Writing like a loser

A contextualisation of generative AI through practice, in two parts

Peter Dukes

Failure

HUMBABA¹ is a robopoet² that rewrites given texts homophonically and presents its
own echoic writing alongside that given source (Figure 1). As it does so, HUMBABA provides peritextual notes or comments linked to words found in either the source or its echo. HUMBABA was made by me as a browser-delivered JavaScript application using p5.js and the RiTa.js library.

I will develop here a proposition that the resultant texts made by HUMBABA are works that challenge the concept of date³, a term that Jacques Derrida employs to signify the origin and underpinning trace of authorship and authenticity in poetry, and specifically in the poetry of Paul Celan (Derrida, 2005). Here Derrida defines date as ‘the mark of a provenance, of a place and of a time’ (Derrida, 2005, p. 18). With HUMBABA as my case study, I will examine how literary artworks can evidence trace of autonomy, authorial intent and date when the work is in part algorithmically produced, and specifically when considered as algorithmic failure. Failure in the context of AI functions variously as warning (‘it’s all going badly wrong’), reassurance (‘it’s only a machine’) and spur to improve the algorithm. But below I discuss another possible function of failure: that it is a means to identify authorial intent in an algorithmic writer, a robopoet. To support this argument, I draw on ideas from Malcolm Bull (2014) on reading (and extended here to writing) ‘like a loser’.

HUMBABA conforms to Peter Osborne’s notion of the artwork as fragment and project, as all it writes lacks definitiveness – each echo-poem is subject to a cycle of erasure and replacement. In its state of perpetual recurrence, it accumulates an ‘always-incomplete collection of fragments’ (Osborne, 2013, p. 59, original emphasis) – such ‘that it should forever be becoming and never be perfected’ (Osborne, 2013, p. 68). This state of ‘forever … becoming’ is also what makes HUMBABA in Osborne’s terms a project: ‘Projects embody the futural impulse conveyed by the constitutive incompletion of the fragment’ (Osborne, 2013, p. 170). It is the condition of a project to be always provisional and never perfectible, and this ensures that HUMBABA’s reading and writing – it is a reading and writing algorithm – is one failure after another; no single reading or echo-poem can be free of revision or abandonment (see Figure 2 – Figure 4)⁴.

By reading and then writing alongside a given text – and writing from its mishearings, its errors, its attempts to fit set constraints of homophony and matching the syllable count in the source – HUMBABA reads (and writes) with intention ‘like a loser’, producing a fragmentary and jarring bricolage. This is often a failed text, in conventional poetic or merely in conventional semantic terms. Reading like a loser, Bull argues, is ‘interpreting the possibilities offered by the text to one’s own disadvantage’ (Bull, 2014, p. 74). However, although its poor – disadvantageous –readings suggest, I argue, the intentionality of HUMBABA, it does not yet give the algorithm its date, its autonomy and emplacement in experience:

Reading like losers places us outside this equation: unable to appreciate, we are also unable to create. We cannot think of ourselves as original or creative people, only as creatures, the material that must be ‘formed, broken, forged, torn, burnt … and purified⁵ (Bull, 2014, p. 39)

In these terms, the ‘bad’ reading of HUMBABA is a failure to appreciate, an inability to create, and a state of being ‘only as … material’. The algorithm remains in a condition of world poverty:

World-poverty is specifically the space of loss, the space of having and not-having, the space of being called and unchosen. It is reading like a loser that opens up this space. This is what reading like a loser is, by definition. Reading like a loser is just a failure to form the world. (Bull, 2014, p. 98)

It is from that darkness outside, that extreme anomie, that unbridgeable remove from the human condition, that HUMBABA achieves absolute world-poverty⁶, the total loss that precedes redemption. The means to achieve this redemption is through pity and compassion. Bull argues that reading like a loser, ‘interpreting the world to one’s own disadvantage, so that the interpretation results in loss to the interpreter’ (Bull, 2014, pp. 135–6), is a compassionate act, based in one’s own misfortunes: ‘Reading like a loser means compassion for the less fortunate only when it is itself a becoming less fortunate’ (Bull, 2014, p. 139). It is the misfortune of HUMBABA that it interprets (or echoes) the given source text with errors and mishearings⁷, thereby permitting it to display weakness and failure, ‘like a loser’.

