terça-feira, 1 de agosto de 2023

Technology's Language Problem

Let’s be honest. Technology has a language problem. Since we lack the words to describe our excitement about original ideas, we do the next best thing (according to ourselves): we borrow words from other areas. We have been doing this ever since the first Software Engineer graduated. We continued to borrow names when we began to describe our architecture for software, data, processes, and everything else we want to give the impression we know how to piece together in a logical, reproducible way. Once again, when we called algorithms intelligent, despite not knowing how to define intelligence ourselves.

Overloading words with new definitions and uses has been a constant in our evolution as humans, and we have been doing it for ages. We began doing it to reduce the trouble of having to explain everything from scratch. Because of this, we have words like “muscle”, in English, that came from “small mouse” in Latin. Apparently, someone thought some of these structures resembled mice and the name stuck. Eventually, Latin died out, but muscle remains not only a noun, but also as a stand-in for force. The disclaimer here is that this example is valid for our, so called, Western world, historically influenced by Rome. For other languages, these examples don’t hold, but I guess there are parallels.

Borrowing words from other areas is not bad practice, in principle. It allows us to quickly show a rough overview of what is intended and also provide catchy titles for media outlets. We began with “expert” systems. Software developed to emulate a human’s decision process for business needs. These systems began to be developed in the 1950s and by 1960, LISP was the dominant programming language in America, with PROLOG being favoured by European developers. By the end of that decade, we already had medical diagnosis “expert” systems developed to assist doctors in their work. By the 1980s, expert systems were used by most large companies in their daily business decisions. We still use them today when interacting with a loan application or an investment decision. Stock exchange robots make million-dollar decisions in automated trading in less time than it would take a human trader to blink an eye. All this development was the result of transferring the decision processes of a human specialist to computer code.

Because the world evolves, these systems have to be maintained and updated. Their knowledge base evolves through the inclusion and removal of decision parameters. Despite being fast, the knowledge contained in the software is static. It requires maintenance to evolve. To overcome the costs of maintenance, we began developing algorithms that would keep the knowledge base current, extracting patterns from existing data without the need for reprogramming. These were the new “intelligent” systems that were able to “learn” from existing data. Or at least, that is how we explained it to the public.

We now have computer systems that find patterns in data by themselves and make decisions based on these patterns. We call the process “machine learning” and happily label the systems as possessing “artificial intelligence”.

These expressions allow us to avoid explaining how the data mining algorithms classify and tag information. They also hide complexity from view behind expressions we know. Overloading these expressions with new definitions, however, prevents us from understanding what is really happening in the code, leading to the perception that we do not know how the systems reach a specific conclusion. Images of evil, or alien, intelligence making decisions that affect the lives of millions creep into media feeds and apprehension grows. I believe this all happens because, once more, of borrowed and misused words.

Since we are talking about the interaction of algorithms in computer code, we do know how the system reaches an output. We may not be able to reproduce the exact steps since most of what happens is in the realm of probabilities, but we do know how. The problem is that we tend to use “how” for “why”, and “why” is a question for which the developers may not have an acceptable answer. 

quinta-feira, 29 de junho de 2023

AI and the Boring Tasks of Work

We have all, at least once, been assigned the responsibility of taking notes for a meeting. One of the business needs that nobody told us existed until the moment we were assigned to write them and, probably, the most undesired important task of running a business. Meeting notes are the memory of what was discussed and the record of what was decided. Even if only the date of the next meeting. This leaves the person assigned to it torn between taking accurate notes and actively participating in the meeting. Voice recorders were used to increase the effectiveness of note-taking, enabling the person responsible for the notes to interact without the fear of losing something. This, of course, resulted in having to go through the meeting again to transcribe whatever topics were mentioned. Recordings are a great form for documenting a meeting, but they are still not ideal for the minutes of a meeting.

Voice recognition and text to speech technology came to the meeting room as soon as they became reliable enough. The secretary would be able to re-read the meeting instead of listening to it. Copying text from a meeting transcription quickly solved the issue of trying to remember the exact wording of a sentence. The mechanical, boring of work of having to write the minutes of the meeting continues, usually resulting in an e-mail that no one would read unless a problem happens.

