Art and Agency: A Reassessment


Anthropology Department University of Durham, Durham DH1 3HN:


In his book, Art and agency, Alfred Gell presents a theory of art based neither on aesthetics nor on visual communication. Art is defined by the distinctive function it performs in advancing social relationships through ‘the abduction of agency’. Art objects are indexes of the artist's or model's agency. This article examines Gell's use of agency, particularly in relation to the ritual art that is central to his argument. Focusing on Gell's employment of Peirce's term ‘index’ (out of his triad of index, icon, and symbol), I note that Peirce's approach deflects attention from signification towards the link between art works and the things to which they refer. I consider what Peirce meant by abduction, and conclude that while Gell makes a good case for the agency of art objects he does not explain the distinctive ways in which art objects extend their maker's or user's agency. Gell lacked the time to make detailed revisions before publication and I acknowledge that, given more time, he might have revised some parts of the book.

It is now five years since the publication of Alfred Gell's highly original, posthumous book, Art and agency (1998). Gell set out to construct a theory of art based neither on aesthetics nor on visual communication. Gell acknowledges the importance of form, balance, and rhythm at various points in his analysis, but argues that they are not to be appreciated in the detached manner implied by the term ‘aesthetics’. His rejection of semiotics is more radical. Drawing on Peirce's concepts of index, icon, and symbol, Gell argues that art objects may be icons or indexes, but never symbols. Indeed, he frequently treats icon and index as synonymous. The argument is dense and ingenious, and highlights many aspects of the social role of art objects that have previously been neglected. In the end, however, I consider Gell's argument to be unsustainable. His efforts to exclude Saussurian symbolism have also been criticized by some of the contributors to a recent volume evaluating Gell's achievement (Pinney & Thomas 2001). In this article I propose to recall what Peirce, Saussure, and Mounin wrote on index, icon, sign, and symbol. Peirce and Saussure employed the terms sign and symbol in very different ways, while Mounin (1970) elaborated a theory of visual communication that acknowledged the special qualities of icons and indexes. Armed with some basic definitions, I then critically re-examine Gell's argument in Art and agency. I argue that Gell was correct to reject a specifically linguistic model for visual communication, but that he was wrong to minimize the importance of cultural convention in shaping the reception or ‘reading’ of art objects.

Gell's approach to the anthropology of art

Gell sets out to provide an anthropological theory of art, rather than one derived from semiotics or art history. His theory is, specifically, a theory based on British social anthropology, that is, on the study of social relationships, rather than on culture (Gell 1998: 7). There are two differences, however, between Gell's anthropology and classic Radcliffe-Brownian structural functionalism. First, the unit of analysis is not status, reproduced as a position in a social structure. Gell's focus is on the agent, and the networks of social relationships constructed through his or her agency. Secondly, ritual is not misguided behaviour that inadvertently has the effect of sustaining the social order, but behaviour to be understood in terms of the participants’ own theory of agency.

Art is defined by the distinctive role it plays in advancing social relationships constructed through agency. Not all objects function as art objects. ‘Agency can be ascribed to “things” without this giving rise to anything particularly recalling the production and circulation of “art” ’ (1998: 23). Art objects can often be recognized intuitively: ‘most of the art objects I shall actually discuss are well-known ones that we have no difficulty in identifying as “art”; for instance, the Mona Lisa’ (7). However, art objects have three diagnostic features. They are usually made so as to be seen (24). Secondly, art objects are indexes of social agency. Smoke is not art, because it is a natural index of fire, unless the fire has been artificially lit, in which case smoke becomes an artefactual index (15). Art objects are further distinguished by being both difficult and captivating: ‘they are difficult to make, difficult to “think”, difficult to transact. They fascinate, compel, and entrap as well as delight the spectator’ (23). ‘Where indexes are very recognizably works of art, [they are] made with technical expertise and imagination of a high order, which exploit the intrinsic mechanisms of visual cognition with subtle psychological insight’ (68). Anthropology is characterized by its interest in unfamiliar modes of thought. The anthropologist's task is to describe forms of thought which are not particularly sound from a philosophical perspective, but which are none the less socially and cognitively practicable (17). Art objects that function as agents in social relations constructed through ritual and magic are therefore particularly instructive for the anthropological study of art.

Existing work on the anthropology of art

Gell contends that the task of an anthropology of art is not, as Price (1989), Coote (1992; 1996), Morphy (1994; 1996), and others suppose, to define the characteristics of each culture's aesthetic. He acknowledges interesting work by Thomas (1991) and Steiner (1994) on the reception of non-Western art in the West, but argues that this is not a genuine anthropology of art, since it does not study art functioning in the context for which it was created (Gell 1998: 8).

Gell rejects Morphy's definition of art, which is that art objects have either semantic or aesthetic properties, or both, which are used for presentational or representational purposes (Morphy 1994: 655). Gell entirely rejects the notion of art as a visual code, arguing that nothing ‘except language itself, has “meaning” in the intended sense’ (Gell 1998: 6). He further argues that aesthetic properties can be assessed only in terms of the intended effect of an art object in its context of use. Gell commends the studies of decorative art by Kaeppler (1978), Price and Price (1980), and Hanson (1983) as among ‘the more interesting studies which have been produced by anthropologists’ (Gell 1998: 73). However, Gell rejects the idea that decoration is valued for its own sake. Decorated artefacts are aspects of their owners’ personhood. Hanson is criticized for failing to identify a specific correlation between Maori art style and the Maori exchange networks through which personal identity is negotiated (Gell 1998: 160).

