Friday, 1 May 2015

The age of perception: a $20 trillion dollar per year opportunity

During the last two to three years, the world has changed.

Perception has been solved. Computers were bad at it. Now they are good at it.

The main culprit is the new generation of artificial neural nets branded as deep learning. Great leaps have been made in image, video, speech and text semantic analysis. The improvements have been stunning in a field used to steady incremental improvement. It has been, and continues to be, simply fantastic scientific drama.

What's the likely economic impact?

On 30 April 2015, PWC released a good whitepaper on STEM with support from the Centre of Policy Studies (CoPS) at Victoria University. CoPS rolled out their famous computable general equilibrium (CGE) modelling to support the economic considerations. As summarised by PWC, the key findings of the report included:

  • 44 per cent or 5.1 million current Australian jobs are at risk from digital disruption in 20 years
  • 75 per cent of the fast growing occupations require STEM skills
  • Changing 1 per cent of the workforce into STEM roles would add $57.4 billion to GDP
  • Top three occupations least at risk in the workforce of the future are doctors, nurses and teachers
  • The top three occupations at risk are accountants, cashiers and administration workers

Much of the report is about the probabilities of disruption to various occupations. Those probabilities were derived from a 17 September 2013 Oxford University paper, "The future of employment: how susceptible are jobs to computerisation" by Carl Frey and Michael Osborne. They reviewed 702 US occupations and found 47% of the total US employment at risk. Not too dissimilar to the 44% of Australia population at risk found by PWC/CoPS.

What does at risk mean? In their conclusion on page 44 Frey & Osborne say, "We refer to these as jobs at risk – i.e. jobs we expect could be automated relatively soon, perhaps over the next decade or two."  Frey and Osborne suggest we do the following to stay relevant, "For workers to win the
race, however, they will have to acquire creative and social skills."

PWC quote an impact of a benefit of some $57 billion being added to GDP for an increase of 1 percent in STEM roles for Australia. A one percent workforce increase is pretty conservative given Australia is over 30 percentage points down on Singapore STEM graduate output from their universities. I understand PWC's reluctance to headline the true scale as it is enormous and difficult to seem credible. It is a revolution on the scale of the industrial revolution.Luddites will be misunderstood again. Opposing the moon and tides is no real choice as you'll be naked when the tide goes out.

Let's meander through the numbers as applied to the Australian top automatable jobs. We'll then multiply by a population factor 14.6 (321M US to 22M Oz). I'll crudely use the Australian Average Weekly Earnings from the Australian Bureau of Statistics which is, um, less than ideal. Then for completeness we'll just scale to the world's population which would have more roles affected but at less value, but let's not allow another complete fudge to stop us. The scale is suggestive. 50% either way and the big numbers are still big. It is a revolution.
Employment categoryProbability of automationWorkers impacted$Billion
per year
Accounting clerks/bookkeepers97.5%263,34821.1
Checkout operators/cashiers96.9%128,74510.3
General office administration96.1%284,17122.7
Wood machinists93.4%31,0812.5
Financial and insurance93.1%128,42510.3
Farm, forestry and92.5%106,0178.5
Personal assistants and92.4%137,91711.0
Sales administration workers91.1%56,9644.6
Keyboard operators87.1%59,8524.8
Hospitality administration and85.5%248,86219.9
Sales assistants and85.2%698,78055.9
Real estate sales85.2%70,6735.7
Factory process workers84.6%52,6314.2
Fabrication trades workers84.3%90,0397.2
Clerical and office83.8%114,7109.2
Printing trades workers82.9%23,9301.9
Mobile plant operators82.8%127,29810.2
Food preparation assistants82.5%154,43812.4
Food process workers82.2%63,0725.0
Glaziers, plasterers and81.4%60,9774.9
Food trades workers80.7%173,63913.9
Automobile, bus and80.5%94,9467.6
Machine operators80.1%83,7576.7

Australian Total


US population multiplier

US Total


$4 trillion of effect per year for the USA. That is quite a sizeable disruptive opportunity. So, the guess is that at around 20-30% of GDP represented by such services will be disrupted by automation. World GDP in 2014 was around $77 trillion, so perhaps we may be looking at a $15 to $25 trillion per annum dollar disruption across the world. That is quite some blue sky opportunity. And risk. You'd better put your head down and be part of the revolution rather than a casualty.