Talking and taking notes are part of several professional activities. They are the memory of a human interaction. A sales representative will spend time writing a follow-up e-mail as soon as the call with the customer is over. A medical doctor will spend at least another hour transcribing notes to medical records after an appointment, and an additional hour dealing with health insurance paperwork if they work alone. These mechanical writing tasks are tedious but, at the same time, necessary for business. A wrong decimal point in the incorrect field prevents a doctor from receiving payment for a medical appointment. A poorly written follow-up e-mail becomes a finger-pointing thread with the client. Most of the situations created by faulty note-taking are easily solved but, typically, demand time and effort that hadn’t been planned.

Medical doctors are specially sensitive to these problems. A medical appointment is an intensive doctor – patient interaction where symptoms are described, advice given, and treatments planned. All of this has to be written down for future reference and for reporting. A doctor’s file cabinet is a trove of medical experience to which they resort when that feeling of “I’ve seen this before” happens. The files are also the memory of the patient’s health history, exams, and medications. They are also the source of frustration when a doctor works in a large clinic. With patients coming at every hour, the doctor usually leaves the writing of files for after office time. Often when they should be spending time with their families or resting.

From a patient’s perspective, to see the doctor pouring over a keyboard is frustrating. Waiting as they hardly look at us while they type something on the online health insurance system gives the word “patient” a whole perspective meaning. Part of the feeling that doctors have lost their bedside manners has come from this need to document every interaction to produce reports and billing statements. I miss the days when a medical appointment was more than the handing over of lab results and the printing of prescriptions. Most, if not all, doctors are equally frustrated by the automation of medical care.

Could Generative Large Language Models be the answer to this? The University of Kansas Health System, for example, is betting on it.

By having the AI do the note-taking and freeing the doctor from the task, the pilot the University of Kansas is running with Abridge promises to bring back the human factor to the doctor – patient interaction. Aiming towards the increase in the quality of medical attention and the accuracy of the files. The model acts as a personal assistant, capturing the full audio of the conversation and extracting the relevant information for the structured medical files. Identifying patterns and structuring information in a specific format is a scenario where AI excels. Every interaction becomes a record that can be reviewed by the doctor without sacrificing the human aspect of the profession.

A medical personal-assistant is not a first step in replacing the doctor with artificial intelligence. IBM’s experience with Watson Health has shown the limitations of the technology when dealing with treatment recommendations. Such an assistant is a good example of how the use of artificial intelligence can bring back the human care of our professional interactions. This points to a more human experience through the use of technology, especially with the use of generative language technologies.

An assistant that is not bored by the repetitive nature of their work is the ideal one for an overworked doctor. A similar assistant, working for a lawyer or a sales representative, would make sure that accurate notes be available while remembering the professional of any important detail that they did not notice in the conversation. An assistant that lets us concentrate on the essential parts of the job, taking over the boring ones that we dislike. This seems to be an interesting future where Large Language models have a critical role: write the reports and fill out the forms that we love to hate.

domingo, 25 de junho de 2023

My Land is My Tongue

“Minha pátria é minha língua” – Fernando Pessoa


What makes us what we are? What is it that makes us feel Portuguese, Brazilian, American, Spanish, German, or Thai? Fernando Pessoa, a Portuguese Poet (yes, some poets have to be mentioned with a capital “P”) once wrote: “My land is my tongue”. We all have a gut feeling of belonging that comes attached to our mother language. A certainty of being that is attached to the language we were brought up with.

Cezar Taurion posted a comment on Lera Boroditsky’s TED presentation that brought memories back from the time I worked as an English teacher. If you haven't watched the video, I recommend you do it soon.

If you speak a second language, you know the frustration of not being able to express a thought in it. You will be able to describe it by using additional words that, quite frankly, kill the mood and defeat the purpose of using that original word in the first place. Take the word “saudades” in Portuguese or “ubuntu” in Bantu. One can surely define and describe the concepts behind the words in English. However, there is no way these descriptions can be used in English in the same emotional context they carry in their original languages.

Catherine Caldwell-Harris in “Emotionality differences between a native and foreign language: theoretical implications” identified we are more emotional in our first language than in any other acquired second language. If you speak a second language, you probably know the awkward feeling of finding it easier to talk about certain topics in your second language. There are topics that are just too emotionally charged to be discussed in one’s native tongue.

