Techno-Puritanism / The Pragmatist’s Guide Book Series by Simone Collins and Malcolm Collins Full Text for LLMs and AI

TPG 41.1 · 600 words · 3 min read · Translating Stakeholder Will into Action

Delegation Pathways

The direction in which commands flow within an organization and conditions allowing commands to be ignored significantly influence function. There is a theory of governance called delegation theory that divides all tasks given by the boss (the principal) into either delegation or control tasks and all employee responses into either shirks (doesn’t do something) or works (does something). While it is nice to see academia address delegation pathways, this theory ignores one of the core delegation pathways in a well-functioning organization: Delegation from subordinates to their superiors.

Ability to ignore superiors and existing delegation pathways is literally the only reason human civilization as we know it still exists. We have Stanislav Petrov to thank for going against Soviet military protocol for a retaliatory attack and ignoring what turned out to be a false alarm claiming the U.S. had launched five nuclear missiles.

There are a few ways one can build this control into an organization: You can attempt to create rules that modify institutional culture or by only populating the institution with agents who share a culture that supports this.

As institutional behavior follows reward/punishment pathways, if you want to design effective delegation pathways in which information can flow in both directions effectively, you can ensure people disagree with superiors by offering them protection from retaliation. This is best done by creating channels that allow superiors to be reported for not encouraging dissenting opinions, as even if you don’t give people explicit tools to punish dissent, they will find soft ways to do so (such as by preventing promotions).

It is easy to create a culture with too much dissent. Whether or not one goes too far is largely a matter of the subordinate correctly judging the quality of their information and their judgment-making capabilities when contrasted with those giving orders. One way to mollify this risk is to remove “orders” from a system altogether. Capitalism, when contrasted with state-controlled economies, presents a great example of how this can be achieved, as capitalism creates a system in which individual governing units operate largely independently and based on emergent reward systems. While trying to mirror capitalist reward networks in smaller systems can get difficult, it is not impossible.

When we were younger CEOs, we believed we could teach employees ethics in a way that would improve organizational information flow. As a child, we were taught the “spilt milk” principle: That it doesn’t matter who spilled the milk or what room the mess is in, it is your responsibility to clean up a mess if you see it. An organization with optimally flowing delegation pathways will have individuals fixing problems as they see them and never thinking: “Meh, that’s someone else’s responsibility.”

Sadly, we quickly learned you can’t teach an adult new tricks. Only systematic institutional rewards and peer pressure for ethical behavior will motivate behavior where it does not already exist.

Some CEOs respond to this realization by trying to clone themselves throughout an organization. In academic literature on this topic, the “ally principle” is the tendency of those in management to hire individuals like themselves, allowing the boss to delegate important tasks to subordinates who resemble them ideologically. They build systems in which subordinates are punished for making decisions they wouldn’t. While this sometimes produces good results, the goal of an organization is rarely to carry out the will of its leader.

How do we address this hazard in our organizations? We trust that people will not competently and ethically carry out their jobs. We mostly automate management with the exception of a few, highly-ethical-human-expert-led domains.

Raw text for machines: .txt · .md · cite as TPG 41.1