Juan Diego Barrera SandovalAndrea Iorio is a Brazil-based Italian tech expert who has quickly become one of the regions’ main keynote speakers and authors on technological transformation. After a business career aiding brands like Groupon, L’oreal Brazil, and Tinder LatAm, he decided to dedicate himself to written and public speaking advocacy and consulting.
Nowadays, he hosts NVIDIA’s “Vem A.I.” podcast and works with other tech giants such as IBM to help costumers understand digital transformation’s potential and the implications of new developments such as AI.
If you or your peers are still sceptical about AI, he’s the go-to guy to convince anyone that it’s just a tool, and it can be as good or as bad as its users.
“I am optimistic about what the future can be if we use AI properly, but it really depends on our own decision and posture,” Iorio told us.
Some will be thinking ‘AI is a threat to me in the workplace. I’ll try to fight it off being smarter and more efficient’. This is not going to work. The other option is assuming ‘AI is an ally, and I’m going to use it to improve my lifestyle, improve the quality of the time I have and reduce time spent on repetitive tasks so that I can focus on the creative ones.’
We interviewed Iorio, and he gave us key insights into the state of AI technology adoption in our region.
How important is the tech industry becoming for Latin America?
Technologically, Latin America is not as mature as Asia. But, compared to developed markets — such as the US and some parts of Europe—, this region leap-frogged in many aspects related to digital transformation.
Let us use an analogy: China has not had a very strong physical retail phase because it went from a small stores dynamic directly to e-commerce. I am not saying the same thing happened with Latin America, where there’s very big physical retailers. Rather, in many sectors, the rate of acceleration has been higher than it was in more mature markets. This is exactly because disruptors came in, such as very good founders. And that happened because people here are very enabled tech talent. Every year, just between Brazil and Mexico, 600,000 new software developers graduated from very good universities.
Virtually every sector in the region had very traditional companies leading the way in each country because of the generalized underdevelopment. Their inefficiencies opened opportunities for startups to disrupt. That has made the region attractive for venture capital.
Over the last ten years, the region produced the highest rates of unicorns. You name it: Nu Bank, Kavak…
There is a great opportunity for new competitors. A part of it has been already seized, but there is so much when it comes to opportunities related to AI now: cloud computing, cyber security and so on.
What has been the biggest surprise this year?
I am very surprised by the speed at which AI has been adopted. I do loads of work for companies, but also governments. And it is interesting to see even they are now using chatbots and conversational AI. There is also the generative-AI side where companies are building their language models.
For example, the state of Rio Grande do Sul, at the south of Brazil, is recognized as the most digitized region of the country. When it comes to digitalization of its public services: 99% of the services are now accessible and can be paid for online. But now they have a challenge, which is: even though everything is digitized, not everyone uses those new digitized systems. And so, they have to educate.
I am seeing lots of adoption, but the challenge now is definitely learning how to reap the benefits.
Back in 2023, Mordor Intelligence surveyed tech leaders in Brazil and 40% of them declared they were already using AI solutions. This surprised me because this technology was not mature two or three years ago. So, when we look at the rate of adoption, that is really high. AI’s adoption was extremely fast!
But sometimes, although people are very open to AI, companies or the public sector throw out tools their users or customers do not really know how to use. I think educating on their usage is the next step.
That quick rate of adoption doesn’t seem to be met with growing discussions on AI’s issues, like sustainability…
There is a not very clear understanding of the impact AI can have when it comes to climate change. It happened also with the use of crypto currency.
We need to look at the electricity needed in order to train these large language models we are talking about, because it can get as bad as the levels of consumption of entire developing countries.
The challenge is: if we want to continue to improve AI algorithms, we need to continuously train them and train them on bigger and bigger data sets. And the cap of the possibilities and benefits of training new models will be the environmental and climate impact related to the energy consumption, especially considered that oftentimes data centers are located in countries that do not have sustainable energy sources.
That is interesting for Latin America. Specially for countries like Brazil or countries in Central America that do have a sustainable energy matrix.
But when you look at the effort for the development of large language models, its basically global and done by a handful of companies that may look for those resources to operate: Open AI, Anthropic, Perplexity, Microsoft and so on know it is so extremely expensive that they’ve been looking for funding from Saudi Arabia, so additional expenses may not be a priority for them right now.
If that great limiter is not talked about as a main concern, what is?
I would say that the main concern is data security. The journey to AI requires lots of data. The more companies implement clear data strategies, the more risked and exposed their data pools turn out to be. This is why the cyber security market is increasing very fast. Their numbers are doubling, but when it comes to the number of attacks, they are also growing in double digits.
There’s a readon for it. AI has a double sword effect in cyber security: on the one side, it makes cyber security more reactive and predictive. Just like the effect it has on the health sector, where it can better diagnose and predict future outbreaks. But on the other side, it makes attacks more sophisticated. It is what they call adversarial AI. When you put those two in balance, companies fear that accelerating AI adoption will make them more vulnerable to attacks. Therefore, budgets for cyber security need to increase.