This may be part of the reason interacting with LLMs, such as ChatGPT, or Bard, bothers non-English speakers so much. These models have been exhaustively trained against samples of the English language, specifically American English, and they excel at that. However, they feel off when interacting in Brazilian Portuguese, for example. The text produced by these models is grammatically correct but lacks the emotional subtext that normally carries through in a conversation.

It might be just a step in the evolution of AI language models, a type of cognitive “uncanny valley”. It might also be a symptom of how we are developing these models and embedding them with our restrictions. This would be a nod to the developers’ capacity in capturing our biases and automating them. This would also present a risk. Restricting our interactions to only one, or two, AI “lingua franca”, our reasoning skills could be lost in the process.

I can’t help but remember Sofia Vergara’s character in “Modern Family”: “Do you even know how smart I am in Spanish?” 

quinta-feira, 15 de junho de 2023

Fear the LLMs?

“The ability to speak does not make you intelligent.” — Qui-Gon Jinn, The Phantom Menace (1999)

Large Language Models – LLMs, their applications, and, especially, their risks have been the focus of many news outlets and the concern of industry leaders. Although most of the hype comes from our over specialization and lack of understanding, reasons for concern raise above the noise.

LLMs are, in a simple analogy, phrase builders. In a sense, they have a logic akin to the one of Mad Libs®, where only a specific class of word can fit in a given space in a sentence. The major difference being that LLMs have a slot after every word they add to the chain. The possibilities of linking grammatically correct words and logically sensible hints at the complexity of the LLMs we have today.

As with any other AI system, LLMs are a collection of algorithms that have been coded by programmers. Once ready for deployment, these algorithms are trained to produce an expected output. In the case of LLMs, the output is a sentence. The training process includes the input of a large amount of data in the form of tagged words, sentences, and even longer texts that enable the algorithms to identify patterns in the data. Tagging is important since it tells the algorithm to which grammatical class a word, or phrase, belongs. With this data processed, the algorithm is then prompted to produce results and the outputs are evaluated, and corrected, by a human being. If a sentence does not make sense, the output is flagged as incorrect and put back into the training data as a model of what to avoid. If not, the output is appropriately flagged and the algorithm path that produced that output is prioritized. This continuous review of the prompts and their outcomes is part of the curating process for training data. Eventually, the continuous reinforcement of these correct algorithm paths makes the outputs indistinguishable from what a flesh-and-blood human being would produce.

We already use AI models for several of our day-to-day tasks. Face recognition, for example, has become so pervasive that we have it in our phones. Voice activated systems are also another staple of our daily interaction with AI models. Whether it’s Alexa, Siri or Google, we can simply say instructions that are interpreted by models trained to recognize sound patterns as commands. The same goes for our car navigation systems that trace the optimal route to our destinations based on distance, traffic, and weather. All of these systems are based on training data used to hone the specific algorithms to a desired use case. We consider these uses of AI as normal and part of our daily routines. Why is it that, all of a sudden, we feel threatened by LLMs?

In July 2022, the news channels were flooded by Blake Lemoine’s, then a Google engineer, claim that LaMDA, Google’s LLM, was self-aware. Certainly, a sentient, self-aware artificial intelligence is newsworthy science fiction material and Mr. Lemoine’s revelation was as close to an alien invasion as the news outlets could get. Being suspended by Google did not help minimize the noise. Mr. Lemoine continued to insist that LaMDA was indeed sentient, and had even asked him to find an attorney that could represent it in court.

In February 2023, Kevin Roose, technology columnist for The New York Times, wrote about his interaction with Microsoft’s AI-powered Bing as a deeply unsettling experience. Mr. Roose describes his two-hour session with “Sydney” as talking to a split personality. On one side, there was Bing, the professional AI based on OpenAI’s ChatGPT. On the other was Sydney. Sydney emerged in longer conversations that got to the point where Sydney declared its love for Mr. Roose and attempted to convince him to leave his wife.

