By JOSÉ MACHADO MOITA NETO*
There is no artificial intelligence that does not have educational, cultural, social, economic and political consequences.
Controversies surrounding ChatGPT, which seem to fuel a marketing, provoke a reflection that goes beyond this specific technological application, directing the gaze to the broader field of Artificial Intelligence (AI). who had the opportunity to read Artificial intelligence by Kai-Fu Lee may have been swept up in technological optimism or reflected on the issues of power, domination, and educational and cultural transformations absent in that text.
The position of an educator is always, in the face of almost everything, one of restrained optimism. There is nothing spectacular that does not deserve criticism or, at least, an alert to possible and undesirable developments. The purpose of this text is to navigate through possibilities of Artificial Intelligence, while some applications may be considered useless or dangerous.
human bots
The first point of this reflection is the existence of human “bots”: people who behave, in some practices, as if they were robots. This comes from training, standardization, standardization of behavior, even required by companies. As an example, some have already suffered the unpleasantness of wanting to quickly acquire an item in a supermarket, anticipating the information that will be requested. The item is delivered stating that it will be paid by credit card (one time or in cash), thinking that this will save time and also help the cashier to do his job faster. This is not true. It will process the purchase and then ask the question: credit or debit? If the answer is credit, then it will ask if it is in cash or in installments. This information had already been provided previously, but he did not pay attention or, if he did, he must follow a very strict protocol, which makes him a true human “bot” (human robot).
So, if we have this admissible behavior in society, this super standardization, and this is just one of the examples, it is not surprising the enormous success that an Artificial Intelligence can achieve precisely because humans quickly standardize. Uniform behaviors and patterns are easily picked up and learned by machines. This is the case, for example, of ChatGPT. Artificial intelligence will overwhelmingly replace “human bots”.
Standardization of human behavior
There are techniques in social engineering for patterning. It is common, for example, for Slogans commercials or emotional manifestations follow a certain pattern dictated by someone, by a set of discourses, in such a way that the behavior of very different people is standardized/formatted around the same type of discourse, the same approach, the same idol and even the same product or brand. Creating mass, standard behavior is excellent from a market point of view for many businesses.
Post-ChatGPT Transparency
There is no artificial intelligence that does not have educational, cultural, social, economic and political consequences. If it did, it would be irrelevant. In this way, discussing the technical-constructive transparency of each mechanism used in an AI defines how society wants to be transformed by it, including tracing ethical limits. In this post-chatGPT provocation, there is an issue of transparency that can be simplified by a catchphrase that in the past was heard to flaunt a social position “do you know who you are talking to?”.
What should be required of any Artificial Intelligence is that its first response be a question to the user asked by the Artificial Intelligence itself: “do you know who you are talking to?”. The user who answered “no” would hear a quick introduction, in which the dataset with which the Artificial Intelligence was trained and the period that the database used comprises would be indicated. In addition to cases of indications and contraindications for its use, such as a medicine leaflet. Everything that comes out right or wrong in the answer will obviously depend on this database or its characteristics. Any Artificial Intelligence that is evolutionary and continues to learn from its users should contain a special warning due to the already experienced risk of unwanted evolutions that lead to incorporating programmed attacks by users to change it.
Let's assume that a database is trained from the programming of the Globo between the 1970s and 1980s. By doing this, you can store all this programming and, based on it, train an Artificial Intelligence. This way, it will include all programming for that particular period. A question about climate change can be answered with: “I was not trained for this” or, even, a question involving recent cultural changes can be presented with a prejudiced answer that would be perfectly adequate in that decade. Therefore, we can classify an answer about the role of women in society as anachronistic; because we know that, in that historical period, that was the way it was said and that was how a certain question was treated.
With this, the argument I advance is that Artificial Intelligence today, precisely because it wants to cover a very large range, is very concerned with problems that involve many sets of data. Thus, it faces all kinds of problems starting from the definition of the data set itself and its representativeness. In this way, AI models are facing the same problems as all other areas of science faced with extremely large/wide models.