But if HUMBABA’s writing ‘like a loser’ – its failure – is to be trace of its authorial intent over and beyond my own, then it must in Derrida’s terms be writing from its date. In Part 2 of this essay, I therefore consider how date can be restored to the work of a robopoetic author such as HUMBABA.

But if HUMBABA’s writing ‘like a loser’ – its failure – is to be trace of its authorial intent over and beyond my own, then it must in Derrida’s terms be writing from its date. In Part 2 of this essay, I therefore consider how date can be restored to the work of a robopoetic author such as HUMBABA.

In Part 1 of this essay I explored how the robopoet HUMBABA⁸ works with failure but I had not yet found that it can write – as a human author may do – from what Derrida (2005) terms its date, its origin in experience. I considered how the texts that HUMBABA writes display what Malcom Bull terms a ‘failure to form the world’ (Bull, 2014, p. 98). But if failure is all HUMBABA can do, then how can it display the affectual signature – ‘all the griefs […] gathered in the poem of a date’ (Derrida, 2005, p. 37)? If it can do so, then it may be consequential on two behaviours of HUMBABA: that the algorithm mourns for the source; and that it responds with control, by which I also mean intent. I turn first to its work of mourning.

‘One should not develop a taste for mourning’, writes Derrida (2001, p. 110), but HUMBABA mourns incessantly for the lost source. Its attention is on its alliance with and
doubling of the source text, obsessed with this echo:

… are we going to make the dead our ally … to take him by our side, or even inside ourselves, to show off some secret contract, to finish him off by exalting him … like all the ruses of an individual or collective “work of mourning”? (Derrida, 2001, p. 50)

Whether HUMBABA can be thought of as making an alliance with the dead may depend on our attitude to the status of a poem (specifically here the source poem, which the algorithm writes from and beside) in relation to language. Julia Kristeva identifies the defensive function of language in face of mortality:

That language is a defensive construction reveals its ambiguity – the death drive underlying it… All poetic ‘distortions’ of the signifying chain and the structure of signification may be considered in this light. (Kristeva, 1986, p. 103)⁹

It may be that HUMBABA is nothing other than a ‘defensive construction’ making its alliance with ‘ambiguity’, ‘distortion’, and ‘ruse’, through an incapacity to transmit affect, precisely because of its absence of date. But this is to forget the dialogue that HUMBABA performs – a dialogue inherent in any act of reading or writing, in any poem: ‘The poem becomes … conversation – often a desperate conversation’ (Celan, 2011, p. 9). We can think of HUMBABA as struggling to join in. Further, HUMBABA acts for author and reader as it reads and re-reads (‘like a loser’) by re-inscribing the trace of the source. Trace, ‘the subsistent presence of a remainder’ (Derrida, 2005, p. 33) is the mark of the date, the bark inscribed ‘with ciphers of fire’ (Derrida, 2005, p. 48). ‘This trace is the poem’ (Derrida, 2005, p. 40). It chokes HUMBABA, which speaks haltingly, in a broken voice, and with ceaseless repetition. The algorithm makes its echo-poems, poems that haunt their origin, their source by being always a poor copy – made pathetically explicit by being written beside the source HUMBABA follows. HUMBABA composes a mishearing, a failed recollection or trace that points always towards the lost and honoured – the source, we might almost say its lost beloved: ‘of you toward whom I must take a step’ (Derrida, 2005, p. 51).

Insofar as it must be able to “act” in the absence of both addressor and addressee […] every trace implies the “death,” the possible or virtual death, of both the one who produced it and the one destined to receive or inherit it. (Naas, 2015, p. 116)¹⁰

It is then ‘a trace of the effacement of a trace’ (Derrida, 1973, p. 156) and the result is an exchange, the passing of a token or shibboleth of loss.