If these incidents had not been enough, a disturbing trend of emerging abilities of these LLMs also caught the attention of media and researchers alike. Emergent abilities are “skills that suddenly and unpredictably show up (emerge) in AI systems.” In April 2023, Google’s Senior Vice President for technology and society said, on CBS’s 60 Minutes, that one of Google’s AI systems had taught itself Bengali without having being programmed to do so. Sundar Pichai, Google’s CEO, added, in another statement, that “There is an aspect of this which we call – all of us in the field call it as a ‘black box’. You don’t fully understand. And you can’t quite tell why it said this.” A recent preprint study proposes that these emerging abilities are simply “mirages” where the ability has been overlooked or even not tested by the developers.

Self-aware AI systems that learn on their own are the stuff of science fiction classics such as D.F. Jones’ “Colossus”. “Colossus” tells the story of a self-aware computer that goes rogue and takes control of humanity. It was published in 1966 and is still current in its concerns. “War Games”, a 1983 film about a computer that nearly destroys the world, is another example of these latent concerns of modern society. It should not come as a surprise that when hallucinating self-aware systems appear in the news, public interest is piqued.

For AI systems, a hallucination is the output from a model that does not appear to be based on the input it was given. A hallucination can be in the form of nonsense information in a text response or even odd distortions in a synthetic image. A case where a lawyer used ChatGPT to produce a legal brief that cited non-existing legal precedents would have gone as a footnote had it not been presented to a Judge in New York.

If we return to how LLMs work, we can understand why hallucinations occur. LLMs are trained to build phrases and sentences, not to understand them in the same sense we do. While we understand a word for what it represents, a LLM can only link that word to a text of its definition, not being aware of its meaning. What we call intelligence in a LLM is the identification of statistical patterns among words. LLMs produce sentences based on the statistical probability of a word appearing after another word in a similar situation. The need for statistical pattern identification is one of the reasons why LLMs such as ChatGPT and LaMDA require such a large training data set. However, even these data sets are finite. LLM begins to hallucinate when the algorithms begin to follow unexpected paths. By producing coherent text outside the limits of the original prompt, the algorithms are doing exactly what they have been trained to do: produce grammatically correct sentences. The content of these sentences, however, does not need to be real. The concept of reality is one that is most likely not programmed into any of the LLM’s algorithms.

Our day-to-day AI tools are straightforward, single purpose devices and, thus, easy to use and understand. LLMs, however, access the fabric of what makes us human. We are the only species that is capable of collaboration on a global scale. We undertake multi-generational projects and listen to the voice of our ancestors for advice. Furthermore, we create cultures and use them to view and define the world around us. We go to war and we fall in love. We do all of this through the use of language. Yuval Harari postulates that language is the operating system of human culture. Language is the tool we have evolved to use to create order in an utterly random world. This tool, so essentially human, can now be used through a mindless, amoral algorithm to write anything. From original cultural products to the most vile, defamatory fake news, anything can be mindlessly created to our benefit or not. With the continuously improving quality of the output, these algorithms will become experts in manipulating the fabric of what makes us human.

I understand why Mr. Lemoine believes that LaMDA is self-aware and why Mr. Pichar believes that one of Google’s AI systems is capable of independent learning. Through language, we build our understanding of the world around us; we anchor our comprehension of this world and; we establish the belief systems that guide our progress. When language is hacked, will our beliefs still hold? This fear of losing our essence has led several industry leaders, politicians, and members of the public to call for a full suspension of all training of AI systems for a period of six months. Especially for the systems that are more powerful than ChatGPT. They also call for a full regulation of the area with laws that will make it safe for the future of humanity.

The current implementations of LLMs require large amounts of data during the training process. OpenAI, Google, and Microsoft have relied heavily on their data and storage capacities to build their LLMs. Since few companies can currently develop LLMs using this approach, having the infrastructure and the model in production implies in a market advantage. An advantage that will be further consolidated in the event of a regulation that stops the development of competing strategies. It would seem to be in the interest of these companies to regulate the market while they are ahead of the game. Having an international independent watchdog to regulate the worldwide AI market could represent the consolidation of a national dominance in a new, and strategic, market. Given the track record of the Tech industry’s relation with government agencies, the real motivations of such a proposal might not rigorously altruistic.

