Agent, Agent Capability, and agent_in #637
Replies: 9 comments 22 replies
-
|
I definitely agree @CarterBeauBenson , and I want to tag @avsculley here. I think we should think through the full design pattern here in order to clarify how this term fits into the picture. Here are some initial thoughts and one potential start: Thinking about what you've written Carter, we seem to now have two ways of thinking about agency in CCO: a first that is weaker and a second that is stronger.
These two ways may not be opposed but complimentary. Agent Capability seems to be about those dispositions that allow their bearers to form and 'hold' Directive ICEs. Subclasses of this might be 'Planning Capability' or 'Task Allocation Capability' or even 'Tool Use Capability' (I have been working on an experimental ontology of such capabilities here.). In addition, it may be useful to have the class 'Agency' to capture (1) (since Agent Capability is tied to (2) already). Agency might be a Realizable Entity that is realized in processes where its bearer is an agent in that process, and then agent in remains as is. To align with this view, we could modify the definition of Act such that it becomes A Process that has_agent some Material Entity, and a Planned Act becomes An Act that realizes some Agent Capability. We would need to then better define 'Agent Capability' in terms of how Material Entities hold Directive ICEs. Thoughts? |
Beta Was this translation helpful? Give feedback.
-
|
Agent is a Material Entity that is agent in some planned process that realizes some skill or ability? This might make the bar for agency too high. Either AI is not agential or we anthropomorphize AI. First-degree murder: First-degree murder seems to be a planned act. Pinging @johnbeve, @mark-jensen, @neilotte, @APCox |
Beta Was this translation helpful? Give feedback.
-
|
@CarterBeauBenson @cameronmore I've done some work on the difference between planned acts, intentional acts, and intended acts. They are in fact distinct kinds. You can see this starkly in the AI case. AI can perform planned acts and intended acts, but not intentional acts. I'd have to dig up the draft paper to give you more detail than the following but here is what I recall off the top of my head.
I also toyed around with what I called "Plantentional Acts" to capture the plan counterpart to intentional act (and also Planned Plantentional acts). These seem to be relevant in cases of AI agents. If we start with making these distinctions, I think a model of agency will be a lot easier to construct (still not easy). |
Beta Was this translation helpful? Give feedback.
-
|
Is "intended" a little modal and what is intention here? |
Beta Was this translation helpful? Give feedback.
-
|
I cheated a little bit in counting intentions as realizables. While this is a view, it is not my view. But let's suppose its true. An intention could be something like a mental disposition that concretizes a directive ICE. My view is that intentions are conative representations (mental qualities), but this would require an account of the grounding of dispositions in conative representations to get from the representation to the action. I was just trying to illustrate the difference between the kinds of actions as simply as I could think of. I don't know if "intended" is a little modal.. if you mean to ask whether you can intend to do something that you do not do, then yes. |
Beta Was this translation helpful? Give feedback.
-
|
Or if you mean to ask whether there can be intended actions that are not performed, I'm less confident. In one sense, yes; in another sense, no. |
Beta Was this translation helpful? Give feedback.
-
|
The way I was thinking about it would require intended acts to be performed, but I'm not married to the view. |
Beta Was this translation helpful? Give feedback.
-
|
@CarterBeauBenson, are those definitions from a legal source? I find the definition of Second-degree murder puzzling. If you kill someone, premeditated or not, there certainly seems to be intent to cause serious bodily harm. As for planning, I suppose the pertinent question is: how far in advance do you have to plan? I'm not talking about the legal definition, but rather about whether it's useful, from a modeling standpoint, to create a Directive Information Content Entity that prescribes the Act, thereby making it a Planned Act. Once upon a time, a Information Content Entity was supposed to generically depend on an Information Bearing Entity. I submit that, for many acts of First-degree murder, the murderer is unlikely to express their thoughts in any way that could be considered an Independent Continuant. |
Beta Was this translation helpful? Give feedback.
-
|
These are inspired from https://en.wikipedia.org/wiki/Murder_in_United_States_law. There are different definitions from each state, some of which do not actually acknowledge murderous distinctions. I think it still points to something. There is a difference between premeditated actions and intentional actions done in the moment. Brain (from Pinkie and the Brain) has a plan in each episode to take over the world. They are a structured set of interrelated actions that the success of which add or subtract from the overall goal. There are other sorts of actions, like a swerve to miss an animal in the road, that are intentional, but without a plan. As for the last point, I think the IBE in question that the plans generically depend on is the murderer's brain. |
Beta Was this translation helpful? Give feedback.
