Figure 6. Barriers to content creation across content types. (© Google, used with permission.)
WHAT (EXPERIENCE OF) ASSISTANCE DO WORKERS WANT
Mapping the core-peripheral distinction to assistance
The core-peripheral distinction and Davenport's model of knowledge work allowed the research team to start making sense of the experience of knowledge work. However, by itself it didn't explain the support people wanted from AI or human assistance.
When discussing assistance, respondents expressed clear preferences for receiving different kinds of assistance depending on the type of task they received assistance with. The more peripheral a task was to them, the more they wanted to completely offload it from their responsibility. With core tasks, on the other hand, respondents preferred assistance that enhanced their execution of the task, without removing it from their oversight.
Seven specific types of assistance emerged from our research: remove, short-cut, anticipate, synthesize, scrutinize, improve and inspire. They can each be placed on the spectrum of assistance between offloading assistance and enhancing assistance. The following chart illustrates this with specific examples.
Figure 7. Spectrum of Assistance. (© Google, used with permission.)
Offloading peripheral work
“I feel more busy than I should be… I get 100's of emails a day and most of them are bulls*!t”
Ingrid, Analyst, International Bank
As Ingrid illustrates, offloading peripheral work is often less clear-cut than outsourcing expense claims. The spectrum reveals that the experience of assistance that workers require is nuanced and can vary task-to-task within a workflow. Offloading does not necessarily mean total removal of the task; it can also be about speeding up the task (short-cut), pre-empting what is required (anticipating) or simplifying complexity (synthesise).
As discussed earlier, just because workers are not always rewarded by (or find meaning in) peripheral work this doesn't mean it's not significant and high risk. Ingrid's ‘bullsh*t’ emails still require a thoughtful response. But if how she arrives at that thoughtful response can be expedited then that would be of immense value to her.
Importantly for peripheral work, it's not critical for Ingrid to feel a sense of personal agency over the task. She doesn't need to know or understand how the assistance works, and she doesn't need to take credit for it, she only cares that it's correct and produces a satisfactory outcome. Ironically this means that trust is a more important factor for peripheral work even if it is regarded as lower value work. This is because if the worker is willing to relinquish oversight they must place more trust in the agent that is working on their behalf.
Enhancing core work
“They employed me for my personality and for my thinking. You can't teach strategic thinking — you either have that type of brain or you don't”
Peter, Strategist, global CPG firm
Unlike peripheral work, core work is directly linked to how worker performance is measured, and often to their sense of value, identity and self-esteem. Because of this workers want to feel like they are in total control of all work they define as core.
In Peter's case, he feels like he is employed because he has the ‘type of brain’ which is uniquely suited to his role. It is clear he derives a significant amount of self-worth from his belief about his skills, so any task which truly utilises them - such as developing a recommendation a new direction for a brand - must be responded to entirely by ‘himself’. Any form of assistance received during the execution of these types of tasks must be experienced as an augmentation or extension of his own capabilities. If he felt these tasks were being done ‘for’ him this would not only, in his view, dilute the quality of the work, but pose an existential threat to his personal sense of value. Peter is open to his work being ‘scrutinized’, ‘improved’ and even ‘inspired’. But it ultimately must remain his work, and by asking for assistance this must never be called into question.
There is a tension inherent in the concept of Core Work. As work becomes more collaborative it becomes more difficult for individuals to define and account for their specific contribution, reducing feelings of agency and ownership. For example, we noted a desire from several participants for an ‘audit’ trail for content they have personally contributed. Often as content is shared throughout an organisation individual contributions become adapted and merged into larger documents. It therefore becomes very difficult for an individual to know the impact their contribution is making and, by extension, take credit for that impact. Potential design implications of this for AI are discussed below.
Interestingly, even though core work is of higher value to the worker there is less need for them to trust the assistance they receive. Because workers want to remain deeply involved in their core work they have more capacity to evaluate, accept or dismiss any assistance that they solicit or receive.
Design principles for assistance
The above can be summarised in a simple set of design principles to inform how AI-driven assistance is ideally experienced by knowledge workers. As is outlined in the impact section below, this is one of the frameworks which is guiding product teams across GSuite. These fundamental distinctions in how assistance should be experienced has important implications for who ‘owns’ different forms of AI and how it should be implemented within organisations.
Figure 8. Assistance design principles. (© Google, used with permission.)
Two types of AI
The paradox of peripheral work is that while it is perceived to be of lower value, trust in the assistance is more critical because workers are delegating work that is still regarded as their responsibility. For example, you may want to delegate filing your expenses, but if a false claim is made on your behalf then that puts your reputation at risk. Therefore trust in AI acting on your behalf must be exceptionally high. Trust is established when an assistant completes a task satisfactorily on a repeat basis. In these circumstances oversight is gradually withdrawn.
