Re: [FRIAM] What is an agent [was: Philosophy and Science}
Thank you Dave, Yes, one of the fourteeners I should climb, and before I get too old to do it. Eric > On Jul 16, 2023, at 3:51 AM, Prof David West wrote: > > If you have not read it — I highly recommend The Tree of Knowledge by > Humberto Maturana and Francisco Varela. Self organization from simple to > complex via a single mechanism. > > On Fri, Jul 14, 2023, at 7:30 PM, David Eric Smith wrote: >> I have had a version of this problem for several years, because I want to >> start with small-molecule chemistry on early planets, and eventually talk >> about biospheres full of evolving actors. I have wanted to have a rough >> category system for how many qualitative kinds of transitions I should need >> to account for, and to explain within ordinary materials by the action of >> random processes. Just because I am not a(n analytical) philosopher, I have >> no ambition to shoehorn the universe into a system or suppose that my >> categories subsume all questions even I might someday care about, or that >> they are sure to have unambiguous boundaries. I just want a kind of sketch >> that seems like it will carry some weight. For now. >> >> Autonomy: One early division to me would be between matter that responds >> “passively” to its environment moment-by-moment, and as a result takes on an >> internal state that is an effectively given function of the surroundings at >> the time, versus one that has some protection for some internal variables >> from the constant outside harassment, and a source of autonomous dynamics >> for those internal variables. One could bring in words like “energy”, but I >> would rather not for a variety of reasons. Often, though, when others do, I >> will understand why and be willing to go along with the choice. >> >> Control: The category of things with autonomous internal degrees of freedom >> that have some immunity from the slings and arrows of the immediate >> surroundings is extremely broad. Within it there could be very many >> different kinds of organizations that, if we lack a better word, we might >> call “architectures”. One family of architectures that I recognize is that >> of control systems. Major components include whatever is controlled (in >> chem-eng used to be called “the plant”), a “model” in the sense of Conant >> and Ashby, “sensors” to respond to the plant and signal the model, and >> “effectors” to get an output from the model and somehow influence the plant. >> One could ask when the organization of some material system is well >> described by this control-loop architecture. I think the control-loop >> architecture entails some degree of autonomy, else the whole system is >> adequately described by passive response to the environment. But probably a >> sophist could find counterexamples. >> >> One could ask whether having the control-loop architecture counts as having >> agency. By discriminating among states of the world according to their >> relation to states indexed in the model, and then acting on the world (even >> by so little as acting on one’s own position in the world), one could be >> said to express some sort of “goal”, and in that sense to have “had” such a >> goal. >> >> Is that enough for agency? Maybe. Or maybe not. >> >> Reflection: The controller’s model could, in the previous level, be >> anything. So again very broad. Presumably a subset of control systems have >> models that incorporate some notion of a a “self”, so they could not only >> specifically model the conditions of the world, but also the condition of >> the self and of the self relative to the world, and then all of these >> variables become eligible targets for control actions. >> >> Conterfactuals and simulation: autonomy need not be limited to the receiving >> of signals and responding to them with control commands. It could include >> producing values for counterfactual states within the controller’s model, of >> playing out representations of the consequences of control signals (another >> level of reflection, this time on the dynamics of the command loop), and >> then choosing according to a meta-criterion. Here I have in mind something >> like the simulation that goes on in the tactical look-ahead in combinatorial >> games. We now have a couple levels of representation between wherever the >> criteria are hard-coded and wherever the control signal (the “choice”) acts. >> They are all still control loops, but it seems likely that control loops >> can have different enough major categories of design that there is a place >> for names for such intermediate layers of abstraction to distinguish some >> kinds as having them, from others that don’t. >> >> How much internal reflective representation does one want to require to >> satisfy one or another concept of agency? None of them, in particular? A >> particular subset? >> >> For different purposes I can see arguing for
Re: [FRIAM] What is an agent [was: Philosophy and Science}
Eric, I hadn't seen your mail until David quoted it. What you say reminds me of a project I worked on for a couple of years in the Robotics Institute at Carnegie. Under the global title of Factory of the Future I coordinated a project to automate and optimize a fluorescent lamp factory. There were a sequence of processes that a lamp (bulb) went through from sand to melted glass to cutting into five foot tubes. Then white "paint" flowed through the tube after which it was baked. To make a long story short, electrodes were added to the ends, the tube was cured and tested by running a current through it. Our approach was to saturate the sequence of machines with sensors including visual, chemical, viscosity, electrical, etc. One goal was to reduce "shrinkage" or rejection of bulbs. The existing rate was about 10% as I recall. It was known that there were interactions of the elements of the sequence of processes. Therefore we had hopes that we could have the processors, actuators, and sensors take advantage of an understanding of those interactions. The Dutch firm Phillips bought all of Westinghouse's lamp manufacturing operations and they cancelled our work during the very early design stage. Disclaimers: my memory is less than perfect about events from the early eighties. Frank --- Frank C. Wimberly 140 Calle Ojo Feliz, Santa Fe, NM 87505 505 670-9918 Santa Fe, NM On Sat, Jul 15, 2023, 12:52 PM Prof David West wrote: > If you have not read it — I highly recommend The Tree of Knowledge by > Humberto Maturana and Francisco Varela. Self organization from simple to > complex via a single mechanism. > > On Fri, Jul 14, 2023, at 7:30 PM, David Eric Smith wrote: > > I have had a version of this problem for several years, because I want to > start with small-molecule chemistry on early planets, and eventually talk > about biospheres full of evolving actors. I have wanted to have a rough > category system for how many qualitative kinds of transitions I should need > to account for, and to explain within ordinary materials by the action of > random processes. Just because I am not a(n analytical) philosopher, I > have no ambition to shoehorn the universe into a system or suppose that my > categories subsume all questions even I might someday care about, or that > they are sure to have unambiguous boundaries. I just want a kind of sketch > that seems like it will carry some weight. For now. > > Autonomy: One early division