-
|
I think plans require something extra. Sometimes I think of them as directive ICEs that prescribe intended actions. Here, "intended action" is defined as in my above comment: Intended act = An act prescribed by a directive ICE which is concretized by an intention. |
Beta Was this translation helpful? Give feedback.
-
|
@CarterBeauBenson that may make 'Agent' a high bar. If we go along with a stronger and weaker sense of agency, we might have an Agent simply be an agent_in some pracess, but a subclass of Agent might be an Intentional Agent that is an agent_in a Planned (or as we decide, Intentional) process? |
Beta Was this translation helpful? Give feedback.
-
|
@CarterBeauBenson I'm going to let others better versed in BFO foundations than I am decide whether a brain can be an IBE. I will admit that, if a brain is an IBE, it affirmatively answers a question I once posed: is there an IBE that's not an IBA? |
Beta Was this translation helpful? Give feedback.
-
|
Interesting. I had always thought of BFO as expressing a scientist's perspective: we model that which is observable. As far as we know, thoughts are not observable. Regarding a name assigned to a fetus, parents speak it to each other, and that speech is the IBE. But if others tell me this is not how BFO is to be used, I will change my beliefs. (No comments about things like string theory, please. Whether string theory is truly science is a whole 'nother matter.) |
Beta Was this translation helpful? Give feedback.
-
|
Speech seems to me to be a process of some sort that transmits information. Perhaps the air that is manipulated by the waves is the bearer in speech. |
Beta Was this translation helpful? Give feedback.
-
A brain can be an IBE. Consider: I read a poem. I walk into another room. I write down a second copy of a poem. If the second copy of the poem is an IBE bearing a second concretization of the poem ICE, how did it come into existence. The answer is that another IBE that concretized the poem walked between rooms. From what we know, the brain is where memories are kept. IBEs that are not IBAs: The definition of IBA is "A Material Artifact that carries an Information Content Entity and is designed to do so using a particular format or structure." |
Beta Was this translation helpful? Give feedback.
-
Speech and other information transmissions are the rationale for adding to BFO process as a possible concretization of a GDC. |
Beta Was this translation helpful? Give feedback.
-
So far. But more and more it seems technology has been able to tease out thoughts, stuff like neuralink. In any case, direct observation isn't a requirement for science. You can't observe a quark on its own, only the products of quarks interacting. But scientists think quarks exist.
Speech is a process, so can't be an IBE. At best the varying pressure in sound waves are the carrier and the air is the IBE. However, in my view, processes can concretize GDCs, so that's how I would represent this situation. |
Beta Was this translation helpful? Give feedback.
-
|
Seems to me intended acts can not be realized. I intend to go to the grocery store but have an accident on the way, for example. |
Beta Was this translation helpful? Give feedback.
-
|
I used the word 'realized' in a non-BFO sense, which I shouldn't have done. I should have used the word 'performed.' That said, I think that intended acts can be performed, in the case where you do make it to the grocery store. It would be both an intended and intentional act. It is intended because it is prescribed by the DICE concretized by your intention, and it is intentional because it realizes your intention. Now, whether merely intended actions can be performed, is a different question. I think it is possible in two cases. First, when I intend for a non-human agent to take some action. Second, when I intend to take some action, but then accidentally take the action. But these are edge cases. I'm not sure they're worth spending too much time on until there is a reason to do so. |
Beta Was this translation helpful? Give feedback.
-
|
I was treating "performed" as synonymous with "agent_in" since if you perform the grocery shopping act, you are causally responsible for the grocery shopping act, though as Alan has pointed out, if the groceries were not in the store, then you could not grocery shop. |
Beta Was this translation helpful? Give feedback.
