Creating content with artificial intelligence requires discretion and attention – 03/02/2023 – Education

In recent months, we have witnessed the rapid adoption of creation technologies assisted by artificial intelligence and accessed by natural language, such as ChatGPT by OpenAI, which creates texts with perfect grammatical structure and internal coherence, and MidJourney, by the company of the same name, which returns realistic images from text commands.

These are the best-known examples of a growing wave of applications that make the use of AI available to the general public and that are taking the market by storm in record time. While Instagram took two and a half years to reach the mark of 100 million users and TikTok needed nine months to do so, according to data collected by Reuters, ChatGPT reached the mark in just two months — the fastest accession in the history of users. end-consumer applications.

But all this technology doesn’t come without risks. It is important that users understand the ethical and social implications of using artificial intelligence tools, as well as their limits, so that they can use them consciously and carefully. This points to one of the pillars of media education: the critical and reflective use of technologies. From this point of view, we identified some points of attention that can – and should – be the target of educational actions: misinformation, degradation, bias and standardization.

misinformation: At the moment, AI text generation has control mechanisms that guarantee the quality of the narrative and grammatical structure, but not the reliability or accuracy of the information contained therein. When there is data, for example, not only are the sources not cited, but the data may, by the robot’s own admission, be entirely “hypothetical”. This can help spread false, partial or biased information.

On the other hand, the very language generated by the chatbot, being correct and well structured, ends up generating a false sense of authority in the user. MIT researcher Patty Maes compared GPT Chat to a parrot, which knows how to reproduce coherent grammatical structures, but does not know how to reflect on its content or meaning, which reinforces the need to educate the public for its critical use. The CEO of the company that created ChatGPT, Sam Altman, recently acknowledged that the technology can give “a misleading impression of competence”.

Degradation: A recent article in The New Yorker drew attention to the gradual loss of quality that occurs when an algorithm queries a large amount of information and seeks to reduce it to its most frequent structures. This process discards much of the information, retaining only an approximation of the original. This approximation, which is at the base of the technology used by AI robots, is comparable to the loss of definition of an image reproduced on a photocopier: imperceptible at first glance, but which gets worse as we reproduce already degraded images again. When the already simplified texts go back to the databases that “train” the AIs, the loss of detail and quality increases even more.

Bias: Another way to cause database or image degradation is to feed the system with skewed images or texts. One example is journalistic texts that directly associate favelas with violence, or images with racist or misogynistic connotations. This tends to reinforce clichés, stereotypes or prejudices present in the database, negatively affecting the quality and diversity of the generated content. Absences, biases and cuts are nuances that escape the analysis of algorithms, and we still have a long way to go in this regard.

Standardization: ChatGTP delivers texts that are skillfully constructed from a grammatical point of view and according to the standard norm, but achieving stylistic and creative nuances depends above all on a more skilful handling of the commands. The same happens with MidJourney: the repeated use of the same image banks and the frequency of images of certain styles on those bases ends up generating a uniformity of results and a repetition of visual clichés. In both environments, the quality and originality of the results will always depend on the users’ ability to generate better questions and commands, which in turn requires knowledge about the topic and aesthetic baggage.

In short, when it comes to innovations with disruptive potential and great impact on society, it is essential that responsible companies are clear about the limits of these technologies and that they commit themselves from the initial stages of development, together with the various sectors of society, in mitigate possible negative effects. It is up to us, educators, to prepare current and future generations of users for the conscious, creative and ethical use of this or any other technology.

note: The first paragraph of this text was prepared with the help of ChatGPT and brought some data that we had to check and correct. When asked about the source of the data, the robot replied as follows: “I apologize, but as an AI, I do not have direct access to real-time data sources. The data presented was used only as hypothetical examples to illustrate the answer the question”.