The Center for Data Innovation spoke with Elliot Turner, founder and CEO of AlchemyAPI, a Denver-based firm developing artificial intelligence (AI) technologies. The company has developed advanced tools to analyze large amounts of text which developers can use to create their own software. The company has also recently begun to offer “computer vision” tools that allow users to analyze and recognize images. Both of these applications rely heavily on a technique known as “machine learning” which involves continuously collecting and analyzing new observations to make predictions about the future based on data from the past. One of the most advanced areas of machine learning is known as “deep learning”, an AI-technique to automatically construct complex models to make sense of large datasets.
This interview has been lightly edited.
Travis Korte: For those who may be unfamiliar, can you give a little background on AlchemyAPI, what your company makes, and who uses it?
Elliot Turner: AlchemyAPI is the world’s largest provider of natural language processing and AI technology in the cloud. By incorporating our language understanding and computer vision technology, we enable customers to build AI-driven applications and services via an internet accessible platform. AlchemyAPI serves customers across a wide variety of industry verticals including advertising, business intelligence, publishing, retail, social media monitoring, and more.
TK: Tell me about some interesting ways your text analysis software has been applied or interesting insights it has gleaned.
ET: With more than 35,000 developers worldwide leveraging AlchemyAPI technology to build AI-based applications, we see interesting applications of this technology every day—from targeting ads to understanding geopolitical events to predicting disease outbreaks. We’re automating things that were historically relegated to people—cognitive processes such as reading and understanding documents, drawing conclusions from those documents, and understanding photographs and videos. There is an infinite number of ways that this technology can be used so that humans can be more effective.
TK: AlchemyAPI has also moved into computer vision, and your site mentions the goal of “democratizing” access to computer vision and other technologies driven by deep learning. What does this mean? Does the average person have tasks that can be better solved with computer vision, or are you more referring to small businesses?
ET: We’re trying to make deep learning technology available to the entire world by providing access to universities, individual researchers, small companies and large corporations. We’re seeing leading-edge research being done on top of AlchemyAPI to help drive technology forward. This is not just in the area of computer science; we’ve seen a variety of academic departments—ranging from Economics to New Media Studies—leveraging these capabilities to come to new conclusions.
The average person is already benefitting from deep learning today. For instance, the speech recognition technologies in mobile phones, such as the iPhone and all Android devices, are powered by deep learning. Computer vision changes your mobile phone from something that can merely take a picture to something that gives you the power to understand and act on that photo in a meaningful way, whether that’s connecting you to more information on a product in a store or identifying a plant on a hike.
TK: You turned some heads with your API Hack Day Denver demo this past September, in which your image recognition software identified some obscure objects. How has your technology improved since then?
ET: Our September demo was already impressive but AlchemyAPI has since made a number of key breakthroughs that now enable us to consistently exceed the performance and capability of Google Image Search for computer vision tasks.
TK: Deep learning is a prototypical “hard problem” in applied computer science. In your view, what are the present technical or theoretical limitations that must be overcome for deep learning to become a larger part of our lives?
ET: Deep networks are still a relatively new idea and are not fully understood on a theoretical basis. We are now starting to see more research into the behavior and operation of these networks. These are areas that have to be explored further so that we can maximize what can come out of this as a technology. We’ve already seen amazing progress with deep learning in the past few years—speech recognition systems instantly became a third more accurate than they were before, and computer vision systems are achieving benchmarks never before seen. However, despite all of these strong indicators of success, further theoretical exploration may uncover more opportunities for improvement.