«Tía, deje de mandar eso que no es verdad»

Discourse analysis and misinformation in the AI era.

Daniel Murcia
7 min readDec 30, 2022

Sometimes one shares mental notes since it is the only way to join hunches with others. As we approach a needed space for interdisciplinary work, there comes a time where sole ideas have no value. So, I guess this is my way of crying for help to those who might feel the urge to contribute to the matter. Clio’s been working on tech development which joins the study of language and NLP with the intend to contribute to AI-based products/processes that could be used in the educational field; however, there comes a time in which the rapid progress of tech problems surpasses our creative capacity.

— A bit of context —

On the one hand…

Today (as of 29th of Dec, 2022), I woke up with the news that the Colombian poet Jota Mario Arbelaez was dead. The news given by, perhaps, the biggest radio station (WRadio) of the country was soon to be denied by the poet himself. Yeap, we all thought it was a sort of Colombian April’s Fools day joke, though it wasn’t, and soon the interview turned to a whole lot of excuses to the family. The prestige of the famous radio station was at doubt so his spokesman: Julio Sánchez Cristo managed to provide evidence that they checked in the poet’s Facebook site and Wikipedia where he actually appeared to be dead. From then on, I turned the radio off. The whole morning of the station turned into apologies and arguments that sought to alleviate the misunderstanding.

On the other…

I left home to make a day, but then as I sit to wait for my wife, the news piked my phone announcing that the site Plagio SOS was no longer to be published in 2023. I was following this site since it started to publish all theses that presented some degree of plagiarism. As a scholar, I reckon some morbidness on the way I followed all stories that led to embarrassing cases of declared plagiarism which turn into national scandals of governmental and highly ranked administrators and public figures. My interest has always been on the journalists of Plagio SOS and the way they insist on changing society using public exposure of academic incompetence. They’ve even done so with the current Colombian Minister of Education:

[Why Alejandro Gaviria shouldn’t be minister of education?: Because of his dishonesty. After formal complaints of plagiarism and fraudulent titles in UniAndes, he stayed silent and did not act against academic felony] https://twitter.com/plagiosos/status/1545123275432394754 (translation by author).

The issue is that this platform announced they will no longer continue their work in 2023 claiming that what they publish does not seem to have any effect on academia.

— There is no third hand, so end of context —.

I’ve followed plenty of initiatives to achieve transparency and honesty in published information, but today I became bewildered with the fact that a Colombian news portal of the importance of W Radio is no longer efficient to validate all the news they publish. Hence, I decided to write a bit to expand my thought and share some reflections on the matter. Here is the core of this mental note.

The era of synthetic reality

As an AI enthusiast, I care too much for all latest developments on the field and most of my networks allow me to check on that. Last week the video of simulated Morgan Freeman became viral because this clip is the beginning of deepfakes videos which can alter the states of reality published on media. It seems as if the truth will become partial because our senses will not longer be enough to define the validity of what one hears or watches.

Concept and deepfake by Bob de Jong Freeman’s voice is imitated by the incredibly talented Boet Schouwink.

This fake versions are just some samples of all reality alterations that can be sketched for extremely negative purposes. (You may also listen to the conversation with AI generated Steve Jobs).

Our present is driving us to cast doubt on all information sources to the point that there is a breaking point between real reality, augmented reality, virtual reality, and why not, synthetic reality. Then we’ve arrived to a major gap in validating information.

There is indeed a stronger reflection to be done about Post-truth discourses, but to what we know, Twitter has started to provide interesting tools that seek contextual validation of information where there is an author that generates text and an audience that discrimantes information by asking or providing context:

Note. Sample of a tweet with reader’s context tool. The example doesn’t hold any implication with Fox News’ information.

This plays a great analogy on how AI works, which in Machine Learning terms is called Generative Adversial Networks. In GANs, the objective of the Generator is to create potential outputs, while the Discriminator prevents poor outputs. So far, this is a control system that seeks for better quality of the produced text.

I strongly believe we are soon to be immersed in this discussion. Online information will struggle with a perpetual echo in which AI generated text will resonate in the search engines or GPT-3 text creation and an alternative version of reality will be presented as news. Then, how do we filter reality? How can I trust an informational source?