Gell sets out his vision of the anthropology of art as follows. A programme of elucidating non-Western aesthetics is exclusively cultural, whereas anthropology is a social science not a humanist discipline. The anthropology of art should focus on the social context of art production, circulation, and reception, not on the evaluation of particular works. From its inception, anthropology has been concerned with things which appear as, or ‘do duty as’, persons. Tylor's work on animism initiated this approach, and it continued with Mauss's work on the gift. There is thus a solid basis in anthropological theory for treating art objects in the same way. ‘I view art as a system of action, intended to change the world rather than encode symbolic propositions about it’ (Gell 1998: 6).

What is wrong with existing work


Gell's rejection of aesthetics is less radical than his rejection of art as a visual language. He does not deny that works of art are sometimes intended and received as objects of aesthetic appreciation (Gell 1998: 66), but reminds the reader that ‘the “aesthetic attitude” is a specific historical product of the religious crisis of the Enlightenment and the rise of Western science … [that brought about] the separation between the beautiful and the holy’ (97). Gell argues that aesthetic values vary from culture to culture and are always embedded in a social framework. The anthropology of art should be interested in how aesthetic principles are mobilized in the course of social interaction. ‘Melanesian aesthetics is about efficacy, the capacity to accomplish tasks, not “beauty” ’ (94). Trobrianders attribute outstanding skill to superior magic; Westerners attribute it to artistic inspiration.

I find these arguments uncontroversial. Of course, aesthetic values vary from culture to culture, and their effect may be construed within a different theory of being (Layton 1981: 11-19). Forge long ago documented that Abelam artists discuss form and proportion in their work, but ‘the skilful artist who satisfies his aesthetic sense and produces beauty is rewarded not for the beauty itself but because the beauty … is recognized by the others as power’ (Forge 1967: 82-3). Morphy later showed in greater detail how Yolngu artists are clearly concerned to produce effects on the senses which Europeans would interpret as aesthetic, but ‘what Europeans interpret at a general level as an aesthetic effect Yolngu interpret as a manifestation of ancestral power’ (Morphy 1989: 23).

Nor does Gell reject universals. Acknowledging Washburn and Crowe (1992), he argues (1998: 160) that ‘There exists what amounts to a “universal aesthetic” of patterned surface; the same symmetry configurations … turn up all over the world’. Gell ascribes two functions to the formal qualities of decorative art. The world is filled with decorated objects because, in the first place, decoration is often an essential aspect of the technology of enchantment. Decorative patterns weave their spell because ‘we can never quite understand the complex relationships they embody’ (80). Relationships between the elements of decorative art are, secondly, analogous to social relationships constructed through exchange.

I disagree with Gell's discussion of aesthetics to the extent that he only recognizes cases that support his own agency-orientated approach. Many other social contexts for the appreciation of form can be found in the anthropological literature. Hughes-Freeland (1997) discusses the politicization of aesthetics in Java in relation to the impact of colonialism. The Lega of Central Africa consider objects made of ivory and well-polished wood to be the most important because they are associated with the most skilled and experienced leaders. However, Biebuyck provides ample evidence that smooth and polished forms are appreciated for their own sake (e.g. Biebuyck 1973: 179). Mende of Sierra Leone expect women to be beautiful, delicate, pretty, and groomed. They also expect women to be kind, patient, and loving (Boone 1993: 304). Beauty is dangerous because it is powerful, but the danger is that a beautiful Mende girl may become arrogant and narcissistic. Among the Wola of Papua New Guinea men imitate birds of paradise and incorporate their feathers in ceremonial head-dresses worn at exchange festivals. They do so not because they expect the birds’ qualities to be transferred to themselves by sympathetic magic but because the dancers wish to appear virile and handsome by modelling themselves upon the birds’ behaviour (Sillitoe 1988: 310-11).

The linguistic model

Thomas rightly notes in his foreword to Art and agency that the really radical quality of the book is not so much its rejection of aesthetics as a basis for an anthropological theory, but its repudiation of the view that art is a matter of meaning and communication. ‘Visual art objects are not a part of language … nor do they constitute an alternative language’ (Gell 1998: 6). Art objects are only signs with meanings when they are used as a part of language, that is, as graphic signs. Gell wishes to avoid ‘the slightest imputation that (visual) art is “like language” ’ (14).

The stated motive for Gell's attack is the failure of structuralist semiotic anthropology in the 1970s (163). At that time, Gell writes, ‘it was customary to discuss systems of all kinds as “languages” … Art was the (cultural) “language of visual forms” (164). Gell singles out Faris (1971) and Korn (1978), who constructed vocabularies of visual elements and grammatical rules for combining them into well-formed motifs or compositions. He objects that there is no hierarchy of levels in art equivalent to phoneme, morpheme, and syntax in language. Lines, circles, and zigzags are not visual phonemes. This may be a fair comment, but semiologists have long recognized that language has unique characteristics that distinguish it from other forms of human communication. The French semiologist, Mounin, made much of the unique phoneme-morpheme ‘double articulation’ of language (Mounin 1970: 43-4, 52, 74), just as Saussure, the founder of semiology, emphasized the distinctive arbitrariness of the association of sounds in language with the prototypes which they denote.

