Introduction: Skill Polarisation
The Digital transition will grow some jobs, see others decline, and new jobs developed. AI will expand high- and low-skilled jobs while reducing medium-skill jobs in the workforce. This transition will lead to skill-structure changes known as ‘skill polarisation’. Therefore, it will enhance high-skilled jobs, possibly replace medium-skill jobs through programmable routines especially since they have had the slowest salary growth over the past 30 years. However, it is unlikely that AI will affect low-skilled work with unpredictable patterns of work-related tasks.
The ‘skill polarisation’ anomaly has been closely analysed in the US and Europe. However, explanations for this trend from empirical literature is not consistent. The recent demand for skills requested include creativity and interpersonal skills showing a decline in demand for physical skills[1] .
High-skilled work uses cognitive skills, abstract thinking, coordination of activities, and decision-making (e.g., lawyers, doctors, software engineers, nurses).Low-skilled work uses personal interaction and physical proximity (e.g., food-service workers, childcare workers).Medium-skilled work makes use of routine manual tasks that are more likely to be replaced by technology and globalisation (e.g., administrative support workers, machine operators).
Measuring the Scale of the Upskilling Challenge
The Organisation for Economic Cooperation and Development (OECD) is conducting an international literacy test in its Programme for the International Assessment of Adult Competencies (PIAAC). This tests a representative sample of working-age adults in each participating country.
To understand whether AI and robots are likely to alter the fundamental structure of work, we need to know whether these new technologies will require changes in work skills and whether they are feasible or not. If changes are feasible, the overall effect of AI and robots will look like the changes we have seen with other technological innovations such as the internet. However, if the necessary changes in work skills are not feasible, then this time may be different.
What does this mean for the potential impact of AI on work? - we are faced with two practical questions:
1. How proficient are computers for these other skills, and how many people are more proficient than computers?
2. For other skills, can most people do what computers cannot do, or do most people have trouble with those tasks as well? Jobs requiring high levels of social skills and interaction are often proposed as promising jobs of the future since computers may not be able to interact as much.
AI & Human Skills Proficiency Levels for Literacy & Other skills incl. Social Skills
Literacy is a key skill and a foundation skill that is critical for specialized reasoning and problem-solving skills in specific domains. Therefore, computers may be able to carry out several information related tasks currently performed by the workforce with literacy proficiency at Levels 2 or 3. However, they may not be able to perform many of the information-related tasks performed by workers with literacy proficiency at Levels 4 or 5. Computer performance at PIAAC Level 1 for litracy does not pose a threat to employees, since few adults have low levels of literacy. However, computer performace at Level 3 may be a threat since few adults may be better than that.
Social skills we may overestimate employees abilities and underestimate AI abilities. While some people may be capable of simpler aspects of social interaction including facial recognition or responding to direct requests for information, AI today may also carry out such tasks. However, while AI may not perform more complex social interactions, such as conducting sensitive negotiations or gaining the trust of an angry customer, such complex social interactions also tend to be difficult for certain people. Therefore, as regards social skills and other major skill areas, one needs to assess how computer capabilities compare with human proficiency.
Arguing that people have better skills than AI would not be enough, to continue a work-based economy, people need to develop better skills than those provided by AI.
According to literature on the diffusion of technology, it tends to take a while for industry to adopt and apply new technologies. This would include:
1. Time to learn about AI.
2. Refine it for particular applications.
3. Invest in AI technologies at scale.
Moreover, widespread diffusion may take several decades, which means that we have time to understand what AI capabilities currently exist and anticipate how they are likely to shift the skills needed by the workforce over the coming 10 to 20 years. Moreover, improvements in education are often slow and difficult, which means that even 10 to 20 years of warning may not be enough to develop the skills required. .
Foreseeable Impact of AI on the Future of Work
If human-level general AI is still far in the future, AI is likely to have great economic importance in the meantime as it will steadily increase the number of tasks that AI can do faster, much cheaper, and better than people, with minimal to no error. However, for industry to adapt and meet the demands of the impact AI will have on the Future of Work it has to first predict the tasks required.
To predict future tasks for AI, industry must take note of what today’s computers are often better at doing than people such as: routine data processing like summing up retail store’s transactions, and predictable physical work such as assembly line production.
Two key factors required for today’s machine learning algorithms are:
1. large amounts of data, such as recordings of previous conversations with customers, sensor logs of previous machine operations.
The success of Google Translate was enabled by the availability of vast amounts of textual data in many languages on the web.
2. There must be a pattern to be recognized for instance, machines can carry out credit-risk assessments when using information such as a borrower’s current assets, income, and debt. Moreover, machine learning programs may be trained to recognize patterns in pools of data that are far larger than humans may be able to assess.