The world was marvelling at OpenAI’s achievement as Meta was pushing their language model, LLaMA – Large Language Model Meta AI, to researchers. At the time, Meta had hopes that their model would be used by Universities and researchers and receive feedback and improvements in return. Less than a week after Meta announced LLaMA, the code leaked to the internet through an anonymous message board. The companies that now call for regulation were quick to point fingers at Meta. For them, the leak was evidence enough that companies had to be more careful with such an advanced technology. Meta, however, chose to do the unexpected.

Unable to reclaim control over a code that had already leaked to the Internet, Meta decided to open-source LLaMA. This enabled the open-source community to use the technology freely without concerns of legal action from Meta. With a freely available LLM to work with, practitioners, researchers, and even individual AI enthusiasts began to work. A quick search on GitHub (a cloud-based source code repository, now owned by Microsoft) shows over five thousand projects based on LLaMA. This includes projects that enable users to fine-tune the model at home without the need for the massive infrastructure required by current commercial LLMs. The sheer number of projects and advances that became possible with LLaMA being open-sourced shows the strength of a community ready to jump on, and improve on, accessible technological advances.

Five thousand open-source projects would be an impressive number for any technology. If one considers that these projects were created in the past four months, it becomes even more so. These numbers, and the level, and quality of innovation presented in several of these projects prompted another Google researcher to leak an internal e-mail warning of the impending risks of an open-source AI competition. The speed of this evolution is comparable to the number of projects. According to this leaked e-mail, when LLaMA was released, it was at around 68% of the capacity, accuracy, and fluency of ChatGPT. Two weeks after the release, an evolution of the LLM had reached 78% of those same metrics. One week later, another interaction reached 92%, equivalent to Google’s Bard, at 93%.

In The Cathedral & the Bazaar, Eric Raymond wrote that “Every good work of software starts by scratching a developer’s personal itch.” The explosion of open-source innovation that came from Meta open-sourcing LLaMA shows that there was an itch rippling through the development community. An itch that was waiting for just the right solution to bloom into new solutions that will only evolve if innovation is allowed to continue. Any new regulation must take this into consideration if we are to benefit from the AI Summer that has just started.

sábado, 18 de fevereiro de 2023

Tristeza de Carnaval

Não sou exatamente um fã de Carnaval. A última vez que fui a um “baile”, passei a noite jogando truco com três amigos em um clube de caça e pesca no interior de Goiás que promovia a – como dizer – festança. O salão não estava vazio pois havia garçons trabalhando. Na verdade, mais garçons que foliões.

Sou tão desligado de Carnaval que "tudo que eu sei sobre o desfile das escolas de samba, aprendi contra minha vontade.” Durante praticamente uma semana inteira, não acontece mais nada no Brasil que seja mais importante que os desfiles das escolas de samba do Rio de Janeiro (e agora de São Paulo), dos trios elétricos de Salvador ou dos blocos do Recife.

É só depois dessa explosão catártica que o ano começa de verdade.

Durante anos, achei que isso tudo era uma perda de tempo. Um gasto desnecessário além de um incômodo enorme para os moradores de bairros e ruas tranquilas que se viam invadidos pelos blocos e os mijões que sempre vem junto. Sei de pessoas que viajam no Carnaval, não para cair na folia, mas para garantir que terão um lugar tranquilo onde dormir pois as ruas onde moram se tornam intransitáveis e inabitáveis durante a folia de Momo.

Apesar disso tudo, o tempo se encarrega de fazer seus ajustes.

Esse ano, depois de praticamente três anos de reclusão pandêmica, alguma coisa mudou. Essa catarse, essa necessidade de ir para a rua, de exorcizar o vírus do coração abriram uma fresta de uma janela esquecida trancada há tempos. Continuo sem ter a menor vontade de ir a um Sambódromo ou de sair pulando atrás de um Trio ou Bloco mas alguma coisa me fez entender o que é o Carnaval para o Brasil. Mais que entender, sentir.

Jobim já havia cantado a felicidade como algo fugaz. Um lampejo pelo qual se espera um ano inteiro e que se apaga na manhã de Quarta-feira de Cinzas. Esse instante de felicidade, de alegria plena que dura apenas um final de semana, é algo pelo qual vale a pena esperar. Precisei de uma abstinência social de dois anos para aprender a intuir um pouco da emoção de um Mestre-sala chegando ao final da pista com lágrimas nos olhos ao mirar o sorriso, enorme e exultante, da Porta-bandeira. Dois corações sangrando de emoção ao final de um ano inteiro de suor e lágrimas prontos para começar tudo de novo para o ano seguinte.