In fact, just as Gell accepts the study of aesthetics where it can elucidate the power of art objects as agents, so he also concedes that there is some value in a semiotic approach, provided that it does not rely on a linguistic model. ‘No reasonable person could suppose that art-like relations between people and things do not involve at least some form of semiosis; however one approaches the subject there seems something irreducibly semiotic about art’ (14). Gell uses the terms ‘meaning’ and ‘semiosis’ imprecisely. I return to these issues after outlining Gell's theory of agency.


Gell defines agency in the following terms: ‘Agency is attributable to those persons (and things …) who/which are seen as initiating causal sequences … events caused by acts of mind or will or intention … An agent is the source, the origin, of causal events, independently of the state of the physical universe’ (Gell 1998: 16). On the face of it, art objects cannot therefore be agents in themselves, but act merely as extensions (indexes) of their maker's or their user's agency. As Gell writes, art objects are not self-sufficient agents, but secondary agents that have effect by virtue of being enmeshed in social relationships.

Although Gell does not cite an authority for his use of the term ‘agency’, it is noteworthy that Giddens saw an intimate connection between agency and power. For Giddens, agency does not refer to people's intentions but to their ability to act. Agency is the ability to act in particular ways, where more than one course of action is possible. One ceases to be an agent if one can no longer make a difference, in other words, where only a single course of action is practically available (see Giddens 1984: 9, 15). Each of the various forms of constraint simultaneously enables someone else to exercise agency (Giddens 1984: 173). Gell emphasizes the ritual agency of art. He has in mind the ability of designs to entrap the recipient in competitive exchange; they may also dazzle the opposing warrior or deflect evil spirits. All these contribute to an argument for agency as power.

The idea of art objects as social agents is not new (Layton 1981: 43; Wolff 1981: 24-5). Appadurai wrote, ‘in many historical societies, things have not been so divorced [as in contemporary Western thought] from the capacity of persons to act’ (Appadurai 1986: 4). The issue concerns the type of agency attributed to art objects. How does an art object extend its maker's or user's agency? Gell argues that causal chains initiated by agents come into being as states of mind and are orientated to the states of mind of other actors. If the initiator and recipient do not come into contact, the initiator's agency must be mediated by physical objects. We detect their agency in the disturbances they leave in the material world. A stone chipped in a suggestive way attracts our attention because it may be a prehistoric tool and therefore an index of ancient agency. Kula objects are ultimately attached to powerful persons, but circulate well beyond their physical reach. Kula valuables associated with a participant's name are indexes of his bodily presence. Gell identifies four types of agency that art objects can possess. Sometimes the agency is psychological, as when a spectator is impressed by technical excellence or erotically aroused. Sometimes the art object's agency is physical, as when a supplicant is cured by kissing a religious icon. Works of art are sometimes intended and received as objects of aesthetic appreciation; they may, indeed, sometimes function semiotically, but Gell specifically rejects the notion that they always do so (66).

People and things are agents only if there is a recipient (a ‘patient’) for them to act upon. Patients enter into a social relationship with the index, the art object. The art object in turn refers to a prototype (such as an historic individual, or a deity), either by representing it iconically (Samuel Johnson by Joshua Reynolds) or by an indexical association – such as a stone cast as a thunderbolt by the deity. Chapter 3 of Art and agency is a highly original working-through of the permutations offered by the relationships between these four terms. The fundamental question, then, is how indexes or icons have agency, how they mediate the interaction of states of mind. Gell cites land mines as agents of the evil intent in the minds of Pol Pot's soldiers (Gell 1998: 21). Pol Pot's soldiers used them as extensions of their own agency. This is a misleading parallel. Art objects do not have the same kind of agency as mantraps or poisoned arrows. If Pol Pot's soldiers had spent their time burying pictures of the Mona Lisa, or even pictures of Pol Pot, there would not be so many Cambodians whose lives today have been ruined by shattered limbs. Nor is Gell's book illustrated with diagrams of land-mine firing mechanisms, breech-loading cannon, and so on, but rather with reproductions of paintings, sculptures, tattoos, and decorative designs.

Index and icon

Gell draws on Peirce's theory of semiotics to characterize the way in which art objects function as social agents. An art object is an index of its maker or user. ‘An “index” in Peircean semiotics is a “natural sign” … from which the observer can make a causal inference of some kind, or an inference about the intentions or capabilities of another person’ (Gell 1998: 13). Smoke is an index of fire, of the agency of the person who lit the fire.

The naturalness of the index is crucial to Gell's escape from a linguistic model. In Peirce's theory the index was one of three types of sign: index, icon, and symbol, classified according to the way in which they are linked with the object – Gell's prototype (Peirce 1955: 100-9). An index refers to an object by virtue of being really affected by that object. The shadow on a sundial is an index of the time of day, a weathervane an index of wind direction. An iconic sign, such as a picture of a horse, or an onomatopoeic sound (woof, neigh), has some of the same characteristics of the thing it denotes. An icon is a sign of an object to the extent that it is like that object. A Peircian symbol is arbitrarily associated with what it denotes, and this is the case with the vast majority of the words in any language. The link between the word (symbol) and the object is wholly established by convention.