to me would be between matter that responds > “passively” to its environment moment-by-moment, and as a result takes on > an internal state that is an effectively given function of the surroundings > at the time, versus one that has some protection for some internal > variables from the constant outside harassment, and a source of autonomous > dynamics for those internal variables. One could bring in words like > “energy”, but I would rather not for a variety of reasons. Often, though, > when others do, I will understand why and be willing to go along with the > choice. > > Control: The category of things with autonomous internal degrees of > freedom that have some immunity from the slings and arrows of the immediate > surroundings is extremely broad. Within it there could be very many > different kinds of organizations that, if we lack a better word, we might > call “architectures”. One family of architectures that I recognize is that > of control systems. Major components include whatever is controlled (in > chem-eng used to be called “the plant”), a “model” in the sense of Conant > and Ashby, “sensors” to respond to the plant and signal the model, and > “effectors” to get an output from the model and somehow influence the > plant. One could ask when the organization of some material system is well > described by this control-loop architecture. I think the control-loop > architecture entails some degree of autonomy, else the whole system is > adequately described by passive response to the environment. But probably > a sophist could find counterexamples. > > One could ask whether having the control-loop architecture counts as > having agency. By discriminating among states of the world according to > their relation to states indexed in the model, and then acting on the world > (even by so little as acting on one’s own position in the world), one could > be said to express some sort of “goal”, and in that sense to have “had” > such a goal. > > Is that enough for agency? Maybe. Or maybe not. > > Reflection: The controller’s model could, in the previous level, be > anything. So again very broad. Presumably a subset of control systems > have models that incorporate some notion of a a “self”, so they could not > only specifically model the conditions of the world, but also the condition > of the self and of the self relative to the world, and then all of these > variables become eligible targets for control actions. > > Conterfactuals and simulation: autonomy need
Re: [FRIAM] What is an agent [was: Philosophy and Science}
If you have not read it — I highly recommend The Tree of Knowledge by Humberto Maturana and Francisco Varela. Self organization from simple to complex via a single mechanism. On Fri, Jul 14, 2023, at 7:30 PM, David Eric Smith wrote: > I have had a version of this problem for several years, because I want to > start with small-molecule chemistry on early planets, and eventually talk > about biospheres full of evolving actors. I have wanted to have a rough > category system for how many qualitative kinds of transitions I should need > to account for, and to explain within ordinary materials by the action of > random processes. Just because I am not a(n analytical) philosopher, I have > no ambition to shoehorn the universe into a system or suppose that my > categories subsume all questions even I might someday care about, or that > they are sure to have unambiguous boundaries. I just want a kind of sketch > that seems like it will carry some weight. For now. > > Autonomy: One early division to me would be between matter that responds > “passively” to its environment moment-by-moment, and as a result takes on an > internal state that is an effectively given function of the surroundings at > the time, versus one that has some protection for some internal variables > from the constant outside harassment, and a source of autonomous dynamics for > those internal variables. One could bring in words like “energy”, but I > would rather not for a variety of reasons. Often, though, when others do, I > will understand why and be willing to go along with the choice. > > Control: The category of things with autonomous internal degrees of freedom > that have some immunity from the slings and arrows of the immediate > surroundings is extremely broad. Within it there could be very many > different kinds of organizations that, if we lack a better word, we might > call “architectures”. One family of architectures that I recognize is that > of control systems. Major components include whatever is controlled (in > chem-eng used to be called “the plant”), a “model” in the sense of Conant and > Ashby, “sensors” to respond to the plant and signal the model, and > “effectors” to get an output from the model and somehow influence the plant. > One could ask when the organization of some material system is well described > by this control-loop architecture. I think the control-loop architecture > entails some degree of autonomy, else the whole system is adequately > described by passive response to the environment. But probably a sophist > could find counterexamples. > > One could ask whether having the control-loop architecture counts as having > agency. By discriminating among states of the world according to their > relation to states indexed in the model, and then acting on the world (even > by so little as acting on one’s own position in the world), one could be said > to express some sort of “goal”, and in that sense to have “had” such a goal. > > Is that enough for agency? Maybe. Or maybe not. > > Reflection: The controller’s model could, in the previous level, be anything. > So again very broad. Presumably a subset of control systems have models > that incorporate some notion of a a “self”, so they could not only > specifically model the conditions of the world, but also the condition of the > self and of the self relative to the world, and then all of these variables > become eligible targets for control actions. > > Conterfactuals and simulation: autonomy need not be limited to the receiving > of signals and responding to them with control commands. It could include > producing values for counterfactual states within the controller’s model, of > playing out representations of the consequences of control signals (another > level of reflection, this time on the dynamics of the command loop), and then > choosing according to a meta-criterion. Here I have in mind something like > the simulation that goes on in the tactical look-ahead in combinatorial > games. We now have a couple levels of representation between wherever the > criteria are hard-coded and wherever the control signal (the “choice”) acts. > They are all still control loops, but it seems likely that control loops can > have different enough major categories of design that there is a place for > names for such intermediate layers of abstraction to distinguish some kinds > as having them, from others that don’t. > > How much internal reflective representation does one want to require to > satisfy one or another concept of agency? None of them, in particular? A > particular subset? > > For different purposes I can see arguing for different answers, and I am not > sure how many categories it will be broadly useful to recognize. > > Eric > > >> On Jul 15, 2023, at 8:28 AM, Russ Abbott wrote: >> >> I'm not sure what "closure to efficient cause" means. I considered using as >>
