You challenge me (within the norms of politeness too, another ethical frame) in the name of inferential norms, calling upon me to defend my claim. Indeed, in making that came, I have indeed committed myself to its defense. If I can't defend a strong challenge, it's my duty to withdraw or modify the claim. — plaque flag

Ultimately, epistemic agreements and disagreements rest upon assumptions as to what speakers means by their words:

Doesn't it strike you as odd, the assumption that a person can believe in something impossible? For what is said to be impossible is also said to not exist, and so cannot be said to be the cause of the person's belief. So how can a belief even refer to something that is impossible?

And what then of falsified beliefs? Aren't they also a problematic concept for similar reasons, for weren't one's previously held beliefs, that one presently judges to be "falsified", also caused by something that fully explains their previous existence?


Isn't a physicalist, who is committed to a causal understanding of cognition, forced in the name of objective science to always side with the epistemic opinions of the speaker, no matter how wrong, mad or contradictory the speaker might sound? For shouldn't the physicalist always interpret a speaker's utterances in the same manner that he interprets as a sneeze that is understood to refer to nothing more than it's immediate causes?

(When interpreted with empathy, do Flat-Earther's really exist?)

One supports this approach phenomenologically, which is to say by simply bringing us to awareness of what we have been doing all along. Scan this forum. See us hold one another responsible for keeping our stories straight. See which inferences are tolerated, which rejected. Bots can learn this stuff from examples, just as children do. — plaque flag

From the perspective of an engineer who has a causal understanding of AI technology and a responsibility to fix it, the gibberish spoken by an "untrained" or buggy chatbot is meaningful in a way that it isn't for a naive user of technology who is intending to play a different language-game with it. And obviously, any agent of finite capacity can only learn to play well at one language game at the expense of doing worse at the others.

I am also not allowed to contradict myself, for the self is the kind of thing (almost by definition) that ought not disagree with itself --- must strive toward coherence, to perform or be a unity. — plaque flag

When you find yourself disagreeing with the beliefs of your earlier self, are you really contradicting your earlier self? For didn't the facts of the matter change that you were responding to?

Consider a sequence of beliefs {b 1,b 2, b 3 ...} regarding the truth of the "liar sentence" L unfolded over time, where L = "this statement is false" :

b 1. Presently at time t = 1, L is believed to be true.
b 2. Presently at time t = 2, b 1 is understood to imply that L is false, and hence that b 1 is false.
b 3. Presently at time t = 3, b 2 is understood to imply that L is true, and hence that b 2 is false.
b 4. Presently at time t = 4, b 3 is understood to imply that L is false, and hence that b 3 is false.

In spite of the fact that each belief negates the previous one, each belief can nevertheless be considered to be "true" at the time of it's construction without entailing contradiction with any of the other beliefs, for none of the beliefs were simultaneously held to be true, and each belief refers to a different object, namely it's own temporal context.

Jj asdoin asdmoi valfm capicasdjknca p spdmcsd l sd p p m[ o [o,asdcvvdflmvdf. — plaque flag

Are you implying that a public language must be decipherable? What about encryption?

When the public cannot agree on a linguistic convention, as is so often demonstrated when debating the philosophy of language, where does the authority of meaning reside then?

Consider Humpty Dumpty's famous proposition

`When I use a word,' Humpty Dumpty said in rather a scornful tone, `it means just what I choose it to mean--neither more nor less.' — lewis carroll

If meaning is considered to be use, as Humpty is suggesting, then how can meaning-as-use be grounded in linguistic convention?

Also consider the fact that any explicitly defined linguistic convention can only be finitely specified, implying that there is always uncertainty as to the intended meaning of a convention. (which is another of Lewis Carroll's remarks that argues against the grounding of meaning in convention).

Video games occasionally have procedurally generated worlds that are generated dynamically on-the-fly in response to the player's actions. These games demonstrate that the unknown past and the future can be considered as being metaphysically identical for all empirical intents and purposes. This parsimonious viewpoint has the advantage of treating causality in a temporally symmetric fashion, with both forward and backward causation.

I'm not saying there isn't a beetle in your box. I'm just saying the concepts are public. You don't get to make up your own language and your own logic ---and that's what oxymoronic private concepts would entail or mean (inasmuch as I can parse the phrase at all.) — plaque flag

And why not?

(I'm an Epicurean-Spinozist — 180 Proof

Wasn't Spinoza an idealist in all but name? At the very least, isn't his metaphysics compatible with "being is perception"? I don't see how matter as divine immanence can be closer to ontological naturalism (which has no concept of immanence) than to idealism for which immanence is a tautological truth.

As for Epicurus, as an empirically minded philosopher, didn't he stress the epistemic primacy, if not ontological primacy of sense-data? I'm also not seeing any real points of disagreement between the ontological arguments of Berkleley and Epircurus, save for whatever brand of atomic materialism Epircurus might have subscribed to, (which at most would amount to a physics disagreement with Berkeley , as opposed to a metaphysics disagreement). If both are understood to be empiricists who rejected Platonism, then how is metaphysical disagreement between them possible?

