So for whatever reason, today I feel particularly interested in the mushy ideas in my head. While taking a shower, this rough model of my personal general purpose trust metric fell out. I'll try to describe how I view trust as a whole, keep in mind I'm not talking about any practical implications of these ideas....
Now, I'm not talking about trust as a moral question, but entirely in the sense of resource allocation and access control. Yes, I think about things like this when I think about people in my life. Yes, that's probably insane. Nevertheless, it's reasonable to generalize this a bit. Let's think of this as a system to govern resources and access given to
participants to solve a particular
problem. Here trustworthiness is then defined roughly as function describing the level of
effectiveness a given participant has at solving the problem. Effectiveness can be a negative or positive quantity, basically acquired by summing up the weighted result of
actions in the course of working on a problem.
An effectiveness rating determines where a person falls in a series of roles. I suppose this could be scaled as needed, but in most situations, I can classify a participant as fitting one of 6 possible trust levels, a random and frivolous graph shows them below.
So basically you can see that each of these roles is in order of effectiveness from worst-case to best-case. One thing I find that is relatively controversial in society is whether trust is something that should be given by default and then adjusted accordingly, or that it should be earned.
I'm of the opinion that trust should be earned, however, I think that it's inappropriate to give someone a jail sentence before the crime is committed. Even moreso, I think that artificially limiting a participant's potential effectiveness is well... ineffective. Arguments that this is usually done with security in mind may be valid, but keep in mind my view of trust is as a function of effectiveness, not safety.
So let's say that in general the neutral role lies at zero effectiveness, with some weighted buffer zone on each side that we'll discuss a little more later. Typically, a participant that is neutral is not causing a net gain or loss on the effectiveness scale, but basically balancing out between their effective and counter-productive actions. A participant in this role is most likely kept around because of their future potential to become trusted, or because they're part of the greater ecosystem in which somehow their presence improves the effectiveness of others.
I haven't thought about how those interactions may work except in the fuzziest of ways, so I won't explain them here. Let's just say in this atomic model of trust, those interactions are unimportant, a neutral participant is basically like constants in Calculus, they just fall out of the equation.
However, a new participant's first action can never result in maintaining a neutral effectiveness. It will either improve or diminish things... for this reason, even though their initial score would be zero, they would have the same level of trust as a trusted participant. This is simply because presumably there is a difference in resource allocation between neutral and trusted participants, and the potential to perform good actions is better for those who have the needed resources.
The question of whether or not the potential to perform bad actions is greater for a trusted participant is not exactly relevant at the moment, this has to do with importance of the problem and I'll try to give a rough description of that later. For now, let's assume even if a trusted participant could kill everyone on earth, I might still provide the same resources to an unknown participant with some confidence things will work out. It's good to start with the absurd and work our way back to the practical world, so humor me. :)
Okay, so basically, what we've discussed so far is how new participants enter the system, and the rough trust levels which are laid out. I'll be staying far away from concrete details of what one might do with these roles once they've been assigned, or even what the 'resources and access' might be which this system governs. None of that is important, it will always depend on what the problem is.
We also know that a participant's first action cannot cause them to become neutral. They have the same resources as a trusted participant, and are kept as 'unknown' until they fall into one of the other categories. For this reason, the system gets very confusing if you do not make neutral a very small range. Realistically, this range should roughly represent 0 +/- some confidence factor. This confidence factor would roughly represent how sure you are that you're accurately calculating a participant's effectiveness, which probably would be some function of time and number of actions. The implication here is that it becomes harder and harder to maintain a neutral presence. You're either helping or hurting, and I think this is true in most cases.
Anyway, this brings me to the idea of weighting, and also what most of these roles mean in the context. The key here is that the weighting is external, and roughly represents the importance of the problem to whoever 'owns' it. Ownership simply implies the ability to grant resources, and for simplicity, let's assume this is a single source.
The overall weighting is simply an effectiveness range from worst to best. This range is based on the assumptions of if there are no participants other than the owner, what is the worst and best effectiveness that can be achieved. Using this, you basically have a seed from an authoritative source, and that scales the rest of the system.
Once roles are assigned, the distance between them can be tweaked as needed to fit what are basically risk vs. reward calculations, and this would probably get quite complicated as you tried to optimize different people's potentials. Let's assume that we're doing this as a feedback control system, and setting these levels can basically be done at random, that the effectiveness scales effectively consistently as you tweak each one, and the business is mostly to optimize them. Maybe if this wasn't such a crackrock idea, I would have some way to formalize that, but I don't. The best I can offer is my hazy view on what each role means to me:
-
Malicious: Net effectiveness is worst than the owner's worst case. This means that such a person most likely will not be granted any resources, but in fact, resources may need to be expended to prevent them from causing harm. However, because this system is dynamic, a participant needn't be permanently malicious.... if the confidence factor degrades over time, his score will eventually deflate to something less severe.
-
Untrusted: Net effectiveness is somewhere in the negative field, so more actions are harmful than helpful. Such a participant may have severely restricted resources and is likely only kept around because a) killing them isn't an option or b) they play an important part in the overall ecosystem, which is not described here.
-
Neutral: Net effectiveness is roughly a wash. It seems unlikely that this would really happen much in practice, but it provides a small buffer zone for those who either are extremely inconsistent or have disappeared for a while and need somewhere to come to rest. I mostly leave this role in this discussion for symmetry, it may not be needed.
-
Trusted: Net effectiveness is positive and sufficient for enough progress to be made that the participant is not considered Neutral, and thus is not immediately expendable. A trusted participant has access basically to all the resources needed to work on the problem directly, and has room for growth.
-
Helpful: Someone who raised the bar above the owner's best case estimates. This person eventually pushes the curve higher, and thus deserves all the same rights (in the context of the problem) that the problem owner does. I don't think it has a place here, but in practice, a consistently helpful person may become the defacto or actual problem owner, or in a multi-owner system, at least a partner.
After discussing these rough ideas, I realize I've mainly written them because they were fun to think about, not because I think they're particularly useful. It does vaguely represent how I view trust though, and I'm curious if there is anything practical hidden within there.
The appealing thing about this model is, like my actual outlook on how to trust people and with what, it's extremely decentralized and decomposable. You can break it up into as many parts as you want, change the importance levels of things, and generally go wild with how you tweak the overall system.
This is why I don't need to "like" you to trust that you'll do good work. It's why I don't need to believe you are honest to accept a gift from you. It's also why I probably have some sort of dementia that causes me to write such things. :)
