In 800 days, AI will reach the complexity of philosophy. It’s not months or weeks. It’s days. This means that business leaders who wait to see if “this AI thing” will work out won’t be able to understand the competitive landscape in three years. It will all be too fast.
The prophecy comes from Silvio Meira, the chief scientist of TDS.company, emeritus professor at the Federal University of Pernambuco, and one of the founders of Porto Digital, Recife’s center of excellence in innovation.
Silvio also sits on the boards of companies like Magazine Luiza, MRV, Tempest, and CI&T. “Between Linear B writing (the oldest known form of Greek) and Plato, it took 1,200 years,” says Meira. “Between online Artificial Intelligence and the equivalent of a contemporary Plato, it will take about 1,200 days. Not 1,200 months, not 1,200 weeks. It’s 1,200 days.”
We’re already around day 400, considering the launch of ChatGPT. An epic launch, which has already made history as the information system that fastest reached 100 million users. “100 million people doing what? Training an information system. For free,” says Silvio.
Even with AI having walked a third of the scale he imagined, Silvio himself says we’re still in the Stone Age of AI. When ChatGPT was launched, the model could process 4,000 tokens. A token is a measure that defines the amount of information the AI model can pay attention to in order to maintain coherent interaction with people. More recently, Google launched Gemini, with a capacity of 700,000 tokens. It’s enough tokens to process 60% of the entire Encyclopedia Britannica.
By 2030, Meira says there will be billions of tokens capable of understanding all the content in Portuguese created in the world throughout all times. And by 2040, there will be trillions of tokens. Meanwhile, companies still need to understand the basics: artificial intelligence is not a tool. It’s an extension of intelligence, and that means people can’t be replaced. It will also mean a transformation of businesses, to the point of creating a one-person market. Silvio doesn’t mince words to make his point.
Not even with the small ones… “If you are a small company and you’re not using artificial intelligence that’s in the open cloud, which won’t give you training, configuration, security trouble, you’re an idiot.”
Not even with the big ones… “If you’re a big company, a bank, a big retail network, a big finance business, a big university system, and you’re using an Artificial Intelligence business that’s in the cloud, you’re an idiot too.”
It’s hard to understand, but Silvio even uses chariots to try to explain everything in this conversation with the Brazil Journal.
What has been the most recurring reaction of people to artificial intelligence?
Shock combined with hysteria. Outside the computing area, everyone thinks there was a big bang. But for those who have been following closely, this subject has been talked about for at least the last 20 years. What was perhaps missing was what we could call “glue”, from a scientific point of view. Like a binding. Computing has been in companies since the 1950s and from the point of view of society in general, it was hidden for 45 years.
When did computing appear to people? In 1995. Why? Because the code was published on the internet.
The internet is not a communication environment. It is a connectivity environment, which enables you to publish code on the internet. Many algorithms at the same time. And what we saw in 2022 was the publication of online Artificial Intelligence that led to the information system that had the fastest 100 million users in the entire history of the universe.
100 million people doing what? Training an information system. Let me say it again, they’re not using this system. They’re training it. And for free. And then you made people pay to train a system that will be used later.
We are in a kind of Stone Age of intelligent computing.
It doesn’t seem so primitive though.
Let me make a parallel with writing. What we know as the alphabet, not hieroglyphics and the like, is 3,500 years old. And then it evolved into Linear B. Between Linear B, in 1,500 BC, and Plato it took 1,200 years. But between online Artificial Intelligence and the equivalent of a contemporary Plato, it will take something like 1,200 days. It’s not 1,200 months or 1,200 weeks. It’s 1,200 days!
We are reaching a point where I can go to an artificial intelligence and say, “create a theory for me, for this, for this scenario here and create a theory that serves this type of goal.” And it does.
This is something that was not expected to be happening now, and there is a surprise, even for people in the field. Who have known this subject for a long time. There’s a ‘holy crap’ moment here.
But we have the issue of errors, hallucinations. Will that end?
Errors won’t end and precisely they will create a very large set of opportunities for evolution. We will never have a model capable of understanding the universe as a whole, that is computationally impossible.
There’s no story of creating something that understands the whole world and tells us the answer to the meaning of life. But it will make us continue to evolve on a large scale and at a speed that perhaps we have never seen.
Is this happening with Google’s Gemini?
ChatGPT came so far ahead and so out of nowhere that no one expected that to happen. What happened is that Microsoft jumped straight into the opportunity, went there and put $10 billion into a defensive move.
Just to remind you, back in the beginning of social media, Microsoft did the same thing with Facebook, it looked like this and thought I’ll never do this so I’ll get in here. It bought a clause to block investments from others. No one can buy Facebook because Microsoft wouldn’t allow it, it has an option to buy it.
And it did the same thing with OpenAI. And then what happened to Google, Facebook, Apple? They have to launch something. In Google’s case, this is even more dramatic because it changes its business model. Imagine us asking the following question: Explain to me the history of written languages. Google responds to me with 50,000 links. ChatGPT explains to me the history of written languages in one answer!