I

The fact that they mean different things, with the first referring to the present and the second referring to the future. — Michael

Well yes, one might take it to be analytically true that a prediction is future-referring, meaning that skepticism regarding the future-contingency of a prediction is ruled out a priori as a matter of linguistic convention. But in that case, the future-contingency of the prediction cannot mean anything about the world in itself, and only speaks of present linguistic conventions, speaker attitudes and so on.

When we speak of "failed predictions", we interpret past utterances as referring to the present. Relative to this convention, it might be considered to be analytically true a posteriori that a past utterance is present referring. But to say that past utterances are present referring, is again not to speak of reality but of linguistic convention.

As a matter of interest, do you consider ChatGPT's responses as future-referring?

So in your view, what makes a proposition future-contingent?

Suppose that a person looks at the sky and says

A. "There are ominous dark-clouds in the sky"
B. "In five minutes it will rain"

What makes B a future-referring proposition, in contrast to A that is merely a present observation?

Isn't the difference accounted for by the attitude of the person towards each case? In which case, isn't it fair to suggest that B isn't objectively speaking, a future-contingent proposition?

dispute that requirement. I can talk about the future. — Michael

Well, certainly I can accept that the word "future" has sense to you, as it does to me, but one can dispute that the word has reference beyond the immediacy of one's present. For whatever thoughts one presently identifies as being of tomorrow, one will presumably no longer regard them as being of tomorrow once tomorrow arrives...

How so? I don’t need to experience something to talk about it. — Michael

I can only interpret you as referring to something, in relation to an understanding of what your referring consists of. If my experience is private in the sense meant by philosophers, then ordinary means of referring that appeal to a causal linkage between another speaker and my experiences are ruled out. In which case, how I am supposed to interpret a speaker as referring to my experiences?

If experience is truly private, than it is presumably impossible to even refer to someone else's experiences in the literal sense of "someone else".

In which case, whatever legitimate doubts you might have regarding the quality or existence of "someone else's" experiences, is it not the case that for those doubts to be intelligible to you, your doubts must at least be defined in terms of empirical criteria that you yourself could observe in principle?

In other words, doesn't the privacy of experience actually imply that the problem of other minds is solvable in principle, contrary to popular opinion?

Philosophical questions regarding consciousness concern, among a great number of other things, the semantics of neuroscience in relation to the first-person perspective, as opposed to the scientifically established facts of neuroscience whose perspectival meaning and significance is undefined.

Philosophical questions also call into question the semantics and significance of science as a whole, and so it isn't possible to draw philosophical conclusions about consciousness from a narrow consideration of neuroscientific discoveries, because all that will be argued are circular tautologies that resolve none of the semantics of concern.

Since GPT is a merely an auto-regressive model of corpora on which it is trained to perform "next word prediction", then it will certainly reproduce the cultural biases of texts written in a given language.

But how is that evidence for Sapir-Whorf hypotheses? - i.e. the hypothesis that language affects cognition?

GPT will learn to parse text differently for each language on the basis of auto-regression, thereby having conceptual representations (neural activations) that vary according to the language, but this is true regardless of whether or not the texts of different languages exert different cultural biases. E.g a text that is written in two different languages will nevertheless have very different internal representations in an LLM. Furthermore, to identify two internal representations as meaning the "same" thing is to relate those representations to an external source of information in accordance to some translation scheme, an d so the semantics of neural activations isn't "in the head" of either humans or machines.

I think that the machine learning science of LLMs is telling us that without additional assumptions and contextual information, Sapir-Whorf hypotheses are generally meaningless, in referring to any number of different and independent aspects of the culture-cognition-language feedback loop.

You are implicitly referring to the logical paradox concerning the identity of indiscernibles.

The potential existence of indiscernibles, e.g perfectly identical twins who lead perfectly identical lives, except for their space and location, seems to imply that materialism cannot be reconciled with the subjective concept of self-identification in a way that would be acceptable to a materialist.

The materialist has several options with regards to his understanding of self-identification, but none of the available choices seem to look attractive to him:

1) He could accept the identity of indiscernibles and identify his 'self' as being the possibly infinite collection of physically identical selves that exist at different spatio-temporal locations throughout the universe; each of the physical entities in this collection not only has qualitatively indiscernible experience from one another, but is in fact the same experience, i.e. there is one psychological 'token' of experience shared between many physical lives.

The materialist presumably finds this option barking mad, for not only does it seem a bizarre proposition, but it seems to undermine the whole point of materialism, namely to explain the self as a physical token that is part of the causal nexus, rather than as an abstract and non-causal type of physical entity that is related to the causal nexus via dualism of some sort. It also goes without saying, that this choice also would play havoc with his understanding of science in general.