Therefore workers liked the idea of offloading both agency and ownership for peripheral work. On the other hand, they wanted to experience any assistance with core work as integrated and indistinguishable from their own efforts. They wanted to maintain and sometimes deepen agency and ownership of these tasks.
In practical terms this meant they liked the idea of their employer organisation as being the agent of peripheral work, while they personally retain control of their core work.
Figure 9. Preference for ownership differs based on type of assistance. (© Google, used with permission.)
Peripheral work = ‘Company AI’
Workers liked the idea of delegating peripheral work to a company-owned tool, meaning the company is responsible for positive and negative outputs rather than the worker. By extension, workers were open to the organisational AI being represented anthropomorphically as an external agent (like Google Assistant, Siri, Alexa etc).
This would be a resource that would be part of organisational infrastructure, and therefore remain in place if a worker were to leave the organisation.
Core work = ‘My AI’
In contrast because workers viewed core work as integral to their value and identity, they preferred the idea of a personal AI that would move with them between organisations. As they invested time training the AI it would become increasingly personalised and indistinguishable from their own capabilities.
In this sense ‘My AI’ should not be experienced as an external anthropomorphic agent but as largely embedded in their workflows and practices, to the point where it is not recognised to be AI as such and indistinguishable from their own capabilities.
The idea of ‘My AI’ can be seen to run counter to the trend of work becoming more collaborative - in this sense a ‘Our AI’ may seem like a better reflection of the way that work is developing. But this runs counter to the aspirations of ownership, autonomy and agency that emerged strongly from the research. As work becomes more complex AI may actually become a tool for maintaining personal agency and autonomy as it helps individuals automatically define and track their specific contributions within the context of the whole.
IMPLICATIONS AND IMPACT
The two concrete contributions the work has had for Google can be summarized as:
- Embedding a new set of taxonomies and frameworks that inform AI-related decision making throughout the GSuite organizationThe frameworks outlined in this case study have been socialised across both the executive and product layers of the organisation, helping teams prioritise and develop strategies. Previous to this work GSuite had many successful products that provided AI-driven assistance for knowledge workers, but lacked a foundational framework with which to categorise and evaluate existing products from a user perspective, nor a clear means for understanding where to innovate in the future. Our work has provided GSuite management with a set of adaptable tools to organise and manage innovation across product teams.
Figure 10. Amy Lokey, VP, User Experience, GSuite, introduces our foundational Core / Peripheral work framework at Qualtrics conference. (Lokey, 2018a)
- Driving product innovation within specific GSuite teamsEach product team at GSuite (Gmail, Calendar, Sheets, Docs) is now using these frameworks to inspire and guide how they integrate AI into their products, with many examples already live
Figure 11. Calendar's new auto Meeting Room allocator is driven from the idea of reducing Peripheral Work. (© Google, used with permission.)
Figure 12. Sheets new ‘Explorer’ feature enables users to generate charts using ‘conversational’ queries, augmenting ‘Core Work’. (© Google, used with permission.)
It's also important to emphasise the ethical dimension here. By highlighting a worker-first perspective of what good ‘Assistance’ is at work, our project has guided Google towards sensitive solutions which help workers excel at their job, by both augmenting their skills and removing aspects of their work that were blocking them from excelling.
Our guidance would have looked different if we'd taken an IT-first perspective. As part of the project we conducted a number of management interviews with IT decision makers and it was clear their priorities were often quite different to individual workers. Their concerns primarily revolved around value for money, and seeing AI as a means to reduce costs - although there was evidence of the increasing role of worker preference in driving decision making (companies like Google and Slack have made influencing workers first central to their ‘bottom-up’ adoption strategies). This is not to underplay the importance of this perspective, but to emphasise the role of this project was to focus on the needs and priorities of the end user.
Public references to our work (see full reference in citations)
- Qualtrics conference keynote, 2018 (Lokey, 2018a)
- Keynote at Google NEXT conference (Lokey, 2018b)
- Interview with Teryn O'Brien for Silicon Angle (O'Brien, 2018)
For Knowledge Work
By the end of 2018 over 5 million businesses are paying to use GSuite worldwide (https://9to5google.com/2019/02/04/g-suite-5-million-businesses/). The influence GSuite has over the way people do work is enormous (especially if we include the consumer side of Gmail, then the number increases to 1.4 bn users.)
By helping workers to focus on Core Work and reduce Peripheral Work, GSuite will contribute to the streamlining and specialization of roles as they are optimised towards leveraging the specific skills and aspirations of the individual. From this perspective Knowledge Work should also become more rewarding and enjoyable as users focus on work that they find most interesting and valuable.
