@Chelonian I am not a compatibilist nor is my view of free will rooted in religious beliefs. Basically I reject any view of God that does not allow agents to have free will. There is no determinism and so there is no point in thinking that free will is compatible with determinism. There are constraints coming from actions of other free agents. These constraints are real.

But those constraints do not imply determinism. They are a kind of karma left over by the actions of free agents.

@FrankHubeny Thanks for that. So, do you think voluntary human actions (e.g. choosing which flavor of ice cream to order in an ice cream shop) are due only to the workings of the brain, or do you think they actually originate in a non-physical "something", like a soul? Or a combination of both?

@shp999 You're really grasping at straws here... you would do better asking me questions than critiquing something you haven't the context to critique

Artificial intelligence using NN's is not what I'm talking about, though NN's may or may not be used here.

What are "intentions"?

Intentions come to play in goal-oriented behaviors; this isn't merely "accomplishing something", it's a property of entity that abstractly carves its environment up in a particular way

In particular, the entity (such as us) is capable of analyzing the environment in real time, and directing its behavior to attain some goal that it has. What's key here is that in complex environments, unanticipated things pop up... an entity that exhibits goal based behaviors will, in real time, use this to adjust its behaviors in order to attain that goal

We can use a NN trained to play chess, for example. Chess playing is quite advanced; we can certainly build chess engines capable of beating most players

Such chess playing programs, though, do not qualify; they are in extremely simplified environments. The only "decisions" made by these engines is to make a move, and there's not much to do here when they make a move

So they "accomplish something", and they are very sophisticated, but they are not what I'm talking about

A SDC, by contrast, needs to navigate a complex environment... it's not merely "making a move" in a simplified fashion... it has to constantly monitor its environment, using a particular high level model of some goal state it needs to accomplish to adjust its behavior

The chess playing program lacks intention; it doesn't need it. The model of the goal used to direct behavior which an SDC requires is intention; it's something that the SDC needs.

Whether or not each involves use of an NN isn't relevant

What's relevant is whether or not we have goal-oriented behavior

@shp999 I hold no opinion on what "free will" means, so it's a bit meaningless to tell me that this isn't the kind of intentions we apply to "free will"; however, people who believe in free will almost always think we have it

In terms of that thing we have, "intention" plays a role in our sense of control; given a simple physical act, like walking to the kitchen, the intention guides our behavior and a part of our mind constantly monitors its environment and uses it to adjust the behaviors accordingly to achieve the goal

This breaks down in feed forward models in the predictions of each broken down step and feed backward models in the monitoring of the effects and comparison to those predictions

If those predictions vary slightly, our behaviors adjust

Insofar as the predictions match the outcome, we have a sense of control. If the prediction diverges greatly, it can metaphorically "set off an alarm", and we sense it as lack of control

""Intentions" of any deterministic device are defined by their creators (and/or users)." This is extremely questionable. I may tell my SDC where I want to go, but I'm not going to tell it when or whether to change lanes, reroute around that moose that walked into the road, etc

Neither is this "programmed"

Again, if the SDC hits a person and breaks his leg, you can't charge the programmer with battery

I suppose I can chock this up to your not knowing what I mean by intention, but the kinds of things the programmer programmed something like a SDC to do are not intentions

...at least in the same way

@shp999 No, blame requires much more than this

You can charge both the programmer and the company; you just can't charge them with battery

For blame you need an entity that makes decisions, and can make them in response to a moral context. The morality here we're discussing is usually human-centric (as evidenced by the special place harm against humans takes in discussions)

That's more complex, but it doesn't critically hinge on whether or not the entity's mechanics involve original causes

In fact, such an appeal to mechanics is entirely irrelevant

It's put on a pedestal but for no logical reason. We care about real concerns here for blame, such as whether or not the entity "knew what it was doing", "knew right from wrong", etc.

Sure

@shp999 You're attaching this way too much to NN's

There might be a practical matter whereby NN's are the only way to program such entities, or we might discover some other cleverness in the future. Even within NN's there have been some pretty major leaps (still NN's but different in a game-changing way) within the relatively recent past

Well if I drive a car, I'm responsible for what it does; just bare responsible

If the car makes a right turn it's because I intended it (hopefully)

(We can bar corner cases, like skidding/etc)

Those types of things I can relinquish, if I had a SDC

It doesn't make sense to say that the SDC is responsible in this way, however, unless the SDC also had goal oriented behavior

But given it does, it can be assigned bare responsibility; it's the SDC that is responsible for that right turn (again, hopefully)

But just that type of responsibility is all you need to bootstrap the mechanics to assign moral responsibility... outside of these additional details you mentioned

Namely, there's an entity that makes meaningful decisions to attain goals; whatever goes into those decision processes can be assigned responsibility for the outcomes related to that goal

Put another way, what I'm claiming is that given we are agents in this sense, even if our mechanics were deterministic, nothing about us changes with respect to our moral responsibility, because this particular aspect of our mechanics is a red herring

We still make decisions, we still value things, we still have to live a life and interact with others. Those decisions still have outcomes. The only thing that changes is some particular class of philosophers using the term "free will" won't apply it to us.

I have to take care of the cat, because I've assumed control over it and it now depends on me. I've taken on that responsibility. This isn't quite "fact", it's rather an extremely pragmatic social construct... were I to take in cats and not feed them, something must be done about it.

We don't want cats to die anyway; but when the cat's life and death (and humane treatment/etc) critically relies on the decisions that an agent capable of involving in these pragmatic concerns is involved, we want to make sure this agent's aware, in a social context, that we are implicitly consigning him with a duty. He should feed the cat.