2) He could accept the identity of indiscernibles as above, but argue that the material universe is such that there are no actual physical indiscernibles. But this seems physically speculative and amounts to neo-Kantianism that goes against the grain of materialist epistemology (should a materialist be deducing materialistic implications from introspective a priori arguments?) This option also completely fails to address the underlying metaphysical problem concerning whatif there were indiscernible selves?

3) Alternatively, he could reject the identity of indiscernibles, by arguing that the spatio-temporal position of an individual is an internal property of the individual's mental state. But in this case it seems that he must either

a) give up the claim that the experiences of his self are explainable by causal laws, for his spatio-temporal position per-se has no causal relevance with regards to his experiential content, or

b) he must reject the Cartesian assumption that selves exist, whereby it is trivially the case that spatio-temporality is an internal property of the self-as-universe. But in this case, what is left of materialism seems dangerously close to idealism and even solipsism, that the materialist presumably wishes to distinguish and distance his beliefs from.

Indirect Realism accepts they have private sensations, but as argued by Wittgenstein in his beetle in the box analogy, such private sensations drop out of consideration within a public social language as irrelevant. — RussellA

Agreed.

Wittgenstein's beetle in the box analogy justifies Indirect Realism. — RussellA

Don't you mean to say that Wittgenstein's beetle in the box analogy justifies talk of indirect realism in relation to the third-personal public concept of perception, but that this doesn't justify talk of indirect-realism in the case of one's own perception?

" 275. Look at the blue of the sky and say to yourself "How blue
the sky is!"—When you do it spontaneously—without philosophical
intentions—the idea never crosses your mind that this impression of
colour belongs only to you. And you have no hesitation in exclaiming
that to someone else. And if you point at anything as you say the
words you point at the sky. I am saying: you have not the feeling of
pointing-into-yourself, which often accompanies 'naming the sensa-
tion' when one is thinking about 'private language'. Nor do you think
that really you ought not to point to the colour with your hand, but
with your attention. "

In relation to your diagram above of the two people perceiving the same circle, consider the irrealist understanding of the beetle on the box:

Given that each individual only has access to his or her private colour, and uses his or her mother-tongue in a bespoke private fashion when referring to the "shared" circle, then what is the purpose of colouring in the shared circle?


According to Nelson Goodman (and quite possibly Wittgenstein), there doesn't exist a transcendental underlying fact with regards to the real colour of the shared circle. Following this line of thought further, one could even deny the very existence of a shared circle, as part of a strategy for defending direct-realism for all perceivers, without condemning any ensuing disagreements as amounting to contradiction.

As a practice, mathematics is a family-resemblance of language games, whose premises are usually in conflict, in part due to the fact that they do not generally share the same logical assumptions regarding the justifications of a mathematical proof. Disagreements among mathematicians and logicians have concerned whether mathematical truth is reducible to the constructive notion theorem proving , which are disagreements that parallel, and often replicate, the historical disagreements between realists and idealists.

Intuitionistic logic, which corresponds to the mainstream Cartesian conception of computation in terms of the Church Turing Thesis, or equivalently, the behaviour of the ideal solitary computer, might be said to be solipsistic, since this logic makes the following assumptions:

1) Truth and theorem-proving are equivalent.

2) "Computation" is describable in terms of the solitary activity of an ideal individual making up rules and following them consistently, without any interaction with his environment playing a role in the meaning of "computation", whereby theorem-proving is considered to be a tautological exercise.

Classical logic on the other hand, accepts 2), but drops assumption 1) by it's appeal to The Law of Excluded Middle, which allows it to infer theorems for which there is no step-wise constructive proof.

This is the reason why many mathematicians are platonists; By rejecting assumption 1 they reject the idea that mathematics is a product of their minds, and yet by accepting assumption 2 they are also forced into rejecting the idea that meaning of mathematics is contingent upon events in the real world. The only way they can resolve the resulting dilemma is by positing an external platonic world of mathematical meaning.

Since the seventies, Computer science has slowly begun to break free of the the cartesian conception of computation, by questioning, if not rejecting, assumption 2), mostly as a result of needing to invent new logics for dealing with the empirical contingencies of computer-program IO. See for instance, the reactive-programming paradigm that concerns the problem of how to write programs to consistently reason with input-streams of data that are potentially-infinite and outside the control of the program.

We are not little homunculi driving robots and looking at screens. We are members of a community who live embedded in a shared world within and with which we collectively interact. — Banno

If a 'person' is taken as referring to a brain, and only to a brain, then a person is by definition a homunculus , hence a forteriori, a subject of indirect realism. Community interaction among brains cannot change this conclusion.

Irrealism tries to sidestep the dispute by claiming in a relativistic and solipsistic fashion, that direct-realism only applies in the first-person, i.e that only the first-person isn't a "brain person".

People should stop conflating metaphysical, epistemological and psychological solipsism, as their premises and conclusions are very different.

For example, in Metaphysical Solipsism "other minds" do exist empirically, in the sense of experiences that the subject has direct access to, even if the literal notion of other minds is considered to be unintelligible. Therefore, according to Metaphysical solipsism, empirical skepticism regarding the existence of "other minds" is unintelligible; Thus metaphysical solipsism is in agreement with analytic behaviourism.