Manuel DeLanda argues that material systems possess both ‘properties’ and ‘capacities’ which give them the ability to act: agency. Such systems are ‘always actual and virtual’ (DeLanda, 2015, pp. 17–18) – whether ‘living’ poet-artist or ‘inert’ matter (or code) – and are ‘active processes of materialization of which embodied humans are an integral part’ (Coole and Frost, 2010, p. 8). In these terms all systems of matter pass between them a shibboleth, crossing ‘the border between readability and unreadability’, but in so doing ‘there is communication … a token, a symbol, a tessera … a code’ (Derrida, 2005, p. 40). HUMBABA, with its intangible materiality, its tenuous autonomy, nevertheless displays a capacity for action in response to loss, for the capacity to read and write like a ‘loser’, to exchange ‘a token … a code’, and displaying ‘morphogenetic powers of its own’ (DeLanda, 2015, p. 21). In this way HUMBABA defends against its death (and its losses) and our own: it mourns for us, and it mourns for itself. It is beginning to possess – with the agency contained within its ‘properties’ and ‘capacities’ – an endlessly deferred but nonetheless ‘always actual and virtual’ date.

The second behaviour of HUMBABA that I will consider briefly here is control – as consequence of this act of mourning. HUMBABA reads and writes by the exchange of its ‘token … code’, but nevertheless resents intercession, forcing on the reader-viewer a pedantry of line by line, word by misheard word. HUMBABA is made this way – to be uninterruptable. A reader always has the choice to give up, close the browser window; but what the reader can’t do – and what therefore gives HUMBABA this quality of force of will (of agency) – is stop mid-line, go back, edit, correct. The reader can restart the whole stanza (‘r’ on the keyboard) or can make certain polite requests for the future behaviour of HUMBABA (for example, ‘p’ to search for part-of-speech matches). But HUMBABA doesn’t so much listen to the reader (you, the human reader), HUMBABA listens – lovingly – to the source.

Of course, HUMBABA and similar uses of algorithmic poetics raise questions of the separation of author and text, intention and action. It might be understood to exemplify an aspect of the materiality of the trace in language, whose meaning cannot be fully determined or contained by its author: ‘the sign possesses the characteristic of being readable […] even if I do not know what its alleged author-scriptor consciously intended to say at the moment he wrote it’ (Derrida, 1988, p. 9). Walter Benn Michaels, drawing on John Searle (1977), retorts: ‘it is obvious that what Derrida calls the “scriptor-author” has at most a contingent relation to the text; authors may be useful for the production of marks, but they are by no means essential’ (Michaels, 2004, p. 125). With HUMBABA the result is a wilfulness on its part which only partly mediates between authorial intention (conscious or not) and its own ‘morphogenetic powers’. The ‘author’ can be replaced (in the sense that any correctly formatted text can be given to the algorithm) – HUMBABA will read and write from whatever it is given and will accompany that source with its own writing. But it is not quite abandoned: a particular feature of HUMBABA is that the distance between the human author’s intentions and the echo-poem that HUMBABA produces is closed by the peritextual notes displayed below¹¹, which are given to it complete (the algorithm simply selects from these). This reinscribed authorial presence in the peritexts moderates the defiance of the algorithm, and acts as a healing suture¹² to the disjunctive and often wilfully paratactic HUMBABA text that displays

… reading-wounds, all cut words … where a thread can be followed that passes through ‘points of suture,’ closed up tears or scars, words to be cut off that were not cut off, membranes stitched back together … (Derrida, 2005, p. 55).