No one likes the guillotine.

At the bottom of this post I've listed the job categories from the PWC report they claim will be less affected. Not quite sure why DBA is in there, but there you go. Debate away.

Personally I think this is going to happen much faster than Fey and Osborne thought (one or two decades) due to the current perception revolution. Let's meander through this revolution in science.

The Age of Perception

Computers have promised since Arthur Samuel's checker program in the 1950's to give us the Jetson age with household robots and cute metal dogs with wagging tails. Sixty years later we're still waiting. A major problem has simply been that computers cannot perceive. No sense from their senses. They have struggled to visually label objects, turn sound into text and never in a crowded room, nor have they been all that good at working out if a story on Reuters is good or bad for a company.

Five years ago they still couldn't do any of that acceptably well. Now they can. The dawn of the Age of Perception has passed and the Age of Perception has begun. Computers can perceive.
Atlas, platform for the DARPA challenge
from Boston Dynamics
which is now owned by Google.

In some cases human performance has been exceeded. Indeed, human level performance is under threat across the entire perception spectrum. Don't forget the perceptions we don't have, such as radar, ultrasonics, infra-red. It is not just wider fields and ranges. The practical harnessing of this powerful array of new perceptive powers has hardly begun.

Geoffrey HintonYann LeCun and Andrew Ng are three notable stars who survived and emerged from the second multi-year neural winter as part of a rapidly growing constellation. It's not just for the big firms. Hundreds of start-ups are scrambling to benefit. Robotic systems at last have the promise of very useful perception.

We're not talking about truly intelligent systems yet, but we will see seemingly intelligent systems thanks to the discovery of the new perceptive powers. I'm referring to useful semantic analysis from sensed data as perception. You have to be careful not to confuse perception, such as semantic classification of a video feed, with true higher level cognitive reasoning. Many tasks just need good perception and adequate heuristics.

You may have used a deep net in the last week and not noticed. The Android or Apple speech recognition on your phone are types of deep learning nets. This is not yet another promise of maybe in a few years we'll really crack it. The future is now. There is a tsunami of application potential building from this great tectonic rumble that is already affecting our lives.

The long awaited automation, or semi-automated, revolution is happening. It is not just about going from cheap floor vacuums to self-driving cars. Our newest children may never know what it is like to pick up the toys of their own children, wash dishes, clean rooms, iron clothes, weed the garden, or do perfunctory cooking.

So why has Google been snapping up robotic and deep learning companies?

Well, it is probably a $1 to $10 trillion dollar a year opportunity, just in the US. Maybe it's the $20 trillion dollar global opportunity. Maybe it's just the fight to stay relevant. Buggy whips indeed.

This is going to affect you and your family and the choices you need to make. Governments need to plan for the disruption. Societies' economic fabrics with be dissolved and rebuilt simultaneously by the invisible hand.

Fight or flight? Ignorance is flight. The choice is yours.



Note: PWC used computable general equilibrium (CGE) modelling to arrive at their conclusions but it still amounts to crazy arsed intelligent guessing at the end of the day.

I also liked the list of jobs that are unlikely to to be automated. It is hard to agree with all of the probabilities above and below, but it is interesting and a fun debate. Read the Oxford paper. You'd better get yourself and your kids one of these jobs:

Employment category Probability of automation
Medical practitioners 0.4%
Education, health and welfare managers 0.7%
Midwives and nurses 0.9%
Advertising, Public Relations and
sales managers 1.5%
Database and systems administrators, and
ICT Security Specialists 3.0%
Education professionals 3.3%
ICT managers 3.5%
Tertiary-level teachers 3.6%
School teachers 4.0%
Engineering professionals 4.2%
Legal professionals 6.5%
Social and welfare workers 6.8%
Accommodation and hospitality managers 7.2%
Construction, distribution and
production managers 8.2%
Child carers 8.4%
ICT network and support professionals 9.7%

No comments:

Post a Comment