Gell is clearly using index in the way it was defined by Peirce. So too in his use of icon: ‘I believe that iconic representation is based on the actual resemblance in form between depictions and the entities they depict’ (Gell 1998: 25). While symbol does not figure much in Gell's argument, he does use it in the Peircian sense. ‘I do not believe that iconic representation is based on symbolic “convention” (comparable to the “conventions” which dictate that “dog” means “canine animal” in English)’ (25).

Conflation of index and icon

Gell decided to avoid not only a linguistic model, but also any appeal to the role of culture in educating the eye (Gell 1998: 2, quoting Price 1989). Index, icon, and symbol can therefore be ranked in order of decreasing usefulness to Gell. The index and icon have an intrinsic connection with the objects they denote, unlike the Peircian symbol, which depends entirely on cultural convention. ‘A symbol is a sign which refers to the object that it denotes by virtue of a law, usually an association of general ideas’ (Peirce 1955: 102). A symbol is no use to Gell. An icon is also relatively unhelpful, because it is never identical to its prototype, a point given only minimal attention by Gell: ‘without pausing to unravel the very difficult question as to the nature of the relationship between real and depicted persons’ (15). ‘It is true that some “representations” are very schematic … Recognition on the basis of very under-specified cues is a well-explored part of the process of visual perception’ (25). This process is, in my assessment, both central to a comparative anthropology of art and irredeemably cultural. Iconicity is mediated by cultural convention. Representational styles select which aspects of the world they depict according to cultural tradition, and the chosen aspects are organized in conventional ways. When Constable began to depict foreground vegetation with fresh green pigments, his contemporaries were shocked that he had abandoned the warm browns and blues conventionally used to convey depth (E. Gombrich 1960: 40-1). Among the cultures of the Northwest Coast of North America, two-dimensional split representation strives to depict all the diagnostic traits of a totemic species, but lays them out as if the body were split down the line of the back or chest (Boas 1955: 221-30). Central Australian art depicts animals and people according to the marks they leave in the sand as they run, walk, or sit (Munn 1973: 132-45).

Rather than address the conventional character of icons, Gell circumvents the issue by treating icons as a subset of indexes:

One may need to be told that a given index is an iconic representation of a particular pictorial subject (26);
Prototypes: entities held, by abduction, to be represented in the index, often by virtue of visual resemblance, but not necessarily (27);
‘iconic’ … that is, indexes physically resembling a prototype (97).

The clearest conflation of index and icon occurs in Gell's discussion of Lucretius. Gell proposes that the convergence of images (icons), and parts (indexes) of things, can be approached from a philosophical angle through the doctrine of ‘emanations’, which ‘ “primitive” people anticipated, in their confusion’ (Gell 1998: 104). This Epicurian doctrine was clearly stated by Lucretius: idols (simulacra) of things are like films peeled off the surface of things, which fly to and fro through the air. ‘Pictures of things … are emitted from things off their surface’, like the smoke which logs of wood emit (passages from De rerum natura bk 4, ll. 30-60, quoted Gell 1998: 105).

There is something special about the index because, according to Peirce, an index ‘necessarily has some quality in common with the object’ (Peirce 1955: 102). The link is not constructed by cultural convention. ‘Indexes are not part of a calculus (a set of tautologies, like mathematics) nor are they components of a natural or artificial language in which terms have meaning established by convention’ (Gell 1998: 13). A Kula valuable ‘does not “stand for” someone important, in a symbolic way; to all intents and purposes it is an important person in that age, influence, and something like “wisdom” inheres in its physical substance … mind and reality are one’ (231). Maori meeting-houses, according to Thomas, ‘were not“symbols” … but vehicles of a collectivity's power. They simultaneously indexed a group's own vitality and ideally or effectively disempowered others’ (Thomas 1995: 103, cited Gell 1998: 251).


In order to avoid treating art as a medium of communication, Gell introduces the term ‘abduction’. Abductions are inferential schemes, as when we infer the same type of agency in a real and a depicted person's smile (Gell 1998: 15). For Gell, abduction is a form of inference that does not derive from knowledge of cultural convention. Abduction was defined by Eco, following Peirce, as the process that occurs when ‘we find some very curious circumstances, which would be explained by the supposition that it was a case of some general rule, and thereupon adopt that supposition’ (Eco 1976: 131, citing Peirce 1931-58, 2: 624, quoted Gell 1998: 14). Abduction is a form of inference practised in ‘the grey area where semiotic inference (of meaning from signs) merges with hypothetical inferences of a non-semiotic (or not conventionally semiotic) kind’ (Gell 1998: 14).

Abduction is, however, a concept taken from semiotics. As defined by Peirce, and used by Eco, abduction says nothing about the presence or absence of conventional rules. It characterizes the logical procedure a person can adopt when they think they have detected a pattern in events and act upon that supposition (Eco 1990: 59). Eco quotes two examples from Peirce. One is a purely statistical question of probability concerning bags of coloured beans. The other reads as follows:

I once landed at a seaport in a Turkish province; and … met a man on horseback, surrounded by four horsemen holding a canopy over his head. As the governor of the province was the only personage I could think of who would be so greatly honoured, I inferred that it was he. This was an hypothesis (Eco 1976: 131, citing Peirce 1931-58, 2: 265).