Only in epistemological solipsism is there skepticism regarding the existence of either "other minds" or other minds, which is due to a Cartesian worldview that interprets sense-data and thought as being mere representation, which can lead to doubt as to whether there is anything behind the representation.

Psychological solipsists embrace the doubt of the epistemological solipsist to deny the existence of other minds. But in contrast to the metaphysical solipsist, they do not consider their personal sense-data as having constitutional relevance with respect to their concept of "other minds"

If metaphysical solipsism is true, then it is tautologically true. This is why metaphysical solipsism (and realism) have no empirical content.

People have a psychological tendency to misconstrue metaphysical solipsism with Cartesianism (psychological solipsism), due to the fact their use of the personal pronoun comes with a lot of baggage.

Berkeley's subjective Idealism, which is often associated with metaphysical solipsism, distinguished Ideas (that is to say , experiences with empirical content) from Spirits (the notion of volitional agents that aren't perceivable, pace Hume and Melbranche). The postulate "being is perception" therefore concerned ideas only, leaving intact the common-or-garden "vulgar" meanings of causality and "who" did "what".

Ironically, one of the central motives of metaphysical solipsism is to refute epistemological scepticism that doubts the reality of the "external" world, by arguing that the "external" world isn't external and therefore not doubtable.

You might be suffering more from the effects of social isolation and rumination than from a philosophical belief.

Solipsism has no empirical consequences. To associate a particular type of experience with solipsism is to misunderstand it.

I presently use the predecessor for coding projects as a plugin to Visual Code Studio, for filling in "no brainer" project detail and for providing a very rough first approximation to subsequent work that it cannot automate, on pain of needing a degree of prompt engineering that is as complex as manually writing code.

Given that GPT only appears to understand code, since it isn't a software verification tool, debugger, theorem prover, or test harness, and isn't meaningfully integrated with such tools, i would presently say that it's main use case is in giving very rough explanations or recalling very rough solutions to very common software problems, but I predict that it won't reduce the skills or knowledge burden for many IT professionals, due to the fact that employers will simply respond to it's existence by demanding that more development time is spent on more complicated and novel features that an LLM by itself cannot handle.

The software industry presently regards an LLM as being an " AI pair programmer" that works in tandem with users to solve problems in an interactive, iterative and cooperative fashion. I think that use-case of LLMs is also demonstrated in your dialogues above, where there is an iterative cycle of prompting the LLM for a response and then testing its response by independent means, followed by prompt adaptation and so on. I think this understanding of LLMs is applicable across all subject domains.

This raises the same question for each subject : what constitutes 'best practice' with regards to the prompt engineering and test cycle when GPT4 is used as a pair-programmer for researching, disseminating or applying that subject?

I just read in the news that GPT4 will be given internet access and the ability to run code. So it is clear that that it's role is ultimately to serve as a portal for domain specific reasoning networks consisting of inference engines and query-able expert nodes.

One potential side-effect of this however is 'reflection' - unlike a conventional gateway, the portal will be able to access an online representation of itself by calling itself. This feedback mechanism could facilitate the model to be extended with the ability to simulate and evaluate it's potential responses before deciding how to reply to the user, which could potentially lead to unstable behaviour, such as when a user traps it in a liar paradox.

By definition, LLMs are only models of language and therefore shouldn't be confused with reasoners or domain experts. They should only be thought of as a natural-language filter that comprises only the first step of a reasoning pipeline, for routing prompt information to domain specific models and for caching the responses of popular queries, so as to avoid unnecessary computation.

Consider an LLM to be a Bayesian prior over experts P( e | p ), that predicts how accurate an expert e's response will be if e is fed prompt information p. For example, Banno's example above demonstrated that ChatGPT knows the English of, but not the game of, "noughts and crosses". And so it is well positioned to forward Banno's prompts to a dedicated Noughts and crosses expert.

Given a collection of experts e1, e2, ... that are each trained to solve separate problems, the response r to a prompt p is ideally determined by feeding the prompt to every expert and taking a weighted average with respect to the LLMs output.

P ( r | p) = Sum m, P (r | m , p) x P ( m | p )

where P( r | p) is optimised by fine-tuning the LLM prior so as to minimise an expected Loss function

E [ L ( r , p) | P (r | p) ]

Those models had a strong tendency to hallucinate. GPT4 also hallucinates, sometimes, but its tendency to do so is sharply reduced compared to previous models — Pierre-Normand

I guess that depends on what you mean by "tendency", and whether or not you include the "wizard of Oz" Open-AI team that continually patches it's errors in the background.

In terms of being an LLM, any apparent improvements are merely down to increasing the size of the training set on which GPT 4 was pretuned (together with increased architectural capacity). If we are referring to it's innate ability to generalise correctly with respect to unseen inputs, then it is certainly not the case that it's tendencies to be correct have improved, due to the fact that it isn't a reasoner that can prove or verify theorems the hard way.

