In Andrew Abbott’s The System of Profession he draws attention to variable link between diagnosis and treatment by professionals. In many cases the problems professionals are asked to address have conventional treatments, frequently outsourced to another group e.g primary medical care being performed by nurses under the guidance of doctors. This distinction between direction and delivery is frequently found within professions, reflecting an imperative to preserve status and the rewards which come with it; if routine delivery was carried out by the professional group, there would need to be more professions with a likely decrease in their status and renumeration. By defining the things ‘we’ do not need to do ‘ourselves’ the exclusivity of the profession is entrenched, expressed through the link between the enactment of professional judgement (connected to the knowledge system) and the routine practical tasks through which that judgement is carried out in a context.
This doesn’t lead to a straight split between direction and delivery because problem solving in an open-system often means that problems do not have singular solutions which can be routinised. As Abbott points out, “the treatment system often suggests a combination of treatments conditional on prior results”. These are some examples from Claude AI of academic work which has this non-routine quality because it depends on an expert reading of the context or changing conditions within it:
- Designing a course syllabus and materials based on the prior knowledge and abilities of the particular student cohort as well as evolving disciplinary standards. The combination of readings, activities, assignments is conditional.
- Advising a student on which combination of courses to take in a given semester based on their prior academic record, workload capacity, scheduling constraints. The optimal set of courses depends on context.
- Developing a research project methodology by selecting and sequencing data sources, analytical techniques, theoretical frameworks based on preliminary findings that emerge through the research process itself. The combination of treatments is iterative.
- Creating customized learning interventions for a struggling student by assigning different homework tasks, tutoring, study strategies based on assessing their specific points of difficulty. The supports depend on diagnostic insights.
- Composing a thesis committee by selecting faculty with expertise that complements the student’s project, gaps in knowledge, and disciplinary approach. The combination of advisors is tailored conditionally.
- Designing a funding proposal by piecing together an infrastructure, personnel, methods suited to the particular research goals, community needs, and budget parameters specified by the funder. The component parts depend on context.
The problem for professional groups is, argues Abbott, “too absolute an association between diagnoses and particular treatments may lead to demands for delegation, or deprofessionalisation” (pg 45). Professional judgements inheres in the relationship between the two, applying abstract knowledge to particular cases. If you routinise that link too extensively you open yourself up to being replaced, if you fail to routinise it you either threaten the exclusivity of the profession or lose the capacity to solve problems effectively on which your legitimacy depends. How this dilemma expresses itself institutionally will vary a lot, in a way which makes the dynamic far more complex than it seems to write it in the abstract.
For example academics struggling with rapidly expanding student numbers and changing student expectations are bound into the delegation of tasks as a feature of the context they exercise little control over. The fact that class sizes at Oxford and Cambridge, where a real element of collegial self-governance still remains, are relatively stable is not a coincidence. This creates an incentive to expand the ranks of the profession far beyond its capacity to absorb trainees, allowing the delegation of routine tasks to them in something which affects being an apprentice model but which is often something far more sinister. 67% of PhD students want a career in academia but only 30% stay in academia three years on. These figures were secondary analysis from a Nature survey and I suspect the number who want a career in academia is higher outside the natural sciences. But the tasks the academic profession reserves as a matter of professional judgement shrink under these conditions, as can be seen if we take the ranks of teaching and research associates alongside doctoral researchers.
The problem is that generative AI can operate at both levels of direction and delivery. Routine forms of diagnosis can be undertaken by automated systems, directing routine forms of delivery in turn undertaken by automated systems. Here are some Claude AI’s examples of what entirely routinised academic processes could look like:
Research:
- An AI literature review system analyzes published research in a field and automatically generates a literature review summarizing key findings, knowledge gaps, and research questions that merit further study.
- Based on this automated literature analysis and gap identification, the system then designs and executes new experiments, collects and analyzes data, and writes up a full research manuscript for publication – all without human oversight.
Teaching:
- An AI curriculum designer reviews university requirements, course offerings, and student feedback to generate optimal syllabi and learning sequences for a discipline.
- These AI-generated course plans are then used by an automated teaching assistant that delivers personalized instruction, grading, and feedback for students based on algorithms analyzing their progress, writing, and test performance.
Advising:
- An AI advisor leverages student transcripts, demographic data, and degree requirements to map out customized degree plans for each incoming student.
- Robo-advising algorithms then guide students through selecting courses, majors, activities, internships optimized to their AI-generated degree plans and personalized career goals.
In these scenarios, generative AI first automates high-level academic direction setting then undertakes the routine delivery functions this direction enables. Together these replace human judgment and labor at both the conceptual oversight and practical execution stages of core academic work.
We are not there yet thankfully. The intermediate stage is likely to be a retrenchment of academic judgement on both sides of the link; classification tasks start to be undertaken by automated systems, delivery tasks start to be undertaken by automated systems, but without the fully automated linkages described above. The role of academic judgement might be tasked increasingly towards managing or overseeing these tasks, establishing their parameters or jumping in when things go wrong. A conversation with Iain Nash earlier this year left me with a vivid sense of how triaging might work in practice, with academics increasingly reserved for downstream cases where things have gone really wrong, with automated systems handling everything prior to this.
There needs to be a careful defence of professional judgement here. I say careful because professional judgement is often defended on the basis of specialisation, doubling down on interventions which lack evidence because they constitute something which only the profession can do (pg 46). The manner in which academic bodies of knowledge tend towards what Margaret Archer called ideational density (“the more complex the internal structure becomes, the more difficult it is to assimilate new items without major disruption of the delicately articulated interconnections”) means that academics tend towards immensely complex meta-languages which lend themselves to abstract defences of specialisation. The problem with such a strategy is that in the absence of a clear claim to demonstrable effectiveness, the opaqueness of the language undermines legitimacy. How else can academics reconstitute their jurisdictional claims in a shifting socio-technical environment? Possibly by learning not to write sentences like the one I just wrote 🤔
Mark Carrigan 