It would be a matter of further investigation to discover whether an elaborate protocol of coloured canopies, numbers of attendants, and so on existed to signal status, or whether the governor, on a whim, used his superior authority to call up four horsemen to shelter him from the sun on a particularly hot day (see also Keen 2001: 32-3).

Mounin distinguishes between systematic forms of communication, such as language, musical notation, and marine signal flags, and apparently non-systematic forms, such as the plastic arts and advertisements. Communication through non-arbitrary signs does not require the same degree of systematization as language. Advertisements may use a range of colours to catch the consumer's eye; a series of different advertisements can entice the consumer into buying the same product. A photograph of a refrigerator can choose a variety of products to advertise its capabilities (Mounin 1970: 38). In the face of uncertainty about the nature or precision of rules, abduction (in the semiotic sense) may indeed be the appropriate mode of interpretation, although some art traditions are more highly structured than others.

Agency, not signification

Gell recognizes the difficulty of replacing a theory based on art objects as visual signifiers with one based on art objects as agents. ‘To assert … the idol is not a “depiction” of the god, but the body of the god is all very well, but I accept that any such assertion constitutes a paradox’ (1998: 99). There are two principal stages in Gell's theory of agency in art. The first considers agency in individual objects, the second considers the artist's œuvre as an aspect of his social identity.

The individual object

Gell's theory of agency is directed towards the way in which actors’ states of mind are altered. Abduction is a cognitive operation. The agency of art objects therefore derives from the way in which they affect the mind of the recipient (the patient).

Gell's first step is to cite Frazer's theory of magic. The power of the icon is exemplified through what Frazer called ‘homeopathic magic’. Homeopathic magic aims to destroy an enemy by destroying an image of him, or cure someone by giving them medicine made of healthily coloured objects (Frazer 1994: 29). The power of the index is exemplified by contagious magic. Contagious magic works on fragments a person gives off, such as hair or nail clippings (Frazer 1994: 37). ‘The kind of leverage which one obtains over a person or thing by having access to their image is comparable, or really identical, to the leverage which can be obtained by having access to some physical part of them’ (Gell 1998: 105).

During the 1960s, around the time that Gell and I were students, there was a debate about the difference between ritual and science, and how best to render magical actions rational (Beattie 1966; Goody 1961; Horton 1960; 1964; Lienhardt 1961). Gell cites one of the principal contributors. Beattie criticized Frazer for attributing causal intentions to magical practices. Beattie argued that magic is in fact symbolic or expressive. He concluded that spirit mediums among the Nyoro put on a dramatic performance that was satisfying in its own right as a theatrical representation of the everyday hazards of Nyoro life, and not intended to change them (Beattie 1966: 72). Gell challenges Beattie's attempt to deny a means-end relationship in ritual or magic: ‘Magic is possible because intentions cause events to happen in the vicinity of agents’ (Gell 1998: 101). Magic registers and publicizes the strength of desire and therefore has an effect just as, if there were no breakfast-desiring agents, hens would not have been domesticated, saucepans invented, and eggs cooked. ‘The real causal explanation for why there are any boiled eggs is that I, and other breakfasters, intend that eggs should exist’ (101). If one were taking a semiotic approach similar to that advocated by Beattie, contagious magic would become a case of metonymy: calling the whole by the name of a part to dramatize the enemy's suffering. But Gell rejects this: ‘These exuviae do not stand metonymically for the victim; they are physically detached fragments of the victim's “distributed personhood” ’ (104). Like the classic Peircian index, the nail clipping is an index of the person as smoke is of fire. A Malangan funerary carving is a skin for the deceased (cf. Lucretius), which creates new skins in the memories of the onlookers. ‘Thus memory becomes a socially engineered medium for the transmission of the power to change the world’ (227). According to Gell, all this happens without visual communication.

Gell then moves to a more complex case, the worship of images. Because ‘fertility can be represented – i.e. objectified in an index – it comes under the control of those who control the index, the priests’ (107). Hindu images are worshipped to gain blessing (darshan) that is conveyed through the eyes of the image [icon]. Ancient Indian philosophers held that the eyes send out invisible beams. Darshan is ‘the gift of appearance’ imagined as a material transfer. Some Indian philosophers compare darshan as ‘seeing’ to the way in which the blind use a stick to ascertain the shape of objects. In relation to images, there is thus no distinction between similarity and contact (that is, icon and index).

Freedberg asked whether religious and other images are efficacious because the rituals of consecration have endowed them with power, or because they are images linked by mimesis to what they represent. Freedberg chose the second solution and argued that they are images with signifying functions (Freedberg 1989; Gell 1998: 150). Gell acknowledges Freedberg's very pertinent objection that collapsing art objects into real persons risks losing sight of the specificity of art but opts for the first, on the tendentious grounds that the anthropologist deals primarily with human beings themselves playing a role, rather than portraits and effigies. He goes on to discuss a case where the deity is embodied in a living person, a living icon. The transition between a person and an image is ‘insensible’; the position they occupy in networks of agency ‘may be regarded as almost entirely equivalent’ (153).

