Seminar with Hannes Bajohr "Artificial and Post-Artificial Texts: The Reader’s Expectation after AI"

Welcome everyone. Thank you so much for joining our second bids seminar of the um spring semester. Uh uh my name is Kirsty Whitaker. I'm the executive director for bids, the Berkeley Institute for Data Science. Um really really excited to have so many of you here today to listen to what is going to be I think an extraordinarily relevant question of how we receive and sort of respond to artificial and postartificial texts. Um our speaker today is Hannes Bayor. He's an assistant professor in the department of German um here at UC Berkeley. He's published extensively on the impact of digital writing technologies on language and literature. And he is not only a theoretician but also a practitioner of digital literature. He's published numerous generative works including in 2023 a novel co-written with a self-trained large language model. Um I have edited my introduction. He is also extremely prestigious has won many many prizes. has many many um trophies all over his cabinet walls. But we will have a listen to his presentation. He's going to speak for about sort of 25 minutes and then we'll have lots of time afterwards for questions. Um if you're able to hold your question until the end, that just means we have a nice complete talk. But if there's something that is terribly important for you to interrupt on, please raise your hand and I will run over to you uh with the microphone. A little thank you as well to everyone who's joining remotely. Thank you to all of our remote participants on Zoom and um thank you Hannis and Hickas away. Thank you. Thanks Kirsty for the uh introduction. Thanks for being able to speak here. Um I want to go and start right away because I I hope I will be on time with my manuscript here. So within uh three years basically we have a situation where the widespread availability of generative textual AI uh shows that we are at the brink of a somewhat new time uh one in which the text we encounter may entirely be generated by a machine uh sometimes or even most of the time. So practitioners of the hermeneutic disciplines which is um German literature etc. uh we are called to consider what impact the current rapid advances in machine learning uh might have on the ways in which we understand and interpret text. So in this talk I want to contribute to an answer by rephrasing and thus delimiting this question namely what will be the impact of AI on the reader expectations of unknown text. What happens when we are confronted with artificial text as alongside natural ones, that is those written in the traditional way? How, in other words, do we read a text when we can no longer be sure that it was not written by an AI? And what direction might this increasing doubt as to a text origin take if at some likely point the distinction between natural and artificial itself becomes obsolete so that we no longer even seek to differentiate and read post artificial text instead. that there is a standard expectation that readers have when confronted with unknown texts at all that they are namely natural written by humans becomes observable only once there is an alternative to it where they're also artificial text generated by a machine. This distinction between natural and artificial text was introduced by German philosopher and author Max Benza in his 1962 essay on natural and artificial poetry. In this essay, Benzu considers how non-intentional computerenerated literature differs from intentional literature written by humans. To him, this is clear in the case of natural poetry. For Ben's meaning in language arises from what he calls a poetic personal poetic consciousness that links the speaker's ego to the world, embedding being into science and grounding text and reality. Without this consciousness, words become mere empty symbols devoid of intrinsic meaning. And it is precisely this case that Bender's second category, artificial poetry, describes. Benz defines artificial poetry as literary texts created through the execution of rules or algorithms lacking any connection to consciousness, ego or the world. He calls this a material origin. If such texts mean anything, it is purely accidental and only for a human reader. And we know this from the symbol grounding problem and and arguments like this. Yet far from defending a romantic notion of human creative power, Benza seeks to determine what can still be said aesthetically about a text when traditional categories like meaning, connotation or reference are set aside and that makes him very much a modernist. Benz himself was involved in several experiments with artificial poetry. The most famous one was certainly the stochastic texts which his student tutz produced on the Souza Z22 mainframe computer at the University of Stoutgart in 1959 in Germany in which can be considered the first German language experiment with digital literature recombining lists of words from Kafka's castle in a random way or pseudo random way. It includes phrases like not every castle is old, not every day is old, or not every tower is large, or not every look is free. Very much appetic sentences that in in themselves don't seem very uh poetic, but he published it uh in a selection in Ben's literary magazine, also in 59. The stochastic texts were artificial poetry in Benza's sense. No consciousness or world anywhere to be found. That the computer itself could actually be the author of this text seemed absurd to both Benza and Lutz. But both also knew how it had been