10,000 Hours of Training

What do Bill Gates and the Beatles have in common? Apart from being extremely successful in their respective fields, both, according to Malcolm Gladwell, have 10,000 hours or even more of practice to thank for their success. In his book, Outliers: The Story of Success, Gladwell attributes the Beatles’ musical success to playing over a thousand overnight shows in Hamburg. Meanwhile, in Gladwell’s interview with Bill Gates, the Microsoft founder says that having access to a high school computer a time when they were not commonplace helped him succeed. Gladwell asserts that this early exposure to computing meant that by the time computers were commonplace, Gates had exceeded the requisite 10,000 hours of practice for success.

Machine Minds

Apart from making Bill Gates the richest man in the world for a significant part of the past few decades, computers have brought on many changes since they were introduced. While Bill Gates and numerous other programmers have long had to give very specific instructions to computers to automate tasks, software development has evolved such that computers may be programmed to learn how to do tasks instead of being explicitly programmed. It’s clichéd, but machine learning truly has brought an entirely new paradigm to software development and it is set to change our relationship with knowledge and practice altogether.

Computers have a history of beating humans at games, from Deep Blue defeating Garry Kasparov around two decades ago to AlphaGo beating Lee Sedol and AlphaGo Master besting Ke Jie in more recent years. The illustration that accompanies this article was not painted by a human but was created using a technique called neural style transfer where a computer “paints” an image based on the style of a reference image. This is even before going into the ubiquitous use of artificial intelligence in more “mundane” fields such as natural language processing and image recognition.

Machines are more suited to the rigorous and time-consuming training that has traditionally been linked with success.

This phenomenon does not disprove the idea that prolonged training (or in Malcolm Gladwell’s words, er, numbers, 10,000 hours) leads to the skill necessary for success but instead validates it as artificial intelligence is fueled by machine learning which literally translates to machines spending hours training using data to solve problems. This only shows that machines are more suited to the rigorous and time-consuming training that has traditionally been linked with success.

Human Success in the Age of the Machine

What then must we humans do to succeed when practice is something that a machine can invariably do better? Success happens when talent meets opportunity, and there are fortunately still areas where humans can add value and have the opportunity to succeed.

Having attempted to automate much of his production line, Elon Musk once remarked that “humans are underrated.” There are complex and varied reasons why this statement rings true but it wouldn’t be too much of an oversimplification to say that this is because machines need data to learn. To give a concrete example, a machine that has been programmed or trained to distinguish between Tesla Model 3’s and Model X’s can be trained on a photos of red Model 3’s and white Model X’s and may thus mistake any white object as a Model X. This may seem like a trivial example but several incorrectly trained machines in a fully-automated factory will lead to disaster without human intervention.

This may be explained as humans gaining an edge over machines not due to our depth of knowledge in a specific field but due to our breadth of knowledge across several fields. Of course, one may argue that as artificial intelligence goes beyond artificial narrow intelligence (ANI) which is can beat humans at specific tasks to artificial general intelligence (AGI) that can show expertise across all fields this statement is invalidated but that time has not come yet.

For now, the machines are here and they can already practice better than any human and for longer than any human. Fortunately for us, the machines are not necessarily our competitors but are generally at our disposal. Machines have always been here to automate structured and repetitive processes — what has changed is that the task of learning now falls under this realm of processes that can be automated thanks to advances in machine learning.

The onus therefore lies on us to best utilize machines to increase productivity and improve quality of life. This means that the more tedious parts of the task of learning can be outsourced to machines. This is not to say that the pursuit of knowledge should be abandoned entirely but rather that machines can be leveraged to greatly accelerate this process.

Artificial intelligence is already being used extensively in the sciences. Astronomers utilize classification algorithms in conjunction with computer vision to quickly process data collected by telescopes and identify patterns. CERN relies heavily on a combination of neural networks and gradient-boosted models to efficiently distinguish between particle signals and background noise in the huge amounts of data generated by experiments using particle accelerators. The alternative would require scientists to spend hours or days or even weeks of precious time to run the analysis manually. The cost of time lost developing artificial intelligence is far less than the price of human effort wasted. Soon, every organization will need to leverage artificial intelligence to stay competitive.

With machines doing the heavy lifting, humans are free to focus on other tasks. In the sciences, this may involve brainstorming theories and allowing machines to validate the underlying mathematics. In a business setting, this means that repetitive tasks where performance scales with practice are best passed off to machines while humans do more strategic work in guiding the overall company direction — the worker of tomorrow is not a craftsman but a manager of machines.

Taking this idea further, one may say that years of experience on the job will no longer matter as heavily as they used to as long as one displays the necessary skills in critical thinking and adaptability that are commonly associated with managerial roles. Previously, employees progressed on the career ladder on the basis of tenure but, with machines gaining work experience on behalf of human workers, the best employees of tomorrow will merit their positions through business savvy and intelligent application of automation.

In a world where learning itself is increasingly automated, it isn’t just how many hours of effort and work experience that is important but also the skill to identify where effort pays off. The age of craftsmen learning their métier through years of practice is coming to an end but this need not be the end of human productivity and expertise. On the contrary, machine learning has the potential to democratize expertise as long as the data and computing power required is made available.