-
|
These are great questions, Carter--thanks for raising them. I regret that I didn't see them when they were first posted. A few thoughts occasioned by some of your questions: Re 1: What of value in the present definition would be lost if the definition were replaced by something like this? (Def) A Realizable Entity that is realized by a Planned Act performed by this Realizable Entity's bearer. I think there's no even apparent circularity in (Def). Admittedly "performed" is left unanalyzed in (Def). Definitions have to bottom out somewhere, though, and the present Agent Capability definition is clearly intended to imply, though as you point out it doesn't seem to actually imply, that a Material Entity must perform a Planned Act in order to count as realizing an Agent Capability that inheres in it. Re 5: Insisting explicitly, as in (Def), that the Planned Acts that realize an Agent Capability must be things performed by the Agent Capability's bearer would rule out cases like the one you mention. Re 6: In another discussion thread I noted out that nearly every Realizable Entity definition in CCO defines the entity in question EITHER as a RE that inheres in its bearer in virtue of xyz OR as a RE that is realized by processes of such-and-such sort. The current Agent Capability definition is sort of a hybrid of these two definition types: It defines an Agent Capability as one that inheres in its bearer in virtue of a certain fact, but this fact is about (among other things) what its realizations are. It's tempting to think that "in-virtue-of" picks out a "because" or sourcehood relation. But on reflection in these definitions that might not always be what's intended. The present definition, for example, is as follows: Agent Capability = A Realizable Entity that inheres in a Material Entity in virtue of that Material Entity's capacity to realize it in some Planned Act. If you read "in virtue of" as "because," then you get: Agent Capability = A Realizable Entity that inheres in a Material Entity because this Material Entity is capable of realizing this Realizable Entity in some Planned Act. But I doubt any Realizable Entity inheres in an entity because this entity can realize this RE. Compare: fragility1 inheres in glass1 because glass1 can realize fragility1. That sounds really bad to me. Not exactly circular--just getting the order of explanation wrong. If glass1 can realize fragility1 then this is surely partly because fragility1 inheres in glass1. Your suggestion that "in virtue of" is really meant to pick out something like the realization conditions for the RE in question is intriguing. I'd like to sample other RE definitions that employ the same language to see how often it might be used that way. |
Beta Was this translation helpful? Give feedback.
-
A full definition of a realizable entity has 3 aspects: what the bearer is, a process type, and a way that the bearer participates in the process, since the bearer of a realizable entity always participates in the realization. There is more structure to functions in that functions are the product of design or evolution. There's more structure to roles in that there is some "blessing/dubbing" that sanctions the existence of the role. No need to go to nonspecific terms like performs when a good definition of a realizable says explicitly what performing means, whereas performing says very little beyond participation. The only thing we haven't formalized are "ways of participating in processes". So we have to be descriptive. |
Beta Was this translation helpful? Give feedback.
-
|
"Re 5: Insisting explicitly, as in (Def), that the Planned Acts that realize an Agent Capability must be things performed by the Agent Capability's bearer would rule out cases like the one you mention." I do not see where this interpretation is coming in based on The definition of "agent capability" does not say anything about what makes it special, only that it is a realizable entity that can be realized in a planned act. It does not say the ME must be causally responsible or something like that in the planned act. |
Beta Was this translation helpful? Give feedback.
-
|
Causality seems to me to be impossible to define clearly. What participant in a process is not causally involved? No participant can be ignored. Without any of them there is not that process. The best we have in BFO, IMO, are realizable entities. A participant is causally involved because they are realizing a realizable entity. Agent in tells me (maybe) who to blame. But even that is dubious given the uncertainty about how far back in the causal chain to really determine that. Intention and planning seems like a much stronger basis for resting definitions on. Maybe intention less so. Intention seems like a kind of a attitude towards a plan concretized as a realizable entity. I think we have a pretty good intuitive sense of what planning is. e.g. there is a recipe as written down, and a recipe as remembered - so it can be copied from memory. But there's also potentially a manifestation of the recipe that is actually doing what is said. What we have that lets us do what the recipe is a realizable. It's something that is manifested in a correlated process. Me intending to make the recipe adds is a choice that I will try to realize the realizable entity. Maybe the realizable entity changes type to "strengthen" it. agent_in. I would have it go away in favor saying instead what realizable is realized by what. Work on classifying realizables. Maybe it will become apparent which when bearers of realizables are agents, but I have the sense that that judgement will be socially defined. I'm not sure cross-culture that judgements of agency will be the same. So better from the bottom up. |
Beta Was this translation helpful? Give feedback.
-
|
I think agent_in because I think of that as the above "performed". Participation is weak, everything participates. An instance of a play has participants that are observers, actors, chairs, the stage, props. I suppose that the actors and the observers are engaged in different acts that could be occurrent parts of the play, but I want to say that the actors are agents in the the play, while the audience is merely participating. Or put another way, Will Smith and Chris Rock were both participants in an slap act, but Will was the agent_in it. If there is some other analysis, one that picks out performance, I would be happy to adopt that. |
Beta Was this translation helpful? Give feedback.
