Filosofia no Buteco

A seguinte deliciosa crônica foi escrita com "ajuda" do modelo de linguagem da OpenAI ChatGPT na quarta-feira, 10 de junho de 2026.

Comando: Os ilustres doutores honoris causa Endométrio Augusto e Paramécio de Souza estão em um buteco bebendo uma gelada e conversando sobre filosofias.

Realismo fantástico acadêmico

Endométrio Augusto: Meu caro Paramécio, a grande questão da existência é se somos agentes da realidade ou apenas células dentro de um organismo maior.

Paramécio de Souza: Concordo parcialmente, Endométrio. Mas veja, até um paramécio, embora microscópico e diminuto, toma decisões em sua escala. Talvez a liberdade, ou libertinagem, seja proporcional ao tamanho do universo que conseguimos perceber.

Endométrio Augusto: Então a filosofia seria o metabolismo da consciência?

Paramécio de Souza: Exatamente. E este boteco é o nosso laboratório epistemológico.

Nesse ínterim, o dono do buteco pensa: "Dois doutores discutindo o sentido da vida e ainda não decidiram se querem outra porção de torresmo!"

Comando: E daí o Dr Endométrio chega para o seu homólogo Dr Paramécio e fala: pô, eu sou agente do ciclo menstrual, conquanto vós sois um agente que pode ser facilmente fagocitado por amebas!

Rivalidade biológica existencial

Endométrio Augusto: Pô, meu caro Paramécio, eu sou um dos eventos mais complexos, e controversos, da fisiologia humana: o ciclo menstrual. Tenho papel na preparação do organismo para uma possível gestação. E vós? Sois um agente microscópico que pode facilmente ser engolido por uma ameba!

Paramécio de Souza: Calma lá, Endométrio! Ser pequeno não significa ser insignificante. Eu represento bilhões de anos de evolução. Tenho cílios, movimento próprio e capacidade de sobreviver onde vós nem conseguiríeis existir.

Endométrio Augusto: Mas eu participo de uma narrativa de altíssima complexidade!

Paramécio de Souza: E eu participo da narrativa de que a vida começou simples. Antes de haver doutores honoris causa em botecos, já havia organismos unicelulares resolvendo problemas.

O delicioso dono do bar, trazendo outra rodada, conclui:

"Os dois estão certos. Um tem diploma no corpo humano, o outro tem currículo de bilhões de anos."

Diplomacia entre Diplodocus

A diplomacia dos Diplodocus não tem pressa, afinal de contas, a diplomacia, bem como a estratégia, são pratos servidos frios. E também a diplomacia não pode ter pressa, porque, os Diplodocus contam com vinte e cinco metros de envergadura e um pescoço que leva éons para se curvar.

Imagem gerada pelo modelo de linguagem da Meta AI Muse Spark em 9 de junho de AD 2026.

O cenário

É o fim da estiagem no que seria hoje o estado americano do Colorado, há cerca de 150 milhões de anos. O nível do rio baixara e virara uma espécie de espelho. De um lado, a Manada do Vale. Do outro, a Manada da Cordilheira. Na intersecção, samambaias gigantes são suficientes para nutrir filhotes por semanas, mas não para os dois clãs ao maesmo tempo.

Ninguém ruge. Diplodocus não briga; ele tem endosso.

Os embaixadores

O macho maior não tem vez entre o clã de Diplodocus. Quem manda são as matriarcas com maior bagagem etária, por que apenas elas se lembram da últma Grande Estiagem.

Elas se reunem nas margens do rio, como na imagem (gerada pelo modelo de linguagem Muse Spark da Meta AI), e fazem o primeiro aceno para a diplomacia: erguem o pescoço o máximo que conseguem, depois baixam paulatinamente (lentamente) até ficarem quase paralelas ao nível da água. É o equivalente a tirar o chapéu e mostrar a garganta: Veja, venho em paz, não vou pisotear.

No entorno, os jovens Diplodocus atuam como assessores, mastigando devagar para não incomodar seus superiores com barulhos incômodos.

.

Como se negocia sem palavras

1. Vibração: Diplodocus falava pelo chão; batidas leves de cauda emitiam infrassom que viajava quilômetros. Neste delicioso caso, a batida é curta; duas vezes: "Temos fome."
2. Oferta: A matriarca da esquerda arrasta com sua deliciosa pata uma porção de coníferas frescas para a margem. Não é presente; é evidência de boa vontade.
3. Tempo: As matriarcas dos clãs Diplodocus ficam imóveis por cerca de uma hora. Para um animal que atinge os setenta anos de idade, esperar faz parte do processo de barganha (negociação).
.

