Commentary on ‘Socially Extended Scientific Knowledge’
By Mark Sprevak
2 November 2022
One of the most interesting turnarounds in the paper is when an objection to socially extended knowledge from K. Brad Wray is considered, answered, and then turned into an objection to distributed (group) scientific knowledge. Wray’s objection centres on how cases of malpractice inside scientific teams are handled. Generally, such cases are resolved in terms of responsibility being assigned to individual researchers: the finger of blame invariably points to a single person. This suggests that the epistemic credit in creation of scientific knowledge is likely to lie with an individual too. Duncan Pritchard argues that Wray’s cases do not pose even a prima facie problem to socially extended knowledge. However, I am not sure that Wray does not have a reply.
Imagine a minimal case of a scientific team, consisting of a lab leader (Head) and a research assistant (RA). Most scientific research teams have a more complicated internal structure, but this basic setup will serve to illustrate the point.
When RA arrives on their first day, they begin to learn how to operate the instruments, analyse data, run experiments. Over time, they become able to do this more adeptly and so facilitate Head’s overall research goals. As with the other resources in the lab, RA’s cognition and behavioural capacities facilitate the creation of true beliefs and knowledge. As Pritchard describes it, this setup is a case of socially facilitated scientific knowledge. As Pritchard also makes clear, the relevant states of knowledge belong to Head – the role of RA, like that of other technological devices in the lab, is to facilitate the acquisition of Head’s knowledge. (One might even imagine that RA lacks the relevant theoretical concepts and background knowledge to entertain the relevant knowledge states themself.)
Imagine that, over the years, the dependence of Head on RA grows and the interactions between them more fluid. Head’s internal cognitive processes might become completely intertwined with what what RA can do. If a sufficient degree of integration is achieved, we have, by Pritchard’s lights, a case of socially extended scientific knowledge.
Now suppose that at some point after this integration process takes place, RA becomes overly ambitious, disillusioned, bored, or acquires another undesirable motivation that leads them to start falsifying data. Note that this is typically the way such cases proceed: it is rare for a junior researcher to commence their career by falsifying data. They first need to serve an apprenticeship in order to be treated by their collaborators as trusted team members. Once they are ‘in’, they then have scope to misbehave. After they have been integrated into a team – such that any normal oversight procedures, delays, and checks are no longer required – real epistemic poisoning can start.
Wray’s observation is that whenever such wrongdoing is brought to light, blame is normally assigned to a single individual (RA). Wray suggests that this indicates that scientific knowledge was never socially extended in the first place. Pritchard points out that Wray’s observation is compatible with knowledge being socially extended, because socially extended knowledge is always ascribed to an individual. All that happens is that, in this case, an individual (Head) makes use of information-processing resources that extend outside their brain and that lie partly inside another human (RA). What Pritchard says here is right: socially extended knowledge does remain the property of an individual. However, what is puzzling is, if knowledge is extended in the way that Pritchard suggests, why RA would be blamed for any cognitive failure rather than Head.
On the view of socially extended knowledge proposed by Pritchard, Head seems the most plausible candidate for epistemic responsibility. Head is the subject who acquires and holds the relevant belief states. Head is the one whose cognitive processes are responsible for acquiring those beliefs. RA functions – like an instrument or technological aid – as simply a conduit to acquire and maintain Head’s belief states. RA’s brain, sense organs, and limbs are subsumed to, they are repurposed as implementations of, Head’s cognitive processes. If there is something that has gone wrong in generating or maintaining knowledge, then it is Head’s extended cognitive processes that are to blame. Why then does RA get the blame?
Pritchard suggests what might be happening in this case is that there is a ‘recalibration of where cognitive responsibly lies’: we discover that a member of the team we thought was playing a role was not actually playing that role: ‘there are fewer collaborators who are proper parts of the socially extended cognitive process than initially thought’. But it seems hard to sustain this line of argument. It is not that when the malpractice came to light, we then discovered that RA was never part of the team in the first place, nor that RA silently dropped out of the research team once they started falsifying data. RA became, and remained, part of Head’s extended cognitive process after integration took place. It is just that they start feed poor quality information into it. Blaming them for this, however, is puzzling. On Pritchard’s picture, the knowledge acquisition processes are owned by Head. It is Head’s cognitive processes – which coincidently happens to employ bits of RA’s brain and body – that have gone wrong.
In short, the worry expressed by Wray’s example may not be as it appears. The worry is not that blame for malpractice attaches to individual researchers rather than to the group. The worry is rather that blame attaches to the wrong individuals. If RA engages in malpractice, we blame them, not the leader (Head). But if RA’s contribution is literally part of Head’s cognitive processes, then surely the finger of blame should point at Head – it is their cognitive process that is at fault. It looks like either we have been doing it all wrong (and our leaders have been getting off scot-free), or the socially extended picture is not handling these cases correctly.