18:59 PM

How AI is accelerating tectonic shifts in Brazil and what lies ahead for 2021

Brazil is the largest of the Latin American economies and is leading in its use of AI






Above: São Paulo business district - AI is predicted to boost the economies of Latin America.

A guest post by Anderson Thees, managing partner at Redpoint eventures 

Several years ago, Marcelo Sales took a year-long sabbatical to help scope out the next big thing. Sales is a prominent entrepreneur in Brazil, who co-founded Movile - which is now one of Latin America's largest tech startups, back in 2008. As he traveled through Asia and Europe, he studied robots and built his own robot, became intrigued with 3D printers and more. Sales realized that AI was something truly powerful that would change the world.

Fast-forward nearly five years to 2020, and we now know that's certainly the case. AI in the form of perceptual smart machines (along with blockchain and IoT) was a key trend Gartner spotted back in April 2016 - calling it "one of the most disruptive technologies over the next 10 years." Even in a challenging year due to the global pandemic, the AI market's worldwide revenues are expected to clock in at about $156.5 billion during 2020 (a 12.3% jump from 2019) and then nearly double in value by 2024, according to IDC.

In Brazil, Latin America's largest economy and most advanced tech startup ecosystem, local AI innovation was still fairly untapped in 2016. When Sales returned from his global travels, excited by the huge AI opportunity, he began asking everyone in his extensive network who was doing cutting-edge AI research in Brazil. There was not much of an ecosystem he could find, except for some brilliant researchers and graduate students scattered throughout Brazil. It was disturbing because it set up Brazil for not competing in what some have predicted to be an AI-led economy in the near future. Since that troubling realization, Brazil has come roaring back from a vacuum and emerged as a "world-class AI innovation hub."

During the annual BayBrazil Conference, which was new-normally virtual this year and spread out over the late summer, Sales and other renowned AI experts gathered online to discuss the tectonic shifts the pandemic has only accelerated tied to AI innovation solving huge challenges in a world that s barely recognizable after the chaos of 2020. And, to debate and predict what would be changing the next as we progress forward in the 2020s.

Now in its tenth year, BayBrazil is a Silicon Valley-based nonprofit organization dedicated to gathering and connecting startup founders, investors and private-sector leaders to engage in dialogues on the roles of Brazil and U.S. entrepreneurs in the global economy.

The conference attracted the likes of Google's Peter Norvig, an industry legend and longtime AI researcher. This year's line-up of AI-focused sessions included senior executives and scientists from Ashoka, Brazil's largest K-12 education group (Grupo SEB), CyberLabs (Sales' latest startup), Pinherio Neto, Pinterest Labs and Stanford University.

Here are three key insights from them about what's next for AI:

It'll "Take a Village" to Add Humanity Back into AI-led Decision Making

The ability for today's smart machines to process massive amounts of data and find patterns or discrepancies that help humans to more rapidly diagnose patients, discover potential life-saving treatments, prevent quite costly money laundering, or unearth the most relevant legal precedents in stacks of law books is impressive. But, while this new wave of intelligent automation promises to displace many people's jobs, the need for humans in this brave new world is not going away.

"Machines are, in some sense, really good at doing pattern recognitions," said Jure Leskovec, chief scientist at Pinterest and Stanford professor, during his BayBrazil Conference fireside chat. "In healthcare, a certain doctor can only examine evidence and make so many diagnoses in his or her lifetime," but Leskovec argued that medical doctors would be part of the diagnoses and suggested treatments for many years to come.

"I think they're very complementary and I see an opportunity for humans who can think much more holistically to contribute, and in the end, make smarter decisions with the help of these machines that can collect, analyze and make sense of very large amounts of data."

Pushed further on that topic, and how to instill a sense of humanity across sectors that are increasingly being impacted by AI technologies, Leskovec said: "What we're realizing, and I think it really brings people from different disciplines together, is AI technology isn't developed and used in isolation or by itself, apart from the society to which it belongs.

