Humans are employed thanks to their utility. Humans’ utility can be broken down into three parts:

• Cognitive: Finding the solutions to informational problems
• Persuasive: Encouraging others to do things
• Manual: Manipulating physical objects

All of these may soon be better performed by machines.

Humans will cease to have cognitive employment due to machine intelligence

Historically computers have worked best on tasks that fitted the stored program model: executing well defined repetitive actions with mathematical precision.

We’re seeing the emergence of a new paradigm: systems that take noisy, poorly-structured data and find patterns in them (‘artificial general intelligence’, AGI, for lack of a better term). This is often finding a pattern over time or space, e.g. discovering the grammar of written text. Once this pattern is learnt, it can be applied: new text following the grammar could be generated and text could be tested as to whether it follows the grammar. Given text, the system can guess what will come next.

This gets more interesting when two different data occur together — e.g. text and audio, or multiple languages. Then the system can learn to produce one given the other, e.g. creating subtitles, or translating languages.

AGI is powered by deep nets, a refinement of earlier neural networks. Thanks to improved architecture and greater computational resources, deep nets are managing to impressive things. Without being told how to do them, they’re learning to caption photos, play computer games and much more.

There is a lot of data in the world for AGIs to learn from.

Deep nets are rapidly finding applications across all industries.

Deep nets and human brains have a similar computation architecture. They both learn structure from streams of data. They both have layers, finding higher level patterns in the patterns they’ve already learned (for example, learning sentence grammar after spotting words from sequences of letters).

There is nothing special or unreproducable about the human brain. Its density may be hard to replicate from an engineering perspective, but it’s solvable.

AGIs work every hour of the day. They are consistent in judgement. They can be cloned. They can have their storage, inputs and outputs doubled when needed.

In the 30s John Maynard Keynes articulated technological unemployment: “This means unemployment due to our discovery of means of economizing the use of labor outrunning the pace at which we can find new uses for labor.”

Our technological capability is growing exponentially. Deep nets are growing in ability more rapidly than biological systems. At some point AGI will learn new jobs faster than humans.

AGIs aid in the creation of more powerful AGIs, further compounding the exponential progress.

Any job performed through an electronic device could have its human agent switched for an AGI.

There is no reason any cognitive task humans perform could not be performed by another comparative system. Given technology is progressing faster than biology, AGI will therefore beat human intelligence across all cognitive tasks.

When deep nets learn new jobs faster, they will be the first candidates for those new jobs. When they work more reliably and cloneably, they will be the preferred candidates. It is easier to train then duplicate one admin bot than train a hundred humans and retain them.

It is highly likely a large technology company will develop and own the technology powering AGI. As the technology overtakes human employment, the company will capture a larger part of our economy than any has before, and become more powerful than any before. AGI will become a fundamental infrastructure.

Many countries will have their own AGI infrastructure to avoid being dependent upon eachother.

Persuasion may be undervalued

Persuasion, convincing others to do something you want, is the most valuable skill since it can leverage all other skills.

Much more research has gone into building machines with cognitive and manual abilities than with persuasive abilities. Either persuasion is undervalued, or the perceived chances of engineering persuasion are low.

Sales, management and teaching can be provided to some extent through text messaging. As AGI improves, it can learn and mimic these interactions, therefore persuasion can be delivered to some extent by AGI.

The simplest artificial persuasion implementation is helplessness — the furby convinced children to care for it, despite having very limited functionality.

Persuasion often consists of two parts: Discovering what incentives you can motivate another with, and educating another that they can receive those incentives by carrying out the action you desire.

The first part, discovery, is easy to imagine being done by a current machine learning system. The second part, education, is harder given humans’ mistrust of aliens. At the extreme, if a robot today offered one enough money in return for performing an action (e.g. learning something, or buying a product) it would have some success at persuasion. Therefore building persuasion is a matter of closing the gap.

The robots are coming

The cost of robotics is shrinking. House cleaning robots are now sold on Amazon, CES is flanked by elderly assistance robots and Mountain View, CA is being circled by self-driving cars. Self-checkouts have replaced store clerks, CCTV with AI is replacing police officers, beacons are replacing waiters, RFID and contactless payments will further remove checkout staff, droids and drones are replacing warehouse workers. It’s easy to imagine supply chains from farm to household that do not involve any humans at all (replacing one of the USA’s biggest job sectors).

Endgame

Human salaries are driven by talent scarcity. AGIs/robots can be cloned, therefore provide a limitless talent supply. Human salaries will be driven down to the marginal cost of launching another AGI, or the running cost of a robot.

We have seen this happen before. Once, clerks were employed to add decimal numbers. Then we created calculators. Now, it is hard to imagine adding numbers as viable employment.

Human consumption will continue to grow. However this increase in demand will be met by AGI and robots before human labour, therefore the demand will not increase human wages.

As human salaries drop below a level that provides subsistence, the population will be forced to rely on social security payouts instead of employment.

Human unemployment will lead to depression of many individuals.

The majority of the humans will have a decrease in real wages. Economic activity will slow down as a result.

A small number of companies will own the IP and capital that provides the majority of the world’s labour.

Humans will campaign for higher taxes to be levied upon those companies. Humans will campaign for an increase in social security payouts. Governments will slowly do both until greater social and economic stability is achieved.

Despite the improvement in conditions, most humans will still be marginalised in a way comparable to feudal systems.

The situation may lead to revolution.

AGI and robotics are such powerful sources of labour that even revolution will not eradicate them.