Will Machine Learning Replace Human Know-How?

Will Machine Learning Replace Human Know-how?

January 23, 2018


Kevin Vecmanis, P.Eng

I was walking back from the Vancouver Resource Investors Conference yesterday and I saw a sign that said,  “Will Machine Learning Replace Human Know-how?”.

I have seen and read a lot of articles online exploring this hypothetical question, but I wanted to take the time to explore this question and its potential answers through the lens of first principles (economic fundamentals).  Amazon launched their first fully autonomous grocery store this week.  This is an amazing proof of concept that demonstrates the abilities of cognitive automation. People should take note of this because it’s a major step in the unstoppable trend of automation.  In this article I want to discuss the economics of this trend.  First, a quick discussion of how society even got to this point.

Automation is a trend, not a threshold.

Automation and mechanization has been an ongoing economic trend that was kick-started in earnest during the industrial revolution. Technology, in essence, has always had one of two distinct and potential impacts on an economy (in some cases both):

1. It shifts the demand curve for labour to the left within the industries being automated.

2. It enhances the productivity of the workforce, defined by Output/Unit of labour.

Rising productivity is the measure for a rising standard of living – producing more with the resources we have available.  With rising productivity comes a rising standard of living, rising real wages, and all other societal and economic benefits you would expect from a prosperous society.  There are those that are negative on technological trends and society as a whole.  My counterargument to that is that, as a whole, the data suggests that we live in the most peaceful and prosperous era humanity has ever known.

Human labour, whether its physical or knowledge-based, can be subdivided into two mutually exclusive and collectively exhaustive categories:

1. Repetitive tasks.

2. Non-repetitive tasks.

Since the industrial revolution, repetitive tasks have been the “low-hanging fruit” from a capital investment standpoint.  Advances in technology have largely facilitated easy replacement of repetitive labour tasks that have freed-up the labour force to focus on more complex and higher value non-repetitive tasks.

The only repetitive tasks that haven’t been automated are those for which technology has not advanced sufficiently to displace the ability of the human worker.  Or, the total life-cycle cost of the technology still exceeds that of employing a human in that task. Technology will never fully replace human beings in the workforce because that presupposes there’s a finite number of things humans can do.  This simply isn’t true.  Faced with boredom and a lack of meaning, humans have a seemingly infinite capacity to create new industries, interests, and tasks for themselves.  In order for this to happen, our time needs to be sufficiently freed from the mundane tasks of life.

Our visual systems and the tactile dexterity in our hands are probably the last bastion of human ability that is keeping us employed in the realm of non-repetitive tasks.  If you’ve spent time training machine learning systems to recognize anything visual, you can appreciate how far this gap still is.  But this gap is closing rapidly.  The automated Amazon grocery store is proof of this.  Amazon says that the system is so robust that intentional theft is almost impossible without being detected.

In the technology world there’s a rather cold and technocratic term for people – “wetware”.  That is, insofar as people still remain in business processes, these processes are comprised of software, hardware, and wetware.

Take a company like Uber, for example.  The only reason “wetware”, the driver, is used is because there are tasks that still can’t be skillfully assumed by software and hardware.  The visual system of a person is still cheaper and more advanced than our artificial intelligence (AI) counterparts. One day that will not be true, and human drivers will disappear altogether.  This is nothing new, it’s just the inexorable procession of automation running its course.  We have replaced mostly every human job that existed 50-100 years ago with robotics and automation.  There’s still plenty for us to do today.

This is the paradigm shift:  It was once thought that only repetitive tasks could be automated.  Artificial intelligence and machine learning now put all non-repetitive tasks at “risk” of being automated as well. (By “automated”, I refer to a human being replaced or supplemented with an automaton, just to keep the definition straight).

That might sound scary, but lets revisit the first principles again. With the introduction of sufficiently advanced artificial intelligence, there are now three potential economic impacts on the labour force:

1. The supply curve for knowledge workers & labour can now shift to the right (humans + intelligent electronic agents).

2. The demand curve for human labour can shift to the left (human abilities no longer meeting job requirements).

3. The aggregate productivity of the workforce can increase (Total factor production).

Let’s unpack each of these:

The supply curve for knowledge workers & labour can now shift to the right.

What does this mean?  It means that the supply of workers, human or otherwise, that can execute knowledge-based, non-repetitive tasks is going to increase.  All else being equal, when the supply curve for anything shifts to the right, prices for those things must fall in order for an equilibrium to be reached between supply and demand.  Minimum wage does nothing to mitigate this, as the value of one’s labour must always exceed its cost in order for it to persist long term.  Because we are always in competition with hardware and software, whose cost is falling and capabilities rising (both exponentially), legislating a higher wage for people only enhances the business case for automation.  It actually has the effect of outlawing low-value human labour – a challenge for youth trying to gain experience in the workforce.