But at what point are we on your AI evolution scale?
We are something like 400 days after the starting point. And we’re still not understanding a lot of things, starting with what human intelligence is. Until recently, we had two dimensions of this society’s intelligence: the intelligence of each of us and social intelligence, which is the intelligence of groups or networks of people.
And then comes artificial intelligence. It is not a technology. It is not a platform. It is a new dimension of intelligence, and I think people are getting lost here by thinking there’s a tool where you put the processes as they already exist. No.
We now have another set of intelligences at our disposal, which, in addition to being each of us and us in a network, are intelligent and disembodied agents.
Where we are, is just the beginning.
Because what defines the competence of these intelligent agents is a measure called tokens per interaction. What is the maximum amount of information to which the model can pay attention coherently, which maintains an information flow as if two people were talking? We are currently at thousands of tokens per interaction. The current version of Gemini, the paid one, works with 700,000 tokens.
In a year and a half, we evolved to hundreds of thousands of tokens. If you look at what Gemini is capable of, it is capable of doing things out of this world.
But I’m talking about the other world, compared to what ChatGPT does. Why? Because its attention zone is much larger. It processes a much larger set of data. The informational context zone is the size of 60% of the entire Encyclopedia Britannica. We’re talking about 40 million words.
And we’re going to a model of billions of tokens where it captures almost all the relevant information ever produced by humanity. So conversational context with these models, for example, there at the end of the decade, looking at 2030, this intelligence will be accessible in the conversational interface, not in the NASA or Google or Microsoft laboratory.
So you’ll have the equivalent, for example, of all the literature in Portuguese, everything newspapers have ever published, everything that has ever been done in Portuguese worldwide will be available in the conversational interface for you to elaborate things with it.
In the medium term, we’re talking about 2040, there will be models that deal with trillions of tokens per interaction. These models will start doing completely on their own the models of the world. And as the world changes around them, they will extend not probabilistically, they will not guess the answers with a certain probability, they will change the rules themselves to understand what this new world outside is.
And how are companies dealing with it?
Companies tend to dramatically simplify AI and start having some non-trivial problems.
Want to see one? Air Canada’s Artificial Intelligence-enabled chatbot. It was put online to take people out of front-end service and reduce 24-hour service costs. What did the chatbot do? Out of nowhere, it gave birth to a refund policy for a customer. And the customer demanded it to be followed. The cause went to court and Air Canada claimed it was a system error. The court said there was no system error, that the company put the robot in the air and had to pay the customer according to the refund policy the chatbot created.
What’s the problem? The problem is that the company didn’t understand what AI is.
If you decide to put a robot in the front line, suddenly, the customer will ask questions that can lead to completely absurd answers, for which the business will be responsible.
Companies have not understood that you are not looking at some classic information processing environment, which is deterministic like when you put CRM, an inventory system, or a logistics system where you write some rules for that digital behavior.
What we’re talking about here now is a system that you don’t know a priori what it’s going to respond. I myself have several intelligent agents – little robots I created – that write text, and one of them is a law expert.
The other day it surprised me by creating a piece of the Brazilian Constitution. I had given him the entire Constitution – and yet he invented a new piece. So there’s a fundamental difference here. This is not an exact information system, it’s a creative information system and this creativity problem is what will make the difference.
Business leaders must all be pulling their hair out not knowing what to do.
The opportunity now is not to say, ‘I’m going to fire my call center and put an AI agent in.’ Far from it. The opportunity now is to find out how I can use people and intelligent agents in networks. Not to replace repetitive cognitive jobs with AI. This can increase the complexity of business problems. The question is to expand intelligence. Because if it’s a simple replacement, the CEO himself may be replaced at some point.
This is a point that makes everyone nervous, since AI replaces a higher technical level. We’re not talking about the factory floor as in other technological revolutions, are we?
What we found is that the factory floor employee is very difficult to replace. For example, the driver on the streets of New Delhi or in the outskirts of Rio de Janeiro, São Paulo, and Recife. These people are extremely difficult to replace. Why? Because they work with a mutating and very difficult to encode context. Imagine being an autonomous car in Maré, Rio de Janeiro, or in the Complexo do Alemão.
It’s totally different from the streets, the cities, the interior of Germany, where everything is predictable. All the traffic signs are the same size, and they’re all the same height.
But in what context will companies use AI?
We will start using artificial intelligence to create markets for one person. I’ll be able to capture so much information about the person and keep that information consistent with an interactive discourse, that eventually each person can be treated as a market for a specific purpose.
There will no longer be categories like “university professors who are 70 years old”. There will be the Silvio Meira category. And the type of service I’ll have is completely different from the type of service another computer professor – who is also 70 years old and lives in Recife in the same neighborhood as me – will have.
I have a law for companies and clouds. If you’re a micro, small, and medium-sized enterprise and you’re not in the cloud, you’re an idiot. If you’re a large or gigantic company, and you’re in someone’s public cloud, you’re an idiot too.