produced. Whether its artificial origin can be recognized when this knowledge is missing is less clear. The readers of Agenlick were not compelled to ask this question. An accompanying essay enlightened them as to the details of its creation. But when the following year Lutz generated a second poem according to the same pattern and published it in the December issue of the youth magazine Ya and N, there was no explanation to be found. The poem was placed on page three along the miscellaneia just like any other poem. Only the author's name, Electronis, might have allowed one to guess who or what was behind the text. The next issue solved, but most readers had not even identified as a riddle that a computer had written the poem. Clearly, Lutz was having fun here, as is evident from the ironic captions under a photo of the author, the Soua Z22, and a second poem in the poem's handwriting, as the text reads, that is in a teletype printer. On the same page, he published uh a series of letters to the editor. their writers uh without knowing how it had come about were quite divided in their assessment of the poem. Quote, "Perhaps you should reconsider whether you want to open the columns of your paper to such modern poets complained." One while one another was on the contrary quite impressed by the avantgard stance. Finally, something modern evident in these reactions is what I would like to call the standard expectation of unknown texts. The electronis poem was artificial poetry in Benz's definition, but because its readers were unaware of the conditions of its production, they took it for a natural text and assumed it was written by a human with the aim of communicating meaning. The standard expectation of unknown texts then can be captured as a relationship between two elements which sometimes is extended by a third. First, that the text has an originator that a human or sometimes more than one human wrote it. and second that the text has intentional and semantic content that a communicative will to meaning or sometimes uh just understood as a reference to the world is expressed in it. In some cases there's also a third uh stance the text connection to an author function that this constellation can and should be subsumed under the name of an author which organizes the attribution and circulation of some texts but not all. I leave this aside. This is more thing for literary theory. We can summarize the standard expectation as such. We assume a text was written by a human who wants to say something. That there is a standard expectation at all has only become apparent since there's been an alternative to it. And Benz's conceptual distinction and Lutz's practical demonstration have been illustrative here. As the insensed letters show, to recognize a text as violating that standard as artificial always requires additional information. If it is not provided, the human origin is usually assumed. Passing off an artificial text as a natural one, one could say, is also the poor principle of artificial intelligence as such. Nine years earlier, in an article that became the founding document of artificial intelligence to some degree, Alen Turing had pondered whether computers could ever be intelligent. He rejected this question and replaced it with another. If we assume that intelligence is a property of humans, then all we need to find out is whether a human would consider the computer itself to be human and thus intelligent. The point is not that the answers in the conversation of the imitation game have to be correct, but that they sound human. Lying and bluffing are explicitly allowed. If one wants to examine the reader's expectations of artificial text during his test is still a helpful starting point. First because the setup assumes a teletype conversation and thus equates intelligence with written communication and second because the goal of this communication is to misrepresent signs that are meaningless to the machine as meaningful to humans. To put it bluntly, the essence of AI understood this way is to pass off artificial text as natural ones. It is only worthwhile to make this attempt at all, however, because the standard expectation of a known text is that of a human origin. Artificial intelligence as a project, not in each of its actual instances, is therefore based on the principle of deception from the start. This is an argument by media scholar Simon Natal. I would like to call this position strong deception. Because this conception of AI is essentially deceitful, we can ask whether the expectations of a text could ever change under these conditions. And I think not. It insists that artificial and natural text remain neatly separated so that one can be mistaken for the other. If it is suddenly revealed that a text that a natural text is in fact an artificial one, its readers will feel dis will feel cheated and they will feel so not without reason. Deception here turns into disappointment. We don't know how Tolutz's readers reacted to the revelation that the computer had written the poem, but one can guess if one considers recent cases in which the artist subsequently turned out to be a machine. Most recently, this happened to Japanese author Ria Kudan who disclosed in an interview that she had generated portions of her prizewinning novel using Chipt. Public scone followed and she was accused of fraud. There are many such examples and although these disappointed expectations are usually exaggerated they reveal what was actually expected namely artificial sorry namely natural and not artificial texts. However, neither