However, for this vision of knowledge work to be realised there are a couple of questions that warrant further exploration:
How do we enhance Core work in a collaborative environment?
Work is simultaneously becoming more complex and more collaborative. Given each worker has a personal incentive to focus on Core Work this may lead to tensions as work overlaps and workers compete to do the same high-value work. And more importantly, given that Knowledge Work requires increasingly complex forms of collaboration, it may become more difficult to define and quantify unique contributions and, by extension, the nature of “your core work” vs “my core work”.
One outcome may be that role definitions becoming increasingly collective and integrated, so that Core Work is not defined in such individualistic terms. Alternatively AI may actually help workers parse, define and measure their contributions in this more complex environment. This is an area that we would like to explore further.
How will the removal of Peripheral work affect Core work?
Just like other workers, it is easy for ethnographers to segregate the work we do between ‘Core’ and ‘Peripheral’. For example, we may want to minimise the logistical burden inherent in conducting global fieldwork. From organising transportation to syncing meetings across time zones there are many tasks that seem to detract from time spent on what we commonly think of as our core work (namely field research, pattern recognition, meeting with clients).
However, there is a danger in minimising work that is perceived to be Peripheral. Last year at EPIC we outlined the danger of ‘AirSpace’ - the idea that global platforms like Google, Uber and Airbnb are making ethnography ‘frictionless’ and thereby reducing its richly textured scope to an extended interview (Hoy, 2018). To put it simply, sometimes getting stuck on public transport may feel like Peripheral Work, but it can also lead to the most unanticipated, abductive insights. In this sense, the work we perceive to be Peripheral may be reframed as Core.
In this sense removing the rough edges of Knowledge Work may not always be a good thing if it restricts our idea of what our work is, or could be. And this may be a challenge that extends to other Knowledge Workers too. This is another area we would like to dig deeper into.
There are some important learnings from this project on how we study AI as ethnographers. In the context of work, we found framing ‘assistance’ in human rather than technological terms was an important way for us to begin our conversations with participants. This enabled us to put pre-existing ideas about AI (from media narratives about jobs being automated to consumer instantiations such as Siri or Google Assistant) to one side and focus on the everyday support they would appreciate at work. It was only once we established these ground rules that we introduced the idea of technology.
Secondly, performing ethnography with multiple workers in the same team enabled us to better understand the distinctions and tensions between individual autonomy and teamwork, and how one person's Core work can be another person's Peripheral work. Also, we could triangulate between the claims of different workers and observe team dynamics, enabling us to build up a truer picture of everyday work.
Citation: 2019 EPIC Proceedings pp 125–143, ISSN 1559-8918, https://www.epicpeople.org/epic
Cross, R, Taylor, S, and Zehner, D, July-Aug 2018, “Collaboration Without Burnout”, in Harvard Business Review, Harvard Business School Press, Boston.
Davenport, T.H. and Prusak, 1998, Working Knowledge: How Organizations Manage What They Know. Harvard Business School Press, Boston.
Davenport, T, 2005, Thinking for a Living: How to Get Better Performances And Results from Knowledge Workers, Harvard Business School Press, Boston.
Davis, J, et al., 2018, “The Future of Work” in Vanguard Megatrends, Accessed October 28th, 2019 https://pressroom.vanguard.com/nonindexed/Vanguard-Research-Megatrends-Series-Future-of-Work-October2018v2.pdf
Garimella, K, 2018, “Job Loss From AI? There's More To Fear!”, in Forbes, Accessed October 28th, 2019 https://www.forbes.com/sites/cognitiveworld/2018/08/07/job-loss-from-ai-theres-more-to-fear/#359b388b23eb
Hoy, T, 2018. “Doing Ethnography in AirSpace: The Promise and Danger of ‘Frictionless’ Global Research.” Ethnographic Praxis in Industry Conference Proceedings
Lokey, A, 2018a. “Making Magic from the Mundane.” Keynote Presentation at Qualtrics Conference Accessed October 28th, 2019 https://www.qualtrics.com/events/x4-px-breakthrough-sessions/making-magic-from-the-mundane/?
Lokey, A, 2018b. “Product Innovation Keynote.” Keynote Presentation at Google NEXT Conference Accessed October 28th, 2019 https://youtu.be/PZ1Lqxfs1yw?t=3796
O'Brien, T, 2018. “G Suite UX exec becomes ‘chief empathizer’ to fully engage users.” Interview with Silicon Angle, Accessed October 28th, 2019 https://youtu.be/PZ1Lqxfs1yw?t=3796