# Having Thought  ## Metadata - Author: [[Haugeland]] - Full Title: Having Thought - Category: #articles - Summary: insert summary - My notes: - Summary: The content discusses the concept of perception and the interconnectedness of different components of cognition. It suggests that the understanding of phenomena and the generation of solutions rely on the combination of separate elements. The content also touches on the idea that personal involvement and self-awareness play a role in perception and understanding. It concludes by mentioning the importance of language in categorizing and describing experiences. - URL: https://readwise.io/reader/document_raw_content/136690636 - Source File: Having Thought by Haugeland Essays.pdf ## LLM Chats ## NotebookLM ## LLM Audio ## Highlights > ficial intelligence is that computers don’t give a damn. Or so I will argue by considering th ([View Highlight](https://read.readwise.io/read/01hnfgwzh5m8w5g4fx62wv7csa)) > Natural languages, logical calculi, and computer program- ming languages, as well as numerous more specialized notations, are all interestingly different species; but they are generically alike in being broadly language-like or logical in character. By contrast, pictures, though equally representational, are not linguistic at all, even in this broad sense; rather, they, along with maps, scale models, analog com- puters, and at least some graphs, charts, and diagrams, are species in another genus of broadly image-like or iconic representations. So the level of generality intended for representational genera is that of logi- ([View Highlight](https://read.readwise.io/read/01hnfh2mtm88f3qyrrzfn6d126)) > d ([View Highlight](https://read.readwise.io/read/01hnfh2pdr04f3epmhgwxe2db0)) > b ([View Highlight](https://read.readwise.io/read/01hnfh3g3y3n1gbgayc34154zr)) > . Second: besides the two familiar genera, are there any others? In particular, is so-called distributed representation a separate genus, on a par with logical and iconic representation, yet as different from each as they are from one another? And, if so, what is its generic essence? ([View Highlight](https://read.readwise.io/read/01hnfh498zsarbenfpfmh72xsp)) > A sophisticated system (organism) designed (evolved) to maximize some end (such as survival) must in general adjust its behavior to specific features, structures, or configurations of its environment in ways that could not have been fully prearranged in its design. If the relevant features are reliably present and manifest to the system (via some signal) whenever the adjustments must be made, then they need not be represented. Thus, plants that track the sun with their leaves needn’t represent it or its position, because the tracking can be guided directly by the sun itself. But if the relevant features are not always present (manifest), then they can, at least in some cases, be repre- sented; that is, something else can stand in for them, with the power to guide behavior in their stead. That which stands in for something else in this way is a representation; that which it stands in for is its content;1 and its standing in for that content is representing it. ([View Highlight](https://read.readwise.io/read/01hnfh4ebf3gar216399ad791b)) > Here, however, we will reserve the term ‘representation’ for those stand-ins that function in virtue of a general representational scheme such that: (i) a variety of possible contents can be represented by a corresponding variety of possible representations; (ii) what any given representation (item, pattern, state, event, …) represents is determined in some consistent or systematic way by the scheme;2 and (iii) there are proper (and improper) ways of producing, maintaining, modifying, and/or using the various representations under various environmental and other conditions. ( ([View Highlight](https://read.readwise.io/read/01hnfh4kev5s4svgka22zmdc3w)) > . One cannot simply say, for instance, that linguistic tokens are essentially digital, whereas images are analog, since, in the first place, speech is to some extent analog, and in the second, images are often handled digitally. More to the point, however, paradigm representa- tions of any genus—logical, iconic, distributed, or whatever—can be recorded without loss3 in strings of bits; and certainly no properties of bit strings as such can differentiate in general between the sorts of representation they record. And, on the face of it, there seems equally little hope of differentiating the genera on the basis of their repre- sented contents. Thus, a police officer’s report might well include both descriptions and photographs of one and the same crime scene; and, prima facie, there could be distributed representations of it as well. ([View Highlight](https://read.readwise.io/read/01hnfh4r9zpga89b3hwaa14s49)) > Logical representations are distinctive in that they have a (genera- tive) compositional semantics. This means that complete logical tokens (sentences, well-formed formulae, production rules, and so on) are complex structures, each with a recursively specifiable syntax and de- terminate atomic constituents, such that the semantic significance of the whole is determined by its syntax and the semantic significances of its constituents (perhaps with some situational parameters). ‘Semantic significance’ is here used as an umbrella term, meant to include, without being limited to, the represented contents (what the tokens represent). The respective contributions of the possible struc- tures and constituents are fixed arbitrarily by the scheme (for in- stance, a language); but, given these (and any relevant parameters), the significance of the