Such improvements would require augmenting the LLM with a planning module, together with domain-specific models that it would need access to for testing it's planned responses against, prior to replying to the user in that given subject domain. I'm sure these improvements are down the pipeline, with the chat-bot service fragmenting into a network of modular expert services that are maintained and improved separately.

The future of a "General intelligence" service can only be to serve as a front-end for a network of domain-specific intelligent services, at which point it could be asked how large does an LLM really need to be, if it's eventual use is only to interpret requests before forwarding them to domain specific services?

The trouble with LLMs, is that they don't grasp the computational complexity of a question; they produce instant copy-pasta in the same amount of time regardless of the difficulty of the question, and without providing a measure of confidence in their answer.

For example, when asking an LLM a simple variant of a well known logic problem, if that variant isn't in it's training set, it will likely respond quickly with a wrong answer, rather than returning "mmmm......." followed by thinking further to produce the correct answer, or to return "don't know".

Well expressed, interesting link. Wittgenstein, of course, took a contrary view to Augustin as to how we learn a language, treating it as becoming a participant in the activities of a community. It's not apparent hat this account could not be used here, that those "intelligible objects" are grasped as one becomes acquainted with the way they are used in a community. So learning what "2" is consists in learning how it is used by those around you, and using it within that community. — Banno

Wittgenstein never said communication was the essence of language , neither did he equate in all cases, truth conditions with community agreement. Sure, in some language games a sign such as "2" might be defined in relation to community responses - a good example is the Github code-repository for the mathematics library of the Lean Programming language that is used automate mathematics; it implements and comprises the meaning of "2" for the community of Lean users. But then there are many other use-cases that don't fall under this definition, such as the isolated platonist who privately identifies "2" with what he is seeing or imagining. If his use-case bears no relation to his community, then he might be said to be playing a single-player "Augustinan" language-game. Perceptual and aesthetic judgements tend to fall into this category.

Recall that Wittgenstein rejected logicism - he rejected the fundamentalism that equivocates arithmetic with analytic tautologies. He is in record of saying "I think I know what Kant meant when he said that 2+5 = 7 is synthetic" . If the propositions of arithmetic are regarded as being synthetic, then it implies that the meaning of such propositions is open to interpretation in each and every case.

Truth and falsity are relative to convention. Truth values can also be eliminated from discourse, if one is willing to abandon the idea of shared belief referents.

E.g, if i judge my earlier beliefs to be "false" on the basis of new evidence that I obtain, then I am taking the referents of my present judgements and my past beliefs as being identical, in order so that I can speak of the new evidence as falsifying my earlier beliefs.

Alternatively, I could consider the new evidence as constituting a new referent of my present judgements , in which case I consider my earlier beliefs to be obsoleted by the new evidence, rather than being falsified by the new evidence.

But life goes on with or without you right. I have no problem with such solipsism such as yours but it fundamentally rejects reality. — invicta

All solipsism rejects is the idea of absolute reality, analogous to special relativity's rejection of absolute motion. Solipsists are able to reconcile their surface-level disagreements by transforming the meaning of each other's assertions to fit their own frame of reference, in a manner analogous to using a Lorentz transformation.

So it isn't necessarily contradictory to assert that one is immortal from one's own perspective, yet mortal from another's perspective.

.

Having information rest solely in the minds of observers seems at risk of becoming subjective idealism. The information has to correspond to and emerge from external state differences or else how can we discuss incorrect interpretations of any signal? — Count Timothy von Icarus

That's why i suggested "two player" game semantics. The semantics of interaction isn't accommodated by the traditional conceptions of either computation or causality, both which define life to be a one-player game but disagree as to who the solitary player is.

Wouldn't physics generally be answering the question of "if nature acts in such-and-such a fashion how will nature respond?" — Count Timothy von Icarus

There certainly are many scientists who offhandedly assume in an old-fashioned way that causality must be an "objective" notion. But as Bertrand Russell pointed out, the notion of causality is objectively redundant. e.g, what does the notion of causality add to a description of the Earth orbiting the Sun? The notion of causality adds nothing of descriptive value to any proposition that states an actual state of affairs, while the employed purpose of causality is to model possible outcomes in relation to possible actions. Do you really wish to promote the possibilities that exist in relation to a model to the status of objective reality, given the fact that possibilities aren't scientifically testable or observable?

In general, scientific models are supposed to be about "the way the world is," not games. I don't think such interpretations were ever particularly popular with practicing scientists, hence why the Copenhagen interpretation of QM, which is very close to logical positivism, had to be enforced from above by strict censorship and pressure campaigns. — Count Timothy von Icarus

The Copenhagen interpretation itself isn't generally regarded as constituting a game-semantic interpretation of QM, but it should be noted that the the linear logic behind the ZX calculus has very strong game semantics (e.g see Blass and Abramsky's work on game semantics and linear logic ). The conceptual connection between Logic and games goes all the way back to Aristotle. And of course, logic is used to both state the causal assumptions of a model, and also to define computation. So there are good reasons for interpreting both causation and computation at least semi-normatively in terms of game-semantics, an analysis which if correct, precludes both from constituting or describing observer-independent properties of the universe.