Defensive AI

The work of Kissinger et al. (2021) has inspired me to think about content moderation AI algorithms that will have as a task to filter information. I know this sounds like an Orwellian regulatory policy, but “antidisinformation AI” is currently being cooked in research labs as a solution to what we foresee will be another flaw of an AI era. At first glance we could trust algorithms to discriminate information efficiently and, perhaps, it will prevent the major news channels to disseminate flawed information with great reliability. But then questions come to mind:

*1: If antidisinformation AI makes a mistake surpressing content that is not malign disinformation but in fact authentic, how do we identify it? Can we know enough, and in time, to correct it? Alternatively, do we have the right to read, or even a legitimate interest in reading, AI-generated “false” information?

Which then leads me to believe we’ll have a greater segmentation of information to that of 1) mainstream free information, 2) Freemium information, and 3) Premium information which is corroborated by AI. It all seems like a subscription plan and depending on your payment affordance, you may access to a version of reality.

Well it all seems very much a futuristic anime scene, but we should then turn to how we access information these days. Do you pay for the information you consume? Or are you informed by the WhatsApp chain messages and Instagram choices personalized for you? Our choices of information explain very much how we interpret reality.

One of the projects I deeply admire is the Detox Information Project (DIP). This Colombian project seeks for an educational proposal of information digital literacy which embraces behavioral sciences to counterattack fake news and polarization in Colombia. They have redirected to interesting data about Colombian news consumption that explains the dimension of this matter.

Colombians information sources:

2022 Digital News Report. Reuters Institute for the Study of Journalism University of Oxford. https://reutersinstitute.politics.ox.ac.uk/digital-news-report/2022/colombia

As it appears, most Colombian know about the current affairs by using their phones and so digital information processing comes to a level of importance that could ignite several interdisciplinary projects. Ideally, education could just be enough to develop informational and digital literacy skills in the population so that we could get past the WhatsApp messaging chain, as for now, it is not enough to scold our relatives: «Tía, deje de mandar eso que no es verdad». [Aunt, that’s not true, stop sending it!]

My view is that interdisciplinary work is at it’s best moment. The work of Van Dijk on Media Discourses and its influence on Social Cognition has not been connected to an AI defense system in the news or how about training defensive AI using a DHA approach (Reisigl and Wodak, 2009; Stubbs, 2015) of discourse to define:

How linguistic data that refer to people, objects, phenomena / events, processes and actions are openly articulated in online discourses to correlate the variability of what they say about them across different excerpts in multiple media?

This process can drop results on the veracity probability of information and so we’ll have the chance to determine the type of reality we consume. It is very much like the food we consume, information could be also a matter of decision.

Poltec (2021). Norma de etiquetado nutricional de los alimentos en Colombia: https://www.poltecsas.com/post/norma-de-etiquetado-nutricional-de-los-alimentos-en-colombia-aqu%C3%AD-te-lo-contamos

However, “*2: the human mind has never function in the manner in which the internet demands”. I still have the feeling some would just feel comfortable with the easy access info to make on-the-go decisions. My intention was to leave a mental note which could spark interesting conversations on the problems of tomorrow which seem to be happening now.

This article was written like in the old ways, no GPT tech was used in the construction.

— Pd. I cannot imagine how many times they checked the actual death of Pele.


*1: Kissinger et al. (2021, p.116).

*2: Kissinger et al. (2021, p.).

Kissinger, H.A., Schimdt, E., & Huttenlocher, D. (2021). The age of AI and our human future. Little, Brown and Company.

Lara, F. (29 de diciembre de 2022). Jotamario Arbeláez sobre rumores de su muerte: “es de muy mal gusto”. La W Radio. https://www.wradio.com.co/2022/12/29/es-de-muy-mal-gusto-jotamario-arbelaez-sobre-rumores-de-su-muerte/

Ramirez, J. (20 de diciembre de 2022). El portal PlagioSOS, que denunció que Jennifer Arias había plagiado la tesis, anunció su cierre. Revista Cambio. https://cambiocolombia.com/articulo/poder/el-portal-plagiosos-que-denuncio-que-jennifer-arias-habia-plagiado-la-tesis-anuncio

Stubbs, M. (2015). Computer-Assisted Methods of Analyzing Textual and Intertextual Competence. In D. Tannen, H.E. Hamilton, & D. Schiffrin (Eds), The Handbook of Discourse Analysis (2nd ed., pp.486–504). John Wiley & Sons, Inc.



Daniel Murcia

Educator | Applied Linguist | Researcher AI/Natural Language Processing in Language Assessment