University of Michigan professors Jason Mars and Lingjia Tang wanted to improve upon artificially intelligent personal assistants like Siri and Google Now, but those programs were locked up in the massive tech companies that created them.
So Mars and Tang built their own.
Mars and Tang are cofounders of Clinc
, an Ann Arbor-based artificial intelligence (AI) startup. Clinc is home to Lucida, the open-source AI Mars, Tang, and their team of student researchers debuted in 2015. Mars says Lucida was intended to be a sandbox in which researchers and developers could address a rather crucial problem in the AI field: the fact that today's AIs simply aren't built to handle a lot of queries.
"If everyone used Siri all day long, we couldn't even come near to supporting that kind of computational load," Mars says. "It turns out that these applications work well because people really aren't using them that much."
And people aren't using them that much because most AIs still don't work that well. At this point we've probably all had the frustrating experience of repeatedly yelling things at Siri or another AI assistant, trying to word our command in a way the AI will understand. Mars says that's because most major AI assistants are hard-coded with a fairly specific vocabulary, forcing users to adapt their speech to the AI's programming. With Lucida, Mars and his team worked in the opposite direction. They fed massive amounts of human-generated data into their AI and programmed it with models that work like the human brain, allowing Lucida to learn from user interactions over time.
Mars says the goal is to build something like Data, the uncannily lifelike android on "Star Trek."
"You want something intelligent enough that when you talk to it, you actually feel like it's intelligent," he says. "You don't want to feel bounded in the way you interact."
From Quake to Clinc
As the driving force behind Lucida and Clinc, Mars' interests in programming and artificial intelligence have essentially been one and the same since he was in high school.
"I always made the analogy that computer programs were like little life forms," he says. "You can start with basically an empty document, an empty source code file with no text, and…you can summon, essentially, a little being into existence. That was the curiosity and fascination, when I was 15 years old, that got me into programming."
Mars created his first AI around age 15 or 16, a program he called the Quake Random Map Generation Engine. The C++ program was designed to build additional levels for the classic 1996 video game "Quake," removing the limitations that the game's fixed 30 levels imposed on players. As he'd do with Lucida years later, Mars posted his code online for others to use. It drew thousands of downloads.
"I was having a blast," Mars says.
As he went on to undergraduate studies at the University of Pittsburgh and later the University of Virginia's Ph.D. program, Mars says he was "delighted" to discover that he could "tinker" with his fascinations in the field of research.
"It was clear that the next big shift in human technology, after the Internet and after cellphones, was artificial intelligence," he says. "We knew what we wanted, we know what we want, but the open question is: how do we get there? How do we realize true artificial intelligence that you actually believe is intelligent?"
Mars met Tang while they were both completing their Ph.D.s at the University of Virginia in 2005. Although they're both young, they've already begun to attract a dedicated group of academic acolytes. Michael Laurenzano, Clinc's chief technology officer, studied under Mars and Tang while they briefly worked at the University of California, San Diego, in 2012 and 2013. But when the duo decided to take faculty jobs at U-M in 2013, Laurenzano followed them.
"I had to uproot my life, move my family, sell my house, the whole nine yards, just to study with them as a Ph.D. student," he says. "The belief I had in them, and the opportunity I sensed to work with them as a student – historically, looking back, I think that decision was one of the best decisions I've made in my life."
The emphatic response Lucida has drawn so far seems to further validate Laurenzano's choice. Tang says the Lucida team has received inquiries from numerous industries, ranging from law enforcement to automotive to finance, requesting an industry-specific AI assistant. Perhaps most excitingly for the Lucida team, their project also recently drew major investment. Last month Clinc announced that it had closed a $1.2 million round of seed funding and earned a $225,000 grant from the National Science Foundation (NSF).
That's allowed the company to increase its staff from five to 12, scaling up an operation that was previously driven by Mars, Tang, and their team of Ph.D. student researchers.
"As a few guys sitting around the lab developing things, there's only so much we can do," Laurenzano says. "With NSF support we're able to hire additional developers and infrastructure specialists to make the software super high-quality, super easy to wield, so that others can take it and go build awesome things."
The Clinc team is also venturing into some major new non-open-source projects, utilizing the same basic principles Lucida was built on. Tang says Clinc is focusing on creating assistants that specialize in a particular domain, rather than building one end-all, be-all AI.
"If you look at Apple's Siri or Google's Google Now, I think they try to give you one assistant that can do everything," she says. "It ended up to be one assistant that can do maybe weather well and I don't even know what else, because it's so limited in all the domains it's supposed to provide."
Clinc's first such specialized AI will be what Mars describes as "a Data for your personal finance." More details on that are expected early next month. But that's not the only project in the works at Clinc right now. Tang says the team is also currently working on an assistant for Ford that would provide an intelligent dialogue with drivers about how to operate a vehicle's special features.
It's a big moment for Clinc, one that Mars doesn't hesitate to compare to the moment one of Clinc's biggest competitors in the AI space had years ago. Mars notes that when Google arrived on the scene, it was just one among a host of search engines. Today, of course, it's the only one – and with good reason.
"Because [Google's] technology kind of pushed the envelope in a big way, a small, tiny nobody of a company was able to topple everything else that existed in the market," he says. "That's why startups work. That's why being able to take really, really smart people and start from scratch gives you an opportunity to build something that's quite new."