Online retailers like Amazon use AI to detect subtle patterns in vast amounts of data related to which type of customers buy which kind of products[1] . While, no amount of machine learning may predict the exact path a shopper will take through a shopping mall.
It is difficult to predict what future technologies will/will not be able to do. The examples provided are mere guidelines and rough judgments of key tasks AI can do today.
While it is simple to assert that “computers can do tasks that are precisely specifiable in ways computers can execute,” it is difficult to provide general rules for what is and is not precisely specifiable in this way, making it even harder to predict the effects of future progress in AI[1] .
We do not know whether categorical statements that we believe are accurate today will actually be true in the future: “computers will never be really creative” or “computers will never be able to understand human emotions”. In the 10 to 20 years computers may be able to do all things as well as people do.
For instance, computers have already generated strikingly artistic images[2] , including a painting that was sold for more than $400,000 at auction. Similarly, they have also proved capable of reading human emotions in different circumstances.
To understand the difficulty in predicting what machines will be able to do in the future is to recognize a key difference between classical AI and modern machine learning:
● Classical AI (1950s-1980s) uses the “represent and reason” approach by representing the knowledge required and then reasons logically using the knowledge.
● Machine learning, uses a “function fitting” approach, as it tries to find the parameters for complex mathematical functions that allow the system to predict the input-output relationships observed in the training data.
While both approaches are useful, neither can “think”, to date, in the ways humans do.
The classical approach is similar to how people think consciously, however, it has not been very successful at reflecting different real-world human cognitive activities.
The machine learning approach seems more similar to the unconscious human mental processes, like visual perception, however it poses challenges to reasoning and interpretability.
An interesting direction would be to combine the classical and more recent approaches[1] , however, it is difficult to knows the exact limitations of either approach separately or when combined.
Nonetheless, while it is difficult to predict precisely what tasks AI may do in the future, one may still outline some possibilities as we will see important opportunities for AI in a number of tasks.
Potential Impact of AI on the Future of Work
The potential impact of AI on the future of work revolves around 3 elements, namely sensing, deciding & creating:
1. Sensing: AI systems today can already analyse different images, sounds, motions, and other inputs.
Some Interesting facts:
1. AI at times may be better than physicians at interpreting X-rays and similar medical images.
2. Amazon Go stores enable people to pay for their shopping without having to stand at the cashier by having sensors and cameras track what items were selected and charge them on departure.
3. Airbus also uses different kinds of maintenance records, motion sensor logs, and other data to anticipate when preventive maintenance on an aircraft is required .
These examples barely hint at the innovative ways AI may be used to sense various aspects of the world around us.
1. Deciding: AI may carry out better decision-making that people when decisions are based on inputs and the desired result are codified with precision, even more so when ample data on similar past cases is made available.
AI already carries out credit risk decisions, and with medical cases, as large amounts of data become available, various automated medical diagnoses may also become feasible.
2. Creating: Creation involves patterns that AI can learn. News articles as regards Little League baseball games were successfully generated through innovative technologies from basic statistics about the game . Soon, it may also become possible to generate customised versions of legal contracts, sales proposals, and other documentation.
The GPT-3 system may flexibly generate human-like text, despite being unable to really “understand” what it is developing. Hence, eventually it may increasingly enhance other text generation software. AI today is also becoming useful to generate detailed, innovative designs for objects such as heat exchangers.
For tasks like these, machines are likely to perform work currently done by people, but for the foreseeable future, human work is likely to become even more valuable in tasks that humans usually do better than machines (McKinsey, 2017 and Malone, 2018, 273-281), including those that require social skills, unpredictable physical skills, common sense, or the general intelligence needed to deal with new, non-routine situations.
Figure for predicting where machine learning may be useful - page 17 (Source: Thomas Malone in MIT Sloan and CSAIL 2019, Module 2): https://workofthefuture.mit.edu/wp-content/uploads/2020/12/2020-Research-Brief-Malone-Rus-Laubacher2.pdf
Source: https://www.youtube.com/watch?v=m5uut9pya4g
Source: https://www.youtube.com/watch?v=UdE-W30oOXo
Source: https://www.youtube.com/watch?v=a-7Azih0D98
Source: https://www.youtube.com/watch?v=Cx5aNwnZYDc
Curious about this? have a look at these documentaries:
Source: https://www.youtube.com/watch?v=-ePZ7OdY-Dw
Source: https://www.youtube.com/watch?v=s0dMTAQM4cw
Source: https://www.youtube.com/watch?v=SN2BZswEWUA
Source: https://blogs.salleurl.edu/en/ai-data-centers
Source: https://medium.com/@sadievrenseker/ai-thinking-vs-human-thinking-bd40d34b629c