the need separation of Ben's ideal types nor this expectation itself can remain intact as soon as the exception becomes the rule. That is as soon as we are surrounded by text whose origin is unclear. At first stand, such examples seem to suggest that the reader's expectations of unknown text have not changed since lots time. We assume human origins and communicative intent, which is why deception can be a useful strategy. But in fact, I believe that the expectations are nevertheless already in the process of shifting. And this has become ever more clear in the last three years because the number of computerenerated text is constantly increasing. And because we ourselves are writing ever more with, alongside and through language technologies, we are on the way to a new expectation or rather a new doubt. The more artificial text there is, the more the standard dissolves and the question of their origin must arise even when we normally would not think about it at all. That there nevertheless has been a shift can be explained by the fact that the examples of texts that I've been using so far are quite special ones. They are literary texts, texts that are marked as exceptional in our cultural tradition. They appear to be intended and worked through to the smallest detail and they more than other texts are read as having an author. Despite all the attempts of the literary avantgards to create text without a voice and despite more than 60 years of literary scholarship proclaiming the death of the author, the standard expectation of specifically literary texts includes all of these three elements, namely that they have authors that are humans with some kind of communicative intent. I will come back to this uh what it means for literary writing in the age of AI in a moment. First, however, it is worth taking a look at the other side of the spectrum, at those rather unmarked automated texts that remain in the background, that are merely functional and that do not assert themselves as products of human intent or a strong notion of authorship. For them, the Turing test is simply a false description of reality, and the standard expectation is already in the process of fraying. For there are forms of human machine interactions other than strong reception and other text types than the artificial natural partition would suggest. Especially when engaging with interfaces, we're likely to find ourselves in an intermediate stage between natural and artificial. Here already we can experience a looming shift in the standard expectation. For it is quite possible to know that something has been produced by a non-intelligence machine and at the same time treated as if it were conscious. Um in fact this is quite normal. Natali whom I mentioned before has proposed the term benile deception for this phenomena. In contrast to what I have called strong deception here users are aware that they are being deceived. We understand that Siri is not human and does not have an inner life. But smooth communication with her works only if we treat her at least to some extent as if she had one. Knowing this is not a contradiction that suddenly and unexpectedly destroys an illusion as in the examples of competitions in which an AI participates surreptitiously. Instead, banel deception becomes a condition of functionality. If I do not play along, Siri will just not do what I want. The situation is similar with written text. It starts with a dialogue box on the computer screen. After all, the question, would you like to save your changes? Enables an interaction that is basically similar to one with a human being. The answer yes has a different effect than the answer no. And both lie on a continuum of meaning that connects natural language with data processing without one suspecting any meaningfully strong notions of intent behind it. This would already lower the expectations of unmarked text where we act as if we expect human meaning and a conscious interest in communication. We kind of bracket the question that there really must be such things involved. This bracketing means discarding the third element namely authorship while entertaining the second one intent in an assertive modality and possibly the first one human origin in a fictional one. This bracketing, however, does not always proceed smoothly. But null deception is an as if that demands of us the ability to hold a conviction and its opposite simultaneously. The self-contradictory position quickly gives gives rise to a doubt. The more convincing artificial texts become, and the more the aesthetic impression they make on us suggests something like humanlike intent, the more difficult does it become to feel comfortable in the limbo into which benell reception lures us. If artificial texts become indistinguishable from natural ones and if moreover we know that computers are capable of writing them, a new standard expectation of unknown texts lies before us. It is the doubt about their origins. Rather than taking a human source for granted or simply deferring the question, the first thing we would want to know about text would be, was it made by a human or a machine? With the increasing integration of LLMs into existing software, it becomes increasingly difficult for readers to clearly classify such text as either humanmade or machine generated. The stakes may seem relatively low when it comes to say LLM written marketing pros. But what about uh the lawyer's letter that might be automatically generated even though it's about my own personal case? What about political articles