The idea in a nutshell : It’s possible that AI will create jobs as well as displace them. People have skills which can’t (yet) be ‘emulated’ :o) by technology. They’re leaders, adaptable, benefit form common sense, use their imagination and have ‘ethics.’ Computers don’t. In practical terms, current AI and other software programmes are narrow to the point that they can conduct a ‘task’, not a whole job. Which has some interesting ramifications in itself….

Massive job losses from AI

Most articles in the press at the moment suggest we are heading for massive job losses form the rollout of Artificial Intelligence. Admin ( including thinking admin like Legal Research, Accountancy ), logistics and transport jobs seem most at risk. These jobs have an overlapping characteristic which I cover below. There are two sides to every story, however. We should consider both. What factors will work against artificial intelligence replacing us all with robots.

Flaws in the argument suggesting all jobs will disappear

Without wanting to rehash the fantastic podcast they have which we transcribed in full and have an abridged version of here (this is largely owned by the speakers in that podcast. I have added a couple of points of my own) Criticism of existing studies includes these elements:

  • There are no estimates of jobs created : In the podcast, they’d studied a number of real world examples of software rollouts which replaced jobs. Having witnessed that, the researchers estimated 14 jobs would be created ( they were at pains to stress that it was a ‘best guess’ and no more ) out of 20 which would be displaced. That estimate suggests that 6 out of 20 or 30% of people do not have a job which is replaced.
  • There are few empirical result studies : The researchers had studied a large number of software rollouts and job automation. In this context, with such a nascent technology was around 15. This is such a new field that there are not really the statistically significant studies around to indicate what exactly will happen.
  • They rarely provide timeframes for the work : They say ‘45% of jobs will go ‘ but they don’t state the period over which that change will happen. It is obviously one thing if that attrition happens over 100 years and far more difficult to accommodate such a change in a shorter time like the next 5 years.


  • Limited examples : The same examples pop up in the press to prove the ubiquitousness of the change. The ‘usual suspects’ – The Google Car, Emma the journalism bot, Watson winning Jeopardy, Google winning . Terminator 2 and so on. These items create the impression that AI is talking over every job.
  • There is no consideration of the jobs created in parallel : The presenters also noted that there would be new jobs created by the technological world. They cleverly point out that anti hacking software and even companies didn’t exist before the internet. They also talk about the opportunity for additional human work from regulation. Finally, they raise the importance of analytics employees to interpret the huge amounts of data being created.
  • Forecasts include unjust assumptions : The practical implementation of AI shows it is being used to automate tasks not jobs. If 20% of all tasks are automated does that mean that 20% of all jobs are automated.
  • People see AI as an opportunity, not a threat : Importantly, 80% of managers see AI as an opportunity which islikely to create jobs. Although, as the article points out, they have an incentive to do so.

Jobs more likely to be safe will include

The jobs which will be around in the future are those which have a broad base to them. These are jobs which require workers to employ multiple different, parallel activities in unison. For example, creativity, critical thinking, business casing, innovation.

Elements of each job might be replaced by process automation. For example, there are AI programmes which will book your appointments for you these days and manage your calendar like a personal assistant would. But, if you are influencing, dreaming and then building an idea or improvements to an existing idea, those other bits won’t be replaced for some time. Software can automate well established models or procedures. They can’t automate stuff that’s hard to predict.

What computers can’t do yet :

  • Can’t negotiate or persuade : Managers, nurses and entrepreneurs all need to motivate people to achieve a new goal. Computers don’t have this ability yet and aren’t likely to get it for some time.
  • Creative arts which humans : There is a whole swathe of the writing, drawing and painting side of life which computers can’t do yet. That’s not to say some people aren’t trying.
  • Humans have skills robots don’t : Think of leadership, adaptability, self-awareness, critical thinking, tacit-knowing, conflict resolution, managing emotion, imagination, composite skill use, ethical judgment. In the podcast, the speaker makes the point that we are a long way from robots having these skills.

People to manage and maintain the internet of things – or the robots that do.

Below : The CSIRO outlines 4 scenarios about the future of work in Australia.


This all hinges on how ‘narrow’ AI is at the moment

The influence of AI on employment is clearly not a case of either ‘displace’ or ‘create’. AI will displace some employment, starting with low value jobs. It will also create jobs and make people a lot more productive (and therefore valuable to the companies which hire them)

Real results from real trials suggest that it is – narrow AI is really narrow – it’s task narrow at the moment. In the podcast they actually didn’t talk about it as AI, they talked about workflow automation. The jobs which are listed as most at risk appear to be the ones which are mostly single task.

Admin jobs, accounting tasks, logistics and transport jobs all have significant elements to them which are single tasks. Accounting has a lot of maths. Transport and logistics have a significant component of ‘drive’ to them. Legal has a significant component of ‘review / synthesise / identify ideas in large amounts of data. Admin has a lot of filing and simple procedure work. That seems to be why these jobs are at such risk of replacement. Automating the task largely automates the job.

What happens when we’re doing less drudge work ?

So what happens when the boring bit of our jobs are automated ? There’s an ‘alpha’ factor to the automation of drudgery. I.e. an ‘unknown benefit’ of people being freed up from the drudge work. It frees peoples’ thinking to the point that it happens at a new level.

When the economy created enough capacity that we could support people beyond feeding them ( i.e. we moved from agrarian, hand to mouth economy ) we started working on science, art, literature – the things which might now be considered the best bits of humanity. It created room to do the right thing in a broader context. We freed slaves and slowly started accepting the bits of society which had been previously downtrodden. It was the process of civilisation.

What will the fact that people are doing less drudge work ( no mindless, repetitive work ) do for the production of new ideas ?