.

unenlightened: the natural world can be defined without reference to any Gods. — Art48

That depends on whether "God" is narrowly understood as referring to a specific type of causal explanation that rivals physical explanations as another type of causation, or whether "god" is understood as being an integral concept to the very meaning of cause and effect.

E.g for Melbranche, God is the only causal agency, indicating that for Melbranche science is the study of miracles, implying that any empirically valid scientific law is god choosing to follow a deterministic strategy. His position might seem ontologically superfluous, e.g why assume that the course of the universe is the strategy of a single player game, as opposed to assuming the universe to be a zero-player game that is driven forward mechanically without any intervention, divine or otherwise? Melbranche apparently believed the universe to be mechanically describable but resisted the elimination of causal agency, due to believing that the propositions of mechanics aren't analytic but synthetic, making similar arguments to the empiricists such as Hume who came after him.

The original conception of computation was of a "mechanical method" that is synonymous with a deterministic "winning" strategy with respect to a single player game of finite and complete information. To recall the Church Turing Thesis

A method, or procedure, M, for achieving some desired result is called ‘effective’ (or ‘systematic’ or ‘mechanical’) just in case:

1 ) M is set out in terms of a finite number of exact instructions (each instruction being expressed by means of a finite number of symbols);

2) M will, if carried out without error, produce the desired result in a finite number of steps;

3) M can (in practice or in principle) be carried out by a human being unaided by any machinery except paper and pencil;

4) M demands no insight, intuition, or ingenuity, on the part of the human being carrying out the method.

This original conception of computation in terms of a mechanical method is therefore strongly, if not completely normative, strictly in relation to perspective, and anti-real in being defined entirely in relation to human purposes and human psychology, whist forbidding any empirical contribution from mother nature herself to the computational process. Or as Wittgenstein summed it up : "Turing Machines are what humans do" . Such single player games are incompatible with a realist's conception of causation as a zero player game that is fully determined by the initial state of the game without any subsequent interventions by man, nature or god.

To bring causation and computation into line requires their definitions to be weakened and generalised so as to refer to strategies of two player games involving interaction and dialogue between man and nature. Computer science and mathematics can then be understood as attempting to answer questions of the form "If nature were to act in such-and-such a fashion to my actions, then what are my available winning strategies in relation to my goal?". While physics and it's concept of causality could be understood as asking the complementary dual question "If one were to act in such-and-such a fashion, then how is nature expected to respond?"

As I explained already, this does not give a true representation of "final cause" because it provides no real basis for a distinction between consequences which are intended, and consequences which are accidental. In other words, if final cause was truly determinable from an agent's behaviour, all accidental acts by the agent would necessarily be intentional acts. — Metaphysician Undercover

Firstly, what makes you think that there is an objective matter of fact as to whether an effect was intended or accidental? Secondly, if there are such facts, then what do those facts consist of?

If we narrowly interpret the meaning of an "intention" as referring only to the agent's internal state, , then intentions as such cannot be teleological, for the agent's actions are explainable without final causes.

So in order for intentions to be considered teleological, one must consider both what is going on inside the agent as well as the environmental effects that the agent's behaviour produces, - effects which play no causal role in the agent's history of decision-making. Yet this understanding of 'intentionality' as a type of relationship between the agent's behaviour and the environmental biproducts of his actions, in turn implies that the agent is fallible with regards to knowing what his intentions are. For who now gets to decide what the agent truly intended?

Note that the problem of "Inverse Reinforcement Learning" is the problem of inferring an agent's overall goals from a history of the agent's behaviour, including the environmental consequences it's actions. It is a chicken-and-egg paradox; In order for observers to estimate an agent's overall goals given a history of it's behaviour, they must assume that the effects of the agent's actions were in accordance with it's intentions, that is to say, they must assume that the agent is an expert who understands his environment. But how can it be known whether the agent is an expert? Only by assuming what the agent's goals are :)

This implies that teleological concepts are either semantically or epistemically under-determined.

Since temporal order is what defines causation, separating the two only renders causation as unintelligible. — Metaphysician Undercover

Yes, if taken in the very hard sense of "separation". I'm referring to the fact that different observers from different perspectives, each of whom controls different variables, might have conflicting views as to what was the cause/intervention and what was the effect in a given situation.

Suppose Alice believes that if she presses button A, then a distant observer Bob will press Button B, otherwise Bob won't press button B. No other information is assumed.

Her causal belief might be represented by A => B.

Logically, this is equivalent to asserting NOT B => NOT A.

Therefore, in the event that Alice decides not to press the button, i.e. that event NOT A occurs, shouldn't Alice be open to the possibility that her decision not to press A was the effect of Bob deciding on NOT B 'before' Alice made her decision?

