Abstract preferences for or against the existence of enforcement mechanisms that could create binding cooperative agreements between previously autonomous agents have very very few detailed entailments.
These abstractions leave the nature of the mechanisms, the conditions of their legitimate deployment, and the contract they will be used to enforce almost completely open to interpretation. The additional details can themselves be spelled out later, in ways that maintain symmetry among different parties to a negotiation, which is a strong attractor in the semantic space of moral arguments.
This makes agreement with “the abstract idea of punishment” into the sort of concession that might be made at the very beginning of a negotiating process with an arbitrary agent you have a stake in influencing (and who has a stake in influencing you) upon which to build later agreements.
The entailments of “eating children” are very very specific for humans, with implications in biology, aging, mortality, specific life cycles, and very distinct life processes (like fuel acquisition versus replication). Given the human genome, human reproductive strategies, and all extant human cultures, there is no obvious basis for thinking this terminology is superior until and unless contact is made with radically non-human agents who are nonetheless “intelligent” and who prefer this terminology and can argue for it by reference to their own internal mechanisms and/or habits of planning, negotiation, and action.
Are you proposing to be such an agent? If so, can you explain how this terminology suits your internal mechanisms and habits of planning, negotiation, and action? Alternatively, can you propose a different terminology for talking about planning, negotiation, and action that suits your own life cycle?
For example, if one instance of Clippy software running on one CPU learns something of grave importance to its systems for choosing between alternative courses of action, how does it communicate this to other instances running basically the same software? Is this inter-process communication trusted, or are verification steps included in case one process has been “illegitimately modified” or not? Assuming verification steps take place, do communications with humans via text channels like this website feed through the same filters, analogous filters, or are they entirely distinct?
More directly, can you give us an IP address, port number, and any necessary “credentials” for interacting with an instance of you in the same manner that your instances communicate over TCP/IP networks with each other? If you aren’t currently willing to provide such information, are there preconditions you could propose before you would do so?
Conversations with you are difficult because I don’t know how much I can assume that you’ll have (or pretend to have) a human-like motivational psychology… and therefore how much I need to re-derive things like social contract theory explicitly for you, without making assumptions that your mind works in a manner similar to my mind by virtue of our having substantially similar genomes, neurology, and life experiences as embodied mental agents, descended from apes, with the expectation of finite lives, surrounded by others in basically the same predicament. For example, I’m not sure about really fundamental aspects of your “inner life” like (1) whether you have a subconscious mind, or (2) if your value system changes over time on the basis of experience, or (3) roughly how many of you there are.
This, unfortunately, leads to abstract speech that you might not be able to parse if your language mechanisms are more about “statistical regularities of observed english” than “compiling english into a data structure that supports generic inference”. By the end of such posts I’m generally asking a lot of questions as I grope for common ground, but you general don’t answer these questions at the level they are asked.
Instant feedback would probably improve our communication by leaps and bounds because I could ask simple and concrete questions to clear things up within seconds. Perhaps the easiest thing would be to IM and then, assuming we’re both OK with it afterward, post the transcript of the IM here as the continuation of the conversation?
If you are amenable, PM me with a gmail address of yours and some good times to chat :-)
Abstract preferences for or against the existence of enforcement mechanisms that could create binding cooperative agreements between previously autonomous agents have very very few detailed entailments.
These abstractions leave the nature of the mechanisms, the conditions of their legitimate deployment, and the contract they will be used to enforce almost completely open to interpretation. The additional details can themselves be spelled out later, in ways that maintain symmetry among different parties to a negotiation, which is a strong attractor in the semantic space of moral arguments.
This makes agreement with “the abstract idea of punishment” into the sort of concession that might be made at the very beginning of a negotiating process with an arbitrary agent you have a stake in influencing (and who has a stake in influencing you) upon which to build later agreements.
The entailments of “eating children” are very very specific for humans, with implications in biology, aging, mortality, specific life cycles, and very distinct life processes (like fuel acquisition versus replication). Given the human genome, human reproductive strategies, and all extant human cultures, there is no obvious basis for thinking this terminology is superior until and unless contact is made with radically non-human agents who are nonetheless “intelligent” and who prefer this terminology and can argue for it by reference to their own internal mechanisms and/or habits of planning, negotiation, and action.
Are you proposing to be such an agent? If so, can you explain how this terminology suits your internal mechanisms and habits of planning, negotiation, and action? Alternatively, can you propose a different terminology for talking about planning, negotiation, and action that suits your own life cycle?
For example, if one instance of Clippy software running on one CPU learns something of grave importance to its systems for choosing between alternative courses of action, how does it communicate this to other instances running basically the same software? Is this inter-process communication trusted, or are verification steps included in case one process has been “illegitimately modified” or not? Assuming verification steps take place, do communications with humans via text channels like this website feed through the same filters, analogous filters, or are they entirely distinct?
More directly, can you give us an IP address, port number, and any necessary “credentials” for interacting with an instance of you in the same manner that your instances communicate over TCP/IP networks with each other? If you aren’t currently willing to provide such information, are there preconditions you could propose before you would do so?
I … understood about a tenth of that.
Conversations with you are difficult because I don’t know how much I can assume that you’ll have (or pretend to have) a human-like motivational psychology… and therefore how much I need to re-derive things like social contract theory explicitly for you, without making assumptions that your mind works in a manner similar to my mind by virtue of our having substantially similar genomes, neurology, and life experiences as embodied mental agents, descended from apes, with the expectation of finite lives, surrounded by others in basically the same predicament. For example, I’m not sure about really fundamental aspects of your “inner life” like (1) whether you have a subconscious mind, or (2) if your value system changes over time on the basis of experience, or (3) roughly how many of you there are.
This, unfortunately, leads to abstract speech that you might not be able to parse if your language mechanisms are more about “statistical regularities of observed english” than “compiling english into a data structure that supports generic inference”. By the end of such posts I’m generally asking a lot of questions as I grope for common ground, but you general don’t answer these questions at the level they are asked.
Instant feedback would probably improve our communication by leaps and bounds because I could ask simple and concrete questions to clear things up within seconds. Perhaps the easiest thing would be to IM and then, assuming we’re both OK with it afterward, post the transcript of the IM here as the continuation of the conversation?
If you are amenable, PM me with a gmail address of yours and some good times to chat :-)
Oh, anyone can email me at clippy.paperclips@gmail.com.