The period of technological change we are currently living through–the 4th Industrial Revolution–is defined and shaped by digitalisation and technologies such as the Internet, software, artificial intelligence (AI) and robotics. Similar to the transition from mechanical and analogue electronics to digital electronics in the late 20th century’s Digital Revolution, manual tasks are being increasingly automated in our 'digital work' futures.
Automation can replace both physical and cognitive tasks, particularly those that are predictable and routine. Historically, substituting machine-driven labour for human physical work accelerated from the 16th century on and gained momentum in 19th-century steam-powered mills, factories and farms. Continuing into the latter part of the 20th century, robots in vehicle manufacturing became common, initially performing simple, repetitive tasks to enhance output, standardise production, and control costs. While this mechanisation boosted output and standardised quality, there has historically always been push back about the removal of human-labour and artisanal skills from these types of work1.
Today, automation in the workplace relies on a diverse range of technologies, including software, robotics, and artificial intelligence to execute such jobs as operationalising administrative tasks or programming machines to swiftly deliver parts in factories. Artificial intelligence has particularly gained attention recently with the surge of advanced machine learning systems like large language models (LLMs), interacted with via conversational interfaces like ChatGPT to generate new content from a simple prompt. These new techniques and technologies enable machines to perform tasks requiring human intelligence, such as learning from data, recognizing patterns, making decisions, and solving complex problems.
AI’s relevance in the workplace then becomes less about manual human tasks and more about human intelligence. Instead of following ‘routine’ instructions, AI can now perform more interpersonal duties like human planning, problem-solving, or perception; our AI-infused digital tools can now advance our cognition as well as our physicality.
Unlike previous technological revolutions, where opposition to mechanisation focused on socialist concerns like unemployment, today's unease towards automation has a more existential tone. Emerging generative AI systems alter how we create information but also the locus of judgement. Automated, data-driven decisions may replace workers' or managers' own judgement, mediate employment relationships, and even facilitate design choices or market transactions2.
In essence, our cognition is increasingly being automated, sparking a knee-jerk reaction away from technologically-determined futures where machines appropriate our agency. Especially in the realm of work, our sense of agency—the capacity for independent action and free choices—is closely tied to a sense of ownership. Losing it results in diminished purpose and engagement.
These new AI technologies can serve as intellectual extensions, enhancing both our physical and cognitive abilities. As McCarty & McQuaid write: “[Tools] exist so that we may do more, see better, gather information, transform things, make decisions, investigate new frontiers, interact more fluidly and precisely, achieve higher forms of aesthetic satisfaction—extend our reach”3. Yet, as Marshall McLuhan astutely noted, these tools can also shape us and our worlds, for better or worse, if not handled carefully. The question then becomes: how can we increasingly integrate these cognitive automation technologies into our digital work while preserving a sense of agency in our tasks?
How can AI help us in our work today?
AI technologies have already been integrated into our daily lives for decades, in such applications as email spam filters, facial recognition, voice assistants and so on. These existing AI experiences use what is called Artificial Narrow Intelligence (ANI); systems focused on narrow tasks. The futures discussed in this publication, on the other hand, envision capabilities provided by Artificial General Intelligence (AGI); a highly advanced, currently unattained version of AI that combines various ANI technologies such as natural language processing, knowledge representation, automated reasoning, machine learning, computer vision, speech recognition, and robotics4.
If realised, AGI would possess the ability to communicate, store information, perceive the world, manipulate objects, reason and adapt; many of the skills core to our human cognition. Although still distant, the integration of 'AI-infused platforms' like ChatGPT, IBM Watson, AWS Sagemaker, and Microsoft Copilot into our digital workflows is a first step on this journey to further optimising our work performance, enhancing our efficiency, providing advanced insights, and even generating decisions with minimal user input.
At their most basic level, these AI-infused tools are ‘assistance systems’ capable of reducing the need for specific skills in certain tasks and easing our cognitive load, i.e. automating and increasing the efficiency of lower-level tasks to free us up to think and make decisions about the higher-level, creative or emotional aspects of our work. This allows for 'high discretion' augmentation and intensification, where our productivity is enhanced by quickly automating the ‘boilerplate’ elements of our work. For example, a designer can use generative software to rapidly create a wide array of designs to jump start a brainstorming session with stakeholders, a programmer can generate boilerplate code segments so that they can develop prototypes faster, and a radiographer could use AI to better identify a cancerous growth so they can spend more time focusing on their patient5.
Much like the scenario from Future 1 in this book ('Village People’), these ‘assistive systems’ can automate dull and time-consuming tasks, giving us the agency to direct our work efforts to more purposeful, impactful and societal needs.
More advanced AI-infused tools can take this assistance further in task-specific ‘recommendation systems’. For example, AI-powered tools can help designers refine their creations by generating alternative designs, proposing optimisations, or even introducing new approaches. Manufacturers can leverage recommender AI to analyse sensor data to predict breakdowns, identify patterns, proactively perform maintenance, create detailed instructions, or order spare parts. AI-infused programming software can provide insights and guidance to developer teams, enhancing coding practices, maintaining code consistency across simultaneous project work and enabling individuals without in-depth programming knowledge to contribute to technical projects.