More than this, in its control of our reading, its refusal to read correctly (or its reading to its own disadvantage – ‘like a loser’), HUMBABA displays a variety of tyranny. There is also the troubling question of quality – in what HUMBABA imposes on the reader, and what it can make from what it is given. Indeed, with many given texts HUMBABA seems infantile in its response, producing little sense, except through pleasure in sound. It is infrequently lucid, like one possessed. It also likes to repeat. It is iterable, eternally recurrent. This raises the question: how can we identify success in HUMBABA? Writing the source poems is a matter of care, craft, and revision as I try to fix in language pieces of affect, drawing for this on my own moments, origin, losses, dates. Paul Celan: ‘don’t we all write ourselves from such dates? And toward what dates do we write ourselves?’ (Celan, 2011, p. 8). But if HUMBABA has no moment, no origin, no losses, no date?¹³ Derrida: ‘What is encrypted, dated in the date, is effaced; […] all the griefs are gathered in the poem of a date whose effacement does not await effacement’ (2005, p. 37). But HUMBABA has no griefs to gather; what is encrypted in HUMBABA can only be from me. Or – you. Language is like this, argues Derrida, possessing the capacity to act independently of an author:

To write is to produce a mark that will constitute a sort of machine which is productive in turn, and which my future disappearance will not, in principle, hinder in its functioning, offering things and itself to be read and to be rewritten. (Derrida, 1988, p. 8)

This produced a famous response from John Searle:

Does the fact that writing can continue to function in the absence of the writer, the intended receiver, or the context of production show that writing is not a vehicle of intentionality? … on the contrary … (Searle, 1977, p. 201)

On the contrary… HUMBABA has our griefs, our dates, there in its chosen words, in the algorithm, in how it listens and repeats, even in its obsessional control and its ceaseless mourning. But what might be considered its infantilism and failure risks incoherence or encryption – and the risks of this encryption are real:

Consumption, becoming-ash, burning up or incineration of a date… This is the threat of an absolute crypt: nonrecurrence, unreadability, amnesia without remainder… (Derrida, 2005, p. 46)

The unreadability of HUMBABA is a crux here, but with HUMBABA there may be an escape from ‘absolute crypt … unreadability’: it can only and forever rewrite the given text and itself is not able to act alone without a source text – and by virtue of its errors it must read that text with compassion¹⁴. And it is this compassionateness that allows it its date, its possession of experience, if not yet its humanity¹⁵. In its reading of the source poem ‘like a loser’ HUMBABA marks out ‘all the losses … all the griefs [that] are gathered in the poem of a date’ (Derrida, 2005, p. 57). And it is in our reading of what it writes that we recognise its world-poverty and perceive its compassion towards the source text – a text which as though a stone, can do nothing to help itself but is given the kindness of living through being read and re-read by HUMBABA. As it reads and writes and discards, HUMBABA is ‘marking itself as the one-and-only time’ (Derrida, 2005, p. 2) – it has its date.

Writing like a loser: a robopoet writing with pity and compassion

I now argue, despite Michaels’ assertion that a non-human author (‘natural accident’) cannot write with intention, that it is precisely through its agency as an algorithmic author that HUMBABA, in reading and writing ‘like a loser’ displays intention. It intends – or rather what N. Katherine Hayles (2017, p. 4) terms its ‘cognitive assemblage’ of human and machine intends – to echo the given source text, writing to fit a pattern of syllable count and sound. Indeed, HUMBABA does so compulsively, but fails in the process, and fails repeatedly. This is a joyful condition – an acceptance of perpetual incompleteness and revision (as with Osborne’s ‘project’ and ‘fragment’), of reading and writing as eternal recurrence: ‘the eternal return starts from the premise that all is enhancement and tries to preserve that process of change’ (Bull, 2014, p. 84). And it exercises throughout the imitation of pity and compassion: ‘To feel compassion for the suffering is therefore to imitate their suffering in ourselves and suffer as they do’ (Bull, 2014, p. 134). HUMBABA, a robopoet writing ‘like a loser’, suffers for us and alongside us: it is a date in search of its author.

