In this episode, Melody welcomes Michael Nguyen, VP of Global Data Practice and Partnerships at Innodata, Inc. Michael shares insights into all that AI is (and isn’t), the advances that have been made in ground truth data collection, and what it will take to overcome the biases that are still present in machine learning. 

GROUND TRUTH DATA IN MACHINE LEARNING

Ground truth data is a collection of data that can’t be manufactured, it has to be captured.  Primarily used by companies who develop products for AI/ML, there are several critical components to consider when building a ground truth data set, including identifying what type of data you actually need. Michael points out data that isn’t usually readily available, such as human interaction data, and how his team overcomes the hurdles to secure it. 

THE HUMAN SIDE OF AI

Some data are very easy to come by, while some are significantly more difficult to secure.  Any data collection that has to do with humans is more sensitive, and Michael highlights the ways that they address concerns and protect privacy for anyone who is sharing data. Object and environment data collection is pretty straightforward, while speech, simulation, and human data are much more difficult aspects of AI to capture. The only way to mitigate bias with facial recognition is to capture as much data as possible. 

THE EVOLUTION AND FUTURE OF AI

In the wake of the pandemic, one major focus of AI is turning to healthcare. From robots doing surgery to self-driving cars, AI will continue to improve the quality of life for everyone and take care of things that humans are not best suited to take care of. 

Michael’s Insights

“Basically anything you can think of, AI can be a part of it.” [3:50] 

“When it comes to human interaction, it’s really hard to create synthetic data because everyone is different.” [11:24] 

“I don’t think you can eliminate biases in AI, but what you can do is get as much data as possible.” [24:53] 

[29:18] “AI can help us do a lot of things we can’t do ourselves, especially in healthcare.”

Michael’s Bio

Michael Nguyen is the VP of Global Data Practice and Partnerships at Innodata, Inc. He is an action driven and technology focused business development professional with over 20 years of experience delivering state of the art products and services to global enterprises. For the past five years, Michael has been focused on ground truth data for artificial intelligence and machine learning. 

Links

Michael Nugyen LinkedIn

Innodata

Data is the fuel that is going to carry companies into the future. In this episode, Melody is joined by Florin Tufan, co-founder and CEO of Soleadify, a platform that mimics the human approach to research on companies at massive scale, using a combination of artificial intelligence, machine learning, and natural language processing. Together they discuss the opportunities that are made possible with AI, what decision makers need to know about AI, and how AI can be leveraged to solve real world problems, especially in the wake of the pandemic. 

SOLVING PROBLEMS WITH ARTIFICIAL INTELLIGENCE

Florin admits that his initial impression of AI was largely shaped by Hollywood, and he was somewhat disappointed when he realized that these machines with the power to take over the world were not nearly as intimidating as he once assumed. Once he discovered that AI is an evolving process with change happening at a very fundamental level, he recognized the incredible possibilities available with AI.  At Soleadify, they are now stacking models to accomplish their objectives.

INCREASING SELECTIVITY OF STORED DATA 

Nearly 90% of data is considered dark data, which is not only a drain on resources but is also a major liability to the companies that are storing it. Is there really such a thing as too much data? And if so, how can we become more selective about the data we choose to store and how we use models to generate insights from that data? Florin points out that the more data the better only applies when we know the context in which the data was created.

LEVERAGING AI TO SOLVE REAL-WORLD BUSINESS PROBLEMS

At the beginning of the pandemic, Florin and his team decided that, rather than panic about something they didn’t understand, they would focus on something they did understand.  Within a week they built a free resource for finding medical suppliers, which was a major problem at the beginning of the pandemic. Since then, procurement and supply chain has become a major use case at Soleadify. 

AI MYTH BUSTING 

Communicating with non-technical decision makers can be a challenge for any AI technology based company.  Florin reveals what he would do with a magic AI wand - he wishes that  leaders already knew the basic level methodologies as well as more realistic expectations for what is possible with AI. It is harder than most people might expect, but incredible things are possible with AI right now. 

Florin’s Insights

“The more data the better, provided you know the context in which the data was created.” [17:05] 

“A lot of this crisis was a data crisis, of course I see data problems everywhere.” [24:35] 

“I wish the myth of AI was already busted at a larger scale.” [31:55] 

“Everything is right in front of us, and I like to think of data as the fuel.” [44:15] 

Florin’s Bio

Florin Tufan is the CEO and co-founder of Soleadify, a platform helping businesses get accurate and in-depth data on companies of all shapes and sizes with great SMB coverage. Soleadify is a big data startup that leverages advanced scraping and ML technologies to ingest hundreds of billions of webpages every week and perform at scale human-like ML based research to create and maintain a database of 50+ data points on 70M+ companies.

Links

Florin Tufan LinkedIn

Soleadify

Genetic programming is a technique of evolving programs that have the potential to overcome existing limitations and take AI to the next level. In this episode, Melody is joined by Howie Altman, co-founder and CEO of Perceiver AI, where their mission is to elevate the science of artificial intelligence. They have developed a novel form of AI-based on the foundation of genetic programming which uses an evolutionary engine to optimize through a process of natural selection. Howie highlights the genetic programming process, the potential that it has to alter the path of AI, and the extensive applications that it can solve today and in the future.

ARTIFICIAL INTELLIGENCE AND GENETIC PROGRAMMING

There is a gap between AI problems and the technologies currently available to solve them. Most current AI technologies are based on neural networks and deep learning, which are incredibly effective but utilize pattern matching as their main problem-solving tool. Additionally, the black box problem, human bias, blind spots, and the need to start from scratch every time all contribute to limitations that have yet to be solved within AI. Howie highlights the benefits and alternative solutions offered by genetic programming, using the evolution of breeding dogs as a real-life example of the same process.

AN OVERVIEW OF THE GENETIC PROGRAMMING PROCESS

Starting with the genome, Howie walks listeners through the possibilities that are enabled by genetic programming in a matter of seconds. After running data through Perceiver AI, the result is millions of generations of solution candidates that are tested against a fitness function. Success comes in utilizing more than just error rate, but also measuring information gained from generation to generation, identifying stagnation, and implementing extinction when necessary. This process helps to overcome the major issue of plateauing in a matter of minutes or hours, depending on the complexity of the problem being solved.

UTILIZING INTELLIGENCE ARCHITECTURES

At Perceiver AI, they believe that the universe has an infinite number of intelligence architectures, i.e. frameworks by which to reason, judge, understand or problem solve. Perceiver AI aims to come up with the optimal intelligence architecture for any number of scenarios with the data at hand. With sufficient data (and generally, far less data is required than for neural networks/deep learning), problems can be solved with minimal human intervention by identifying all weaker solutions and discarding them while simultaneously evolving along the way. Each step in this evolutionary process of genetic programming takes us closer to achieving the sought after solution.

PRACTICAL APPLICATIONS OF GENETIC PROGRAMMING

From improved crew scheduling and fueling for commercial airlines to minimizing power consumption at telecommunication data centers, there are endless optimization solutions available with genetic programming. Howie highlights carbon credits, fuel savings, CO2 emissions reduction, and waste reduction and offers listeners insights into the variety of real-world possibilities for the future.

Howie’s Insights

“Whereas neural networks are the digital version of the human brain, … genetic programming is the digital version of evolution.” [13:31]

“As long as you can define it mathematically and in code, you can have any goal in AI you want.” [17:22]

“What Perceiver [AI] is doing is coming up with the optimal intelligence architecture for the problem at hand.” [28:20]

“These are practical, real-world applications, not just research-type things that Perceiver [AI] excels in.” [38:24]

“The technology path that we’re on is just going to continue to explode.”...