Traffic Payout

Google started as a research project—and research has remained a core part of our culture. But we also do research differently than many other places. To shed more light on Google’s unique approach to research, Peter Norvig (Director of Research), Slav Petrov (Senior Research Scientist) and I recently published a paper, “Google’s Hybrid Approach to Research,” in the July issue of Communications of the ACM.



In the paper, we describe our hybrid approach to research, which integrates research and development to maximize our impact on users and the speed at which we make progress. Our model allows us to work at unparalleled scale and conduct research in vivo on real systems with millions of users, rather than on artificial prototypes. This yields not only innovative research results and new technologies, but valuable new capabilities for the company—think of MapReduce, Voice Search or open source projects such as Android and Chrome.

Breaking up long-term research projects into shorter-term, measurable components is another aspect of our integrated model. This is not to say our model precludes longer-term objectives, but we try to achieve these in stages. For example, Google Translate is a multi-year project characterized by the need for both research and complex systems, but we’ve achieved many small objectives along the way—such as adding languages over time for a current total of 64, developing features like two-step translation functionality, enabling users to make corrections, and consideration of syntactic structure.

Overall, our success in the areas of systems, speech recognition, language translation, machine learning, market algorithms, computer vision and many other areas has stemmed from our hybrid research approach. While there are risks associated with the close integration of research and development activities—namely the concern that research will take a back seat in favor of shorter-term projects—we mitigate those by focusing on the user and empirical data, maintaining a flexible organizational structure, and engaging with the academic community. We have a portfolio of timescales, with some researchers working with engineers to rapidly iterate on existing products, and others working on forward-looking projects that will benefit people in the future.

We hope “Google’s Hybrid Approach to Research” helps explain our method. We feel it will bring some clarification and transparency to our approach, and perhaps merit consideration by other technology companies and academic labs that organize research differently.

To learn more about what we do and see see real-time applications of our hybrid research model, add Research at Google to your circles on Google+.

Posted by Alfred Spector, Vice President of Research and Special Initiatives