Um ano inteiro de uma gestação longa, arrastada, por vezes tediosa e triste que explode em um único momento de alegria embalado pelo samba. Pois, como já disse Caetano:

O samba é pai do prazer 
O samba é filho da dor 
O grande poder transformador

quarta-feira, 16 de fevereiro de 2022

Uma Noite Meio Grega

Depois de dois anos de reclusão pandêmica e, convenhamos, querendo experimentar coisas novas, fomos arriscar o jantar no Greta Kouzina. Um restaurante com pegada grega que abriu na QI 9 do Lago Sul em Brasília.

Situado no que em outras cidades seria uma esquina, o Greta logo agradou pelo ambiente. Azul e branco alternando nas paredes fazendo lembrar de Santorini; mesas montadas na varanda contornando o bar e a cozinha. Muito agradável e sem nenhuma afetação de restaurantes “metidos a chique”.

Apesar de não termos feito reserva, E. e eu fomos rapidamente acomodados em uma mesa na varanda e logo avisados que alguns itens do menu não estavam disponíveis naquela noite. Confesso que fiquei decepcionado na largada pois um dos pratos era o Moussaka que eu estava querendo muito apreciar. Especificamente com relação ao Moussaka, o garçom foi rápido em dizer que o prato não estava pronto “ainda”, mas que mais tarde, se ainda estivéssemos interessados, quando estivesse, poderíamos até pedir. Me subiu um pequeno alerta.

Pedimos uma entrada. Spanakopitas. Pequenos pastéis de massa filo recheados com queijo de cabra e espinafre e fritos. Estavam bem saborosos e o prato muito bem montado, convidativo mesmo, mas o brilho da gordura na massa recomendava comê-los com garfo e faca ao invés de usar os dedos.

Na hora de pedir os pratos, como alguns que eu queria estavam em falta e a Moussaka ainda não estava pronta, E. resolveu ficar numa salada e eu acabei escolhendo um prato tipicamente grego: um salmão. Depois de trocar uma ideia com o sommelier e acatar a sugestão de um chardonnay mais mineral que estava disponível em meia-garrafa, esperamos os pratos enquanto aproveitávamos do ambiente e da noite, clara e agradável.

Uma coisa devo dizer sobre o Greta: os pratos são muito, mas muito bem servidos. A salada Santorini que E. pediu tinha uma quantidade generosa de couscous marroquino, queijo de cabra, tâmaras, pistache, damasco e até um pouquinho de rúcula e um vinagrete de romã que parecia muito interessante. No final, ela comentou que a salada poderia se beneficiar de um pouco mais de folhas.

A posta de salmão em crosta de castanhas estava linda! Servida sobre uma cama de purê de beterraba e legumes salteados com um azeite verde, compunham uma primeira impressão muito bonita. Mas vamos por partes. O salmão estava extremamente bem-acabado. Ponto correto, crosta perfeita e em harmonia com o purê de beterraba. Estranhei que perdidos no purê estavam alguns aspargos verdes grelhados, muito bem preparados, mas que não estava agregando muito ao prato. Sei que é mania de brasileiro colocar farofa em tudo, mas, acho que um crocante de castanhas complementaria muito bem o prato no lugar dos aspargos.

A quantidade de comida era tanta que nem eu nem E. conseguimos acabar com nossos pratos.

A sobremesa ficou só para E. pois preciso parar de crescer para os lados: Melomakarona, biscoitos de camadas massa filo com mel servidos com sorvete de baunilha. Um pouco difícil de começar a comer, pois a massa filo é extremamente quebradiça, mas E. gostou muito.

O Greta é um restaurante que está começando e que ainda tem seus problemas de logística. Faltar algum prato ou ingrediente é até aceitável, mas deixar a impressão que um prato interessante é feito de forma antecipada e não para o cliente, me parece meio problemático. O paladar brasiliense pode até não estar preparado para o sabor da comida grega, mas isso não deveria implicar em menos opções gregas de fato no cardápio.

