Algorithm or Alchemy?
Published B&T – April 2017
Machines have been replacing people since the industrial revolution.
The pyramids would be knocked up a hell of a lot faster (with much less loss of life) using modern machinery and techniques. (Though it could be argued that the build quality would not be improved upon – check out inner city Sydney units built in the last 5 years.)
Moving from the industrial to the information age, we are now experiencing machines (computers) taking over the thinking once done by people at an ever - increasing rate. And there is no argument that computers are much faster and more accurate than people. They are also a hell of a lot cheaper and don’t carry the “baggage” that many people bring to a job. (Never had to fire a computer for making an unwanted pass at the photocopier).
As computers evolved, they have become smaller, faster and permeated virtually every aspect of our lives. We now have a whole generation who have not experienced life without computers. (Nor could they imagine life without them). I was helping a good friend who is a professional musician and guitar teacher research online guitar courses and came across the following: “After playing guitar for over 12 years, we tried the most popular lessons online, rated each with a custom algorithm and then wrote reviews to help other guitarists choose.” It reminded me of the numerous times I see the use of “algorithm”. No matter what you want done, there is an algorithm available for it:
- A business who monitors and does analyses of social media by: Aggregating all a company’s social media posts and using an algorithm to determine which are the most relevant.
- A business started by a dual rugby league/union international uses “an algorithm” to identify power players with the most potential to change consumer behaviour. (Links brands to sports stars’ social media followers).
- Media planning and buying: “An algorithm is used to select and buy the most cost effective schedule from available online media inventory.”
- It seems that once an “algorithm” is brought into play, then the result it spits out is automatically deemed to be correct.
Should there be a rigorous examination of the said algorithm, rather than blind acceptance of the outcome?
Take the question of programmatic buying. With so much online media inventory, old style calculations would take an eternity. Computers literally calculate at the speed of light; therefore, it is obvious that the numbers are crunched by computer. There are too many combinations and permutations for an individual with a calculator to attempt.
So, the question is not “is programmatic buying” a feasible solution, rather how good is the program (or algorithm) being used? What calculations and assumptions were made and what is the accuracy or validity of each? The same applies to the social media and sport stars’ algorithms.
For the social media algorithm, someone must first make assumptions on which words or phrases are deemed relevant. If any of these assumptions are incorrect, then the resulting algorithm is going to be incorrect.
Identifying sports’ stars with the ability to change human behaviour would logically extend to any celebrity or high profile individual. How the hell does one determine if a person can change human behaviour? Changing human behaviour is far more complex than asking the question ‘If (sport star a) recommended this product, how likely would you be to buy it?” Having a magic algorithm sounds so far more in tune with the digital age. A victory for marketing fashion over function (again).
A basic fact often forgotten, or not understood - that a person does the thinking, a computer does as it is instructed. Quite often the individual writing the code or algorithm knows very little about the topic for which the algorithm is being developed.
Airbnb, is a huge international success. As a disruptive business that would not exist if not for the internet, you would assume they have their act together. We use it for a property on the Gold Coast. They introduced a new algorithm, called “Smart Pricing”. The price of accommodation varies – school holidays, Christmas, New Year, Easter etc. Instead of manually going through every day in each month of the calendar and entering a price, “Smart Pricing” uses an algorithm to calculate the pricing for you automatically. All you do is set a minimum and maximum price for it to work from.
A great idea in principal, saving a lot of time and for some, much guess work. It wasn’t long before we received a request for two nights, 4 months in advance – the price the algorithm quoted wasn’t sufficient to cover costs for cleaning. It was less than one quarter of the minimum price that was set. (Yet the “algorithm” asks to input your minimum and maximum prices). The person writing the code did not know how to do the maths to work out a price variance equation based on demand and a fluctuating occupancy continuum - minimum and maximum price of the property and because Airbnb asks for 100% of the accommodation cost at the time of booking, guests book “short” rather than “long” – a shit load of variables that require a detailed understanding of mathematics, not some code monkey’s basic statistics. In this instance the mistake was so obvious, it was easily picked up.
But what if the calculation is for say, programmatic media planning? A good way to illustrate the difference between “calculating” and “thinking” is a headline and introduction for a recent story in TheDrum: Mercedes-Benz asks agencies to 'review' safe media list as brands respond to 'programmatic terror' dilemma.
A report that some of the world's biggest brands are inadvertently funding extremism and pornography through programmatic ads has prompted Mercedes-Benz to review the platforms and channels it considers safe.
The next “big thing” after programmatic is the development of algorithms to create advertising and content – computer generated creative. Good luck. Determining all the variables at play is close to an impossible task and adding further to the difficulty, the variables influencing the creative output vary from situation to situation, even when working within strict corporate guidelines. The human brain is the most complex piece of “machinery” on earth and many of its workings are still a mystery to research scientists. How the hell does one develop a computer programme (sorry, algorithm) to duplicate something that no one truly understands the workings of? If anything, computers are dumbing us down – the Google syndrome: “I know the answer but have absolutely no understanding of how it is derived.”
We are still a long way from artificial intelligence. (Singularity, when artificial intelligence overtakes human intelligence.) All algorithms are an analysis of a problem, using a pre-determined formula and assumptions developed by a person. Yes, some are incredibly complex and developed from “machine learning”, but the “learning” is programmed. Deduction is not imagination.
An algorithm cannot turn lead into gold any more than it is an infallible, “magic” solution to a problem or calculation to which the path to the answer is not already known. So rather than merely accept a conclusion is correct because an algorithm was used, the question should be “can we see the formula and assumptions used to develop the algorithm”? You don’t have to be a code writer or mathematician to do this. Understanding of and experience in your area of practice will tell you if the assumptions made are correct.
The basics really don’t change. And when it comes to algorithms, programmes or what- ever term is deemed the most fashionable, the absolute fundamental that has been around since the first computer was invented still holds true:
Rubbish In = Rubbish Out.