The world of today is often a messy place, not just because of the havoc we wreak on nature and each other, but also because of the information and misinformation we are bombarded with. Before information technology (IT) and the Internet took over, the availability of too much information was a problem only for privileged elites with access to more books that they could read. These days, most of us are inundated with news, advertising, emails and the like daily, while social media postings and pop-up alerts soak up our time, moving us away from slower and more reliable forms of gathering information. It can be hard to separate the wheat from the chaff in the ensuing information overflow. While new technologies and the breakdown of many traditional media gatekeepers have brought about significant benefits, like allowing for important, previously marginalized, voices to be heard, the sheer volume also creates new problems and exacerbates existing ones. As social creatures, we are epistemically reliant on each other. Yet most of us live in bad epistemic neighborhoods with regard to at least some important issues. Our modern epistemic environment is seldom a healthy one: In addition to the overabundance of information that is coming at us too fast, a significant amount of content is AI slop or misinformation.
Here I will not focus on our overcrowded epistemic world per se but on a more confined question of how organizations can also be messy epistemic environments. Collective agents have a big part to play in shaping our epistemic environments. A lot of the information and misinformation comes from collective entities, such as corporations, governments, and different associations. Not only do these collective agents affect our epistemic surroundings, but they also have complex epistemic inner lives: epistemic structures tied to their organizational structures, policies, and procedures. Groups can have beliefs and can also lie about these beliefs. According to Jennifer Lackey’s influential account, a group lies when it states something it believes to be false while intending to deceive its audience. The group lie needs to be made by the group, either through a coordinated assertion or through authority-based assertion, for example, through a spokesperson. This is an elegant answer, but it can sometimes be hard to access what groups actually believe to be false. After all, groups aren’t unified minds but are made up of various epistemic policies and members with different roles, different access to information, and sometimes genuinely conflicting beliefs.
There are plenty of examples of things going wrong epistemically within organizations, with sometimes very serious consequences. Some of the biggest failings end up sparking official government inquiries. The U.K. Post Office introduced a new IT system called Horizon in 1999, to handle accounting for transactions at its branches. Horizon was riddled with bugs from the get-go. Subpostmasters, the people in charge of Post Office branches, began reporting that the system was producing incorrect figures. Yet the Post Office consistently denied that Horizon could generate erroneous data. Subpostmasters who contacted the Post Office’s helpdesk were told they were the only ones experiencing such discrepancies.
As a result, many subpostmasters believed they were isolated in their struggles, which prevented open dialogue about problems within the Post Office system and contributed to a culture where the failures of the organization were blamed on individual subpostmasters. Hundreds were wrongly accused and prosecuted for theft, fraud, and false accounting. Subpostmasters were condemned as criminals due to a defective computer system, “stripped of their dignity and freedom by a faceless bureaucracy that could not admit that perhaps it had made a mistake.” There appear to be different views within the organization about who exactly knew what and how much about the scope of the problems with Horizon. The public inquiry is still ongoing, but the question whether the Post Office as an organization did or did not know is not just a technical one—in answering questions about ignorance or knowledge of organizations, we need to have a way to establish how to delineate the “real” group belief from all the noise and possible individual disagreements around it.
I have suggested that we should introduce a narrative constraint to help us see whether a group statement is an honest reflection of group belief. This constraint is made up of two parts. First, narrative coherence requires that a new group statement does not contradict what the group already knows (via its relevant expert subgroups) or has previously said, unless there’s a coherent reason for the change: new evidence, a genuine change of circumstances, and so on. Second, narrative intention says that when a group forms a position on something in its area of expertise, it has to do so in good faith, genuinely engaging with the best available evidence, rather than cherry-picking information to reach a predetermined conclusion. What matters is whether the group’s public statements line up with the knowledge its own experts have generated (through a good-faith process), not what a powerful spokesperson says or what PR spin is put on an issue.
It is always easier to be wise after the fact. Benefit of hindsight gives one a clarifying lens while browsing through the pages of reports from disasters. The NASA decision makers responsible for giving the go-ahead for the Challenger launch in January 1986 should have known that their contractor’s engineers had warned that the rubber O-ring pressure seals used in the rocket boosters could become unsafe in cold temperatures. The report on the Space Shuttle Challenger accident found that NASA’s culture at the time prioritized keeping to tight launch schedules over addressing safety concerns. Therefore, the launch was not delayed, despite evidence of O-ring erosion on earlier flights and the presence of cold weather at the time of the launch. The report also found that serious technical concerns were not escalated to top decision makers clearly enough, meaning that the engineers’ warnings were not heeded. All seven crew members onboard the space shuttle died.
