By Sofia Klapp1
1Digital Transformation Manager, Digital Gov, Chile
How artificial intelligence (AI) might disrupt work and society has been a topic of great concern among academics, practitioners and media outlets over the last decade. This chapter is an invitation for leaders to think further about the AI-talent transformation challenge in the context of AI-in-practice while being prepared for an AI-driven future.
Historically, new technologies have been considered a source of economic progress, but also a source of anxiety among the general population, due to their potential to replace human work, creating, as a consequence, unemployment and further inequality. In the past, technological advancements have eliminated some jobs, but have also shown the potential to generate new ones, whilst transforming others.
However, some economists hold the alarming view that in an AI-driven future, a discontinuation of this past trend will take place. Well-known supporters of this position are Brynjolfsson and McAfee, who argue that AI capabilities are different from other past automation technologies because machine learning is capable of performing highly complex cognitive tasks.1 This capacity enables AI to emulate tacit knowledge, which was previously thought to be limited to humans. Also, both authors forecast that AI advancements will evolve so quickly that it holds the potential to perform all kinds of skills, eventually replacing all human jobs, while also potentially preventing the economy from creating new ones.
It is a fact that the global labour market is undergoing a significant transformation and that technology will impact all workers. However, the impact will be not as straightforward as expected. As the economist David Autor explains, these technologies are skill-biased, being able to substitute or augment work depending on its nature. To this, he adds that a job is made up of a set of tasks, and AI can only automate a few.2 Recently, Brynjolfsson, Mitchell and Rock developed a 21-question rubric that assesses task-suitability for machine learning.3 This study concluded that no occupation can be fully automated by machine learning, but that all of them will be impacted to a different extent. Although many high-wage jobs can be affected, low-wage jobs remain more exposed. In this context, job redesign, upskilling and reskilling will be crucial to achieve productivity gains.
It is expected that AI, in combination with other technologies, will create incredible progress and wealth. However, these economists are concerned about the adverse side effects. They argue that, at least in our current economic system, this progress will also have an enormous impact on the distribution of income and wealth. More than just worrying about job substitution, they point out that only a relatively small group of people will often earn most of the income, generating great inequality. Policy initiatives are being worked on at this moment, such as basic income and robot-taxation. They seem to be radical in appearance, but they might be the only way to allow society to handle extreme unemployment and the creation of further inequality.
People tend to believe that technological advancements have a linear and deterministic impact in our societies. However, they ignore the fact that technology is also shaped by society in a dynamic and bidirectional relationship. Our future depends less on the technology itself and more on the choices society makes. As Erik Brynjolfsson exclaims: “The future is not preordained by machines. It’s created by humans.” More than getting stuck in the reaction of fear, we have to thoughtfully manage the transition to the AI-driven world by thinking about how to match workers’ capacity with this new environment. How do we use technology in ways that will create not just prosperity, but shared prosperity?
Great cases have shown AI’s potential to bypass humans in most skills, but the reality is that at this moment AI applications are still limited to specific tasks. However, it is common to find people who believe in AI’s current superpower to perform all tasks better than humans. As Davenport and Ronanki noticed, this generates unrealistic expectations of cognitive technologies but also fear and higher resistance.4 Unfortunately, all these commonly lead to failure and disappointment.
Nowadays, academics and practitioners are starting to notice that the bottleneck for effective implementation is the understanding and management of human skills in coordination with AI solutions. In this context, human–AI labour division is crucial, detailing that organizations need to define which parts of work tasks or processes could be handled by a machine, which by humans and which in collaboration. To achieve this effectively, leaders should not only consider the particular strengths and weaknesses of humans and AI, but also create new AI–human configurations in a symbiotic relationship.
AI has shown a superior capability in processing massive amounts of data, finding unexpected correlation patterns and making predictions, while adjusting and improving its models automatically as soon as it can access new and more data. Additionally, there has been significant advancement in AI-perception tools, such as voice, speech and image recognition. In contrast, humans are still better than AI in tasks that require complex pattern recognition, socio-emotional skills, creativity5 and judgement.6
However, there is still a debate around machine learning capability to emulate tacit knowledge.7 In this context, it is essential to understand what is meant by tacit knowledge and to what extent AI can emulate it, avoiding taking it for granted. Not all types of tacit knowledge can be encoded into data and performed by machines. Tacit knowledge is all kinds of knowledge that people obtain from socially immersed, personal experiences as opposed to formal teaching. As Collins explains, there is a vast range of tacit knowledge, such as social sensibility and improvisation, bounded by specific and dynamic social community conventions that make it impossible for a machine to acquire and always update them.8
It is common to hear statements pointing out the need for human and AI collaboration or for maintaining humanity in the AI loop. But, what do they mean? It is easier to understand which tasks can be performed better by humans and which by AI separately, but there is a lack of understanding of the ones that are somewhere in the middle.
Daugherty and Wilson propose a model to understand these hybrid activities that can be divided into two types: the ones in which humans support AI, and the ones in which AI supports humans.9 Humans need to support AI by training it to perform certain tasks, explain the outcomes of those tasks (e.g. when results are counterintuitive or controversial) and sustain its responsible use (e.g. preventing robots from harming humans). Also, AI supports humans by helping them to expand their abilities by amplifying their cognitive strengths, interacting with customers and employees while freeing them for higher-level tasks, and embodying human skills to extend our physical capabilities. All of these hybrid activities require high levels of business processes and word redesign, employee involvement and experimentation, and data generation and collection, in addition to cultivating specific employee skills.
Sadly, some companies neglect the value of this complementarity, opting for a substitution approach. The above-mentioned authors studied a sample of 1075 companies that were running AI-related projects, showing that companies that manage human-machine collaboration rather than focusing on substitution achieve superior business performance improvements in the long term. Companies often focus on cost-saving by automating tasks and eliminating headcounts because the benefits can be evaluated in the short term. A value creation goal enabled by an AI–human collaboration approach requires higher levels of innovation, while the benefits will be seen further down the road.