or fake news stories? What about the private, personal, intimate email, the love letter? Are those AI products too, in whole or in part? At least one reason for the discomfort these ideas evoke is that people have a stake in what they write, and to varying degrees, they vouch for their words. And while scholars of literature have learned to read without an eye to what an author wants to say, and merely regard the semiotic interplay of signifiers, this is still the mode in which everyone else reads almost any written document. Even if a text ultimately turns out to be inaccurate or misleading, the standard expectation that a recipient brings to reading it involves the assumption that the author is making what Jurgen Haram has called validity claims among which the validity claim to truthfulness or sincerity is the most relevant in this context. Essentially, it means that we have at a we have a basic level of trust that speakers or writers mean what they say rather than try to deceive. This is the reason that reading critically has to be learned at all. Whether or not readers ultimately judge a text as assertion to be true, they tend to assume the existence of a writer who does and that is truthfulness. If truthfulness is thrown into calamity over large language models um when they can generate texts that appear to have been produced and sanctioned by an author, the same obviously can be said of truth in an understanding of correctness. Another of Haramas's validity claims made in speech acts. We know that LLMs are still lacking when it comes to the handling of knowledge understood as reporting pre-established facts or correct data. If knowledge is simply the probability distribution of tokens over training data, it may merely learn from the the form for instance of a specific genre without any uh insight. For instance, the I mean Galactica is always the example here with the scientific papers that were produced. The current situation is thus both a crisis of truth and of truthfulness. To maintain the standard expectation with its separation of natural art and official text and its assumption of human origins, intent and especially for mark literary text authorship is a challenge under these circumstances to say the least. Given the crisis of the standard expectation, it's not unreasonable to suggest that it is already shifting from the conviction that a human being is behind the text to the doubt of whether it not might be a machine after all. But this would also make the distinction between natural and artificial text increasingly obsolete. We would then possibly enter a phase of host artificial text. After first the tacic assumption of the human origin of a text and second the doubt about its origin, it would be the third expectation of unknown texts. For doubt about the origin of a text, like any doubt, cannot be permanent. Humans have an interest in establishing normaly in reducing complexity and uncertainty to tolerable levels. Already mechanisms are put into place to keep that doubt in check by assurance or by decree by digital certificates, watermarks or other security techniques designed to increase confidence that the text at hand is not just plausible nonsense. However, to quote Hegel, one bear assurance is worth just as much as another, and the law does not abolish the crime. The fact is that technical checks can always be circumvented and there's currently no shorefire way to detect AI generated text. In fact, there are good reasons to believe that none are possible in principle. What this means, however, is that once it is possible to question whether a text might have been generated by a machine rather than written by a human, no certificate or law can extinguish that doubt. It can neither be resolved nor as a belief uh born permanently. And then the only solution is to undo its premises. Should political regulation and technical containment fail, then it is not unlikely that the standard expectation itself will become post artificial. Instead of suspecting a human behind the text or being haunted by the skepticism as to whether it was not a machine after all, we simply lose interest in the question. We might then focus only on what the text text says and not on who wrote it. Post artificial text would be agnostic uh about the origin. If the standard expectation of unknown text is shifting, if it is increasingly riddled with doubt, perhaps perhaps even capitulating to an agnostic position, why the ostentatious excitement over generated texts and literary competitions. It is, I think, because literature is slower than other forms of text. And this is because Benzinwithstanding of all text types, it is special. It is marked in that it makes the most emphatic claim to a willful human origin. All the while connecting it more forcefully than any other type of text to the historically grown notion of authorship. I've said already that there are texts today whose origins do not pose a question. A street sign has no author and in our daily life a new site's weather forecast is also practically authorless. Until now, however, we have always assumed that a human being is behind it. But under post artificial reading conditions, nothing much changes if we simply make no assumptions at all. My expectation is that more and more texts will soon be received in this way. Put differently, the zone of unmarked texts is expanding. Not only street signs, but also blog entries, not only weather forecasts, but also information brochures, discussions of Netflix series, and even newspaper, entire newspaper articles