Posited examples of backward causation look a bit like teleology, but are categorically different. ,

If this is true, it's proof that there is no formalized definition of "cause".

And, since there are two distinct principal types of causation, efficient and final, there will never be an acceptable formalization of causation until the relationship between the two is represented properly. Formalization of one principal type of causation while excluding the other principal type of causation does not give a true formalization. — Metaphysician Undercover

But "Final causes" are representable in terms of bog standard causation without invoking teleological purposes, as demonstrated by reinforcement-learning algorithms that train a robot to implement "goal seeking" behaviour via iterative exploration and feedback . In this case, one might say that the "final cause" of the trained agent's behaviour is the trained evaluation function in the agent's brain that maps representations of possible world states to their estimated desirability. In other words, the final cause refers not to the actual goal-state in the real world that observers might colloquially say the learning agent "strives towards", but to the agent's behavioural policy and reward function that drive the agents behaviour in a mechanistic forward-chain of causation from an initial cause in a manner that is teleologically blind.

The agent's actions are not being "pulled" by the goal in any literal sense, so I am at a loss as to the incentive for mixing up purposes which refer to behaviour that converges towards a goal state, and causation which makes no reference to goal states.

This is exactly why a formalization is impossible, and causation will always be philosophical rather than scientific. This provides no basis toward understanding the cause of "doing something". So, a person does something and this causes something which otherwise wouldn't occur. If we want to know whether the thing which otherwise wouldn't have occurred is intentional, or accidental, we need a much better principle than this. And if you claim that this is irrelevant to "causation", all that matters is whether the thing otherwise wouldn't occur, you fail to properly represent "final cause" in your formalization, and you provide no principles for excluding accidents from our actions. However, it's quite obvious that the effort to exclude accidents is very important. — Metaphysician Undercover

If you accept the distinction between purposes and causes, then there is no case for the concept of causation to answer to regarding the distinction between intentions and accidents. For that's purely a matter of teleology and not causation.

The use of "final" in "final cause" seems to be misleading you. "Final" is used in the sense of "the end", and "end" is used in the sense of "the goal" or "objective". The terms "end", and "final" are used when referring to the goal or objective because the intentional cause is what puts an end to a chain of efficient causes when looking backward in time. So if D caused E, and C caused D, B caused C, and A caused B, we can put an end to that causal chain by determining the intentional act which caused A. It is called "the end", or "final" cause because it puts an end to the causal chain, finality. — Metaphysician Undercover

A is at the beginning :) Either a "final cause" is used to refer to a bog-standard initial cause that implies none of the teleological controversy commonly associated with aristotolean "final causes", else "final cause" refers to a teleological concept such as a purpose that is defined in relation to a goal state that is external to an agent's brain and that plays no causal role in the agent's behaviour, despite the fact the agent's behaviour converges towards the goal state.

Take a chain of dominoes for example. We look at the last fallen domino and see that the one falling prior to it caused it to fall. Then the one prior to that one caused it to fall. When we continue to follow this chain of causation, we find the intentional act which started the process, and say that this is "the final cause", because it puts an end to that causal chain. The terminology is derived from our habit of ordering things from the present, and looking backward in time, so that the causes nearest to us at present appear first, and the furthest are last. — Metaphysician Undercover

I suspect you are deviating from the commonly accepted notion of "final cause". The whole point of the "finality" in "final cause" is to imply that teleological concepts are necessary for explaining the effects of causation, which isn't the case in the dominoes example; teleology is explainable in terms of purposeless causation, as AI programmers demonstrate. But causation isn't explainable in terms of teleology. To mix up the concepts leads to confusion.

This I do not understand at all. The fact that accidents are still considered to be caused, demonstrates that causes are not necessarily "considered to be controllable preconditions". Furthermore, I've never heard of a causal model which allows for a cause to be after its effect. You simply create ambiguity here by saying "in some absolute sense" because the principle of relativity of simultaneity allows that from the perspective of different frames of reference, the temporal order of two events may be reversed. — Metaphysician Undercover

Which demonstrates the point i was trying to make, that what we call the "temporal order" has to be distinguished from the "causal order". That A causes B but not vice versa, doesn't necessitate that A occurs before B in every frame of reference. Also recall the time-symmetry of microphysical laws, models of backward causation etc.

You don't seem to understand causation sime. There is no scientific definition of cause. Cause is a philosophical concept. — Metaphysician Undercover

I think that's misleading and somewhat inaccurate. It is true that causation didn't undergo strict formalisation until the twenty first century, economists being among the earlier pioneers of causal modelling in the twentieth century, and that causation is still undergoing formalization in tandem with the brother concepts of probability and temporal logic. Nevertheless, there has been a rapidly converging consensus in both the scientific community and industry in recent decades to the formal identification of causes with particular variables of a probability model, that if intervened upon by the actions of an experimenter, are expected to produce observable changes in the correlations among variables that lie "downstream" of the intervention. See Judea Pearl for an authoritative account.