This is quite important because it means that there's a lot of collaboration and dialogue needed between technologists and others; say social scientists, public-policy people, even philosophers, ethicists, and so on. These communities need to continue coming together."

AI's Role in Solving Widespread Diversity Issues, Human Bias and Distrust

According to Harvard Business Review last year, "human biases are well-documented, from implicit association tests that demonstrate biases that we may not even be aware of, to field experiments that demonstrate how much these biases can affect outcomes." Many AI pundits, including Bernard Marr, have argued that AI picking up humans' biases - a growing societal concern - is by solving the classic "garbage in, garbage out" scenario in which using high-quality data is absolutely paramount.

One classic example is longtime issues with facial and voice recognition, and its accuracy across various ethnicities. As Steve Lohr wrote in 2018 for The New York Times: "Facial recognition is accurate if you're a white guy." Lohr's story cited research from the M.I.T. Media Lab that found when the person in a photo is a white man, facial-recognition software was 99 percent accurate, but there were an escalating number of errors encountered as skin types grew darker via a study on people of different races and gender.

Sales' engineering team at CyberLabs, in partnership with academic and corporate AI labs in Brazil, set out to solve this problem for Brazilians, specifically. No other organization had aimed at addressing the challenge for Brazil, which is one of the world's most diverse populations.

They started with a major data-collection effort across different data sources for anonymized Brazilian faces. Over three years, the team captured more than 13 million Brazilian faces from different angles in order to feed and train a facial-recognition algorithm created just for Brazil. Now, they're in the midst of wrapping up the analysis and teaching their AI engine the various Brazilian dialects for voice recognition, speech to text and text to speech to invent a voice DNA database that can identify a specific Brazilian's voice in just seconds.

"Brazil has a lot of issues with security and trust," said Sales. "One of the major problems we're attacking now is to create a digital database and algorithm to identify Brazilians in a way that people can trust them to buzz them into building, or to securely purchase products or services online or in stores. Due to widespread fraud and security issues, there's a loss of trust. That's why Brazilians pay a lot for credit and overpriced products."

AI's Future Will Impact Almost Everyone's Jobs and Create New Business Models

In a nutshell, yes, some jobs will definitely be going away as a result of AI's ascension, according to Konstanze Frishchen who heads Ashoka's Global Technology & Humanity work. And, it will cause "tectonic shifts" across all types of industries.

For example, "AI has huge implications on the profession of doctors because you could argue that the machine will know way more than the doctor with their "old AI," their brain, and what they learn in medical school and their own biases, their own set of limitations," said Frischen. "The other interesting thing is that machines may tell us even more about our own body and health in the future" via wearables, predictive medicine and more.

"It's not just the manual or administrative jobs that will go away, but the more complicated and sophisticated jobs will also go away or change - and community will perhaps matter even more instead of less," said Frischen. A good example Frischen cited for this is research underway at Singularity University in collaboration with the Farmers' union in the U.S. that is engaging with farmers to help design and construct better robots that are going to plow the fields and plant seeds.

Frischen and others who participated in the roundtable discussion debated new business models that don't necessarily cut out humans from getting a cut of the action. For example: one in which patients, doctors and nurses partake in training AI to make better medical diagnostics and predictions. Frischen predicted there could be a new Spotify-like model that pays out royalties for these experts participating in the training of AI so they have a stake in it too versus a "winner-takes-it-all" model for AI that rises to the top.

One thing is for sure, the early signs of AI and associated research dates back to 1956, four years before the path towards the internet age began in 1960. Like the internet revolution, AI is now poised to change everything, and its influence is only accelerating faster now.

- - -

Anderson Thees is managing partner of Redpoint eventures, the first Silicon Valley VC fund on the ground in Brazil as of 2012.

Disclosures: Anderson Thees has served as a BayBrazil board member. Redpoint eventures is a CyberLabs' investor.