Science fiction about AI never seems to talk about the interesting stuff, tax and geography and work.
tl;dr AI as a pervasive force is just round the corner, we’ve got a few things to think through like business models, taxation, globalisation and what us humans do.
As a teenager I got to see 2001 in the cinema. It was mindblowing. HAL 9000 was the star of the show. As someone who knew computers as the clunky Commodore PET (Personal Electronic Transactor) at school it was a fascinating and almost unimaginable vision. A computer that could reason, had a personality, was everywhere in the ship through an unblinking red eye, a computer that could murder, a computer that you spoke to and it spoke back. No blinking green square cursor, no clickety clack keyboard, just pure intelligence with a human voice.
It was a computer which at the end of the day needed humans to look after it, and do work for and with it. A computer which could run most of the show, but not all of it. One which, at the end of the day, was vulnerable as, despite being omnipotent, it had a corporeal form.
HAL 9000 had a weakness, you could visit it and take it apart. The computers of that era were also very physical, the IBM System 360, announced in 1964, was a computer you could visit and it needed humans to interact with it physically. To change the program tapes for example.
Computers aren’t like that any more. Although obviously they physically exist somewhere, they need less and less physical interaction. Clusters of machines can transcend geographical boundaries, even corporate boundaries with hybrid clouds. People still visit them to repair them, however technological advances in self healing networks, schedulers such as Google’s BORG and shared nothing data architectures, mean they don’t have to unless there is a critical outage. One can imagine that some day we’ll have near complete data centre automation using robots similar to those that pick goods in Amazon warehouse to fix problems and swap out boards.
I’m not going to get all dystopian about not being able to pull out parts of the computer in the way Dave Bowman could, I’m more interested in four interrelated things — jobs, tax, geography and business models.
Let’s look at jobs. There are few jobs that cannot be automated by AI and robotics. Already, we have chatbots that can appeal parking tickets, ones which can assist in prescribing medication, ones which can help prepare your accounts. These potentially replace stable aspirational white collar jobs. The sort of job your parents would have been proud for you to study towards: lawyer, doctor, accountant. These are jobs which normally require an expensive education and have higher salaries that are highly taxed. When you add in automation in farming, autonomous trucks, manufacturing, brick laying, cooking and furthermore the jobs in the industries that support those, one thing is clear: there will be a surplus of labour in almost every strata of society and concomitantly a shortfall of personal tax.
Tax is interrelated with the thorny issue of geography and territory. It’s easier to tax an individual on where they work and earn, than it is to tax a company on where profits are made. Already 2016 has been a year in which the huge amount of tension around immigration has resulted in some of the most extreme politics in living memory, a bonfire added to with with a series of revelations of how large corporations and the super rich obfuscate finances and move transactions around to minimise tax.
Imagine how that will feel, or be made to feel, by extremist politicians and some elements of the press, when many jobs are being lost to automation. Where extreme politicians currently distort facts and whip up fears of jobs being lost to foreign workers, in the future there will be fear of jobs being lost to algorithms and machines, and likely as not foreign algorithms. Even Donald Trump, who I detest, knows that the fast route to people’s feelings are through employment and tariffs, as described by Michael Moore in the video below. The Fascism, as Michael Rosen so eloquently says “arrives as your friend”.
Globalised companies are at the forefront of Machine Learning, with services being run in the cloud wherever data and processing is inexpensive and increasingly where land, electricity and cooling are plentiful and cheap. However these services are rarely taxed in either the territory where the servers are, or where the services are consumed. A local job, and it’s associated tax, is potentially displaced by an algorithm charged by the hour run somewhere distant, written by person working somewhere else with the profits and costs, rolled up into intellectual property licensing, moved to be taxed in the most expedient territory. We can’t stop it. We shouldn’t stop it. But it won’t be comfortable if we don’t plan for it.
We have some time to figure this out, not long though. People don’t yet fully trust algorithms to do these jobs, or they’re not production ready. One reason why these algorithms may not be switched over to is one of business models. Algorithms to research case law can’t be charged for in the way a paralegal or junior lawyer can. Often these more junior members of staff can be highly valuable to a practice. How to charge a client for work performed by an algorithm when you used to charge for a person by the hour with a large margin?
We’re past the point where we can pull the plug either physically, technologically or societally around AI and automation of the workforce, or would want to. Humankind augmented by AI can achieve so much. It’s time to get serious about the financial and societal implications. The whimsical notion of the legality of a robot being able to make money from their actions in Bicentennial Man are long gone. It’s time to understand how to reform an economy where human labour becomes a less valuable commodity and taxable personal income is a scarcity.