The demand curve for human labour can shift to the left.  

The supply curve isn’t the only thing that can shift – the demand curve can shift as well.  In order for fruitful employment arrangements to persist, value must be received from both the employer and the employee.  From the perspective of the employer, the sum whole of risks and costs associated with employing somebody must be less than the value of the labour they are providing. If this equation reverses, it means the costs and risks associated with employment exceed the value provided, the situation becomes untenable, and the agreement won’t persist in the long term.  This is a fundamental law of economics.  The cost of human labour won’t necessarily shift the demand curve, but the risks associated with employment can (and will).

Labour laws are tilted, and becoming more tilted, in favour of the employee rather than the employer.  At a time when the costs and risks of machines and artificial intelligence are falling and their capability is rising, the costs and risks of human labour are rising, and arguably the capability is falling on an aggregate level.  Legislation is the number one driver of these risks.  Things like the minimum wage only serve to prohibit, by legislation, any labourer from entering the workforce who can not deliver more value per hour than the minimum wage threshold.   Again, the costs and risks of your employment must be less than the value that you provide.  In many situations, the sum of those factors simply isn’t low enough to justify putting a person there.  What are the risks of employing people? Many.  We live in an extremely litigious society.  A robot won’t sue you for harassment, discrimination, wrongful dismissal, assuming its gender, or being “triggered”.  It doesn’t need breaks, holidays, or vacation. It won’t show up late or hung over.  If it gets hurt or injured on the job you won’t go to jail.  Its costs can be capitalized.  When it breaks or gets too old you can replace it with a younger, newer model without a human rights violation.

I think most people would rather deal with another person, but legislation is driving a wedge between business owners and employees at a time when automating these roles is becoming easier and easier.  The worker will suffer from this trend, not the business owner.

The aggregate productivity of the workforce can increase (Total factor production):

There is no historical precedent for automation degrading economic prosperity.  Industries and vocations can, and do, get disrupted or eliminated along the way.  We live in a dynamic world where people need to adapt – this has always been true.

There is a lot of fear mongering regarding this new brand of automation.  “Hundreds of millions of jobs will be replaced” they say.  Is this true? Probably.  Does this equate to all of those people being unemployed?  Unlikely.  When Oxen were first being used in agrarian societies to enhance agricultural productivity, how many farm workers at the time could imagine that, several generations later, their kids would be programming software for iPhone apps? Or building self-driving cars? Or any of the other myriad roles that exist today that weren’t even conceived of back then.  Mass displacement of workers due to automation has always resulted in the creation of new industries outside the realm of imagination for the generation being displaced.  Computer programming is one of the most sought after technology skills today.  Computer programming wasn’t even an idea 50 years ago, at least in the mainstream.  The most sought after skill 50 years from now likely doesn’t exist right now.  Ponder that for a moment.  The most valuable company is the world, 50 years from now, likely doesn’t exist right now.  The world changes, industries die, new ones are born.

Does it make more sense to halt these advancements so that people can keep working mundane jobs that they hate and readily complain about to anybody that will listen?  I think not.  It’s horrific what people sacrifice today to work jobs that they hate – Time, family, health, friends.  In my prior job I had friends that committed suicide because they felt so trapped and unhappy.  One day, generations from now, I think people will look back on us shaking their heads.  Indeed, people need to feed and shelter themselves and their families, but the extent of the human experience shouldn’t be repetitive monotony, day-in and day-out, until you die or can’t function physically or cognitively.  The robotic automation of dangerous, monotonous, soul-sucking work should be a welcomed trend by everybody.

Whether people are ready or not, the trend in automation is accelerating and I would say we’ve passed the knee of the curve. In my line of work I’m close to these trends and it’s very difficult to keep up.  The decentralization of knowledge has resulted in an explosion of innovation globally.  It’s making people nervous, understandably. Don’t be.  The greatest advancements and mega-trends in human history are unfolding and we have a front-row seat.

Whether people are ready or not, they’re going to be freed from the job that they hate.  They’ll have to figure out how to do something else, like every generation before us.  This should be embraced, because the ability to re-educate ourselves and create new tasks for ourselves is an advantage we have over the machines.  Will machine-learning replace human know-how?  Only if we choose not to do something else.

About the Author:

Kevin Vecmanis is a professional engineer, quantitative market analyst, machine learning practitioner, and a candidate in the CFA program.  He graduated from the University of Western Ontario with a degree in Electrical & Computer engineering in 2008 and currently lives and works in Vancouver, British Columbia.

Be the first to comment on "Will Machine Learning Replace Human Know-How?"

Leave a comment

Your email address will not be published.


This site uses Akismet to reduce spam. Learn how your comment data is processed.