Three years after its founding, Seattle’s Allen Institute for Artificial Intelligence is racking up recognition in the field of AI research – and some of its research will have an impact on the burgeoning AI market.
The institute, known as AI2, was founded by Microsoft co-founder Paul Allen in 2014 with longtime computer science researcher Oren Etzioni as its CEO. Since its founding, AI2 has spawned two spin-offs: Kitt.ai, which was created a little more than a year ago; and Xnor.ai, which made its debut this month.
AI2 has built its workforce up to 75 people, which Etzioni says makes it the largest nonprofit AI research center in North America. And AI2 is building up its street cred as well:
- AI2’s Semantic Scholar team came out on top in one of this month’s test scenarios for ScienceIE, a competition that gauges AI’s ability to extract concepts from scientific papers through machine reading, and then figure out the connections to concepts in other papers. Such specialized machine-reading programs could help scientists deal with the challenges facing the peer-review process.
- Researchers from AI2 and the University of Washington are neck-and-neck with competitors from Microsoft and IBM in a reading-comprehension contest known as the Stanford Question Answering Dataset, or SQuAD. The exercise challenges AI programs, such as AI2’s Project Aristo, to answer more 100,000 crowdsourced questions from a set of Wikipedia articles.
- A research paper written by folks at AI2 and their colleagues won a Deployed Application Award at this month’s Conference on Innovative Applications of Artificial Intelligence conference (a.k.a. IAAI-17). The paper describes an AI platform known as Phase-Mapper, which is designed to accelerate the discovery process for advanced materials.
- Ali Farhadi, who’s an AI2 researcher as well as a computer science professor at UW, has just been awarded a prestigious Sloan Research Fellowship. Past fellows have gone on to win 43 Nobel Prizes and scores of other scientific awards.
Etzioni said the institute is sharpening its focus on the moonshot challenges that artificial intelligence can address. “You can think of us as Seattle’s version of Google X, the nonprofit version,” he said.
AI2’s sweet spot is natural language processing: making sense out of data sets in ways that could be understood by an eighth-grader, or using the Semantic Scholar AI platform to organize tens of millions of scientific papers into a coherent web of knowledge.
Semantic Scholar started out as a smart search tool for computer science papers. Then it was expanded to neuroscience. Now its scope is being widened to encompass all of the biomedical literature in the PubMed database.
“We have what’s become quite a popular search engine, with millions of users – but for science,” Etzioni said. “We’re not trying to monetize it in any way.”
Etzioni said that the combined AI2-UW team created under Farhadi’s leadership is providing a great model for cross-pollination in AI research. “Together he’s built a whole that’s greater than the sum of its parts,” he said.
The work that’s going into Xnor.ai represents yet another frontier. The idea is to do the data-crunching for deep learning on your mobile device, instead of in the cloud.
“If you feel more comfortable having the data locally, then with current technology, you have a major problem,” Etzioni said. “That’s one of the major advantages of Xnor.ai: It enables you to do deep learning with the data stored locally and under your control.”
Today, AI2 has a lot of company in the field of artificial intelligence – ranging from the commercial research operations at Microsoft, Google, Apple and Amazon to other nonprofit organizations as OpenAI and the Partnership on AI. But Etzioni insists that AI2 stands out from the crowd.
“I do think that we have quite a unique niche, because we’re very focused on natural language, natural language understanding, tying natural language with vision, science and scientific patterns,” he said. “Often when we look at techniques that other people are using, and apply them in our domain, they don’t work nearly as well.”
One of Etzioni’s big campaigns is to get across the message that artificial intelligence is not inherently evil, but can bring substantial benefits to society – as long as it’s managed correctly.
“I don’t view AI through rose-colored glasses,” he said. “There is a very genuine concern about job displacement due to software in general, and AI in particular. I just feel like the ‘Terminator’ scenario is a distraction from the real concerns.”
So when will the fruits of AI2’s research start popping up in commercial products? It’s worth noting that Amazon’s Alexa Fund is among the investors in the Kitt.ai spin-off. Will we see the institute’s natural-language tricks of the trade show up in the way Alexa responds to our questions? Or is it Alexa that’ll be teaching AI2 some new tricks?
“On that,” Etzioni replied, “let me just say, stay tuned.”