-
|
Couldn't any realizable's realization be a cause of a process? Including dispositions that are wider than those that inhere in specifically intelligent agents? ie, my cco:Strength's realization is the cause of some Act of Motion? If we extend our notion of what may count as an Agent, all sorts of artifact functions may count as agent capabilities? Following this thread, instead of changing the hierarchy, 'agent capability' becomes a defined class of any realizable entity whose manifestation is_cause_of some process. Even given this different modeling pattern, I would still want a property that identifies the main material entity whose realizable is the cause. (Note, this has to be distinct from simply 'realized in' because an agent could be the main cause but also 'use' a tool whose function is also being realized). |
Beta Was this translation helpful? Give feedback.
-
You can do better than actors "performing" (in the generic sense) and audience "merely" participating. You can say that the actors are realizing an actor role in the same process as the audience is realizing an audience member role. They are on equal footing, just doing different things. That the actors are more important than the audience is strictly perspectival. For someone studying audience response, the audience's activities aren't "mere". |
Beta Was this translation helpful? Give feedback.
-
|
The distinction between performance and non-performance participation has nothing to do with "importance," and is certainly not a "strictly perspectival" matter. The difference between the way in which Will Smith participates in the slapping and the way in which Chris Rock participates in it is totally obvious and objective, even if it's hard to articulate the distinction without running in linguistic circles, e.g., by saying that "WS does the thing and CR has it done to him," or that "WS is the agent and CR is the patient," or that "WS is active with respect to it and CR is passive with respect to it," or... It should be conceded that these are all kinda-sorta just different ways of saying the same thing, and that a genuine analysis of any of this probably isn't in the offing. So what? All this shows, if anything, is that here might be where some well-selected primitives belong. |
Beta Was this translation helpful? Give feedback.
-
|
Does the play realize the acting role or does the act of acting? Assume its the act of acting I see that "realizes" uses "participates" in its definition, so we might get the second triple for free, but I am not seeing anything in the axioms that we could infer that from. Could there be an abbreviated pattern in which can be the shortcut of the above since it is sometimes unclear what roles or dispositions are being realized by a process, but it might still be useful to capture the performer. |
Beta Was this translation helpful? Give feedback.
-
|
You get to define what process a role is realized in, and roles can be realized in different kinds of processes. So an actor role could be realized both in a process that is the single actor doing their thing as well as in the play as a whole. Or you could have two roles, each of which realized in one of the processes. Realizables have several dimensions of expressivity and it isn't a good strategy, IMO, to collapse that all into "agent_in", and nudge-nudge-wink-wink we know what you mean. |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
-
If I remember right, it used to be that CCO:Agent was a defined class such that any Material Object that has the agent_in object property associated with it could also be counted as an agent. For example:
person123 agent_in plannedAct123
meant that person123 is an agent. To me, this meant that person123 had causative power for (from the agent_in property), what would often be, planned acts. Since the person had causative power in a planned act, it is likely that they could, or had the ability to, carry out planned acts. I admit that there was work needed here to specify that a rock could not be an agent in the breaking of a Tesla windshield, since in some sense, it causes the windshield to break. The problem is that the implication here was not explicit. The nice thing about this, however, was that we did not have to also assert person123 as an agent.
Now Agents are defined as "a Material Object that is the bearer of some Agent Capability." This means that in order to get the same sort of inference as the above, I need to specify that
person123 is_bearer_of agentCapability123
Although there is nothing wrong here, it does mean that one has to add an Agent Capability and it not get the free typing when they use "agent_in".
Additionally, "agent_in" does not have a restriction in the domain limiting the sorts of things that can be agents any more than the sorts of things that can participate. For example, Sites can participate, therefore Sites can be the agent_in an action. ICEs can participate so ICEs can be agent_in. This seems to violate the spirit of agency in CCO, but does not violate any laws, since there, to my eyes, is no semantic connection between Agent and agent_in.
Below is the definition of Agent Capability:
Agent Capability = A Realizable Entity that inheres in a Material Entity in virtue of that Material Entity's capacity to realize it in some Planned Act.
Several things should be noted about the quality of this definition.
Agent Capability = A Realizable Entity that inheres in a Material Entity in virtue of that Material Entity's capacity to realize an Agent Capability in some Planned Act.
On my view, it should be implied that a MO has an agent capability if it has realized that capability through the proper use of the object property. Basically, I think causal power in a planned act that has occurred should be at the forefront and the potential to have causal power be in the background. Why is this piece of software an agent? Because it performed an intentional act. Why is this person an agent? Because it performed an intentional act. Why is the Tesla rock not an agent, because it did not perform an intentional act. Then we infer that the person and software have the capability (or some other thought out class) and that the rock does not.
I would like to see three things if possible.
Beta Was this translation helpful? Give feedback.
All reactions