Non-profit artificial intelligence research company OpenAI has open-sourced a public beta of “OpenAI Gym,” a virtual toolkit to help AI engineers develop and compare reinforcement learning algorithms.
OpenAI chose to focus on reinforcement learning – a behaviorist-inspired subfield of machine learning concerned with refining actions (like decision making and motor controls) through rewards – due to the method’s broad application and positive results. In a blog post announcement, OpenAI notes that reinforcement learning has helped train robots to run, jump, make business decisions, and play board games. Google’s DeepMind defeated a top-ranked Go player using deep reinforcement learning algorithms.
The problem with reinforcement learning research is its slow development, according to OpenAI, who recognizes the need for better benchmarks and ease of setup and use. Complicated setup and use can restrict developers who lack the resources to participate. Add to that the lack of standardized testing environments. Unstandardized testing makes it difficult for researchers to compare results equally. OpenAI Gym is their attempt to reduce barriers to entry, help establish more consistent benchmarks, and proffer an environment for reproducible results.
OpenAI’s aim is to democratize AI progress and encourage free collaboration between institutions and researchers by opening patents and discoveries to the public. In opening this research to the public, the company hopes to mitigate existential risk from AI. OpenAI’s reasoning is essentially this: if everyone is empowered, no one person holds all the power.
The initiative was announced last year and includes high profile co-founders such as Sam Alton of Y Combinator, Peter Thiel of PayPal, and Elon Musk of Tesla Motors and SpaceX. Though most feedback has been positive, some AI experts, such as philosopher Nick Bostrom, have suggested that opening AI development to more people could create more chances for someone to do wrong. “If you have a button that could do bad things to the world, you don’t want to give it to everyone,” he told Wired.
The Python-based OpenAI Gym is compatible with any framework, including open source projects such as TensorFlow and Theano. Anyone who wants to participate can access the OpenAI tutorial here. Those who want to contribute but aren’t as versed in AI development can help reproduce results here.
The OpenAI Gym may not prove to be an immediately valuable resource for businesses as it will be for developers. But, hopefully, the rising ride will lift all boats. Companies have begun seeking ways to implement AI into their businesses by either employing programmers or outsourcing development. If OpenAI’s vision plays out as they’ve planned, the Gym will help both forms of development by establishing benchmarks and standardized testing environments.
Image credit: Pixabay