AI’s coexistence with humans and why we should all be concerned

With today’s AI technology, a large proportion of jobs will, can and have been displaced

“The invention of the steam engine, the sewing machine, electricity, have all displaced jobs. And we’ve gotten over it. The challenge of AI is this 40 percent, whether it is 15 or 25 years, is coming faster than the previous revolutions.”
— Lee Kai-Fu

The above quote encapsulates the prophecy that Lee has for this AI Revolution we currently live in — that 40 percent of the world’s jobs are displaceable by AI technology in the next 15 to 25 years, with a key difference being the rate of change of this displacement.

AI Revolution — The Third Industrial Revolution?

In human history, we survived the First Industrial Revolution — a massive transition to new manufacturing processes and revolutionized the way goods were produced. We then thrived during the Second Industrial Revolution , also known as the Technological Revolution — a phase of rapid standardization and industrialization of goods and services, including areas such as railroad networks, sewage systems, electrification, maritime, automobile and many more.

Many now coin the age we live in as the AI Revolution, and potentially what our (future) kids might know as the Third Industrial Revolution , simply because the way goods and services are consumed have massively evolved. The sheer number and types of services provided today have seen a surge in numbers relative to the number of goods produced, because the Internet has made this possible.

Not just the Internet, but the rapid development of peripherals that support the entire Internet ecosystem and subsequently the development of AI.

Back To The Beginning — The Rise of AI

AI is a technology that thrives on problems that has a clear input and output. Typically, these are optimisation problems (supervised learning) such as credit scoring, spam classification and sentiment analysis. When the output isn’t labeled (e.g. spam/not spam, positive/negative), we may also seek to do classification (unsupervised) based on what the algorithm ‘thinks’ each data point should be classified into. The aforementioned are typically traditional machine learning methods (e.g. random forests, support vector machines, logistic/linear regression, density-based models) that have been around and studied on for a long period of time.

What is causing this AI Revolution is the proliferation and wide spread adoption of advances in deep learning technology. It is not the “creation” but the adoption of this technology.

As Lee coins it, we are not in an age of discovery but in an age of implementation . It is easy for people to be misled that AI is currently in its infancy as media continually reports on new groundbreaking results such as AlphaGo , Dota 2 and Starcraft beaten by Google’s DeepMind and Elon Musk’s OpenAI . Back in 2016, Lee Sedol could not even conjure the thought of a computer program beating him at Go . Today, he has retired (in 2019) from Go because he thinks that “ even if [he] becomes № 1, there is an entity that cannot be defeated ” [5].

Following AlphaGo , professional teams in more complex real-time multiplayer online battle arena (MOBA) games such as Dota 2 and Starcraft were also dethroned in what people thought to be impossible tasks. In a turn-based game like Go , a computer program could possibly exhaust all the game’s possible states ( the state of the board or game at any given point in time ) and predict what the next best move should be, training over many iterations of the game and employing various explore-exploit strategies. With MOBA games, these game states exponentiate to (possibly) infinity, including additional complexities like game theory, imperfect information, long-term planning and real-time multi-player interactions [3]. What people did not know however, is that these breakthroughs were in fact different applications of the same underlying technology .

The secret sauce? Deep Learning .

Deep learning employs the use of neural networks — a unique way of parsing inputs through a certain network structure – to train a model to understand the linear and non-linear relationship between input and output. Neural networks are an attempt to mimic the neurons in our brains in a similar structure, with weights acting as the amount of importance we place on each feature (factor) when we make decisions. This ability to capture non-linearities is a key differentiator from traditional machine learning models which are mostly linear in nature.

Why wasn’t the technology adopted earlier?

Even though deep learning has been around since the 2000s, the community did not recognise it till 2016 when a team of AI Researchers led by Yann LeCun (one of the Godfathers of AI) surpassed the benchmark error rate in the largest image recognition competition in the world, ImageNet . Back then, there was little adoption due to a combination of lack of research funding, computational power and recognition.

Today, with GPUs and TPUs powering our AI systems, people are able to take that open source technology and apply it to many different domains. What used to be a field dominated by PhDs now no longer had that same barrier to entry because while advancing technology requires advanced technical skills, application of technology does not require that same level of rigorousness. Herein lies the beauty of open source technology, where building blocks are provided for people to experiment and even improve upon.

With this new technology, we can see the rapid application of deep learning across different industries — from manufacturing to retail to banking, and from blue-collared workers to white. The age of AI is an age of change, an age where disruption may become the new normalcy. Every so often, we see industries being disrupted, operations being topsy-turvied, and our roads potentially being populated with autonomous vehicles.

Who is at risk of ‘displacement’ by technology?

Lee broadly categorizes jobs into two distinct categories — physical and cognitive labour. In both charts below, the vertical-axis describes the level of social involvement in a job.

The first chart represents the risk of replacement for physical-laboured jobs, with the horizontal-axis describing the level of structure in the environment. As described earlier, AI is great at learning tasks where the input has some form of relationship to the output (i.e. approximating f(x) = y ). Needless to say, rule-based jobs are the easiest to replace as it does not require AI but simple automation.