Acordos entre Diplodocus não são assinados; são pastados.

A Manada do Vale atravessa o rio diplomático primeiro, logo durante as primeiras horas da manhã, e se delicia com as copas altas das coníferas. A Manada da Cordilheira chega durante o ocaso (entardecer), e limpa as partes baixas e as deliciosas samambaias gigantes. Nenhuma da manadas pisa o ninho da outra, em sinal de respeito, e cada ninho é marcado discretamente com pegadas circulares de jovens Diplodocus.

Ninguém ganha território. As duas manadas ganham tempo até a próxima precipitação.

É uma especie de política externa baseada em física pura: se dois animais de quinze toneladas decidem brigar, os dois perdem. Então os Diplodocus evoluíram para a alternativa mais elegante, curvar o pescoço primeiro.

Employing Generative Artificial Intelligence with Responsibility

Introduction

I wrote the following delightful chronicle when I was studying Graduate Research Methods during the programme of Master of Business Administration (which I concluded in April 2026). The chronicle, though, was written between August and September 2025, and our instructor had prompted us to explain how we made use of Generative Artificial Intelligence tools. Happy reading!

Artificial Intelligence Report

The employment of artificial intelligence in educational settings can have a broad sense. ChatGPT, Perplexity, Google and Microsoft Translators, and Office 365 use certains types of artificial intelligence. Some use predictive analytics, others use machine learning, while alternate models use natural language processors. Artificial intelligence techniques should not be employed as a substitutor of knowledge, but as an enhancer of the practice of studying.

First parameter: Microsoft Translator

When I am stuck with a complex word or idea, I go to the translator hosted by Microsoft, which is similar to Google's, but this one has three options of translations: Standard, Casual, and Formal, and I can manipulate the translation to my liking.

Furthermore, if I translate a word, let's say, the connector "whereas", from detect language to English, and then set the translation to Casual, Microsoft Translation, which can be accessed from any browser, will explain what the vernacular stands for, or in the present case, give me an alternate way to convey "whereas". I have also included a screenshot concerning Google Translate to see if I get similar results. (See below images for reference. Data is from June 5, 2026.)

Sample of Microsoft Translation collected on June 5, 2026. It translates "whereas" from "detect langauage" to English, where it returns "while". Translation mode is set to "Casual".

Sample of Google Translation collected on June 5, 2026. It translates "whereas" from "detect lanaguage" to English, where it returns "whereas". Translation mode is not available for Google Translator, but it does have a dictionary included.

This experiment provides us information that Microsoft Translator can be useful for paraphrasing, and that Google Translator can be useful for semantics akin to an alternate dictionary, for example, a searcher not satisfied with a definition of a word on Merriam-Webster, then goes to Google Translate to get a second opinion.

Second parameter: AI Overview as a scoping review (Google and Bing, basically)

Whenever I want to understand more of a concept that I am not sure what its definition is, I go to a search engine akin to Microsoft Bing and type just the word related to my query. Searchers on the internet tend to make conversational queries, like instead of searching "watermelon", searchers look for "where can I buy watermelon near me". What I prefer doing is to search a broad term and narrow it down as I read through most of the links displayed on the search engine results page. Now that the two most popular search engines akin to Google and Bing have so-called "AI overviews" in their results page, scoping a concept gets to a different level, because the robot will scan the webpages related to the query and summarize the principal results.

In the following images, I queried "covalent bonds" in Microsoft Bing and in Alphabet Google and these are the results. (See images for reference. Data is from June 5, 2026.)

Sample of a search on Microsoft Bing for "covalent bonds" colected on June 5, 2026.

Sample of Google search for "covalent bonds" depicting an AI Overview section. Data collected on June 5, 2026.

Notice that the search engine results page shows the AI overview statement and beside (in case of Bing) and below (in case of Google) a link directing to Wikipedia. This indicates a strong domain authority concerning Wikipedia, which does not mean Wikipedia is a good source, though.

Third parameter: Conversational AI as scoping reviews (OpenAI ChatGPT, Perplexity AI)

This idea is similar to the second parameter, but it requires more critical reasoning, because conversational artificial intelligence mechanisms do not always display their sources.

