Artificial Intelligence (AI), the pipe dream of cognitive scientists and engineers in the 1950s, has until recently been the domain of science fiction, though once again it has become a huge field of research and technological innovation. Kickstarted by research in Deep Learning, the most cutting edge combination of pursuits in Machine Learning and Neural Networks, there is now a plethora of technological milestones coming out of the field. In 2015, facial recognition by AI software surpassed human level accuracy, and in 2016, DeepMind’s AlphaGO beat a human champion at GO, an incredibly complex game, winning 4 out 5 matches in South Korea. Innovations in the field have quietly moved into our everyday lives in imperceptible ways, from search engines, voice recognition software, movie recommendations, news feeds, and advertising. As the investment increases in the field and more technologies are created, it becomes pertinent to understand the technological capabilities of Artificial Intelligence, and the economic implications they will have.

Deep Learning is the branch which contributes the most research into the automation of human labour. The technique uses layered neural networks, which attempts to mimic the structure of the brain, with each layer using the output of the previous layer as input. This layering provides a hierarchical representation of data from low level features to more complex aspects. These layers are shown large data sets related to their task, with the machine using the data rather than a command to learn the best way to solve the problem. The core philosophical principle guiding the machine learning revolution is that of “tacit knowledge”, that human beings know how to do more things than we can explain or communicate. Hayek used a similar conception of knowledge in “The Use of Knowledge in Society” as part of his critique of socialism, and the justification for his belief in “spontaneous order”. This breakthrough in epistemology inspired a revolution not only in economics, but in the cognitive sciences as well.

These technologies have provided the abilities for machines and software to do things previously unthinkable, mostly thanks to the availability of massive amounts of data for training, and the recent reinvigoration of investment in the field. A way to think of the recent automation capabilities is that in the 19th century we automated the dangerous, in the 20th century the dull, and in the 21st century the routine. Any repetitive task that can be learned through data analysis is susceptible to automation. An oft cited Oxford study puts the percentage of the workforce susceptible to automation at 47%, though this may not be cause for alarm. William Stanley Jevons, in commenting on resource consumption, noted that with new technologies that improve efficiency, the consumption of the improved resource often increases. A corollary when dealing with automation is that with efficiency gains to human labour, demand for human labour can also increase rather than decrease.

A modern example of this paradox is the increase of bank tellers following the introduction of the ATM. With automated tellers, the cost of bank space decreased, enabling banks to buy up more commercial space and open new branches. Relieving human tellers from the mundane duties of counting cash enabled them to provide more contact with customers, dealing with more specific concerns, and upselling for the bank. This more profitable and productive use of tellers by the banks, and the prevalence of new locations, encourages the hiring of more tellers.

This suggests that the fear of technological unemployment might not materialise in as wide-spread a manner as some commentators have suggested. While many tasks are susceptible to automation, a job is not usually just to perform a rote task, but generally composed of various disparate skills. Among these, empathy, critical analysis, and creativity seem far off from automating. Professionals such as lawyers will not need to spend as much time searching through previous cases and drawing up routine contracts, but their skills at presenting defences in court or providing legal advice would be of increasing value as they can spend more of their time improving their communications skills. Similarly, doctors need not memorise the list of potential ailments, or even perform diagnosis, with recent innovations in machine learning being more accurate than many professionals at diagnosing common illnesses; however, this will only increase their value in communicating the steps to recovery to patients, providing advice on how to live healthier lives etc.

The aspects of one’s career that generate the most value are also the most human skills. Technological unemployment may not run rampant, but the nature of work will certainly change. Learning and socialising become ever more important parts of one’s job when the routine tasks get stripped out of it. This has the potential to make continuous education a fixture of the workplace and greatly improve productivity. It also may present a worry of its own, as not everyone is as skilled in these social arenas, suggesting a shock to sectors in which many careers are based around repetitive tasks. While the value of these tasks to the individual employee may have been low, it is possible that without significant retraining they may have never developed the skills that will become essential.

While the long-term future through automation may vastly improve work satisfaction and productivity, enabling faster growth and innovation, this may come at the cost of a short-term shock among less skilled professions. The implications of this shock such as wage bifurcation and unemployment, as well as the policy solutions to such an issue will be discussed in the next post.

Leave a Reply

Your email address will not be published.