Another concern is upskilling. Company leaders think “Okay, I will start using AI so I can automate processes, make better decisions based on data… But then: will I really need my human teams? What is their role? Will they feel demotivated at work? Will I have to fire them? Will I have to replace them?”
A real AI transformation is also people’s transformation. What is the use of having generative AI adopted within the company if people do not know how to prompt it well and how to craft good prompts?
The solutions for both problems revolve around data strategy and human resources strategy.
The perfect example of a way in which both problems have been addressed in the past is Latin America’s cloud computing market. Nowadays, it is projected to grow at a yearly rate of around 15%, potentially reaching almost $100 billion dollars by 2029. This growth is spurred by the pandemics’ impact on remote work, but also very much related to concerns people have when it comes to the security of their data and the scalability of their AI models.
Now that you talk about jobs and upskilling, we’ve recently reported on how India has led the way of showing the world that AI is not necessarily a job destroyer, but that it may very much be a job creator. What are other common misconceptions regarding AI and the state of tech?
A very big misconception is that AI models make better decisions than humans, when in reality an AI model is just as good as the data we feed into it.
There is a black box problem related to AI, which is we, the general public, don’t really understand how it makes decisions. Therefore, we are not able to explain to others when needed. Pretend I work for a financial service company and a person comes to my branch and says: “Look, I’ve been denied a loan. Why? What should I do different?” Am I equipped to tell him what an AI system took into account to determine how and why he fell into a pattern that suggests our bank shouldn’t loan to him? Am I sure that pattern analysis gives me the whole picture, including what I can tell from human interaction?
There is also the idea AI never makes mistakes. It is not true.
People ask me what are the sectors that are going to benefit the most from AI, thinking of its customer services. But actually, the sectors that are going to benefit the most in the short run are the ones where you have the most data because volume makes models robust. For example, not many people know that the health sector is responsible for one-third of the global data. After that, maybe financial services come, and then let’s say retail. The bigger impact will become in the short-term to the businesses and sectors that can better train a model to reduce mistakes.
There are other industries that simply do not have enough and the assumption that it does not matter can easily end up in companies thinking AI is infallible and capable of making less mistakes than humans. Well, that is very much wrong.
What would be some key tips and tricks for someone wanting to start their company on the use AI?
The first step that I would take is to try to understand where AI can be applied. The opposite of that is just rushing and implementing AI tools that sometimes are more sophisticated or more complex than necessary. There is a lot of stuff that can be I do not know like automated without AI. It is easier and cheaper sometimes and it is not that AI has to go everywhere.
For that diagnosis, I developed a framework: the universe of AI solutions can be entirely described using a two axis matrix. One of the axis is the line between generative and predictive AI. The other one goes between backend and frontend application. The resulting four quadrants describe a type in which any AI solutions fall:
- Tell me (backend, predictive): AI used for better predictions and insights.
- Tell my customer (frontend, predictive) : predictive AI used in order to communicate better. This includes chatbots and conversational AI.
- Do for Me (backend, generative): Automations, augmentations, and simulations.
- Do for my customer (frontend, generative): generation of new products and services.
A prime example of good diagnosis and application is a startup from Chile called NotCo, which produces vegan and vegetarian food. They developed an AI algorithm that basically simulates combinations of infinite vegan ingredients, and they create their products based on that. Its actually quite similar to the work of the scientists that were awarded this year’s Chemistry Nobel Price.
After diagnosis, you want to start on a small scale, with lots of tests, remembering how Microsoft launched chatbots that were later on denounced for being discriminatory against customers.
A third step is thinking about scaling. That is when you have to think about the infrastructure behind data: cloud computing, computational power… That is why NVIDIA will not stop growing: because the whole AI Infrastructure runs on GPUs nowadays, which are very good at very specific tasks.
A key part of that is understanding all this process as ongoing digital transformation. Traditional computing depended on CPUs, which were all about acceleration and efficiency of processes, and which were limited by Moore’s law. GPUs run under Huang’s law, and we are yet to see where that switch takes us, but for now we see that automatization can be and exponentially faster process than ever before. The question now is how we are going to use that for specialization.
How do you imagine the future of AI for businesses will be?
We will see AI being stuck for a couple of years because of the high cost of investing in training new models and because of the time needed for it to install in the collective conscience and understanding. On the businesses side, we will see a phase where businesses figure out where is and where is not needed.
But in the long-term, we will see improvements in what we call artificial consciousness: the interaction between models and the real and physical. Through that exchange, we will see AI models improving more, and they will start making decisions on they will not need humans. That is very risky, that will definitely lead to some replacement of the workforce, so everything has to be very well thought beforehand.
But up until then (people say it is going to be around 2040) we are going to live through a phase where AI will slowly become more and more a part of our lives, and we’ll critically understand that it’s not the silver bullet for everything.
In the process, we will see super professional rising: the ones that use AI to automate all of their familiar tasks and gain back time to innovate more, to be more creative, to be more empathic with their teams, to be more empathic with their customers, and so on.
Add comment