As a newish engineering manager, I was once responsible for a large infrastructure migration that did not go very well. We had migrated internal services for several weeks before having to touch our primary, customer-facing application. I was working with a star engineer who I was confident would eventually complete the migration. But we hadn’t done much pre-work to anticipate potential failure modes. I assumed we could just push through any problems we encountered. So, when automation began to fail and Slack filled with the complaints of our distraught engineers, I sent a cordial message to the other engineering managers, asking for their teams’ support while we completed the migration (i.e., please stop complaining!).
A more senior colleague quickly reminded me it had been my responsibility to execute the migration without disrupting everyone’s work, so it was also my responsibility to deal with the fallout if the migration was failing. In other words, I had to accept the stress and deal with the situation.
As a new manager, I had interpreted my task too narrowly, focusing only on the top-level goals and none of the externalities imposed by it. Nobody had told me that I couldn’t disrupt either team while conducting the migration, but they shouldn’t have had to! I had implicitly accepted that it was OK if I disrupted other people’s work for a few days if it allowed me to get my own done.
Only in retrospect did I recognize that I had wanted to stifle criticism because I had already partially failed the migration. Rather than admit to myself that things were not going well—reassessing the parameters of my project and working transparently with peers, stakeholders, etc., on a less disruptive implementation—I shifted responsibility to everyone who had grown critical of the project. If they could please just quiet down a bit, didn’t they understand this would all be over soon?
I wish another responsible party had stepped in to correct my mistakes. A prudent superior might have noticed the collateral damage of my rushed migration and asked that I not undermine the organization’s productivity. But as our own responsibilities increase there are fewer superiors to step in on our behalf. As colleagues detect our failures, they likely will not have sufficient influence over our behavior to force us to change. But they can at least complain? Which might accumulate into sufficient social stigma that we force ourselves to change anyways.
It’s possible the migration could never have been conducted without negative externalities impeding my colleagues’ work, but my colleagues’ complaints were justified exactly because I had failed to prepare for the costs. If I had telegraphed the migration, carefully scheduled downtime, run trainings, measured impact and run against pre-defined thresholds, etc., then I might have more rightly asked a teammate dissatisfied with the migration to just hold on to their pants. If risks are already well-understood and well-aligned, then complaints become unproductive.
I often encounter meta-level discussion in the AI safety community interrogating how one might behave in a game-theoretically optimal manner to encourage openness and increased safety at frontier labs. Frequently, this devolves to reminders to employ basic civility (which ought to be a given?). However, there are often also discussions of when one should be “angry,” how one ought to respond to the psychological welfare of individuals directly involved, etc. I am very sympathetic to the difficulties of everyone involved in such high-paced labor. But I am also reminded of my own experience as a manager, where exhortations for colleagues to “just be nice” when I was screwing up my project were themselves misguided and aggravating.
There is always a risk that criticism will encourage a recipient to turtle up and become quiet or evasive. If nothing is visible to criticize, then all is well! I feel this infantilizes the recipient, who has a personal responsibility to do the “right” thing regardless of criticism, though I acknowledge how incentives might pervert this. But, more importantly, if the externalities of an action are high and arise over a sufficiently short timeline, then one can only withhold information for so long. My colleagues didn’t need to know I was running a migration. They only found out because automation started to fail! If the feared externalities of poor safety practices are real, then companies are limited in their ability to hide those issues.
A counterpoint might be a fossil fuel company that withheld information about climate change or misled the public about the impact. Here, the externalities simply arose over too long a time frame for a self-regulating mechanism to take effect. Perhaps prosocial tools to encourage leaders to disclose such information to the public would have helped… though likely not in this case? Or consider the Iran War, which has had massive (if diffuse) externalities around the world. A more productive relationship between Congress or internal military experts and the executive might have avoided such a catastrophe. But the fault there seems to lie mostly with the Trump administration itself. If the primary actor is already non-cooperative, then the public reception is a matter of little consequence. The public may “reprimand” the Trump administration for the Iran War in public polls or, to some extent, the midterms, but here the corrective may be too late.
(I’m not drawing an equivalency with AI companies on either issue).
In personal relationships, a truism is to only offer advice that’s true, kind and necessary. That’s helpful guidance for staying in a healthy partnership. It’s also a good directive for how to engage publicly and professionally. People should be more kind! Even when discussing politics, existential risk, whatever, you shouldn’t be a jerk. But it’s not fair to litigate kindness in others when they are responding to real costs.
A challenge with extremist politics on the left or right is that they always view the costs as existential. A far-right nativist frets at the dissolution of race. A far-left communist rails against the criminality of executives. Because the perceived costs are terribly high, anger is justified, destruction is justified, etc. The increasingly existential terms in which we discuss current affairs and politics thus lend themselves to an atmosphere of tense, unkind discourse.
But even when dealing with irrationality, can arguments be won on emotive posturing, without addressing the cause? A failure of modern Liberal politics has been an unwillingness to address the faults which have led to increasing extremism. Insisting things are fine, while pointing duly at economic metrics, does little to shift a polity’s lived experience. One would do better to search for the underlying problems, whether that is existential malaise or a simple shortage of housing, rather than waging meta-arguments about the temperature of discourse.
It is good to encourage a kind, rational discourse. But in public/professional life, we should not misdirect attention toward policing aggravation. It’s best to take an expansive view of the problems at hand, while asking principal actors to assume responsibility for their costs.