compound is not at all arbitrary. ([View Highlight](https://read.readwise.io/read/01hnfh4xbrpbknysd8btws9n5c)) > Iconic representations are distinctive in representing their contents by virtue of being somehow isomorphic to them. In many familiar cases, like pictures and scale models, the isomorphism is obvious enough to strike the observer as resemblance. But that’s not required: the isomorphisms that determine the representational contents of graphs, wiring diagrams, and analog computers are often so abstract or mathematical that we wouldn’t naturally call them resemblances (though, of course, in a sense, they are). Note that there are many different kinds of isomorphism, and the ones that are relevant to any particular representation are all and only those determined by the scheme to which it belongs. Thus a picture token that happens to be entirely monochromatic may or may not represent its object as like- wise monochromatic, depending on whether it belongs to a mono- chromatic or full-color scheme. Which isomorphisms a given scheme employs is initially arbitrary or conventional; but once they are fixed, the contents of particular tokens are not arbitrary. ([View Highlight](https://read.readwise.io/read/01hnfh50m4c80sdqyw6qj859mk)) > Distributed representations are distinctive in that each portion of the token participates (in some broadly egalitarian way) in the repre- ([View Highlight](https://read.readwise.io/read/01hnfh59vvrdnaw2v9tspk9x92)) > senting of each portion of the represented contents. For this to make sense, there must be antecedent notions of portions, both of the to- kens and of the contents; and then the idea is that the representings of the content portions are all spread out over the token portions and superimposed on one another—whence the term “distributed”. What these respective portions are, and how various possible content por- tions can be (simultaneously) determined by various possible token portions, is, of course, settled by the particular scheme (which is arbitrary); but, given a scheme, it is not at all arbitrary what any particular token represents.4 ([View Highlight](https://read.readwise.io/read/01hnfh5chnta6mw8chvpsr32fn)) > . In each example, the represented content of the resulting plan or map is manifestly a deter- minate function of its compositional structure and the significances of its primitive components—just the token/content relation that was supposed to be characteristic of logical representations. Yet floor plans and maps are paradigmatically iconic. ([View Highlight](https://read.readwise.io/read/01hnfh5gysd8ymp5xjgzrzzh5j)) > Remembering, then, that any text or icon can be encoded without loss in a finite string of bits, and noting that each finite string of bits corresponds to a unique chord played on Turing’s piano, we conclude that any text or icon can be encoded in such a chord. The waveform produced, however, is a perfect distributed rep- resentation, according to the canonical account: each point on the waveform (instantaneous amplitude) participates in recording all of the bits, and the recording of each bit is spread over all of the wave- form. Since whatever it is that is representational in the original text or icon is somehow captured in patterns of these bits, that represent- ing too is spread out and superimposed in the waveform. ([View Highlight](https://read.readwise.io/read/01hnfh5q905rqk8jdcb84qx5e0)) > What these irritating examples have in common is most evident in the case of a photographed inscription. While the result is, in one sense, clearly an image, it is equally clearly not an image of whatever the inscribed text was about; rather, it is an image of the inscription itself, but in such a way that (assuming the text is still legible in the image) the ability of the original to represent whatever it represented is preserved in the image. Obviously, however, that representing remains essentially logical in character. In other words, the logical ([View Highlight](https://read.readwise.io/read/01hnfh5tkx29qndgzjq4bj4tc5)) > representation has in no sense been “translated” or transformed into an iconic one; instead, it has merely been “recorded” in an iconic medium. Significantly, the recording process, unlike genuine transla- tion, would proceed just as smoothly even if the given inscription were complete gibberish. Evidently then, we must distinguish in this case between the logical representation and its iconic recording. ([View Highlight](https://read.readwise.io/read/01hnfh5ya6rydm7q9592x49phk)) > Thus, the image is not translated or transformed into a logical representation, but only re- corded in a logical medium.7 (And, again, the recording itself would be equally successful even if the “image” were a contentless blob.) ([View Highlight](https://read.readwise.io/read/01hnfh620edy184z4qy089wmtc)) > . What is the difference between recording and representing, and how is it that they are so readily confused? Fundamentally, I think, recording is a process of a certain sort; and to be a record is to be the result of such a process. By contrast, representing is a functional status or role of a certain sort, and to be a representation is to have that status or role. These are not the same because, in general, being the result of a recording process is neither necessary nor sufficient for having a representational status or role. On the other hand, the recording process does produce an out- put which is a record of its input; and representational status does involve one entity standing in for another. This formal relational paral- lel, plus the fact that, in certain cases (such as photography), the product of the