There’s one basic thing here: if you’re a small business and you decide to process your own information, you’re a complete idiot, you’re wasting time on something that’s totally irrelevant to you. But if you’re a business that does thousands or millions, tens of thousands, millions of transactions a day in the cloud, you’re an idiot because you’re throwing away a business opportunity, not only to participate in the cloud with these companies or to do service for these smaller companies.
For artificial intelligence it’s the same thing. If you’re a small business and you’re not using artificial intelligence that’s in the open cloud, that won’t give you training, configuration, security trouble, you’re an idiot. If you’re a large company, a bank, a large retail network, a large financial business, a large university system and you’re using an artificial intelligence business that’s in the cloud, you’re an idiot too.
Why? Because it has transaction costs. You pay for tokens consumed and produced in the interfaces.
Are companies in Brazil effectively doing anything or just watching for now? There are a lot of people doing a lot of things.
All the companies where I sit on the board are working with artificial intelligence models, whether it’s solution design, solution delivery, or solution development. Nobody’s standing still. But the environment I’m seeing everywhere, where it’s being done responsibly, is one of learning, experimenting, and using with caution, to see how these models behave from the standpoint of three things: effectiveness, whether or not it solves the problem without creating another problem; efficiency, solving much more quickly; and economy, cost.
But the fact is, anyone who is standing still now waiting to see if this thing will work in the future, thinking that now AI doesn’t solve anything for their company and will take a look only three years from now, those have a problem because down the road they’ll have to take leaps of years and won’t be able to understand the competitive landscape.
If it’s already difficult to understand now…
That’s it. The way it is, it’s already very complicated. Going forward, it’s going to get much more complicated. Companies, of all sizes and all markets, that will be most successful with artificial intelligence are those that will understand that artificial intelligences did not come to replace people, but to work alongside people and groups of people in favor of business models for problem-solving.
If I can get each person to work for ten, I have to go after 10 times more market – and not fire. If I fire the people from the call center, I have to hire IT to operate the artificial intelligence that will talk to the customers.
Those who know how to make this leap to the future using Artificial Intelligence to empower, to extend, to increase the capacity and reach of their people – also simultaneously changing the level of people – will survive in the end. What we’re talking about now is surviving.
Just to compare, some companies that digitized their business models in the last 25 years went bankrupt. Why did they do that? They took the analog business model and just put a digital coat on it. Sometimes it just gets more expensive to execute the business model. Those that survived, grew, and became giants were those that transformed the business model.
Take Magazine Luiza, where I am a board member. In 2011, the company began the digital transformation process. It transformed functions, methods, foundations, and rewrote everything from scratch digitally. The result is a company today 50 times larger than when this process began. It’s not something you just took and said ‘oh, let’s computerize Magazine Luiza’s store.’ No. Transform the store, transform the seller. Transform the digital Magalu. Transform the shopkeeper. And then you do a completely different business.
Whether the market understands or not, that’s another problem, completely different. The mistake people can make with artificial intelligence is exactly to take their pre-AI business processes and artificialize them. You need to be aware that there is a new dimension of Intelligence and start from scratch.
And it’s very difficult for companies to think.
If it was difficult with digitalization…
But if you don’t think from scratch, you keep running in the past. In the vast majority of contemporary companies, you don’t have a guarantee for next year’s budget from last year’s. You can’t, because otherwise you don’t change, you don’t go after process improvement, you don’t go after change of anything. So what’s happening now is a radical change in the context of intelligences, there’s a new dimension of intelligences, whoever doesn’t try to understand this won’t survive in the next 15 years.
This you can write with all the letters: either people and companies understand Artificial Intelligence as a new dimension of intelligence, not as a set of tools, technology platforms, or they won’t survive.
There’s a starting point here that was the paradigm shift from animal traction to combustion engine. Do you know how many carriage factories, horse-drawn carriage factories, and horse-drawn buses successfully built a car, a pickup truck, and a bus powered by the combustion engine? Zero. Because they were looking at the engine and saying oh, but this thing is noisy, breaks a lot, needs a gas station. And the people from the combustion engine went there slowly, made the engine a little better, made the engine a little less noisy, made an engine that breaks less, made an engine that consumes less fuel, starts to install gas stations…
This transition took about 30 years. When did the internet start? 1995. So much has happened since then that we think it was in 1915. But it was in 95. Next year we’re going to celebrate 30 years of commercial internet.
But it only became broadband in 2005, we’ve only had smartphones since 2007. If we look, there are only 15 years of real internet, the next 15 years that will be accelerated by Artificial Intelligence are going to be the years of real impact of the internet, because then the game will change completely.
Whoever was still surviving even hiding from the internet, from Information Technology somewhere, won’t survive and that’s a fact. As Peter Drucker said, the ultimate goal of innovation is to survive. It’s not to create a product, change a process, none of that. It’s to survive.
