Collaboration marks the future of artificial intelligence
Joseph Spisak manages Facebook’s artificial intelligence team, which undertakes its projects with the open source philosophy – this upholds the sharing of data and code as a means of collaboration between developers.
The concept of open source, the approach that advocates sharing data and code between developers, has been a constant in the technology community for many years. Big companies, however, were concerned about patents and getting ahead of the competition, so they used to view this type of cooperation with suspicion. Nonetheless, in recent years they have begun to adopt these dynamics to boost their teams’ creativity and broaden their learning with the knowledge that mass collaboration can provide.
Facebook started this philosophy by opening up its platform to external developers, who would be able to create specific applications and feed off Zuckerberg’s company’s resources, whilst at the same time Facebook could offer new services and learn from their design. This is the same philosophy that Facebook AI Research follows with their PyTorch platform, which provides a single interface for research on artificial intelligence, working with machine learning and creating new products and services.
Joseph Spisak works as manager of this team and brings this methodology with him from his days as a chip developer specialising in video and mobile technology: “Seven years ago I became interested in deep learning and machine learning thanks to open source groups”.
From that period, Spisak is still aware of the importance of collaboration: “In the field of artificial intelligence you can’t be the leader if you are not open to collaborating“.
In fact, collaboration has gone from being solely between developers to being open to other companies such as Amazon, Google and Microsoft since, in Spisak’s opinion, “the more software and hardware is compatible with PyTorch, the easier it will be for artificial intelligence developers to build, train and promote state-of-the-art deep learning models”.
The latest platform improvements create an easier to use interface that brings together all the information on a single channel and makes it accessible to less specialised people. For this reason, Facebook has partnered with online course platforms so that they will use their platform as the basis of their teachings. Even the UC Berkeley university (California, USA) has used this platform to work with algorithms that automatically identify images.
One of the achievements of PyTorch has been to improve machine translation. For example, in May 2018, developments in PyTorch allowed Facebook to perform 6 billion translations of text per day. This is just one small demonstration of what collaboration between developers, companies and educational institutions can do to boost artificial intelligence.