recording process can also have the status of represent- ing the input, therefore being simultaneously a record and a represen- tation of it, is the basis of the confusion. ([View Highlight](https://read.readwise.io/read/01hnfh66vby4c3qk2nderrm5w9)) > Recording is a generalization of copying. To count as recording, a process must be reversible in the sense that there is a complimentary process—“playing it back”—which reproduces from the record a du- plicate or copy of the original. Since a copy of a copy is a copy, copying is a special case of recording (in which the record and play-back processes are the same). Recording, like copying, is only partial; only certain prespecified aspects of the original are recorded. In case what is being recorded is a representation (as a representation), then the relevant aspect, what must be reproduced on play-back, is its sche- matic type; the criterion for successful recording of representations is preservation of type-identity. ([View Highlight](https://read.readwise.io/read/01hnfh6c8nevz2brxa7zx67ydn)) > . A court recorder’s job is actually two- fold: identifying the spoken words (non-trivial) and then recording them (trivial). The case is essentially similar to producing a typescript from someone’s handwritten scrawl, or even restoring the missing characters in a spotty photocopy: identifying the words takes wits; recording them, once identified, does not. ([View Highlight](https://read.readwise.io/read/01hnfh6rnvwt6dge0ka22kmyqt)) > These considerations further suggest, though by no means estab- lish, a sufficient condition for sameness of representational genus: if the representations of one scheme can be witlessly transformed into equivalent representations from another scheme by a general pro- cedure, then those schemes are species of the same genus.8 T ([View Highlight](https://read.readwise.io/read/01hnfh6w3tbqf05xs2kmr0a6h9)) > Even as a sufficient condition, the proposed test is so far only hypothetically acceptable, since it depends on the plausible but unsupported assumption that cross-generic “translation” always requires wits.9 Only later will we be able to explain and support the plausibility of this assumption, and thereby defend the suggested condition. ([View Highlight](https://read.readwise.io/read/01hnfh6yzv7dympam30cpzb8sf)) > Each canonical account characterizes its genus in terms of a dis- tinctive sort of representing relation between the representations and what they represent—roughly compositional semantics, structural isomorphism, and spread-out superposition. Our diagnoses of the three counter-examples, however, have revealed these distinctive rela- tions to be indicative more of certain recording devices or processes than of the respective representational genera. O ([View Highlight](https://read.readwise.io/read/01hnfh74550j3wcgm4b8ecwbfc)) > 6 Skeletal versus fleshed-out content The first question to ask is: What, strictly speaking, are the contents of various representations? Exactly what objects, features, or configu- rations in the world are represented? Often, in real-life communica- tion, only a little need be indicated, in order to apprise the recipient of much about the world. In practice, a little representation goes a long way. But that’s because the recipient is already in a position to learn much from just a little new information—perhaps by already “know- ing” a lot about the current situation and/or about the world in gen- eral. How much of what derives from this “background familiarity” belongs in the represented contents as such? > In one sense, all or most of what a recipient could be expected to learn from a communicated representation could be deemed “con- tained” therein—that is, as part of the message. Unfortunately, such a liberal attitude is problematic in the context of qualitatively differenti- ating the contents characteristic of the respective genera. For, if the recipient’s broader understanding is mediated by other (presumably internal10) representations, themselves belonging to a genus different from that of the representation being examined, then the contents broadly construed would confusedly exhibit characteristics of both genera. Since it is not possible in general to know about or control for the effects of all such mediating representations, there is no hope of finding the essences of the genera in the character of their contents, unless these contents can be delimited more narrowly. > By way of anticipation, then, let’s call the strict content of a repre- sentation, that not augmented or mediated by any other, its skeletal or bare-bones content. The metaphor is intended to suggest that the full- ([View Highlight](https://read.readwise.io/read/01hnfhwvvfjb1x0c3qt4tp2c81)) - Note: The writer is talking about how things are represented and understood. They ask what exactly is being represented in different ways of communication. Sometimes, only a little bit of information is needed to tell someone a lot about the world. But how much of this background knowledge should be included in the representation itself? The writer says that everything a recipient could learn from a representation could be considered part of the message. However, this is problematic because if the recipient's wider understanding is influenced by other representations, then the contents of the message would show characteristics of both types of representation. To solve this problem, the writer suggests calling the basic content of a representation its skeletal or bare-bones content. This means that it only includes the most important information and doesn't include any extra information from other representations.