Gell claims that the Tylorian concept of animism can be made more serviceable if it is removed from its pejorative Victorian context. None the less, I find this stage in Gell's argument weakened by his use of the Victorian strategies of either imagining oneself in the position of a member of another culture (125), or drawing a parallel between the behaviour of Western children and adults in other cultures (129, 134). Gell's strategy of asking how people believe that art objects have agency is, however, a useful one. The Lucretian doctrine of images as cast-off skins is similar to the Australian one of increasing a species by rubbing or smoking the rock (or painting) which is the transformed body of the totemic ancestor. In both cases the underlying causal theory is unlike those familiar to Euro-Americans, and is based on a different ontology. In Aboriginal theory, energy and matter are interchangeable (Layton 1995); increase rites are not symbolic, but rely on cause and effect. Following Horton, Frazer's mistake was to assume that there was no theory of being (ontology) underpinning magic. In religious thought, Horton argued, social relations are taken as the model on which the working of the world is conceptualized. This is preferred in societies where social relations provide ‘the most markedly ordered and regular area of their experience’ (Horton 1964: 99). Horton's theory of religion implies that people attribute agency to sculptures because their explanatory model of how the world works is an inherently religious one: the world is animated by human-like agencies. In these circumstances, the position that a person and an image occupy in networks of agency may indeed be regarded as similar, if not almost entirely equivalent.

If one wants to go beyond cultural relativism, a provocative philosophical and moral problem is encountered. Do we take an Enlightenment view that our explanation is superior to theirs, or a post-modernist view that all theories of being are equal, and only power enables one to prevail over others? Lienhardt's solution was to focus on examples where influencing the participants’ state of mind was, from both a Western and an indigenous point of view, intrinsic to the success of the action, but, like Beattie, Lienhardt considers that the result (such as peace between warring clans) is achieved by dramatizing the desire. Tambiah takes an explicitly anti-Frazerian approach in his reanalysis of Trobriand spells. Because spells express an understanding of the technical properties of the intervening pragmatic activities, they enable yam cultivation, canoe construction, and so forth to be correctly carried out and thus do indeed ensure practical success (Tambiah 1985: 51). Gell (101) cites Tambiah and it is possible that, if he had been able to revise Art and agency, he would have moved further from a reliance on Frazer and Lucretius.

Gell acknowledges that people in non-Western cultures can tell the difference between an image and a person. Many cases of scepticism among the illiterate and uneducated have been recorded; ‘the devotee does know that the image of the god is only an image’ (Gell 1998: 118). If it were to move or speak that would be a miracle. But this puts into question Gell's claim that the position a person and an image occupy in networks of agency can be considered to be almost entirely equivalent. Why bother with images, if people will do? Why not have someone sit or lie motionless in the place of a statue of the Buddha? Why not station people along the roadside to shout at each passing car ‘road works ahead; elderly people crossing’?

The artist's œuvre

In contrast to the problems with the first stage in Gell's argument about the agency that is inherent in art, I find his second stage brilliantly convincing. Agency derives from a position in a network of social relationships. Since we are constructed by our social relationships, our inner personhood replicates what we are externally (cf. Myers 1986). Personal agency creates the distributed art objects that belong to the corpus; our mind becomes manifest in the objects, traces, and leavings that we generate during our lifetime. There is a structural isomorphy between something internal – mind or consciousness – and something external, the aggregate of artworks (Gell 1988: 222). The circulation of indexes steadily transforms agents’ conscious experience. The right to produce a Malangan sculpture, for example, indexes a wider bundle of rights, including land rights (cf. Morphy 1991: 57-74 on rights to produce bark paintings depicting totemic ancestors). Stage two of Gell's argument can stand independently of stage one. It is consistent with Peirce's theory. Peirce regarded an algebraic formula as an icon ‘in that it makes quantities look alike which are in analogous relations to the problem’ (Peirce 1955: 107).

Despite having rejected the relevance of culture for an anthropological theory of art, culture becomes central to Gell's argument in chapter 8. This may be an inconsistency that Gell would have dealt with if he had had longer to revise the book, but I think Gell is here using culture in a different sense from the one earlier rejected. Gell appears to use culture as a way of describing the characteristic way in which a community organizes its social relationships; a way of doing things which demands agency. The style of an art tradition maps out, and can be used to manipulate, social relationships. Gell cites Fernandez (Gell 1998: 153), who reported that among the Fang the production of aesthetically pleasing carvings is an aspect of culture, but is less valued than bringing about harmony in social relations (Fernandez 1973: 194-6).

Gell exemplifies this approach through an extended discussion of Marquesan art and social exchange. Marquesan art was not representational in the Western sense of depicting objects that existed independently of the art. It was a ritual art intended to make the bearer powerful and invulnerable. The limitless variations on recognizably Marquesan themes within the art shows a passion for creating difference which parallels, or has an ‘elective affinity’ (219) with, the social differences created through exchange. Individual art works do not signify social status, they merely advertise their standing as variant parts of the corpus of Marquesan art, through their recognizable style.

