Think Carefully: Could A Machine Do Your Job?

If just 30% of jobs could be performed by machines and all the displaced workers found replacement other jobs – then this would still cause a massive shock to the global workforce

Marc Andreessen wrote an excellent opinion piece in the Financial Times a few weeks ago on whether robots will eat jobs, or unleash creativity. If you do not subscribe to then here is the essence of Marc’s thesis:

  • Better access to education and skills development will allow displaced workers to find new jobs;
  • In the long term, people will still be better than machines at things like creativity, innovation, exploration, art, science, entertainment and caring for others so jobs like these will not be at risk;
  • Just as most of us have jobs today that did not exist 100 years ago, the same will be true 100 years from now: technology will result in the creation of more than enough new jobs to replace those that might be lost.

Basically, the message was “Don’t worry, everything will be OK - you can trust Silicon Valley.”

There is of course a counter-argument to this, which I’m going to outline below.

But you should know before you read any further that, on this occasion, the counter-argument is pretty terrifying...

How to estimate how many jobs might one day be replaced by machines

If we could get a handle on the type of jobs that might be at risk then we should be able to estimate how many jobs were at risk. We could then take a stab at estimating the impact that machine intelligence might have on the global economy.

I realise that this is a hugely complex question that warrants a year-long study, but I only had a couple of hours – so here’s what I did.

The online employment service provided by the UK government, Prospects, lists 630 job types. I took a random selection of 100 jobs from this list and considered those.

For each of these 100 job types I assigned a score from 1 to 5 based on how vulnerable I thought that job was to replacement by intelligent machines – not now, but in the far future.

The scores were simply based on a quick personal qualitative assessment with a ‘1’ meaning that the job was ‘unlikely’ to be replaced by intelligent machines while a ‘5’ meant that the job was ‘likely’ to be replaced by intelligent machines.

You can see the full table here.

A vision for the future of machine intelligence

In order to make qualitative assessments like these it is important to have a clear view of what machine intelligence might look like in the future. Of course, no matter how clearly you think you can see the future, there is no assurance that things will unfold as you envisage, but you need to make some pretty major assumptions to get anywhere with a task such as this.

And so at this point I should explain something about what I see when I wonder what machine intelligence will look like in the future:

  • I foresee a future where machine consciousness and machine creativity exist, at least to some extent.
  • While I see these capabilities as being part of advanced robots, at least to some extent, I foresee something more fundamental, which is:
  • I see ‘machines’ emerging which combine unimaginably powerful data processing ability (which humans are very bad at) with some level of consciousness, self-awareness and creativity (which humans are very good at). Just as with modern software, these machines would mainly ‘live’ in the cloud but they would also have a presence in electronic devices as well.

My belief is that a machine that combines Google-like data processing ability with limited consciousness and creativity will be vastly more capable than the best human minds (let’s say, Mozart or Einstein) - which are creatively brilliant but can only absorb information at the rate of 300 words a minute (say).

In this view of the future, an intelligent machine would work at a speed that is 1000x faster than any human mind, or even much faster. Time would, in effect, speed up for machine minds and slow down for us. 

The capabilities of these 'machine minds' might be limited to specific areas (vertical markets, if you will) but within their narrow fields, their capabilities would be vastly superior to ours - and they would eclipse minds like those of Einstein, Mozart and Hawking.

If you think that machine intelligence will never reach this point (and I’d remind you that ‘never’ is a very long time...) then you should probably stop reading this post right now.

I don’t want to believe in the feasibility of this sort of technology, but what is going on today in research labs around the globe has forced me to confront my own limited intellectual abilities and ask whether a machine could one day do my job better than me.

Example: management consultant

Take the job of a management consultant. An intelligent machine would be able to:

  • Access and analyse data on millions of different business situations in order to identify past examples of situations that are most relevant for a specific client problem.
  • Analyse thousands of relevant prior consulting assignments and then compare what actually happened with what the consultant proposed.
  • Analyse various strategic options in a kind of virtual business environment: we are on the verge of modelling the global weather system and the human brain itself so there seems to be no reason why the economy, including human behavior, could not also be modeled in the future.

A future advanced intelligent machine could do all this in seconds.

The machine I’m envisaging would have real-time awareness of all relevant facts and an ability to answer any question accurately and almost instantly. The machine would be aware of the best solution in seconds, seemingly without having done any actual work. Yet when the underlying work was examined by a human it would be clear that the machine’s ‘final report’ ran to 100s of thousands of pages of unimaginably complex analysis – which would be incomprehensible to even the best management consultant.

To be clear, I am not talking about machines being used to help management consultants – I am talking about machines that would replace management consultants, and many other professions as well.

If this is right then what would these displaced professionals do?

Marc Andreessen’s suggestion is that these displaced workers would just need to sign up for online training courses and retrain themselves so they could get jobs elsewhere – but what would those jobs be? If the type of intellectual work they used to do is now done by machines then maybe the retraining would involve learning how to perform, um, manual labour (in which case an online training course might not be the best choice).

What sort of jobs might be replaced by machines in the future?

The scoring process for the random selection of 100 jobs produced two interesting results:

  • The types of jobs that appear to be most vulnerable are very different to what I’d imagined;
  • The average 'at risk' score was 3.0 (halfway between 1 and 5) - the significance of which we’ll come to later.

The following table lists two sets of job types which were assessed very differently: the jobs in the left column were all allocated a ‘5’ and were deemed most ‘at risk’ while the job types in the right hand columns were allocated a ‘1’ and were deemed least ‘at risk’:

If the scoring is broadly correct then it has a very surprising implication: machine intelligence would appear to be most likely to replace the sort of jobs that are presently performed by sedentary people who spend long periods of time finding, analysing and interpreting data and information.

