It may seem as if everyone’s already on the bandwagon for artificial intelligence and machine learning, with players ranging from giants like Amazon and Microsoft to startups like Xnor.ai and Canotic — but the head of Seattle’s Allen Institute for Artificial Intelligence, or AI2, says there’s still plenty of room to climb aboard.
“Let me assure you, if you have a machine learning-based startup in mind … you’re not late to the party,” AI2’s CEO, Oren Etzioni, told more than 70 people who gathered Tuesday evening at Create33 in downtown Seattle for a Startup Grind event.
Etzioni had a hand in getting the party started back in 2004, with the launch of a startup called Farecast that used artificial intelligence to predict whether airline fares would rise or fall. The company was acquired by Microsoft in 2008 for $115 million and has since faded into the ether. But Etzioni said the basic approach, which involves analyzing huge amounts of data to identify patterns and solve problems, is just hitting its stride.
The potential applications range from spam detection and voice recognition to health care, construction and self-driving cars.
“It’s really a versatile technology, and we’re going to see more and more startups based on machine learning,” Etzioni said.
He demonstrated one of the applications for the Startup Grind crowd, First, Etzioni played a series of short, narrated video clips advertising vacations, fashions and home loans. Then he asked the audience to guess what innovation was reflected in the clips.
Several attendees guessed that the images were assembled by an AI agent, but Etzioni said AI produced the voice rather than the pictures. The videos served as a sneak peek at a next-generation text-to-speech conversion program produced by one of the stealthy startups working with AI2.
“The goal isn’t to create commercials,” Etzioni said. “But think about somebody who can’t speak. All they can do is type, but they don’t want to sound like ‘Ste-phen Haw-king’ … with apologies to the late Stephen Hawking. This is really quite natural, and all it requires is to type, and you can get a variety of different voices.”
Another startup that’s maturing in AI2’s incubator, Blue Canoe Learning, has developed an app that uses speech recognition and machine learning to help users pronounce their English better.
“Why is pronunciation important? A, there’s a bunch of people I know, and I can’t quite understand them,” Etzioni said. “That’s a smaller thing. But a much bigger thing is, think about the developing world, where, say, if you live in Bangladesh or the Philippines. Your road to the middle class is through a call center, but to get that great job … your English pronunciation has to improve. Well, they have built exactly the technology that supports that.”
Etzioni said Blue Canoe is already selling the software to clients in Asia and internationally, as well as to tech companies in Seattle.
Such applications are examples of how AI2 reflects the mission it was given by its late founder, billionaire entrepreneur-philanthropist Paul Allen — a mission reflected in the institute’s motto, “AI for the Common Good.”
“We don’t try to help people click on ads, or build surveillance products,” Etzioni explained. “We have a variety of companies that are trying to generate AI systems that make the world a better place.”
On other topics:
- Etzioni isn’t worried that AI agents will become sentient and take over the world, but he is concerned about the potential impact of AI and automation on the job market, as well as the potential for all-too-human rogue actors to use AI for malign purposes.
- “Deep fakes” — the ability to generate fake videos or documents that look and sound all too real — represents one of the biggest danger zones for AI applications. As an example, Etzioni pointed to OpenAI’s decision not to release the full version of its GPT-2 text-generation model due to “concerns about malicious applications of the technology.” The trend may force the creation of new methods for authenticating trusted information sources, Etzioni said.
- The essential fuel for AI ventures is data, and especially labeled data, Etzioni said. “If you don’t have data, you can’t build the machine-learning program or startup,” he said. Etzioni recommended a recent article in Harvard Business Review, written by computer scientist Andrew Ng, as providing “really good common sense” for AI entrepreneurs.
- Safer roads are likely to be among the biggest near-term payoffs from AI, thanks to self-driving technology. “What we need are safety devices in our cars, followed by autonomous vehicles. That’s coming, and that’s great,” Etzioni said. The technology should help less experienced drivers as well as elderly people who want to stay mobile.
- Etzioni pointed to a couple of interactive apps created by AI2 to demonstrate how AI works: One of them, accessible through AllenNLP.org, demonstrates natural language processing. Another, at imSitu.org, focuses on image situation recognition. Just this month, AI2 released a game demo called Iconary in which humans cooperate with an AI agent to solve picture puzzles. “Don’t believe in anything you can’t play with yourself,” Etzioni said.
