robots, controls access to the customers and keeps the learning generated by everybody’s applications. Manufacturers supply the bodies. Skill developers supply the capabilities. The store owns the relationship that ties them together.
Everyone has a reason to say yes. The manufacturer gets volume orders. The plumbing developer reaches thousands of homes it could never afford to equip. The household gets one affordable robot and a specialist whenever it needs one. Robots come locked to their store, as phones come locked to theirs.
The store can also outbid any single provider. A plumbing company can only justify a hardware subsidy against plumbing income; the store can justify it against cooking, tutoring, cleaning and repairs through the same machine. The developers’ commissions help pay for the distribution system that progressively acquires power over them.
The central struggle is over who gets to accumulate the competence of the whole system.
A plumbing developer learns about plumbing. The store learns across the household: plumbing, maintenance, appliance installation, cleaning. Pooling experience across machines really can help; Google DeepMind’s Open X-Embodiment research found gains from training on data from many different robots.
The developer is told its code and model remain its own, which may be true. The question nobody answered is what the platform can learn from watching the skill work. Then an update arrives with Basic Household Repairs built in. The specialist keeps the difficult cases. The routine market becomes a feature.
Customers stop choosing apps and simply say: the sink’s leaking, sort it out. The store decides who gets the job, and so who gets the experience to improve. It needn’t ban a competitor. It can offer access on terms that leave a rival permanently behind, or promote a newcomer to weaken an incumbent that has grown too powerful.
The marketplace becomes a manager of its suppliers’ competitive trajectories. Customers stop buying a named provider’s skill and start buying a result from the store, in subscription tiers, with a discount for not being advertised at while the robot works.
If every successful supplier trains its replacement, the best stop joining. Manufacturers who don’t want to become interchangeable bodies, developers who want distribution without becoming disposable, and large customers who want to keep their own operating experience back an open robot operating system.
The store’s reply: pay people to install a RoboStoreEye on their open robots, so it keeps learning anyway.
Original fiction
Explore more of this kind of work
This is an original public scenario. Bring your subject, research or planning question to develop a world and explore its consequences.
Foresight and design fiction workshopsFrom a robot on contract to a loop that learns faster than the people who taught it.
You buy a robot the way you buy a car or a phone: on contract, paying over several years. It comes locked to one company’s RoboStore, a marketplace for robot skills, with a starter set: tutor, cat sitter, chef, carer and cleaner. Each skill is updated constantly and needs a subscription. The robot’s value is the range of jobs it can do, the way a car’s value is the range of journeys it can make.
The subscription is cheaper if the robot uploads its data to the cloud for further training, which is what most people choose. Rival AI companies launch their own stores. Their ambition is to be more than marketplaces: to become the system through which people obtain professional services, concentrating professional knowledge the way medieval guilds once did.
The first training data comes from people. Companies pay professionals to wear sensor suits at work: accelerometers, cameras, strain gauges and touch sensors. Ten thousand professionals doing five hours of manual work a day produce 250,000 hours of training data a week; ten thousand cleaners could produce 12 million hours a year. Poorer countries with laxer privacy laws are targeted first.
Then the robots train themselves. Every deployed robot’s experience improves the models, until the robots are better than the people who supplied the expertise. The loop reinforces itself: more customers, more experience, a better product, more customers. The same loop runs between rival stores, whether a store owns the skills outright or provides the platform and the base AI they run on.
She made medicine in her flat. They sent the bomb squad.
Everyone expectsAI invention engines make discovery faster and cheaper for everyone.
Imagine invention engines generating new drugs, materials, processes and products faster than organisations can use them. The possibility looks like abundance.
Follow how it could happenFree dinner. The cartridges cost a fortune.
Everyone expectsFood printers pipe your own ingredients into shapes.
Why buy a machine to assemble dinner when a factory can assemble it and sell it to you in a tray?
Follow how it could happenMiss a payment, and the cool room won’t let you in.
Everyone expectsCities add trees and air conditioning.
A family pays every month for four places in a cooled shelter ten minutes from home. They have never used them.
Follow how it could happenA subject, a technology, a question or a body of research.