Deontological bars should reference the actor's beliefs

In a recent LessWrong post, user TFD responds to Scott Alexander's exploration of 'deontological bars'—moral lines that AI safety activists may feel they cannot cross. Alexander had framed two opposing camps: those who work with the least irresponsible AI labs to steer them toward safer outcomes, and those who push for pause or ban via activism, sometimes involving messy coalitions. Both sides fear the other is violating a deontological bar: the first group worries that supporting any frontier AI company, even the 'less bad' one, is complicity in potential extinction; the second group fears that mass activism requires alliances with unsavory figures or tactics that betray ethical purity.

TFD argues that deontological bars cannot exist in a vacuum—they require the actor to hold specific beliefs about the action's harmful nature. For example, working for an AI company is only deontologically barred if the worker truly believes that company is likely to cause extreme harm. Similarly, engaging in activism is barred only if the activist believes their methods are dishonest or extreme. Unlike consequentialism, deontology cares about intent. Thus, general, context-free bars against 'supporting AI companies' or 'doing activism' are incoherent. The article calls for nuance: ethical constraints depend on what the individual knows and believes, not on abstract labels.