<strong>Generative Artificial Intelligence: everything you should know</strong>

In February 1996, Garry Kasparov, the world chess champion, played against IBM's Deep Blue . Kasparov lost the first game but eventually rallied to win 4-2. The following year, there was a rematch, with a closer result: 3.5-2.5. The whole world watched in astonishment as a machine could beat one of our idols. Kasparov later wrote that he thought he would win " no matter what ," since " it was just a stupid machine ." After these two games, he softened his opinion: " I could sense a new kind of intelligence on the other side of the table ."

 Image generated by Stable Diffusion with the prompt “ kasparov deep blue machines artificial intelligence ” 

In this context, many wondered if these events might be foreshadowing a world in which the artificial intelligence developed by these machines and programs could pose a threat. Artificial intelligence (AI) is a highly advanced way of automating decision-making to solve problems. The history of computing had developed with very precise programming rules; that is, we told the machine what to do. However, artificial intelligence seeks, in some way, to emulate human functions . The example of chess clearly illustrates what specific artificial intelligences are. They are not general intelligences. However, they do perform tasks for which humans need intelligence. What is new about contextual intelligence systems like Google Maps is that they do not require vast amounts of data and extensive learning time to be optimal (and thus beat Kasparov). The new way to approach this is by learning from the data these programs generate as they make decisions.

But in 2023, we're entering a new era. Generative Artificial Intelligence is a type of AI based on machine learning and deep learning methods. These methods gather information about specific elements, which are then used to generate entirely new and realistic ideas. These systems are what I call the " probabilistic charlatan ." By " charlatan ," I mean someone who knows everything and is always talking. By " probabilistic ," I mean that it's a probabilistic calculation of the most frequent occurrences. That is, if I ask it, "What is 3 + 2?", the dialogue system doesn't have a mechanical process to perform that calculation. Instead, it searches its database of texts to find the most common result. It will say 5 if that's the typical answer in the texts it has. But if, for example, there are jokes like "What is 10 + 1? ", which can have different binary or decimal results, it returns strange results because it gets confused. It might say 11 or 3. It's the " probabilistic charlatan ," where the competition is to have the largest number of conversations in its database to find the results closest to reality. And where hackers or people wanting to sabotage the process might appear, generating random, incorrect texts to mislead these machines.

According to Gartner's latest study on emerging technologies, Generative Artificial Intelligence is one of the most prominent strategic trends of the moment. In fact, it is estimated that by 2025 this type of AI will represent 10% of all data produced, compared to less than 1% today. The report also reveals that companies that adopt these practices will see their value grow up to three times more than those that do not implement their AI approach.

Do these machines threaten to think and reason better than humans? Not so fast. They reach conclusions purely through probability calculations (plus user feedback). Humans don't think the same way. A 5-year-old doesn't think the same way as a 50-year-old. A person with extensive professional and personal experience doesn't think the same way as someone who hasn't left their local context. Life is sometimes made up of those small, anomalous moments where human reasoning can save us from an accident or prevent us from investing where we shouldn't. It's about creativity in unforeseen and complex environments. And that's where we humans still hold the monopoly.

Generative AI offers limitless possibilities. It can be used in activities ranging from software code creation to 24/7 customer service, as well as drug development and research. These countless applications, of course, can also be misused. One of the disadvantages of this AI is the potential for scams, fraud, disinformation, and the creation of fake identities. However, GANs go beyond images and extend to videos as well. The famous and controversial "deepfakes" superimpose one person's face onto another's to falsify their gestures and voice. In this way, it makes people believe they are saying or doing something that never actually happened. Generative and discriminatory networks offer such realistic results that they are indistinguishable from the real thing.

 Image generated by Stable Diffusion with the prompt “ human machine relationship in the future ”

But beyond understanding what they do, it's good to think about why they do it and how we can use it. These tools help us on two levels of work: Creativity and Productivity. What automation of any kind replaces to this day isn't abstract work, but tasks. A task can be " making a shopping list ," " finding scientific articles to write a paper ," " summarizing articles into key ideas, " or " what thesis can I defend on a particular topic ?" The tasks where AI can help us most are those of solving already known problems. Knowledge of geography or history, or mathematical calculations, are relatively easy. So are the steps for building an airplane or a train. However, where it falls short is in defining a question that no one has asked before. If we said that the probabilistic charlatan only knows how to answer what he considers highly probable, logically, when faced with a new field of study or an original question, he won't know how to work. Has creativity died? Rather, I would say that inspired creativity has died, like that of those painters who looked to others to define their own style. There are options to improve humans, but not that painter who defined a new style from scratch.

To find precise information on something to build an argument around, we recommend using perplexity.ai and elicit.org . To articulate ideas in a text and ensure they connect coherently, ChatGPT or neuroflash are unbeatable. For rewriting in different formats (for example, a hypothesis formulated by someone else, but narrated in our own style), I rely on ChatGPT. We can use simplified to create content for social media. Colossyan helps us make videos with professional actors, while Jasper allows us to create images by mixing styles and contexts/locations. For writing software code in any programming language, nothing beats GitHub Copilot . In fact, the name of this last tool reflects how we at Brain & Code envision the future of human-machine coexistence: copilots (technologies) accompanying pilots (humans).

Will Artificial Intelligence surpass Human Intelligence soon? This is the question many people are asking in the wake of the disruptive emergence of generative AI. In image and language recognition, it already has (see paper and graphic representation below). It surpasses us. And in other cognitive tasks? Good question. According to several experts, there is a 50% probability that human-level artificial intelligence will exist (not the current kind, but the "complete" kind). 90% of them believe this will happen within the next 100 years. On metaculus.com (a predictive information marketplace I've mentioned before), you can see the market for estimating Artificial General Intelligence in this thread.

 Machines are beating humans