Jochen Werne

Reflections · Jochen Werne

A case for a new Age of Enlightenment

Independent judgment, technological progress and responsible leadership.

Reflections · Jochen Werne

The more power technology gives us, the more we need institutions that can recognise and correct mistakes. That is the case for a new Enlightenment. Independent judgment demands effort from individuals and an obligation from those in authority to hear evidence that could change a decision.

The freedom to think carries a duty to check

„Aufklärung ist der Ausgang des Menschen aus seiner selbst verschuldeten Unmündigkeit.“

Immanuel Kant, Beantwortung der Frage: Was ist Aufklärung?, 1784

Kant’s selbst verschuldete Unmündigkeit names a failure of resolve. It takes less effort to accept a claim than to examine it, especially when it comes from a group we trust or confirms a conviction we already hold. We depend on others for most of what we know. The question is whether we can ask how they know, compare their account with other evidence and correct ourselves when the evidence changes.

Technology changes the terms of that choice. A language model can translate a safety warning for a multilingual crew. The same capability can make an impersonation more convincing. The purpose matters, and design influences which purposes are easy to pursue. A platform that makes sharing effortless and checking laborious puts a thumb on the scale. Those who design and deploy the tool share responsibility with those who use it. Kant’s original text ↗

There is a difference between an individual’s duty to check and an institution’s duty to make checking possible. A reader can compare sources only if the underlying material is available. A customer can challenge a decision only if there is a meaningful explanation and a place to take the challenge. Asking people to exercise judgment while withholding the means to do so leaves Kant’s demand half fulfilled. A new Enlightenment therefore asks a question of citizens and of those who control access to evidence.

Four passages from Kant worth reading again

Courage and convenience

Kant puts the emphasis on the resolve to use one’s understanding. A manager can ask a language model for the strongest case against a proposal and use the answer to find evidence they had missed. The trap is asking for a verdict and repeating its confident wording, even when the source documents are within reach. The same tool can make inquiry easier or provide a convenient way to end it. Kant on self-incurred immaturity ↗

Guardians and the power to set defaults

Kant describes how guardians cultivate dependence by making independent inquiry seem difficult. Ferguson’s history of networks and hierarchies adds a question about power. Who can reach whom, and who controls the channels? The printing press widened the audience for new ideas. Platforms widen participation today, while their owners can alter the rules of visibility at a scale an individual printer could scarcely imagine. When a feed supplies the same claim from several accounts, the appearance of independent agreement may conceal a single source. A useful first step is to trace where each account got its evidence. Ferguson on networks and hierarchy ↗

Public reason and the duties of office

Kant uses an unusual distinction. Public reason is the freedom to address a wider reading public as a scholar. The private use of reason is the reasoning exercised inside an assigned civic role, where duties can constrain action. His example of the officer helps explain why a crew needs clear command and a way to scrutinise the rules of command. On a bridge, the immediate duty to execute an order can coexist with a standing procedure for challenging an unsafe assumption and a later review of what happened. That is a modern application of the distinction. Kant on public and private reason ↗

Leave the next generation room to revise

Kant argues that one age cannot bind the next to rules that prevent it from correcting errors. Today a board can create a smaller version of that trap by buying a system whose decisions it cannot examine or whose records it cannot take elsewhere. Keeping access to evidence, preserving the option to change suppliers and giving people a route of appeal leave future decision-makers room to act on what they learn. Kant on revision across generations ↗

Connection changes what a society can see

Ferguson’s history of networks and hierarchies offers a way to understand the double effect of communication technology. The printing press widened access to learning and gave polemic a larger audience. Social media lets relatives stay in touch across borders and citizens organise around a common concern. It also lets a rumour travel faster than the people checking it. A personalised feed can repeatedly reward familiar or inflammatory claims until its selection feels like the whole debate.

Evidence supports different parts of that argument to different degrees. Vosoughi, Roy and Aral found that false stories travelled farther and faster than true ones in their historical Twitter dataset. That comparison does not tell us what a particular ranking algorithm caused. In a three-month field experiment around the 2020 US election, changing Facebook and Instagram feeds to chronological order altered exposure and behaviour, while the researchers detected no significant change in the polarisation measures studied. Algorithmic design matters. Political identities, social relationships and institutions need to be part of the explanation too. Study of news diffusion ↗ · Feed experiment ↗

The practical question for society is where to put the effort. Experiments on accuracy prompts suggest that drawing attention to accuracy can improve the quality of what people share in the settings studied. A prompt is a modest intervention. Newsrooms, schools and platforms can also make it easier to find original evidence and harder for a correction to disappear behind the claim it corrects. Citizens can ask what would change their minds before passing a claim on. Accuracy-prompt research ↗

Correction also has an audience problem. The person who saw a claim yesterday may never see the better evidence today. Publishing corrections where the original claim travelled is one practical response. Another is to explain what the new evidence changes, so a reader can update the account of what happened. An institution that invites scrutiny should let people follow a decision from the original grounds to its revision.

