Utilitarianism

So, I've talked about Utilitarianism a bit. So, here's a place to get it out of the forums. I've been pushing the idea of logistic scaling and independently sorting harm and utility. This seems to solve all the classical issues with utilitarianism, at least all the ones I am aware of. So, discuss! Or, see if you can stump me.

9/11 update:

So, I had talked about externalities a bit before. It had been floating around undefined, or at least not well defined. Today in the shower an example came to me.

Say, for the sake of argument you're calculating the utility of a politician. In this case, the politician has two acts: accept large sums of money and appoint a police chief. Putting these together may seem like simple addition. However, when you do so you produce a third term: bribery. It's not really simple addition, but an integration instead. This is what I've been calling externalities.

Comments

[blank check]

I have a similar intuition towards separating loss and gain, but what does logistic scaling mean here? As in a sigmoid? What is that applied to?

[deaincaelo]

2419475

Yes, a sigmoid. Consider that function as your basic units.

[blank check]

2420316

So, you mean as an asymptotic bound on both positive and negative utility?

[deaincaelo]

2420390

It is effectively equivalent to that, yes.

[blank check]

2420399

But it moves as a leaky integrator through time, right? Saving millions of lives doesn't mean that later saving more millions results in a negligible utility gain?

[deaincaelo]

2420502

It presumes each causal chain is calculated independently. You can't claim that saving lives as your 9-5 justifies you murdering people on the weekend. At least, not without the extraordinarily dubious claim that the latter is unmitigatable without sacrificing the former at over unity. Saving a million people now doesn't affect saving a million people later, unless there's some causality where doing both leads to an externality.

[blank check]

2420551

That's already achieved with the separation of utility and disutility, though.

[deaincaelo]

2420586

For simple causal chains, but not necessarily for complex causal chains that may have several non integrable variables of each type. Also, this isn't meant to keep a "running total" of your life or civilization like a video game score. I suppose you could, but it would be much less useful than people seem to think.

[blank check]

2420612

So, not evaluative, but purely decision-based? Only future-facing?

[deaincaelo]

2420626

That's my intention, as evaluative and decision based formulae are not necessarily interchangeable. For example, keeping population as a variable in evaluative math can make some sense, but leads to very strange conclusions when forward facing.

That being said, I haven't found a problem with using this methodology evaluatively. If you see one, please point it out. That will help define the limitations and flaws that may be here.