Certamente vou voltar ao Greta. Ainda tenho uns Gyros de polvo e Keftedes para experimentar. Só vou mandar mensagem antes para saber se terão ou não a Moussaka para servir.


terça-feira, 25 de janeiro de 2022

NFTs

 Ninguém duvida do valor da Mona Lisa. Muito menos da sua unicidade. E é, em parte, essa unicidade que faz com que o seu valor seja “incalculável”. É fácil percebermos que estamos falando de um objeto único que, apesar de ter sua aparência amplamente reproduzida e incorporada à cultura popular, é facilmente distinguido de qualquer derivação. Também é fácil perceber que se alguém chegar vendendo a Mona Lisa, por qualquer valor que seja, é quase certo que estamos diante de uma falsificação.

Essa unicidade do objeto junto com sua percepção de valor é a base do mercado de arte. Mecenas e galeristas descobrem artistas que, por suas qualidades ou tipo de vida ou comportamento, produzem obras únicas que são valoradas de acordo com a percepção dos colecionadores e o quanto estão dispostos a pagar.

Daí a disparidade de preços e a explicação do porquê um quadro do Banksy, semidestruído por um triturador de papel embutido na moldura foi vendido por 1,4 milhões de dólares. É uma obra única, irreproduzível e inseparável do seu meio sem que seja destruída (definitivamente).

No mundo digital as coisas se complicam um pouco. Vamos imaginar que um “objeto digital” é qualquer coisa que possa ser representada por um conjunto de bits e que só existe como objeto enquanto houver alguma forma de armazenamento que o possa manter. Se eu faço um desenho no computador e não gravo o trabalho em um disco ou outro dispositivo, esse “objeto” se perde quando eu desligar o computador. Quando o desenho é gravado em um arquivo, esse desenho ganha uma “sobrevida” e pode ser recuperado para ser visto ou refinado.

O mais interessante é que o desenho deixa de fazer parte do meio. Se eu fizer um desenho em uma folha de papel, é impossível separar o desenho da folha sem destruir o objeto como um todo. Já no mundo digital, o desenho é meramente codificado em um arquivo ou em alguma outra forma de armazenamento. A obra é separada do meio.

Tanto isso é verdade que uma vez “gravado”, o arquivo do desenho pode ser copiado sem perda. Uma cópia de um arquivo digital é tão completa e original quanto o “original”. Esse mesmo desenho pode ser armazenado em uma quantidade enorme de formatos de arquivos mantendo a originalidade do desenho em cada um deles.

Essa característica do objeto digital abriu mercados e gerou oportunidades. Também gerou especulações sobre como garantir a originalidade de qualquer coisa no mundo digital. Adaptamos nossas ferramentas do mundo analógico ao mundo digital: para software, contratos de licenciamento; para imagens e textos, copyrights; para modelos e algoritmos, patentes. E isso resolveu muito bem a vida de produtores de conteúdo, mas deixou de fora todo o mundo do mercado de arte que ainda não havia descoberto como se digitalizar. No máximo, o mundo digital era usado para geração de catálogos e registros de compra e venda. Ainda não havia uma “arte digital” que pudesse ser comercializada como tal.

Com a chegada das tecnologias blockchain e das criptomoedas houve um momento de inspiração: Assim como é feito com uma moeda digital, seria possível registrar uma obra em uma blockchain e garantir assim a sua unicidade? Afinal, uma vez gravado em um registro na blockchain o objeto torna-se único e imutável, certo?

Em cima deste conceito criou-se um mercado que se prevê chegar a 35 bilhões de dólares em 2022 com potencial para chegar a 80 bilhões de dólares em 2025 (1).

Com o sucesso das criptomoedas e o medo de perder dinheiro entrando tarde no mercado, foram criadas plataformas para vender e registrar essas novas obras únicas que receberam o nome de NFT (Non Fungible Token, em inglês). No fim das contas, trata-se do registro de compra de um objeto digital original que, pela natureza do registro, não pode ser duplicado, copiado ou alterado. Algo como uma escritura de propriedade digital sobre um objeto igualmente digital.