would tend to be unmarked. then it is not unlikely that they too become writerless and in many cases authorless. Literary texts on the other hand are still maximally marked today. The consequence of this marketness is that art and literature themselves have recently become the target of the tech industry. Probably nothing would have would prove the performance of AI models better than a convincingly generated novel. Ultimately, however, this goal is still based on the paradigm of strong reception. This seems to to me to be of little use when it is the difference itself that is at issue. More interesting then is the question of the circumstances under which this difference becomes irrelevant. I cannot go into a lot of detail here but maybe give you some bullet points. As has often been remarked the standardization inherent in AI generated texts arises from their tendency to produce statistically probable predictable outputs. This might make them particularly suited for for instance uh genres defined by recurring elements such as uh what we call genre fiction, fantasy, sci-fi etc. Tools like pseudorite already enable authors to create such literature with increasing efficiency potentially transforming them into brandlike entities overseeing largely automated production. In contrast, one might speculate that there are two types of literature most likely to res to resist this trend. The first would be to emphasize human origins through genres that need the person behind the text. Autography, sorry, memoirs or autofiction. Autofiction, the one where the difference between the physical writer and the author or the the narrator is no longer clearly delineated. Here in these cases, it really matters who speaks. The second strategy would highlight unpredictability by an experimental unconventional language use that defies AI's prob probabilistic norms. We can think of digital literature that critically engages with the interplay of natural and artificial elements as seen in the works of Crystal Miller who already writes outside the box of predictability by not using letters at all. Highlighting the artificiality of a text on its text production means offering a form of resistance to homogenization marking marking human machine collaboration as a deliberate and analytical act. This leads us into highly speculative territory. I'm not suggesting that narrative or broadly speaking conventional literature is now doomed or that only experimental or explicitly digital literature is worth pursuing. Nor do I mean to imply that post artificial texts are necessarily bad. One can certainly enjoy reading them, discuss their merits, and unravel their interpretive dimensions. Here I've been primarily interested in analyzing tendencies. And for this purpose, it is worthwhile to consider possible extremes. As far as literature is concerned, there is of course a third possibility that AI may through some technological innovation be steered to produce less probable and more interesting output without losing all the advantages that the power of normalization provides in coherence and meaning production. It is a matter of optimism or lack thereof to consider this a likely or unlikely future. Seeing that sufficiently large OLM are still tied to capital interests, I remain somewhat apprehensive. In this talk, I wanted to try to think about how language is changing in the technical age we inhabit today and which will continue to unfold both without fearing technology but also uh refusing to succumb to it some of its ideologies. In that context, one thing seems certain to me with the increasing penetration of language technologies with the triumph of AI models our expectations as readers will change. Thank you. Thank you. Thank you so much. Um I am curious to know if anyone has any questions. I'm going to come to you with the microphone so that the folks online are able to hear your question as well. effect. As you know, artificial intelligence systems are trained by scraping the language from humans on the internet. Soon you're going to having so much computerenerated um text and they'll be scraping from other computers on the internet which is not unlike an audio system in a feedback loop. I'm wondering if you have done any work uh to uh determine what the likely outcome is to be when computers are st start training on each other. Right. What you're referring to is a is a a phenomenon called model collapse. The idea that uh the more you train an LLM on text that already is a product of LLM, the kind of variety the stat statistical variety of difference is reduced to kind of the the middle of the uh kind of the the distribution. Um I'm I have I have looked at this I have looked at this as a practitioner of literature and the in the novel that you mentioned at the beginning. Um I try to see to what degree you can make narrative with AI and the last chapter is trained on all the output of the first couple of chapters and you can really see how this does collapse in the end. Um, but this is kind of a an artistic way to deal with this. And I'm I'm just trying to to show what happens. As far as I know, and I mean, you all know this much better than me. I guess I'm not a I'm not a computer scientist, this is kind of a extreme possibility that can be mitigated somewhat, right? That that uh if you really only use uh synthetic text, you will have some type of normal collapse. But if you spread it out a little bit or try to mix it up, you can probably go around this. What is interesting in in questions of uh selfless what what Benjamin