In case the penny hasn't dropped: this probably means that your job is at risk.

And my job is certainly at risk.

Oh, and Marc Andreessen’s job is at risk as well. Sorry Marc – but you understand more than most about creative destruction so you'll be OK...

While the initial thought might be that machines will mainly displace low-paid, manual jobs this simple analysis indicates that many of these jobs might actually be safer than professional jobs.

Machine intelligence might be the first technological revolution which mainly impacts the top of the employment pyramid: those at the top being most at risk.

This is clearly a very bold statement that is supported by a very crude analysis but the results are so clear that if it was repeated more accurately (good luck defining the methodology, by the way) then I think the message would be similar.

Interestingly, I read an article in The Telegraph this morning where David Willetts, the UK university minister, argued that in the future “high-level professions are more likely to be taken over by computers than those that involve apparently simple tasks”. I agree.

Will machines ever be able to replicate the ‘spontaneous creativity’ of humans?

Example 1: Music composition

If you’ve not already read this article about David Cope’s ‘intelligent composer’ Emily Howell, then you might like to do so.

Although published in February 2010 it is still a remarkable read: a trained concert pianist and classical composer, David Cope is also an academic who holds the position of Dickerson Emeriti Professor at the University of California at Santa Cruz, where he teaches music theory and composition.

Professor Cope has spent over 30 years developing a computer system, called Emily Howell, that he uses to help him compose original music.

In 1987 (correct date), Professor Cope programmed his computer to compose a selection of pieces ‘in the style of Bach’. When these pieces were first performed, at the University of Illinois at Urbana-Champaign, “they were met with stunned silence”, according to the above article (which includes some embedded audio clips which were composed by Emily Howell).

Imagine what might become technically possible – in the far future:

Let’s imagine that you wanted some suitable music for your partner’s birthday party. Because you know that your partner likes Haydn you ask your home hi-fi system to compose a piece of original harpsichord music inspired by Haydn. The music would then be ‘performed’ over a hi-fi system that included an advanced harpsichord synthesiser.

Guests would be able to experience something akin to what royalty might hear today: a piece of music composed specially for a state occasion that was performed by a virtuoso harpsichordist.

Music lovers might recoil in horror at this scenario, but we should not dwell on whether we think this sort of performance would be acceptable, but how feasible is it from a purely technical standpoint?

Example 1: Advanced mathematics

While reading about this topic over the weekend I found another article in the Telegraph about AI, this time written by Professor Marcus du Sautoy, who currently holds the position of Professor of Mathematics at the University of Oxford.

In the article, Professor du Sautoy says “[A machine] may be able to completely outperform my brain in any computational activity but when I'm doing mathematics my brain is doing so much more than just computation. It is working subconsciously, making intuitive leaps. I'm using my imagination to create new pathways which often involve an aesthetic sensibility to arrive at a new mathematical discovery. It is this kind of activity that many of us feel is unique to the human mind and not reproducible by machines.”

But what if this sort of creativity can be reproduced by machines?

If it turns out – as some leading scientists now believe – that consciousness is simply a state of matter, rather than having a divine origin, then creativity might be the same. If so, then this would mean that in the far future we would expect machine brains to exhibit creativity to a similar or greater extent than human brains.

If Emily Howell can compose original classical music that fools music experts then perhaps machines might one day invent a new field of mathematics - one that might be too complex for human brains to comprehend.

So returning to the somewhat provocative title of this article do I think a machine could do my job?

Um, yes...

I certainly don’t like the thought of it. But thinking carefully about what I do I’m pretty sure that an intelligent machine could not only perform my job – but it would be able to do it far better than I can. 

How many jobs could be replaced by machines?

We can do a very simple calculation by assigning some probability to the numbers ‘1’ and ‘5’ that we used to score each of the random sample of 100 jobs.

Let’s say that a ‘1’ means that the job has 10% chance of being replaced in the far future by a machine and that a ‘5’ means that the job has a 50% chance of being replaced.

Based on the reasoning in this article these numbers do not seem too extreme: remember we are not placing a time limit on when these replacements might occur: we are just trying at this stage to envisage what things might be like in the far future.

The average score for the 100 jobs I looked at was ‘3’ which equates to an overall probability of (10% + 50%) / 2 = 30%.

If the sample of 100 jobs we looked at (see list at the end of this article) is representative of all jobs then this would imply that 30% of all jobs are at risk.

What impact would machines have on the global economy?

According to the World Bank, global GDP in 2013 was USD 74 trillion. If 60% of this represents consumer spending then 30% of the workforce represents roughly 30% x 60% x USD 74 trillion, or USD 13 trillion which is 18% of worldwide GDP.

If a machine economy produced an incremental USD 13 trillion in value then the overall economy would be unaffected in terms of GDP, but this would not explain what the displaced workers would do: as I pointed out above there might not be very much for them to do at all.

At this point I could join up the dots, but I’ll leave that for a future blog post.

So what does it all mean?

The question of whether robots will eat jobs, or unleash creativity is far more complex than Marc Andreessen’s article on suggests: I can see two possible market outcomes:

  • Machines remain subservient to humans and we all prosper with a continuing rise in economic output and a narrowing of the income gap. This is Marc’s view.
  • Machines eventually replace what might be 30% of all jobs with most of the replacement occurring in the professional job market. This would lead to social upheaval and the human mind being surpassed by computers.

Sadly, based on where we are now and where technology is headed it seems that the second outcome is the more likely: in the future, your job will almost certainly be at risk, and so will mine.

Although a longer-range problem, I am coming to believe that machine intelligence presents a far more serious threat to humanity than climate change - and I say that as a professional technology analyst and engineer.

What do you think?