Business leadership means developing judgment across the company

AI changes how a company can organise work. The strategic question is which decisions improve when people and machines work together, and how the company will notice when the arrangement fails. In the consultant experiment discussed in Mind over Machine, AI helped on some tasks and reduced the chance of a correct answer on a task outside its capabilities. Adjacent tasks can look equally tractable and behave very differently. Fluency gives the user little warning of that boundary. Field experiment ↗ · Mind over Machine ↗

Consider a board approving a new fraud model. A higher detection rate may look like progress until the people reviewing rejected transactions show that legitimate customers are being turned away. The review team can identify cases the aggregate figure hides. The board can ask for a baseline, measures for both missed fraud and unfair rejection, and a named owner with authority to suspend the rollout. Training should help staff ask better questions, check evidence and escalate exceptions, as well as use the tools. The company learns when those observations can change the workflow.

In Amy Edmondson’s study of 51 manufacturing teams, psychological safety was associated with learning behaviour. That finding gives an executive team something concrete to examine. Do staff who question a model get help investigating an anomaly? Does a manager report an adverse result when it could put a launch at risk? Training in AI should give people practice with those situations. Give two teams the same difficult cases, ask one to use the new tool, and compare the quality of the decisions and the errors the teams can explain. Carry what they discover into the next release. Edmondson’s team study ↗

Learning offers a test of whether that room is being used. In a field experiment in high-school mathematics, students could perform better during AI-assisted practice while an unrestricted assistant weakened later performance when the tool was removed. A tutor designed with safeguards substantially mitigated that effect. The result is specific to the setting, but the distinction travels. An assistant that helps people ask questions and check their own reasoning can extend their ability to judge. An organisation should measure what people can still do when the assistant is unavailable. Mathematics tutoring experiment ↗

At sea, the right warning has a deadline

A person on watch may pick up a change before the skipper does. Authority is useful because someone must act while there is still room to manoeuvre. It becomes a liability when the people closest to the evidence hesitate to speak or their objections are brushed aside. The US National Transportation Safety Board’s investigation into the loss of El Faro identified several factors, including the use of noncurrent weather information. Investigators also found that the captain’s failure to consider officers’ suggestions adequately contributed to the sinking. That is an investigated case of information failing to shape action in time. NTSB investigation ↗

On an offshore passage, the time to agree when to call the skipper and who cross-checks the forecast is before the weather turns. In a company, intervention thresholds and escalation routes should likewise be settled before a launch. Once an emergency starts, rank and a clear order may be essential. An institution can keep that authority and still require leaders to hear contrary evidence, act on it when warranted and explain the decision afterwards.

Before departure, a crew can agree what conditions would trigger a change of course and who will check the latest forecast against what is happening on deck. Afterwards it can reconstruct which information was available and when. The equivalent in business is to write down the assumptions behind a launch and the conditions that would trigger a pause. Both practices make it possible to learn from a sound decision with a bad outcome, or from a lucky outcome that hid poor judgment.

Diplomacy gives rivals something specific to verify

The Cuban Missile Crisis showed what happens when technology shortens the time available to interpret an adversary’s intentions. Ferguson’s second Cold War argument invites comparison with US–China rivalry. Today, competitors are also linked through production, capital and research. Interdependence can transmit shocks or become leverage. It also leaves concrete interests that both sides may want to protect. The hard question is where they can make a limited promise that each has reason to keep. Cuban Missile Crisis record ↗ · Ferguson on a second Cold War ↗

The Antarctic Treaty is a historical example. In 1959, twelve governments including the United States and Soviet Union agreed on peaceful use and scientific cooperation. Article III provides for sharing scientific observations and results. Article IV preserved differing positions on sovereignty. Article VII allowed inspections. Those provisions gave rivals ways to work together and check compliance without first settling their political disagreements. People-to-people diplomacy can open a channel through shared fieldwork or remembrance. International understanding deepens when people can carry out a common task, compare what they observe and keep talking when they see it differently. Formal agreements add obligations a relationship alone cannot supply. Antarctic Treaty ↗ · From Antarctica to Artificial Intelligence ↗

There is evidence for a modest claim about what shared work can achieve. In a field experiment in post-ISIS Iraq, Salma Mousa randomly assigned Christian football players to teams with or without Muslim teammates. Mixed teams improved several behaviours towards Muslim peers. The effects transferred less consistently to people outside the football setting. An expedition or a remembrance project can create conditions for people to work across divides. The harder task is to build on those relationships afterwards, when the shared task has ended and political incentives may again pull people apart. Field experiment on intergroup contact ↗

What progress would look like

A new Enlightenment is a proposition about how decisions should be made. The empirical work shows that some interventions change what people see, share or decide under specific conditions. The historical cases show how responsibility and verification can be organised. The proposition itself should be judged by results. Can a customer challenge an automated decision? Can a crew member make a warning heard before the last safe moment? Can rival parties check whether an agreement is being kept?

For decisions with serious consequences, record what would justify a change of course, give the people likely to spot an error a route to the decision-maker, and review whether warnings led to timely corrections. A procedure that adds delay without catching serious errors deserves revision. The case for a new Enlightenment rests on making it easier to discover when we are wrong and possible to act while that knowledge still matters.

Three measures would make that proposition open to challenge. Track the time between an early warning and a decision to act. Check which people can use an appeal or escalation route and whose evidence is systematically missed. Then look at outcomes, including the people harmed, the errors prevented and the time spent reviewing the decision. If a procedure makes the record look better while none of those outcomes improves, the procedure itself should be reconsidered.

All perspectives