The idea in a nutshell : Many technology projects, including, potentially, Artificial Intelligence projects, get exponentially cheaper over time. People get more expensive over time. The result is the inevitable replacement of the jobs which can be replaced by AI. And, because of the nature of the costs that AI projects face over time, it happens much sooner than you think after the first AI project rolls out.

The maths of job replacement appears to be scary

At the moment, only pioneering companies are rolling out AI because it’s so expensive (and, often, more expensive than the people it will replace) That said, it is already helping hedge funds perform better, Oil companies find natural resources better, doctors detect cancers better, computer companies help us search better and so on and so on and so on.

However, the cost of rolling these AI projects is likely to halve every year or so – as the computing resources that provide them get cheaper and as the lessons that people need to learn to do them efficiently start to be internalised and applied.

Even allowing only a small increase in the wages of the people in call centres etc. (in the graph below, I used 4% a year) it doesn’t take long until the AI is cheaper than them, by which time, the management team replaces the workers with bots or AI algorithm.

Understandably, it’s the companies which operate in markets where the AI provides a great deal of value and therefore justify the costs, which have started the rollouts. The AI projects might provide a tiny edge in how to bet a lot of money and pay for itself. Or it might cut a lot of jobs and save money that way.

The problem is that, with exponential decay in the cost side of things, things change quickly. The fact that these projects are now making sense even for companies where the value is high, it seems likely that AI will be affordable even for companies where the value is (relatively) low, within a few years.

Below : The chart shows a CONCEPTUAL relationship between the cost of a project and the cost of people who will be replaced by an AI solution.

Note : Obviously I did this in Excel in 5 minutes. It’s just to show visually what might be a relevant factor.

Chart below : A conceptual view of AI’s rollout

AI Cost Benefit Infographic

It is notoriously hard it is to predict exponential growth. An inability to forecast it cost the emperor all the rice in India.

I actually wrote an article for Business Insider under a pseudonym Jacobus Bron. On the growth of data utilisation and the threat that Google Glasses posed to telcos. I got that wrong. The premise of the article, however, was that Vodafone had failed to anticipate two overlapping trends. First, that more and more people would buy iPhones and second, that each one of them would use more data. The result were two overlapping exponential curves which caused them their network problems in 2010 and beyond.

Technology gets a lot cheaper over time

No shit. Anyone who has worked in technology for any time at all knows that the cost of it reduces frequently. It’s very similar to Moore’s law and was demonstrated in the reducing cost of processing a human genome. Much of my thinking on this is owed to Ray Kurtsweil and his books which rocked my world 10 years ago (long, long after they were released).

It’s hard to overstate the potential for change when, in 10 years time, computers will not be 10 or 50 times more powerful than they are now but potentially 1000 times more powerful. (And that’s without quantum computing – a field in which Australia is amongst the world’s leading researchers).

Having your job replaced by tech is not new

This is not a new phenomenon. Bank Tellers faced it when ATMs came in. Economists call it technological Unemployment. I was always told that the word Saboteur came from workers whose jobs were going to be replaced by machines and who through their wooden shoes at the kit that was going to displace them. (Unfortunately, this seems to be untrue).

chart2

Above, this real life example of a repeatable process being optimimised demonstrates that costs fall quickly over time. AI projects which cost many millions of dollars now will be very cheap soon.

High labour costs in Australia are likely to accentuate the problem

Unfortunately, the cost curve of these AI projects likely to come down very quickly with the lessons ‘learned’ (literally) from each implementation around the world being applied to the next. At the same time, Australian wage costs are a higher than anywhere else in the world and b) rising faster as this OECD data quoted in the Sydney Morning Herald which means the curves will intersect faster here than elsewhere.

chart3

In some ways, this is what has always happened

The fact that the (apparently inaccurate) story about saboteurs relates to them throwing shoes in to a machine they resent for replacing them reminds us that technological unemployment has always done this.

There are many reasons that artificial intelligence is different. Perhaps most important from my point of view is the idea that automation has affected each industry differently before now. The printing press replaced people writing books out long hand. Weaving machines replaced weavers (I guess) AI will replace every secretary, call centre and chat operator and driver from every industry.

The estimates of those displaced by AI seem to be found in the 30% – 50%  range.  I think the truth is that it’s far from clear whether the prosperity we are likely to see come form the deployment of Artificial Intelligence will overwhelm that increase in the dependency level or we all spiral from here to misery and unemployment. I have tried to consider some of the complexities in providing decent social outcomes by considering the net effect.

Advertisements