Sense and reference

The distinction between sense and reference was identified by Frege (1960).1 A word or picture simultaneously refers to something in the world, and signifies the meaning of that object in a cognitive system. Saussure's primary concern was with sense, or signification, that is, with how sounds are conventionally related to ideas in the structure of the language. Peirce, on the other hand, classifies signs according to the way in which they refer to objects in the environment. However, sense and reference are difficult to discuss in isolation. Saussure is clearly writing about reference when he describes spoken language as arbitrary in the sense that words do not rely on onomatopoeia, but on purely conventional sounds (Saussure 1959: 69): words do not sound like the things to which they refer. On the other hand, Peirce takes sense into account when he writes, ‘the sign stands for something, its object … Not in all respects, but in reference to a sort of idea’ (Peirce 1955: 99). That is, the sign stands for an idea about the object to which it refers. Saussure is ambiguous in his example of what he calls a symbol, ‘the symbol of justice, a pair of scales, could not be replaced by just any other symbol, such as a chariot’ (Saussure 1959: 68). While he apparently intends that the idea of a pair of scales is a metaphor for the more general idea of balanced judgement, he may also have intended that justice could be represented by a picture (an icon) which looks like the object, a pair of scales.

Eco extended Peirce's approach to reference in his analysis of textual interpretation. A text refers to a universe of discourse, and thereby eliminates some possible misreadings (Eco 1990: 28, 60; cf. Ricoeur 1991: 93; see also Layton 2000). Gell is alluding to the second type of reference, reference to an artistic discourse, in his analysis of the relation of individual objects to the stylistic tradition of Marquesan art. He writes, ‘this is the sense in which any part “stands for” the whole as in synecdoche … “Representing” in this sense is clearly a semiotic relation, in which the object is a sign, and the corpus of stylistically related objects from which it is drawn, is what is signified thereby’ (Gell 1998: 166).

Gell's tendency to downplay sense is consistent with his critique of structuralism and the linguistic model, but it is difficult to see how a discussion of both reference and sense can be avoided, if art is about the ability to influence minds.

Construing indexes and icons

Gell makes a good case for art as agency but, to understand the type of agency that is characteristic of art, we need to explore the distinctive ways in which paintings and carvings affect people. One cannot convincingly argue that art objects have the same kind of efficacy as land mines, or that they function in the same way. In what ways is agency mapped out through social relations by means of art objects? The distribution of commissioned paintings and sculptures – Michelangelo's great works, the tondo of the Holy Family, the tomb of Julius II, the Last Judgement, and the Sistine chapel frescoes – do not map out agency in the same way that Hitler's or Stalin's agency was mapped out in the death-camps, the gulags, and the clash of huge armies. Nor does art map agency in the same way that the career of a powerful industrialist is mapped through the sale of commodities, the construction of factories, and the employment of thousands of workers. Of course, the Renaissance Church's power and wealth is manifest in the distribution of Michelangelo's work. Like Marquesan art, Nazi propaganda posters and the art of Soviet socialist realism also map a parallel course to social action, but (while their aims may be congruent) a picture of valiant workers does not affect the viewer in the same way as ten years’ work in the Siberian salt mines. The difference is similar to the difference between a powerful electric shock and the notice attached to a fence depicting a figure falling back under the blow of a jagged flash of electricity. The first is brute energy that will affect anyone who bumps into the fence; the second conveys a message whose stylistic conventions must be understood if it is to do its job. Art objects can be related to their primary agents (the artist, patron, and prototype) because, to the competent viewer, they embody pertinent information. We can develop Gell's analysis by recognizing three distinct ways in which artefacts can extend their maker's or user's agency. First, unexploded bombs and mantraps act to restrict their victim's agency independently of culturally constructed values or meanings. Secondly, money, treasured items, and personal possessions can be exchanged to create and maintain social obligations in the ways discussed by Marx, Mauss, and Polanyi; cultural conventions, such as the distinction between prestige and utilitarian goods, must be understood to appreciate how exchange has its effects. Finally, art objects depend on being ‘read’ correctly to be effective as secondary agents, demanding a semiological approach.

Gell is right to argue that icons and indexes, unlike spoken words, are not entirely dependent on conventional, cultural structures. They are not, however, entirely natural signs. Icons, and the indexes embodied in art, do depend on cultural convention. It is clear from several of Gell's examples (such as the tree saps used in Ndembu ritual) that the significance of indexes depends on cultural convention. ‘The most common case is for the material index to dictate its form simply on the basis of traditional knowledge’ (Gell 1998: 29), surely a property of culture or habitus. Gell writes that where an idol is an artefact rather than a natural object, ‘the nature of agency exerted by the prototype is to cause the artist to produce a religiously stipulated image according to the conventions for such images’ (99, additional bold typeface).