The most cutting-edge 'autonomous systems' leverage generative AI for self-regulation and adaptability in unfamiliar or dynamic scenarios. These sophisticated AI-infused tools have the capability to continuously learn from extensive databases and user behaviours. For example, such systems could develop a profound understanding of the design patterns, structures, and best practices in a certain project or industry, enabling the designers and developers to automatically translate simple prompts into a sequence of actions that facilitate the execution of complex, multi-step tasks. Closer to the capabilities of AGI, these systems could even be hyper-personalised to our preferences and interaction styles, enabling these systems to act on our behalf, completing personal interactions at work, e.g emails and chat platforms, or making decisions based on preference as well as objective measures. As the future scenario in ‘Desiderata’ (or ‘Fractured Future’) is described: “In the digital world, our digital AI twins realise all our potential - humans are there, present with them, but they are playing the main part.” In this future, these autonomous systems can automate most of our agency. But is this what we want?
Where does human agency still matter?
While the positive potential of these AI-infused technologies create opportunities to improve work processes and productivity, there is still a crucial need for mindful deployment of these tools to ensure the quality of work and well-being of workers.
Firstly, generative AI technologies don’t always get it ‘right’. While large language models (LLMs) like ChatGPT or Google Bard can provide accurate explanations, they may not consistently produce correct results–otherwise known as ‘hallucinations’–when asked to perform the tasks themselves. For example, LLMs cannot accurately complete advanced mathematical operations due to the lack of available example problems in their underlying training dataset6. Therefore, relying on AI-infused technologies like generative AI is more effective for understanding concepts rather than delivering accurate answers. This touches on the issues raised in Future 2; that the increasing use of AI-infused tools will create a world where there are no or contested ‘ground truths’. When it comes to using these tools in the world of work, our agency to validate these generative AI outputs will still be crucial in the creation of quality and truthful outputs.
Secondly, while AI-infused systems can create diverse arrays of alternate solutions, they are not good at selecting the best option. This is where the human skill of understanding context, inferring intention and determining the most appropriate action from a range of subjective options still surpasses these AI technologies. Humans are idiosyncratic beings and, as described in the dichotomous digital-analog future envisioned in ‘Games Without Frontiers’ (‘Human Plus’), having agency to select options based on individual preferences rather than optimised recommendations in work and life can lead to experiences that feel more authentic and purposeful.
Lastly, at a more philosophical level, automating our tasks with AI may help make our work more productive but sometimes inefficiency can lead to more impactful outcomes. While I don’t believe we will ever reach the future scenario described in Future 3, ‘Turn on, click through, drop out’ (‘In Nature, no greed’), where “intellectual work is devalued, and efficiency is considered bad”, my own research into creativity and AI agrees that efficiency isn’t always optimal for innovation7. Having the ability to see a mistake as an opportunity or learn larger life lessons from an inefficient approach are distinctly human skills. Resisting the lure of AI-infused automation in certain situations can give us the agency to grow and discover in completely unexpected ways.
Balancing automation and agency in our digital work futures
So, how can we integrate these AI-infused automation technologies into our tasks while preserving our emotional, irrational, human agency to create truly impactful and purposeful work lives?
On an interface level, we need to design AI-infused tools that amplify the divergent parts of the design process while leaving it to a human to converge on the most appropriate result based on their preferences or interpretation of the context, i.e. use AI to generate a wide range of options but require the user to intentionally select the final result.
When it comes to our workflows, the ubiquity of AI assistance will result in needing to redefine what it means to specialise. With AI-infused tools, workers will be able to automate areas they are weaker at or don’t enjoy as much. This ability to become generalists in our domain will give us the agency to execute our whole vision for a project, not just one part of it. Being able to speed up areas of our workflow that take us longer may also give us more time to be ‘inefficient’ and more creative in areas where we have greater expertise and enjoyment.
Taking this idea to an organisational level, the promise of AI-infused tools executing areas of the workflow beyond our immediate expertise empowers us to do the tasks of a small team. In this scenario, we become managers, not one of many individual contributors. Our work–and job description–then relies on our agency and discretion to direct and approve rather than execute all of the tasks ourselves.
Lastly, the very balance of agency and automation will be determined at a personal, organisational and even governmental policy level. Whether an AI-infused technology will be designed to have intermittent interactions (as with assistive systems), continuous enjoinders (as with recommendation systems) or proactive interventions (as with autonomous systems) will depend on the values of the developers, the preferences of the end users, or the context in which it will be used. The decisions we make now when choosing between these human-AI interaction paradigms–and hence the level of agency over how we want our work to be automated–will shape the tools of our work and ultimately the direction of our digital futures.