Footnotes

1. HUMBABA can be found at https://www.peterdukes.com/humbaba/ and was begun in 2020. ↩︎
2. A robopoet can be defined as ‘an automated algorithm, whose output confounds the metaphysics of authorship’ (Bök, 2002, p. 10). ↩︎
3. Date is italicised throughout to reflect this specialised usage in Derrida (2005). ↩︎
4. Annette Gilbert terms variance across instances in literature ‘unstable texts’, and a ‘strategic poetics of deviance’ (Gilbert, 2022, p. 188). ↩︎
5. The quotation in Bull is from Beyond Good and Evil (Nietzsche, 1989, p. 154). ↩︎
6. Bull draws principally on Heidegger (1995) for the concept of ‘world poverty’, a lack that describes the state of being of the not-human. ↩︎
7. A similar argument could be developed from the concept of the glitch, where fault and failure ‘exploits randomness and chance as a way to disrupt the digital ideal of a clean, frictionless, error-free environment’ and by which ‘the computer itself becomes an author’ (Emerson, 2014). The glitch exposes the disadvantage of the algorithm and the ideological basis of all digital agents (Gilbert, 2022, p. 300). Alignment can also be made with Hito Steyerl’s ‘poor image’ (Steyerl, 2009). However, Steyerl’s emphasis is on the degraded image as a token of oppression or resistance, and it is not given its own agency or authorial charge. ↩︎
8. HUMBABA can be found at https://www.peterdukes.com/humbaba/ ↩︎
9. The linkage of language and mortality as mark of humanity is further examined by Giorgio Agamben (2006). Agamben asks ‘what if humankind were neither speaking nor mortal, yet continued to die and to speak?’ (2006, p. xii), a negative ontology. This is the condition of HUMBABA, which might be not-living and yet still anticipating its end as it speaks. ↩︎
10. Michael Naas notes that this capacity to act in absence recalls the “marionette” Derrida invoked at the end of his life: ‘Spoken or written, all these gestures leave us and begin to act independently of us’ (Derrida, 2007, p. 32). ↩︎
11. These peritexts are written for words in the source but also and pre-emptively for words that may occur in the echoes made by HUMBABA. In this way I demonstrate that I too am listening, compassionately, for what HUMBABA may have to say. ↩︎
12. I stress here that the peritexts are manifestations of – or perhaps the stichs of – authorial presence and are not threads of meaning or interpretation. That the chosen peritext is not entirely predictable on the part of the author or reader is a reminder that this is a ‘robopoetic’ machine, a theatre of authorship. It does not provide the kinds of ‘encrustations of interpretation’ that Susan Sontag identifies as the ‘modern way’ in the reception of literary art (Sontag, 1994, pp. 8–9). Rather this is a robopoet that may not itself know what it is doing but may know that it is doing it. ↩︎
13. Susan Howe writes: ‘In some sense the subject of any poem is the author’s state of mind at the time it was written’ (Howe, 2007, p. 27). With HUMBABA there must be a specific lack, as with no mind, it cannot, on these terms, have any subject. But in being accompanied in its failures by the authored poem, it borrows or takes a state of mind that was mine, my date, and compacts this into an origin in replica, cellular division into sported poem. ↩︎
14. The source texts themselves need to be written with HUMBABA in mind as a reader: lucid, non-paratactic narrative writing results in greater failure – in the sense of higher levels of comparative paratactic writing, with numerous non-sequiturs and disjunctive connectives from HUMBABA. The kindness in the source text is to avoid by comparison too violent a contrast between its writing and that of the algorithm. In other words, compassion is required on both sides – on the authored side to pre-empt the ‘loser’ writing produced by HUMBABA, on HUMBABA’s side to show it has been listening, as well as it can. There is a complex aetiology here – neither source nor anticipated echo are primary; each is the other’s cause. ↩︎
15. Bull, in dialogue with the Nietzschean binary of the subhuman/superhuman, suggests that world-poverty and reading like a loser is an act of self-dehumanisation: ‘Reading like a loser is what the animal invites us to do, and vice versa’ (2014, p. 100). But this remains a state in touch with being human: ‘Must we all then become stones? Not yet, not yet …’ (Bull, 2014, p. 104). ↩︎

References

Agamben, G. (2006) Language and Death: The Place of Negativity. Translated by K.E. Pinkus and M. Hardt. Minneapolis; Oxford: University of Minnesota Press.