Britain calls the eurogoras problem like the the snake that bites his own tail is that um you become you get these models that kind of are very different difficult to change in their outlook they are kind of conservative and uh what I find interesting is how language change novel for instance novel pronouns right how they enter into into data sets and maybe don't or are not able to change anything really in the in the in the the final output because they're just such a little uh set of that data set. This might be mitigated for but that is a decision you have to make right and I think one of the problems that that we have is that um if uh if you simply follow the law of scale then novelty in language change will not really be represented in in in future models. Fernando, hold tight. I'm coming to you. Thanks for the wonderful talk. My name is Fernando. I'm curious. You just mentioned kind of an issue of temporality and and one of the beautiful things about human generated language and literature is that it comes from a time and a place. And to first approximation, most of these models are trained in one giant smoothie. It's a smoothie of all language they can find. Right. I'm curious whether you have any experience or whether there's any work being that has been done on on finding the temporal or spatial character of a language of of of these kinds of texts. We we read literature from the 17th century. It sounds a little weird to modern speakers of any language because our modern languages evolved, but it has a character and it has a beauty of its own. And so I'm curious kind of what's been done in that that direction. That's that's a really good question. I I don't have a good answer to this, but I think you can the question of time is is is a fascinating one because you can ask it in several ways. You can ask to what degree uh does a model as I just said capture language change and then you do have a strong difference between the model trained on 20th century literature and the model trained on 18th century literature. But you can also ask to what degree does it itself have history. And I always wondered about this because one argument would be that as soon as you have kind of a destructive training process through back propagation uh it deletes its own history over time right and you you that that is you only have kind of a what uh medieval scholasticism called the nst stance standing now that is as it were moving forward um it is this is a problem that I also looked at in this in this uh novel because it's kind of one one of the hypothes ES it tried to engage with was by literary scholar Angus Fletcher who's I think at at Chicago who says because AI models or language models are correlative they're unable to produce causal relationships narrative is a series of causal relationships therefore AI cannot produce narrative and what I find interesting in in testing this with this novel is well somewhat right but there is an incredible large um tolerance for bearing the absence of straightforward causality and the role of the reader and this is kind of a an old chestnut of of reader response theory and literary studies is that most of this is produced by us. The act of reception is the one that produces causality and projects it onto a text. So to to which degree causality is present in the model itself is actually quite hard to answer because the reception itself changes this. Um and also we can say you know we are now uh uh uh we have a long history of of texts that are not narrative like you have modernism you have Gretstein try to read that and you know make a make a clear chronology out of it it's difficult but so these uh these reception tendencies are also malleable and this is partly something I'm I'm interested in looking at any other questions yeah and folks online please uh put your hand put your question in the chat and we can ask it too. Go ahead. Thank you. Uh thanks for the great talk. I'm curious if you could comment uh along with in the age of post artificial texts how you mentioned things like memoirs and autofiction are cases where the origin of the author is important. Could you comment on non-fiction more generally where as a reader a lot of like the truth or validity of what you're reading is hard to verify so you lend credibility based on the author's qualifications. Could you comment on how that changes? Yeah, I mean the question is to what degree um we can I mean you always read within a context right you you give uh credibility to texts both on their internal consistency and also the context in which they appear. If you have a publisher of you know well-known and well reggarded scientific text then this is a good marker. This will obviously still be the case but maybe this needs to be ramped up somewhat. uh these kind of contextual uh verifications and justifications of the veracity of the facts in in a text. What I'm interested in here is that we can once you have that doubt that it could be generated by an AI, it is actually very difficult to get it uh to to um uh uh mitigate for that technically but also legally. So I'm I'm trying to think at extremes here, right? If there's nothing you can do um and you cannot outlaw it and there's no tech technical solution, you can still rely on you know um reputability and so on but in the end the doubt will remain and what happens then and then the answer is this idea that well maybe we can we can move a little bit in in our reception. So maybe you will have to do more work in trying to uh kind of verify facts in in text like this. But in