I was immediately taken with Gmail’s smart compose feature when it was launched in summer 2018. At the start this was simply curiosity about the prospect that my emails could be composed in this way. With time it became a practical support to my e-mail load as it learned my writing style and I learned how to leverage it without surrendering to it. In some cases it entirely replaces my writing by offering pre-selected routine responses to simple e-mails which match my writing stye:
It automates routine parts of the writing process, such as ‘Hi X’ at the start of an e-mail, with the tab key substituting for actually writing the words yourself. It gets more interesting when you get into the main body of the e-mail, where the suggestions tend to be hit and miss; sometimes you insert them into the email but at other points you just keep writing and ignore them. I find the phenomenology of this interesting because it feels like selectively leaning into the suggestions without abdicating responsibility for the writing. My experience is that it works best when it just melts into the flow of practice, in the sense that sometimes the next element emerges entirely from you and sometimes it emerges from your reaction to the system prompts.
If you combine this with TextExpander in order to establish your own short cuts then it makes dealing with a heavy e-mail load far less time intensive. The risk is that it leaves your e-mails cold and brusque (which I’m sometimes prone to anyway as someone who reached the conclusion some time ago that I can either reply promptly or in a consistently warm/expansive way but I’ve not got it in me to do both). The practical challenge here though is one found in any form of automation implemented at the individual level: if you’re saving time then what do you use that time for? Do you use it to think more carefully about your responses and respond in a more humane and considerate way? Do you use it to spend more time on things which aren’t triaging the ceaseless flow of electronic communication? Or some combination of the two? These are questions which will become integral parts of professional practice in the coming months and year but which at present lack clear agreement with regards to professional norms and standards.
The problem in taking automated scholarship, which is how I’d see this even if it’s a stretch to label e-mail as scholarship, as a model for automation in higher education is how exceptional it is to have volitional control over the use of an automated system. It’s much more likely these decisions will be imposed through the digital infrastructure upon which the operations of the university depend. Imagine for example what is likely to happen at the intersection between Office 365, Microsoft Teams and Outlook. My instinct is that we shouldn’t surrender the scholarship perspective however because it helps us keep hold of how automation can be used in ways that extend the agency of academic staff, if it is undertaken in a purposive way steered by creative imperatives which are prior to the technical system itself. Furthermore, as Gary Marcus has put it, generative AI is “autocomplete on steroids”.
This is why I’ve found myself thinking recently about the idea of Fully Automated Luxury Universities, to riff on the title of Aaron Bastani’s book. What would an automated university look like which has been designed around the imperatives of freeing up staff to work on the things which matter to them in their own terms? What would teaching look like in a university designed to leverage AI to maximise the quality and quantity of teacher/student interaction? How would the administration of such a university operate in practice and how would the different groups within it work together?
This is a very provisional sketch of a complex set of questions but I’m planning to return to it in order to work up these ideas into something a little more systematic and analytical. The corresponding dystopian picture would be Fully Automated Platform Universities in which the full potential of artificial intelligence to entrench bureaucratisation and austerity, diagnosed by Dan Mcquillan in a not entirely convincing way, comes to be realised in an institutional form in which the prospect of meaningful agency over sociotechnical futures has long since vanished.
The other way into this question would involve identifying the tasks which can be automated. There’s a vast online pool of content about task automation (for example this) much of which reads like it was produced by AI-driven content farms:
Task automation refers to the use of software to complete work activities. Task automation improves the accuracy and consistency of workflows, and powers more efficient processes. Most importantly, task automation streamlines manual processes and minimizes the amount of labor required to produce a specific result.
Businesses often use task automation software to simplify and speed up workflows, and to alleviate the burnout and fatigue that come from completing the same task over and over. Task automation allows workers to spend more of their time on activities that create value for the company, like solving problems, improving processes, and taking care of customers.
https://www.pipefy.com/blog/task-automation/
This moves attention away from the replacement of jobs and turns it to the automation of the tasks which make up those jobs. The problem with such a view is its tendency to assume that a job can be disaggregated into an independent sequence of tasks, as opposed to interdependent clusters of activities untied in part by the normative expectations which someone personifying this occupational role experiences themselves as being subject to. From a sociological perspective this suggests that task replacement/automation will usually carry unintended consequences in which automating one aspect of the job will have knock on effects in others. This is recognised in the legitimating discourse continually invoked (as above: “Task automation allows workers to spend more of their time on”) but the negative implications are less widely recognised.
Mark Carrigan 