In the following cases, I queried "friday, june 5, 2026: hegemony" in Perplexity AI, OpenAI's ChatGPT, and xAI's Grok large language models to see what they return. (See images for reference. Data collected dates from June 5, 2026.)

Sample of a search on Perplexity for "friday, june 5, 2026: hegemony". I personally like to add dates in my conversations with robots, so that I know exactly when I did such observation, like a researcher from the eighteenth or nineteenth century would write a journal entry with his observations.

Sample of a search on ChatGPT for "saturday, june 6, 2026: hegemony".

With this simple, yet powerful example, I demonstrate that both Perplexity AI and Grok tend to be more explainable than OpenAI's ChatGPT, because it actually shows its thought process in clear steps, thus making them better for academic research; at least the groundwork, not the full research.

Fourth parameter: Employment of conversational artificial intelligence mechanisms for role playing

I have tried this scenario with several natural language processors (NLPs), that is, mechanisms that convert text into more text based on the initial text. My portfolio of NLPs include ChatGPT, Google Gemini (that I use on my smartphone for quick responses), and Microsoft's Copilot. On rare occasions I found myself having a conversation with Deepseek and Anthropic's Claude. When I utilise X (formerly Twitter), I like to have xAI's Grok large language model to interpret my own posts and some posts that appear on my news feed.

For the next case, I will employ the following prompt: "Hi, buddy! I live on Planet Earth and I am an explorer who happens to be a K-12 student. I would like to know if I can trust natural language processors and large language processors akin to you for school assignments. Thanks in advance!".

Response provided by ChatGPT on Saturday, June 6, 2026. (Note that ChatGPT calls me "Otto" because I previously told it to do so):

Hi, Otto.

You can trust large language models like me for some parts of schoolwork, but not for all parts.

A useful way to think about it is that an AI is more like a very knowledgeable assistant than an infallible expert. It can:

• Explain concepts.
• Summarize texts.
• Suggest ideas for projects.
• Help organize research.
• Check grammar and clarity.
• Generate practice questions.
• Show different perspectives on a topic.

However, it can also:

• Make factual mistakes.
• Invent sources, quotations, or references.
• Misinterpret questions.
• Present uncertain information with confidence.
• Miss recent developments if not connected to current sources.

For school assignments, a good approach is:

1. Use AI to understand the topic and generate ideas.
2. Consult textbooks, class materials, and reliable sources.
3. Verify important facts independently.
4. Write the final work in your own words.
5. Follow your school's rules regarding AI use.

As an explorer, you might appreciate this comparison: a language model is a bit like a compass. A compass can help you navigate, but it is not the territory itself. You still need maps, observations, and judgment to complete the journey.

In short: trust AI as a helpful tool, but not as the sole authority for your assignments. The more important the assignment, the more carefully you should verify what the AI tells you.

The following link redirects to what Microsoft's Copilot large language model returned: https://copilot.microsoft.com/shares/8Ezri9hSZviuWeuVqaPHV

This exchange happened on Saturday, June 6, 2026. Note that, while OpenAI's ChatGPT compared itself to a compass, Microsoft's Copilot described itself as a guide, only because I said in the beginning of the conversation that I was "an explorer".

Conclusion

With these simple, yet replicable experiments, I tried to demonstrate that any form of artificial intelligence akin to machine learning, predictive analytics, or natural language processors, can be employed to conduct preliminary, scoping research of a broad topic. Artificial intelligence should not be employed as a shortcut in education, but as an enhancing tool, just like Google introduced itself back in the late 1990s, early 2000s.

I grew up with Wikipedia and I had a strong urge to copy, paste information from there to complete my school assignments. Instead, I actually took the time to read and critically evaluate what was written on Wikipedia, because it tends to contradict itself most of the time.

I have studied a lot of topics when I was a secondary school student through Wikipedia articles, including the English revolutions of 1649, 1688, and the eventual bankruptcy filing of the city of Detroit in 2013.

Wikipedia is not perfect, but it helped me understand school subjects, just like artificial intelligence tools can be helpful for students in every level of education, especially beyond secondary school. The limitation in this case is to resist the urge to copy, paste information you receive as an output from artificial intelligence, just like I resisted copying, pasting data from Wikipedia. And if you inspect some outputs provided by generative artificial intelligence models, you will notice that they really are or may be direct reproductions from Wikipedia!