literary texts, this is slightly different question where you just have to step back and let the text be as it were and uh just judge it on its own merits. Um any other questions? I'll I'll ask a question. So, my um I work in data science now, but I um my PhD was in developmental psychology and neuroscience. And I'm wondering um when you think about the reader engaging, how do you think that changes across the lifespan of a a human child? So, so are there responsibilities that we should be taking to support children as they engage in stories and as they engage in sort of understanding that it's not necessarily true because it isn't it is not supposed to be a a fact-based communication. Um, and do you think there's a sort of almost like a change in how we approach like education for human beings and that you can then expand out to like all of us human beings that have existed pre large language models? Um, but maybe to constrain that question, what do you what do you think it means for children? Yeah, I mean um when it when it comes to entertainment and uh there's there's been I mean there's a lot of AI generated children's shows on on YouTube and they are completely incoherent. They're absolutely nonsensical. It's like I don't know if it's it's like a Peppa Pig version of you know it's it's it's it's something that maybe might be a story but it's it's it's kind of unhinged and uh there's there's obviously I didn't I didn't know that. That's amazing. Sorry. So I mean a lot of parents bonk their children in front of a computer and just press autoplay on on YouTube and you get all this and that is obviously a problem right what what kind of you know talking about causality what kind of connections do you make with this as a developing brain and I think that there probably is some some danger there the other thing is maybe you can talk about variety maybe you want to make sure that the kind of media diet is is more varied than this if there's a kind of indeed a streamlining or normalization of of content. Um I mean I'm not a child developing psychologist. I I can't really give a good good answer to this except that yeah obviously media literacy is is important. Um and uh trying to make sure that exactly this these new tools that we need in order to deal with the uncertainty of the origin of something uh are taught that you are aware of this. It could be something that is that is produced and in the way in a way I think in the long run this is training us all collectively as a society to be more critical towards towards something like this if we cannot trust sources anymore. Um one way to mitigate for it is to kind of always be on the lookout for it. Um I don't know what it would mean for education in general but essentially like it becoming more concrete about what it is that we're imbuing with that with our trust. Right. Yeah. And then this is the one I mean you still need to be critical. But on the other hand at some point if if this is really too much to do on a daily basis I think this you know more narrow focus on the text itself seems the natural reaction to it. Does this make sense or is it is it complete, you know, incoherence uh bunched together from a from a bunch of Peppa Pigs episodes? Yeah. Yeah. Any other any other questions? Yes. Uh I'm going to come to the back and then come to you. Yeah. Um, so I guess judging a text based on just like the merits of what it says is a very timeconsuming thing where you have to spend a lot of time like thinking about what um like what is the context for this and and like what are what is it actually saying and what do I think about it? And I think a large part of how we get around that currently is by using like authorship or if not like a specific author like maybe like a a publication or something that we trust to sort of like go through that initial steps of like is this actually worth engaging with and trying to figure out what I think about it. So I'm curious what what do you think that looks like after if we are sort of moving past having that likeation maybe or Yeah. And probably it is it will become much more important in a lot of ways to to have this context. uh but again maybe this model should be should be um kind of divided a little that we we we are on the lookout for this in this time of doubt and we have to be but you as you said it's time consuming you can't do this all the time so what are the modes of reception that we might develop over time in which you can dispense with this by kind of just letting the text be the text and obviously this is much easier with literature I'm a I'm a scholar of literature this this interests me the whole point about literary text is that most of it is if not a lie then fiction, right? It is not true. So the cross-checking it with reality is not really something we do. It becomes um a problem as soon as we have really strong assumptions about what it means for a text to be from a human. So if you if you look at a romantic idea of of literature that says um literary or poetic works are expressions of some kind of um the the the inner life of an author then this probably has to go and then then we move to to a a reception technique that really looks at the formal characteristics of whatever it does with us as on a reception side and so on. So mostly I'm concerned with literary texts but yes there must be strategies to deal with with factual texts as well. Of course did you want to ask your question? Yeah. Thank you. My question is a bit similar to the past question that was asked, but like in terms of like