AUDIE CORNISH, HOST:
Apple, Google and Amazon are all racing to build computers we can talk to, that'll understand us. Steve Henn of our Planet Money team says they face competition from a surprising place - small entrepreneurs using software they're getting for free.
MARA: Hello. I am MARA. I'm a mostly audio running assistant.
STEVE HENN, BYLINE: MARA, the mostly audio running assistant, is an app that was built by this guy.
JOEL WETZEL: Let's go for a five-mile run.
MARA: Do you want to warm up first?
WETZEL: Yes, I do.
HENN: Joel Wetzel is a software developer in Seattle.
MARA: OK. I am ready.
STEVE HENN, BYLINE: But, MARA is not something he built at work.
WETZEL: Play something by the Beatles.
HENN: Joel built his talking app on the side - weekends and evenings. Lots of really smart people in Silicon Valley believe that the company that finally teaches computers to really listen to us - to understand human speech, could end up ruling the technology universe.
WETZEL: Let's a play little more Beatles.
HENN: But, right now, there's one problem. Despite the fact that giants like Apple and Google have spent hundreds of millions of dollars, and collected billions of pieces of data, today, most voice-recognition programs kind of suck. Call my mom.
SIRI: Sorry there is no mom number for Stephen Henn.
HENN: (Laughter) I've tried to teach Siri that I do actually have a mom.
SIRI: You can use one of these instead.
HENN: So how can that runner, Joel Wetzel, working on a hobby project by himself on nights and weekends, possibly compete when Google and Apple and Microsoft are struggling? The key, is Joel Wetzel isn't really working on this alone.
LAURENT LANDOWSKI: Play Michael Jackson.
HENN: This is Laurent Landowski, one of the cofounders of a company called Wit AI.
LANDOWSKI: So that's just a very basic set up.
WETZEL: It's good Michael, too.
LANDOWSKI: It's a good one - yes, it's a good one.
HENN: Wit AI makes the voice-recognition software Joel uses in his app. And they gave it to him for free. They're giving lots of computer programmers the same software for free.
ALEX LEBRUN: More than 5,000 developers use the platform from 12-year-old kids to very large consumer electionics companies.
HENN: Alex Lebrun is Wit AI's CEO.
LEBRUN: So if it works with our French accents, it will work with everything in the world.
HENN: In some settings, he claims his software understands you better than Siri - better than Google. So why give it away for free? Because right now the goal is to get as many people talking to this software as possible. All those runners talking to MARA...
MARA: I'm a multi-audio running assistant.
HENN: ...Are actually teaching MARA to get better. It's kind of like training a dog to understand commands. The more language it gets to hear, the more positive reinforcement it receives, the smarter it will get. So, meet my dog Zephyr.
Why don't you get the treats 'cause Zephyr doesn't listen without treats.
Zeph knows a half-dozen commands.
UNIDENTIFIED WOMAN #1: Zeph, Focus. Yes.
HENN: And when she gets one right, she gets a reward. By giving its software away for free, Wit AI has essentially lined up an army of volunteer dog trainers. Those 5,000 developers are tweaking the software, improving it and teaching it not just commands, but syntax and grammar and other things Zephyr will never understand.
LEBRUN: Now, I think we are still at the dog level. But some dogs are very smart. So in about six months we - the AI learned about 120,000 expressions in English alone.
HENN: OK. Did you hear that, Zeph? A hundred and twenty-thousand commands. So far, the software is in robots, an automatic barista and, soon, a smart watch. And it's giving even the tiniest start-up a tool to take on the tech industry's giants. Steve Henn, NPR News, Silicon Valley.
Zephyr, turn on the TV.
UNIDENTIFIED CHILD: She wants to play.
HENN: Turn on the TV, Zephyr.
NPR transcripts are created on a rush deadline by Verb8tm, Inc., an NPR contractor, and produced using a proprietary transcription process developed with NPR. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.