The future of work is not about lost jobs; it is about the loss of human capability to find its way. Madhavi Venkatesan writes.
Much of the discussion surrounding artificial intelligence (AI) and the future of work has focused on employment and productivity. Common questions include: which jobs will AI eliminate and which ones will it create; will AI substitute for workers or complement them; and, perhaps most importantly from an economic perspective, will AI allow us to produce more with fewer resources?
These are reasonable questions. But they reflect the way we have traditionally evaluated technological advancement: through efficiency, productivity, and economic output. What if these measures are insufficient? What if the more important question is not simply what AI allows us to do, but what our reliance on AI may eventually leave us unable to do?
Human beings have always adapted to their environment. Over evolutionary time, changes in climate, geography, disease, food availability, and the natural world contributed to biological adaptation. In this sense, the environment has not simply been something external to human development. It has helped shape who we have become. Technology, meanwhile, has always altered this relationship.
A technology that began by helping us do something eventually changed how we did it – and, in some instances, whether we retained the ability to do it independently.
Calculators reduced the need to perform some calculations mentally. The development of the Global Positioning System (GPS) changed how we navigate. Mobile phones, once expensive conveniences, have become so integrated into daily life that it is increasingly difficult to participate in society without one. In each case, a technology that began by helping us do something eventually changed how we did it – and, in some instances, whether we retained the ability to do it independently.
AI may represent a more fundamental shift. Unlike many earlier technologies, AI does not simply alter the physical environment in which we operate. It increasingly mediates the cognitive environment through which we understand that world. It writes, summarises, analyses, calculates, recommends, and creates. Increasingly, it also informs our decisions.
We are therefore doing something potentially unprecedented. We are constructing a synthetic cognitive environment while simultaneously adapting ourselves to the environment we are creating. The implications extend beyond the familiar concern that AI may eliminate jobs.
Emerging research suggests that the relationship between generative AI and human capability is more complicated than measures of productivity alone imply. Studies of AI-assisted knowledge work and analytical writing identify a tension between task performance and cognitive engagement: generative AI can reduce the effort required to produce or refine written work while also shifting, and in some cases reducing, engagement in the cognitive processes involved in generating, analysing, and evaluating that work. Research drawing upon neuroplasticity raises a related concern. Passive reliance on AI may have different implications for cognition than active questioning, evaluation, and collaboration with it.
What happens when the process and the output become separated?
A worker using AI may produce a better report in half the time. Productivity has increased. The gain is measurable and rewarded. But suppose that repeated reliance on AI reduces the worker’s opportunity – or eventually the worker’s ability – to independently research the subject, organise the argument, assess the evidence, and reach a conclusion. Has productivity actually increased? Perhaps. But something else may also have been lost.
We measure productivity gain; we do not measure capability loss. This is an externality of AI that our current economic measures are poorly equipped to recognise. It is, in effect, an unpriced cost of productivity.
William Stanley Jevons observed that increasing the efficiency of coal use could increase rather than reduce consumption by making coal more economical. Applied to AI, greater efficiency and lower costs similarly encourage greater use. But this interpretation may stop too soon. AI differs from coal in an important respect: increased consumption may change the capabilities and behavior of the person consuming it.
As AI becomes more accessible, we use it more. As we use it more, its presence becomes normalised. As its use becomes normalised, we may cease practicing some of the capabilities it replaces. As those capabilities diminish, functioning without AI becomes increasingly difficult.
What happens when the lights go out?
The AI economy assumes continuous access to the infrastructure upon which AI depends. Yet AI is not detached from the natural world. It requires electricity, data centers, cooling systems, telecommunications networks, semiconductor manufacturing, global supply chains, and people with the expertise necessary to maintain these interconnected systems.
At the same time, climate change is increasing pressure on many of these systems. Extreme heat, hurricanes, wildfires, flooding, and severe storms threaten electrical infrastructure and contribute to power disruptions. We are therefore increasing our dependence on electricity and digital infrastructure at precisely the moment when climate instability is raising questions about the resilience of that infrastructure. This is not to suggest that a power outage will bring technological society to an end. It is to ask whether increasing efficiency is simultaneously reducing redundancy.
A person who uses GPS but can navigate without it possesses both technological efficiency and human redundancy. A person who cannot navigate without GPS possesses technological dependency.
Human history suggests that knowledge should not be assumed to be permanent. Knowledge and skills survive because they are practiced, preserved, and transmitted. When the institutions and communities responsible for that transmission disappear, capabilities once taken for granted can be diminished or lost. This presents a paradox.
Human history suggests that knowledge should not be assumed to be permanent.
The more capable our technology becomes, the less necessary it may appear for humans to maintain the capabilities the technology replaces. Yet the fewer capabilities we retain, the more dependent we become upon the infrastructure that supports the technology.
We are simultaneously creating a civilisation increasingly dependent upon complex technological systems while destabilising the natural environment within which those systems must operate. Our environment is no longer simply natural. It is natural, built, digital, and increasingly synthetic. If humans adapt to the environments they inhabit, then the sustainability of AI cannot be evaluated solely by its energy consumption, carbon emissions, or material footprint. We must also consider whether the environment it creates is sustainable for human capability.
The future of work is consequently about more than whether humans will have jobs. It is about whether humans will continue to develop and retain the capabilities necessary to understand, evaluate, recreate, and take responsibility for the work being done. Efficiency and resilience are not the same thing. A system can be extraordinarily efficient and extraordinarily fragile.
For most of human history, we adapted to environments we largely inherited. We are now constructing an increasingly synthetic environment to which we – and future generations – will adapt. In doing so, we may be changing not only what humans do, but what humans retain the capacity to do. As we consider a future in which artificial intelligence can do increasingly more for us, perhaps the question is not simply what AI will be capable of doing. Perhaps we should also ask: will humans retain the capacity for resilience if the lights go out?