Theoretical path of other sciences
In Chemistry, we have theories for Kinetics that are still taught today, more because of their historical character, and are applicable to all chemical reactions. Today, we know that a single theory capable of explaining qualitatively and quantitatively the kinetics of a chemical reaction is unfeasible. Also, now we know that the theories of the past were important to explain the qualitative aspects involved in the speed of a reaction, but they do not attend from the quantitative point of view. And for that there are other theories, but for very specific reactions. This goes for the theories about the behavior of gases, the general theories are part of the historical framework and give qualitative information about the pressure, temperature, volume under the gaseous behavior. However, they are far from representing a model for a specific gas in an applicable situation with significant/accurate quantitative results. In all sciences, it is possible to demonstrate that the theories initially used had an unmeasured scope.
Currently, Artificial Intelligence models correspond to these generic theories and will only become good and accurate when they lose their generality and move towards a specialization. Imagine that we could have an Artificial Intelligence based on all of Machado de Assis' work, gathering in the database, in addition to his work, all the newspapers of his time and everything that corresponds to his experience. So, this Artificial Intelligence, in a way, will pass the universe of Machado de Assis. And, therefore, a question asked of this Artificial Intelligence would refer to the language of Machado de Assis, in the aspects, including the semantic aspects of Machado de Assis's time and with the categories he used, his differentiation in terms of class society and the environment in which he lived. Such an Artificial Intelligence would have an impressive didactic potential for the student who asked Machado de Assis about that historical period. It would be like having a living ancestor telling stories from his time.
Creation of Custom artificial intelligence
Today, it is within the reach of any researcher, for example, to create quite complex statistical models, whose algebra is often not understood, but it is possible to interpret them in accordance with what one wants. Similarly, it is possible to create an IDLE (Integrated Development and Learning Environment) for students to create their Artificial Intelligences. In this case, the business model would not be the sale of an Artificial Intelligence, but databases adapted for import into IDLE builders of Artificial Intelligences. The training step would be simplified by these algorithms, as it is today in the case of statistics, provided that the databases had been carefully pre-treated.
The possible future of a artificial intelligence
The future of Artificial Intelligence will be built tailored to the user, in which the user himself provides and uses it as he wishes, training from the database he wants. Imagine accessing an Artificial Intelligence that will work with all the programming of the Radio Pioneira of Teresina with the interviews done, music played, throughout the historical period that it has a database. Let's assume that it is from the 1960s to 2010 or 2020. Thus, we could have an AI of this entire period or, even, have an AI of even a more restricted period such as the 60s-80s, 80s-90s, 90s- 2000 and so on in this fragmentation. This means that each person would build his own Artificial Intelligence from a certain database that he would choose.
Indeed, this specialization would be of great value because we would also have “bubble phenomena”. Someone doing an AI in the 80s and 90s would not have all the answers needed for today's world, but the AI would put them inside that cognitive bubble that would correspond to that period. But, nothing would prevent that person, not accessing the AI knowledge of that decade, from using the next decade. Therefore, it may be that someone prefers to keep consulting Artificial Intelligence from the 60s-80s, another from the 2000s-2010s, and so on, because they are people who are stopped in time, who are fixated on those songs that were played at that time. period, in the way of seeing the world of that period.
Fundamentals of communication for Artificial Intelligence
Given the above, the most important thing, also from an ethical point of view, when answering the question: “do you know who you are talking to?” would be the emission of the answer: “you are talking to the Artificial Intelligence based on such a database that was trained and produced from such information, in a certain period of time.” In this way, those basic information of 5W2H journalism (What, Why, Who, Where, When, How, How much) would be in the information given by Artificial Intelligence that each one could create or take one already created within the Artificial Intelligence market, this is the possible future. This would solve the current problems of Artificial Intelligence, but it would bring another ethical dimension and would confine in a certain historical period all the anachronisms that Artificial Intelligences have today.
What is the problem with Artificial Intelligence that does not satisfy most people's requirements? The quick way they can get politically correct or discriminatory by being based on certain views of the past. An Artificial Intelligence will always look to the past, as it trains from what it has been given. In this way, when we declare what the database is and there is no pretense of generality, it specializes in a certain historical time, within a context, it can reach anyone, as today we do not need a professional called photographer. Each individual can create their own photographs, just as we don't need laboratories to develop images, we can project them and show them everywhere.