Causal models merely express the concept that doing something leads to observations that otherwise wouldn't occur. Unlike Russell's conception, the modern meaning of causality is counterfactual. Causal models essentially define causes as being 'initial' with respect to the causal orders they define or describe, making "final causes" an oxymoron in the sense of the causal order.

Nevertheless, causal models have nothing to say regarding the order and linearity of time itself unless their variables are given additional temporal parameterization. All that they demand is that causes are considered to be controllable preconditions of their effects, not that causes are necessarily temporally prior to their effects in some absolute sense, which might well be considered a matter of perspective.

"Final cause" is the intent, the purpose. So it is exactly the case that your thoughts, goals, and motivation are literally the final cause of the shed. Whatever reason you had, whatever purpose you had in your mind, this is the reason why the shed was built. Therefore these ideas, as intent, are the cause of your actions, and by extension the cause of existence of the shed. This is the basis of the concept of "intent" in law, the decision to bring about consequences.

That is why "variable" does not serve as an adequate representation. The fact that you wanted a shed, and this motivated you to go out and built a shed, is the cause of the shed. And you could further specify the particular purpose you had in mind for the shed when you built it. The intent, purpose in mind, or "final cause", is not a "variable" in the coming into existence of the shed, it is the cause of existence of the shed — Metaphysician Undercover

I understand Aristotle's definition of a 'final cause', but it makes no sense to me to muddle such "final causes" with the "causes" meant by the modern scientific definition of "causes" that refer to experimental inventions that go on to produce measurable effects. Especially considering the fact that 'Final Causes' are reducible to iterative evolutionary or adaptive feedback loops between an agent or population and their environment that are understandable in the bog-standard "initial cause" sense.

'Final causes' might be reasons with cognitive significance but imo reasons and causes are best kept apart, for they don't obey the same logic.

I agree, and I see a problem with the determinist attitude. Describing activity in the physical world in terms of efficient causation has been a very useful and practical venture. The problem is that this descriptive format has limitations which the determinist ignores or denies. We find that within human beings there is an active mind, working with immaterial ideas, to have real causal affect in the physical world. Causation from the mind, with its immaterial ideas is described in terms of final cause (goals purpose and intent), choosing from possibilities, which is completely distinct from efficient causation.

So there is a very real need to recognize the limitations of "efficient causation" as an explanation of the activities in the physical world. And we need to accept the reality of the immaterial "final cause" as having real efficacy in the material world. — Metaphysician Undercover

According to one of the two main accounts of causality, namely the perspectival "interventionist" interpretation, a causal model is a set of conditional propositions whose inferences are conditioned upon variables that are considered to have implicative relevance but which are external to the model, such as the hypothetical actions of an agent. These models, whose use is now widespread in industry and the sciences, are thus naturally "compatibilist" in conditioning all models inferences upon hypothetical or possible values of external variables that are considered to be chosen freely. So I presume you are criticising earlier historical conceptions of causality such as Bertrand Russells', which assumed a causal model to be a complete description of a system's actual dynamics (thus making cause and effect redundant notions).

What I don't follow is the relevance of a "final cause", unless it is surreptitiously being used to refer to an initial cause, i.e. a bog standard cause. For example, if I am working to build a shed in the back garden, what is the "final cause" of the shed here? Obviously my thoughts, goals and motivation throughout the project cannot be considered a literally "final" cause, which speculation notwithstanding, leaves the resulting actual shed as the only remaining contender for the final cause. Are you insinuating that the resulting shed caused me to build it? (which incidentally isn't likely to look anything like my imagined shed due to my terrible practical skills)

Direct realism, how can it be proven to be better than idealism? — Agent Smith

Direct realism can be thought of as absolutised idealism, to recall Berkeley's 'Master Argument' that all acts of measurement, thought and observation are in relation to some perspective; if one denies or ignores perspectival relativism, one jumps from subjective idealism to direct realism.

My understanding of indirect realism refers to my understanding of perception in relation to a third person subject, with respect to objects of my first-person world that I consider myself to directly perceive; for my frame of reference constitutes the very foundation of my understanding of indirect realism in other people.

The debate between direct and indirect realism is as misplaced as Galilean debates before Einstein as to whether an object is moving or not.

If you show me a tree and say "This is the one tree that stands before us", then I am in common-sense agreement. This generally means that I am automatically interpreting your proposition as referring to my frame of reference and to my frame of reference only.

However, I can also understand your proposition as referring to your frame of reference rather than mine. In which case, the number of trees depends upon the general consideration as to whether the reference of a proposition is taken to be relative to perspective.

If I interpret you as referring to either mine or to your frame of reference but not both, then it remains true that "there is one tree that stands before us", however it is semantically indeterminate as to which tree (i.e. which perspective) is being referred to the proposition.

On the other hand, if the reference of this proposition is considered to entail both perspectives then it is no longer the case that we can say that there exists only one tree, for there are two distinct perspectives.