Editor’s note: Artificial intelligence is being called this age’s most profound opportunity or gravest threat. Thought leaders from across North Carolina and beyond gathered last week in Durham to discuss the implications of AI on the future of work and beyond. WRAL TechWire’s co-founder Allan Maurer attended the event, and in this the second of several articles he provides a detailed look at AI from the viewpoint of Triangle companies from startups to global leader IBM. This story focuses on a new venture, Tanjo.ai, led by serial entrepreneur Richard Boyd.
DURHAM – One reason pollsters got the 2016 U.S. election results and the British Brexit vote wrong, is that “Everybody lies,” said Richard Boyd, co-founder and CEO of Tanjo.ai, in a TED-style talk to the State of Tech event in Durham.
Before explaining how AI could help separate truth from lies in marketing research, Boyd referred to the book “Everybody Lies” by data scientist Seth Stephens Davidowitz.
Davidowitz, who studied Google search and Adwords data to write the book, said in a Guardian article:
“Everybody lies. They lie about how many drinks they had, how often they go to the gym, whether they read that book….People lie to friends. They lie to bosses. They lie to kids. They lie to parents, doctors, husbands and wives. And they damn sure lie to surveys.”
Copyright, HarperCollins Publishers
Boyd, who with partner David Smith, created Tanjoai animated personas that can’t lie to address that problem when it comes to marketing research. The company recently joined Nielsen’s Connected Partner Program, getting access to its robust retail and shopper data. Tanjo’s PersonaPanels use machine learning to transform big data sources into computer modeled consumers, or animated personas.
Boyd and Smith are no strangers to machine learning, AI and virtual reality. They built and sold two successful computer gaming and simulation companies. Those included Red Storm Entertainment, with best-selling novelist Tom Clancy, Timeline Computer Entertainment with Michael Crichton, and a virtual set and camera system used by Titanic director James Cameron for the movie “The Abyss.”
They spent six years at Lockheed Martin where they founded Virtual Worlds Labs, its innovation group focused on virtual reality, augmented reality and AI. They left Lockheed in 2014 to build their own machine learning brain and start Tanjo. “Our aerospace and gaming backgrounds are part of what makes us unique,” Boyd said. “We wanted to make a game world out of synthetic people where we could test ideas.”
Getting to the truth
Using analysis of big data sources such as Nielson’s from 120 million households, Tanjo creates its PersonaPanels, The technology uses machine learning to discern “Our true values, so we build better products and relationships built on truth, not polls or focus groups,” Boyd said. “We are building a synthetic population of people. We purchase data, census data, consumer behavior data, buying behavior data, to build a deep model of human behavior.”
“You submit ideas and have them react. What do they think about your offering? You can test your ideas and see them react in real time. They don’t have biases. They don’t lie.”
You might ask it, as an example, “How people buy chicken. There are 120 million households in the U.S., but not 120 million ways to buy chicken. but it’s not five, either. You build in millions of points of view and ask, if we take this action, how will they react?” Companies today spend billions on market research using flawed methods such as focus groups and surveys of people with a variety of biases.
In a personal note, Boyd suggested another potential use for the technology. Boyd said he used Tanjo’s technology to create an animated persona of his late father, Col. James C. Boyd, from his military records and letters. “I can go visit him every day and see what he thinks about what’s going on,” Boyd said.
While computer simulations may replace people in some marketing research, there is a need for balance in determining what we should do with our effort and attention and what we should “Turn over to machines,” Boyd said. “It’s one of the most critical questions for the 21st century and the answer is changing every month.”
In the past, Boyd noted, computers could only do what humans told them to. One way machine learning is solving tough problems, he said, “Is by dumping in massive, massive numbers of examples and letting it use its own judgement. That’s how they solved some computer vision problems, such as getting an X-box to recognize a living room, something easy for humans but computationally very difficult. Microsoft “Solved it with a super computer and a massive set of examples.”
Boyd noted how machine learning has changed when competing against human players in complex games.
Warehoused content is not knowledge
Google’s Deepmind learns to play Atari video games by watching people play, then comes up with rules and beats humans, he said. Back in 1997, IBM’s chess-playing supercomputer beat the world chess champion, quite a feat in itself. But more recently, Google’s AlphaGo, in an even more impressive feat, defeated the world’s top Go player in what the New York Times called “humankind’s most complicated board game.”
AlphaGo uses advanced new techniques that allowed it to learn from experience by playing a large number of games with itself. It was so effective that the Chinese master it defeated and others now see it as a teaching tool, opening up the possibility of unorthodox moves and new ways of thinking about the game.
Right now, Boyd said, “Most data in large organizations is undiscoverable and underutilized. “Warehoused content is not curated knowledge.” A lot of it offers “Low-hanging fruit” for machine learning and AI systems that can deliver significant returns on investment.
J.P. Morgan, for instance, he said, was spending $360,000 a year on lawyer time to review contracts. “That’s a $300 million a year expense,” Boyd said. “After a one-time investment on a computer system that cost half million to implement, the system does that. So you get a $240 million annuity from that one investment.
Next, it’s about human behavior
“Sometimes,” Boyd said, “It’s just interesting to see what a machine thinks about your data. The old idea was you come up with a hypothesis and test it. With our system, you put data in and tell it to find all the hypothesis it thinks might be true.” Going through that process with a client turned up “Insights they didn’t know about their own customers.”
Sometimes, though, the machine comes up with something interesting only to the machine and not to humans. In one project, “We had our system read every legal opinion in the U.S. history. The first thing it said was it found a high correlation for the word “the.” If you really give it unsupervised license to read everything, you may get a whole set of uninteresting correlations, but often, it does find things that don’t ever occur to human beings.”
Next, Boyd said, “It’s about psychology and human behavior. Tanjo is looking at “How we understand human behavior better.”