Proposed use of Artificial intelligence in education
I believe that Artificial Intelligence will have this other bias and, when it arrives in a school environment, a practice that could be done would be to ask students to create a database to train an Artificial Intelligence based on Romanticism. So, each student would look for all possible elements of Romanticism, its historical period, the main works, would compile all these databases and each one would add to their training base one thing or another that would differentiate them. That is, we would be truly incorporating the characteristics of Artificial Intelligence into the Education model and we would not be just simple users pressing buttons to get simple answers. This means saying that we should have better teachers, more trained for these possibilities, easier and more popular algorithms to be used and, also, we would have a very specific regulation that would be the necessary documentation of each Artificial Intelligence.
Between creativity and standardization
Man will never stop being two things at the same time: he will always tend towards a uniformity based on his social life, which greatly favors Artificial Intelligence and will also be creative, different from the other/neighbor, which will leave always a gap in Artificial Intelligences. However, it is clear that it will bring advantages to society as a whole, as did the fact of the press, as did the great encyclopedias, dictionaries and the various means that the community has to access and interpret information. Obviously, it will never stop being a mimicry of a certain period and will always need the user to use his roleplaying ability.
When asking an Artificial Intelligence based on data from Machado de Assis to write a text on climate change, this text will have the style of Machado de Assis and the depth corresponding to his time. Perhaps, narrate some situation, associated with climate change, that happened in a certain period, which would simply be portrayed as full or flooding, intense cold or only at the end would declare that the information presented is insufficient to complete the idea, since this category did not exist in that historical period. However, we would have many other interesting things to explore.
Conclusion
The explosion of text, news and information about using ChatGPT qualifies more as a major release of marketing than a change of pace in artificial intelligence research and development. Substantial developments in artificial intelligence will only happen when generic models, which would equally target all consumer markets for a given language, are abandoned.[1]
*José Machado Moita Neto is a retired professor at the Federal University of Piauí (UFPI) and a researcher at UFDPar.
References
It's still early to say that I'm the biggest revolution since the internet, says ChatGPT, https://impactomais.com.br/ciencia-e-tecnologia/ainda-e-cedo-para-dizer-que-sou-a-maior-revolucao-desde-a-internet-diz-chatgpt/.
AI advance with ChatGPT warns of student cheating, https://www1.folha.uol.com.br/mercado/2022/12/avanco-da-ia-com-chatgpt-alerta-sobre-trapaca-de-estudantes.shtml.
ChatGPT: how to use the robot in everyday life, https://g1.globo.com/economia/tecnologia/video/chatgpt-como-usar-o-robo-no-dia-a-dia-11287993.ghtml.
ChatGPT: tool controlled by artificial intelligence generates controversy when creating texts, poems and even song lyrics,
https://g1.globo.com/fantastico/noticia/2023/01/29/chatgpt-ferramenta-controlada-por-inteligencia-
artificial-generates-controversy-by-creating-texts-poems-and-even-lyrics.ghtml.
Lab that created ChatGPT was founded not for profit and could be worth $29-billion, https://www1.folha.uol.com.br/mercado/2023/01/laboratorio-que-criou-chatgpt-foi-fundado-sem-visar-lucro-e-pode-valer-us-29-bi.shtml.
Nature 613, 612 (2023), https://doi.org/10.1038/d41586-023-00191-1
NYC education department blocks ChatGPT on school devices, networks, https://ny.chalkbeat.org/2023/1/3/23537987/nyc-schools-ban-chatgpt-writing-artificial-intelligence#:~:text=%E2%80%9CDue%20to%20concerns%20about%20negative,education%20department%20spokesperson%20Jenna%20Lyle.
ChatGPT and the debate on advances in the field of artificial intelligence – Ao Ponto – podcast by O Globo newspaper,
Concerned about ChatGPT universities begin to review teaching methods, https://www1.folha.uol.com.br/mundo/2023/01/preocupadas-com-chatgpt-universidades-comecam-a-rever-metodos-de-ensino.shtml.
When technology can get in the way. https://cidadeverde.com/cienciaviva/121159/Quem-a-tecnologia-pode-atrapalhar.
Trying to ban ChatGPT in schools will be a waste of time, experts say.
Note
[1] I would like to thank my friend and collaborator Prof.Dr. Regis Casimiro (IFRN) who did the reading and contributed with indications of recent sources of discussion on the subject.
The A Terra é Redonda website exists thanks to our readers and supporters.
Help us keep this idea going.
Click here and find how