Icons are different from indexes. Icons are not a product or part of the prototype. Icons resemble their prototypes but are not identical to them. ‘The semantic rule for the use of icons is that they denote those objects which have the characteristics they themselves have –or more usually a certain specified set of their characteristics’ (Morris 1938: 24, additional emphasis). Here, too, culture intervenes. ‘The way in which such dissimilar reptiles as tortoises and lizards have become visually synonymous in Marquesan art partly reflects their similar symbolic associations’ (Gell 1998: 180 n.). The image must often ‘look like’ what it represents, but according to cultural convention. Totemic art in northern Australia, and on the Northwest Coast of North America, illustrates this well (Boas 1955: 186-209; Morphy 1991: 155-64; Taylor 1996: 147-68). To paint or carve another group's totem would be to claim their land. Gell asserts: ‘culture may dictate the practical and/or symbolic significance of artefacts, and their icongraphic interpretation, but the only factor which governs the visual appearance of artefacts is their relationship to other artefacts in the same style’ (Gell 1998: 216). This may be true of Marquesan art, but it is demonstrably not the case where two or more art styles co-exist in a community. Lawal's study of three Yoruba art styles for representing the head includes good examples (Lawal 1985). The difference between silhouette and geometric art among the Yolgnu, or ‘X-ray’ and geometric infill in Western Arnhem Land rock art, is partly a matter of stylistic convention but it also relates importantly to the kind of information the art is intended to convey (Morphy 1991: 176-80; Taylor 1996: 224-38).

One of Gell's fundamental objections to the linguistic model was its reliance on structuralism. However, his conclusion to the Marquesan case study is that the multiplicity of variant forms ‘suggest an overwhelming need to establish difference and a recognition of the merely relative character of all differences’ (220). This is surely a structuralist proposition. Shortly afterwards, Gell refers to ‘the structural isomorphy between something “internal” (mind or consciousness) and something “external”– aggregates of artworks’ (222).

Gell writes: ‘Semiotic/interpretative theories of art give prominence to the fact that what a person sees in a picture, or, even more, gleans from an utterance or a text, is a function of their previous experience, their mind-set, their culture, etc.’(33). He construes this approach as a denial of the intrinsic agency of art objects, implying that it stems from the postmodern argument that the reader or viewer can make whatever they choose of objects in a gallery. The connection with current Western notions of individualism is, as he says, obvious. But this is not adequate to dismiss the semiotic argument that even to construe the artist's agency correctly the viewer must rely on ‘their previous experience, their mind-set, their culture’. Campbell spells this out very clearly in her analysis of the art on Vakutan Kula canoes. The process by which specific animals in the Vakutan environment are represented is mimetic. The system by which form and meaning converge is symbolic, the means aesthetic (Campbell 2001: 133). In chapter 4 of Art and agency, Gell criticizes Western misunderstandings of West African nail fetishes. He writes, ‘the apparent rhyme between these carvings and Western images of suffering and violation is fortuitous’ (59). The point is that we interpret them according to our experience of the Christian tradition, whereas those who were intended to view them rely on their own cultural tradition. As an unrepentant semiologist, I believe that what Gell has identified as the distinctive features of art cannot be understood except by recognizing the status of art as a culturally constructed medium of visual expression. We should also recognize that a specifically linguistic model is of very limited value in explaining how art objects can extend their maker's or user's agency.

Art and agency revisité


Dans son livre Art and Agency, Alfred Gell présente une théorie de l’art qui ne se base ni sur l’esthétique, ni sur la communication visuelle. Il définit l’art par sa fonction distinctive dans l’établissement de relations sociales, par « l’abduction de l’intentionnalité (agency) ». Les objets d’art sont des index de l’intentionnalité de l’artiste ou du modèle. Le présent article analyse l’utilisation par Gell de l’intentionnalité, notamment dans le cadre de l’art rituel qui constitue un axe central de son raisonnement. En se concentrant sur l’usage par Gell du terme « index » de Peirce (dans la trichotomie index, icône, symbole), l’auteur note que l’approche de Peirce prête moins d’attention à la signification qu’au lien entre les œuvres d’art et les objets auxquels elles font référence. Il examine ce que Peirce entendait par « abduction » et en conclut que si Gell s’en tire bien sur l’intentionnalité des objets d’art, il n’explique pas de quelle manière distinctive ceux-ci prolongent l’intentionnalité de leur créateur ou de leur utilisateur. Gell n’a pas eu le temps d’apporter des révisions détaillées à son ouvrage avant publication, et l’auteur estime que s’il avait eu davantage de temps, il en aurait peut-être remanié certaines parties.


  1.  1 Ricoeur writes of Frege's famous article, ‘Über Sinn und Bedeutung’, that ‘[These are] expressions which Peter Geach and Max Black have translated as “sense” and “reference” ’ (Ricoeur 1974: 86, citing Frege 1960). Harrison similarly writes, ‘Frege introduces the notions of sense and reference by considering pairs of descriptive expressions which uniquely specify the same object … “The Evening Star” and “The Morning Star” … denote, or refer to, the same object … a certain planet (Venus)’ (Harrison 1979: 55). In the third edition of Frege's papers, however, Geach and Black translate Frege's Bedeutung as ‘meaning’. They explain that ‘Philosophical technicalities, like “referent” … would give a misleading impression of Frege's style’ (Frege 1980: ix). The key passage now reads, ‘The meaning of “evening star” would be the same as that of “morning star”, but not the sense’ (Frege 1980: 57). It remains clear that what Frege intends by ‘meaning’ in this passage is what is now conventionally called ‘reference’ (see e.g. Quine 1960: ch. 3).