An even more deadly disaster was the Grenfell Tower fire in June 2017 in the U.K., which saw seventy-two people die when a small kitchen fire spread rapidly up and across the apartment building via the combustible external cladding. The dangerously unsafe material was installed on the exterior walls during a refurbishment. When the final report of the Grenfell Tower Inquiry came out many years later, it concluded that all the deaths could have been avoided. Instead of its being a tragic accident, the fire, according to the inquiry, was the result of a long series of failures by the regulators, government, the local body responsible for the management of the building, contractors involved in the refurbishment, and companies manufacturing or selling cladding and insulation with dishonest claims about the safety of the materials. The inquiry pointed out that they all should have known and acted better.
It should be noted that ignorance in organizations is not uniformly negative. Sometimes it is harmless or irrelevant and simply part of life. Under normal circumstances, there is no reason for a university to know what the most popular sock color is among its employees, for example. Other times, ignorance enables good things to come about, such as privacy, expertise, and efficient division of labor. However, as I have argued in earlier work on culpable collective ignorance, collective agents have a lot of control over their own epistemic structure and thus ignorance does not automatically excuse their actions. I utilized Holly Smith’s idea of tracing culpable ignorance: An act done under ignorance, although not culpable on its own, can nevertheless be traceable to an earlier blameworthy “benighting” act, such as deficient investigation or deficient inference. By applying this idea to collectives, the benighting act can occur at the organizational level through fragmentation of information across roles, rather than being reducible to any single employee’s failure. This allows blame to attach to the organization itself, even absent any bad will in individuals, since bureaucratic compartmentalization can systematically produce ignorant, harm-causing behavior in otherwise well-intentioned employees. The importance of good epistemic structures is not just about reliable channels for spreading information within an organization, but it is also about creating spaces where information from different areas can be brought together.
In the cases discussed, there were clear and serious epistemic faults at the organizational level, leading to devastating outcomes. At NASA, there was a need for improved communication between engineers and managers, while with Grenfell there was poor communication and coordination at various levels among the people who should have known about the dangers of the scaffolding used. The inquiry also found that Grenfell residents’ concerns about safety were routinely ignored by the management body of the local government authority responsible for the high-rise social housing block. The culpability for the poor epistemic structures in these disasters seems clear, even if it is not necessarily straightforward. But what about structures that are inherited from a previous regime, including persistent bad epistemic norms, or epistemic structures whose defects only became visible through hindsight?
People build systems to try to make sense of information overflows. In his recent book, The Unaccountability Machine, journalist and economist Dan Davies asks tough questions about who is to blame when a big system makes terrible decisions. He suggests that the decision-making systems of many large organizations hinder individual accountability by design, creating accountability sinks that sever the relationship between consequences and decision makers. When it comes to truly complex systems, Davies argues that we are led astray if we try to open the black box of organizational decision making to establish responsibility. Instead, we should assign responsibility towards the system as a whole.
I argue, in a similar vein, that when it results from negligent, suppressed, or badly designed information flows, moral responsibility can and should be located at the collective level rather than dissolved among individuals who each, taken alone, could be excused. Where exactly to draw the line is hard to distinguish, though, at least at a general level. How to make a distinction between culpable and merely unlucky organizational failure? Reasonable foreseeability is an area where the answers must lie, but what counts as reasonable foreseeability for an organization is a genuinely difficult question to answer. Another tough question to tackle is how we should approach blame assigned to collective entities in both moral and legal terms. Unlike Davies, I do not think that most of us are just like cogs in the machine. Systems can be improved and the status quo can be questioned. What is certain is that these questions offer fertile ground for philosophers to engage with responsibility puzzles in large-scale collective contexts.
Säde Hormio
Säde Hormio is Academy Research Fellow and University Lecturer in practical philosophy at the University of Helsinki. Her research focuses on social epistemology and collective responsibility. Recent work includes “Group Lies and the Narrative Constraint” (2024) in Episteme and co-authored “Ethics of Doxastic Conduct” (forthcoming) in Ergo.