author branding because I know like as like a big reader, sometimes I'll be like, "Oh my god, there's a new book by like Marissa Mayor coming out or like JK Rowling and people will just read that book because that specific author produced it without even knowing what the book is about." Um, so I was just wondering what that would look like in like a world where it doesn't really matter who the author is. Yeah. And and I think this is a great example of what what Michelle Fuku called the author function. That is the author function. And the author function means that a bunch of texts clusters around a name and that name orders these texts. So if we have um uh texts that are connected to I don't know Shakespeare and one of them is different, right? If you have an author, this difference needs to be explained. Why is it different? if it is from by by fakes. If you have something that is just the output of a language model and one is different then it's just the mistake right that is kind of the difference here what the author function does the thing is now that author functions are not tied to natural persons right you can have someone like um Tom Clansancy I believe most of these are written by committee right this is not one person there's one person that gives their name to it and it acts kind of like a brand as you said um this is probably similar to the other things in that corpus and this obviously you can still do when you when you have kind of a literary production that is highly automatized uh you have the name uh kind of as a marker of coherence of the corpus itself. Uh so I think in that interestingly in that particular case not that much changes. Uh what changes is um things as I said kind of the reception expectations as to whether the author name stands for an experience of a of a real human being. um which which might be more the case with um well specific types of poetry specific uh uh uppercase L literature um rather than than things that are mostly written for consumer market but you can you can really look what are you expecting from from from the literary text when you read it and how is this impact uh by these changes thank you I think probably last question and then we'll finish thank you so much. Good for Oh gosh, a little nervous. Um, I was wondering if uh well, like the first thing that came to my mind when like from hearing your talk was the concept of like misinformation and like I was wondering if you would classify like deceiving a reader as um like with AI text as misinformation and then it's kind of like a two-parter. um in the electronless I think that's what it was called situation like what is the minimum amount of credit that you would need to attribute to AI generation to like avoid um being deemed as misinformation. Yeah. The question I mean maybe you could ask do we have a right to know in a way how was it produced? I think some some legislation does exactly do this. I think the EU AI act has kind of provision for AI generated content if it is you know such and such percentage of the whole thing needs to be marked and I think there is something um to be said for this the question that I tried to ask here is what if this marking can cannot be trusted because it's so easy to circumvent right um you could also ask the same for misinformation we could imagine a type of reception of factual texts uh in which some kind of cross-checking factchecking is slotted in between me and the text, right? So, you would always read it through some kind of system that that checks these things. Um, that tool might might not work perfectly. It also means that we give ourselves more over to to certain technologies. So, this is also a question of I think autonomy of of readers. Can we do can we as you know human readers deal with texts directly or do we now always need some kind of third party to to check for that? Um and the the final thing is to what too much credit how much credit do I want to give? Again, literature is different than a factual text. There can be a great variety in poetics, right? You can be a romantic poet and say this is my this is what what my inner life looks like or this is some kind of inspiration. That is actually not me. Shelley writes about this a lot. The the genius in the classic sense is one who does not actually express themselves but either nature or God, right? God speaks through through the genius. Um but this is this is dependent on the poetics that we have and there is a completely different experimental uh line of literature that works with random processes that works with found text that works with all kinds of distancing mechanisms that make sure that no subjectivity was involved there. So you have to kind of know in which context and in which stance towards literature you have to read it. And the whole point here was um it was one was assumed namely traditional poem but the other one namely the artificial text was used and in this clash this becomes an interesting case. If it says underneath how it was made then it would be an instance of digital literature that is slightly experimental and so on. But that I think that's what what makes it such an interesting case. This this clash between these two poetics. Grab my microphone back and come back into the frame. So I'm going to stand awkwardly close to you. Um I just want to thank everyone online. Thank you so much for joining us. Everyone who joined in person, thank you so much for being here. And could we just give Hannah's um a big round of applause and say thank you so much for the for Thank you.