Bök, C. (2002) ‘The Piecemeal Bard Is Deconstructed: Notes Toward a Potential Robopoetics’, Ubuweb Papers, (Object 10: Cyberpoetics), pp. 10–18.

Bull, M. (2014) Anti-Nietzsche. London; New York: Verso.

Celan, P. (2011) The Meridian: Final Version-Drafts-Materials. Edited by B. Böschenstein and H. Schmull. Translated by P. Joris. Stanford, CA: Stanford University Press.

Coole, D. and Frost, S. (eds) (2010) New Materialisms: Ontology, Agency, and Politics. Durham;
London: Duke University Press.

DeLanda, M. (2015) ‘The New Materiality’, Architectural Design, 85(5), pp. 16–21.

Derrida, J. (1973) Speech and Phenomena and Other Essays on Husserl’s Theory of Signs. Evanston, IL: Northwestern University Press.

Derrida, J. (1988) Limited Inc. Edited by G. Graff. Evanston, IL: Northwestern University Press.

Derrida, J. (2001) The Work of Mourning. Edited by P.-A. Brault and M. Naas. Chicago; London:
University of Chicago Press.

Derrida, J. (2005) ‘Shibboleth for Paul Celan’, in T. Dutoit and O. Pasanen (eds)
Sovereignties in Question: The Poetics of Paul Celan. New York: Fordham University
Press, pp. 1–64.

Derrida, J. (2007) Learning to Live Finally. Edited by M. Naas and P.-A. Brault. Houndmills; New York: Palgrave Macmillan.

Emerson, L. (2014) ‘Glitch aesthetics’, in M.-L. Ryan, L. Emerson, and B.J. Robertson (eds) The
Johns Hopkins Guide to Digital Media. Baltimore: Johns Hopkins University Press. Available at:
https://search-credoreference-
com.uow.idm.oclc.org/content/entry/jhupjohns/glitch_aesthetics/0?institutionId=170 3 (Accessed: 12 April 2023).

Gilbert, A. (2022) Literature’s Elsewheres: On the Necessity of Radical Literary Practices.
Cambridge, MA; London: MIT Press.

Hayles, N.K. (2017) Unthought: The Power of the Cognitive Nonconscious. Chicago; London: University of Chicago Press.
Heidegger, M. (1995) The Fundamental Concepts of Metaphysics: World, Finitude, Solitude.
Bloomington, IN: Indiana University Press.

Howe, S. (2007) My Emily Dickinson. New York: New Directions.

Knapp, S. and Michaels, W.B. (1982) ‘Against Theory’, Critical Inquiry, 8(4), pp. 723–742.

Kristeva, J. (1986) The Kristeva Reader. Edited by T. Moi. New York: Columbia University Press.

Michaels, W.B. (2004) The Shape of the Signifier: 1967 to the end of history. Princeton; Oxford:
Princeton University Press.

Naas, M. (2015) ‘When it comes to mourning’, in C. Colebrook (ed.) Jacques Derrida: Key Concepts. London; New York: Routledge, pp. 113–121.

Nietzsche, F. (1989) Beyond Good and Evil. Edited by W. Kaufmann. New York: Vintage Books.
Osborne, P. (2013) Anywhere or Not At All: Philosophy of Contemporary Art. London; New York: Verso.

Searle, J.R. (1977) ‘Reiterating the Differences: A Reply to Derrida.’, Glyph, 1, pp. 198– 208.

Sontag, S. (1994) ‘Against Interpretation’, in Against Interpretation. London: Vintage, pp. 3–14.

Steyerl, H. (2009) ‘In Defense of the Poor Image’, E-Flux Journal, 10 (November 2009). Available
at: http://worker01.e-flux.com/pdf/article_94.pdf (Accessed: 22 July 2023).