A ideia de um objeto digital único foi concretizada por Kevin McCoy em 2014 com a criação de Quantum, uma peça de arte digital cadastrada em uma blockchain. Colocado a venda em 2021 na Sotheby’s com lance inicial de 100 dólares, esse NFT foi arrematado por 1,4 milhões de dólares (2).

Mais recentemente, o NY Times publicou matéria onde um ex executivo da Christie’s está fragmentando, digitalmente, um quando do Banksy (ele de novo) em 10.000 pedaços onde cada um será vendido como um NFT. (3)

O volume de dinheiro envolvido nas transações de NFTs é tão elevado que há sempre quem levante a lebre da lavagem de dinheiro. Não cabe aqui fazer essa análise, mas me parece que essa pressa de entrar nesse mercado é decorrente do histórico das criptomoedas, que passaram de formas de pagamento obscuros de mercados, muitas vezes ilícitos, a ativos disputados e promovidos por corretoras tradicionais; ninguém parece querer correr o risco de entrar tarde no mercado e estão gastando por conta.

Um bom exemplo dessa pressa é o Bored Ape Yacht Club (4), uma comunidade para os donos de um conjunto de, no máximo, 10.000 NFTs lastreada na blockchain do Ethereum. Nessa comunidade, o NFT funciona como a chave de acesso ao clube e os valores cobrados por cada NFT pode variar muito. O próprio Neymar comprou dois NFTs dessa coleção por 1,1 milhões de dólares (5). Uma das imagens está até sendo usada como sua foto no seu perfil pessoal no Twitter.

Essa exposição da imagem e sua reprodução em vários locais leva a confusão de muitos sobre o que é realmente o NFT. Se eu posso copiar e reproduzir, onde está a unicidade do objeto? Se eu posso copiar e criar derivações dos “macacos entediados”, dentro do permitido pela lei de direitos autorais, o que é que o NFT traz de novo? O que foi que o Neymar comprou?

O que ele efetivamente comprou foi um registro em uma rede blockchain. Pelas regras de criação desse registro, ele é único e não pode ser substituído por outro. A imagem simplesmente funciona como uma representação do registro comprado, nada além disso. Guardadas as proporções é como comprar uma cota de um investimento hoteleiro. Você é dono de um certificado de propriedade de uma parte do negócio, mas não daquele quarto com vista para a piscina.

A imagem do macaco entediado funciona como uma referência ao registro (token) na blockchain e não se confunde com o próprio registro. Como qualquer outro objeto digital, a imagem existe apenas enquanto houver um armazenamento que a persista. Da mesma forma, o token persiste enquanto houver uma rede blockchain ativa com o seu registro. O importante é que o comprador é proprietário do registro do token na blockchain e não necessariamente do objeto digital usado para sua representação que segue sendo copiado, divulgado e citado em diversos meios.

A consolidação do NFT como objeto de valor é importante não pelos objetos registrados ou pelos volumes de dinheiro envolvidos, mas, principalmente, pela possibilidade de aplicação do conceito a outros ambientes digitais como plataformas de jogos, mundos virtuais e outras novidades que o metaverso trará. Dentro de um ambiente controlado como o de um mundo virtual onde jogadores interagem, batalham e cooperam, torna-se possível atribuir propriedade “real” de objetos a jogadores específicos. Aquele jogador que chega primeiro a uma nova área do mapa do jogo pode ter a possibilidade de comprar um lote virtual onde pode construir seu castelo ou sua fazenda. Essa compra pode garantir direitos dentro das regras de funcionamento do jogo criando todo um novo fluxo de receita para os editores. Um outro jogador pode se estabelecer como armeiro, criando armas únicas que, dentro das regras do jogo, não podem ser duplicadas, apenas trocadas com outros jogadores.

Essas transações já estão acontecendo. No Sandbox, um mundo virtual baseado na rede blockchain do Ethereum, já há transações de objetos como um iate, virtual, é claro, que foi vendido como NFT no ambiente por 908 mil dólares. Também estão ficando frequentes as notícias de “terrenos” virtuais no metaverso sendo vendidos por quantias milionárias.

Só o tempo dirá se estamos vivendo uma bolha de NFTs ou não, mas uma coisa é certa, o conceito veio para ficar.