This is similar, in some ways, to the word suggestions that texting software such as iMessage (or the BlackBerry typing system) already use. But being able to read a whole e-mail and find the right way to respond to it is a much more complex undertaking.
Google engineers began working on the idea "some months ago," the company said in another blog post Tuesday describing the research. Eventually, the researchers built a special neural network — a huge network of machines designed to mimic the human brain — to run the heart of the Inbox feature.
Google attempted to head off potential privacy concerns raised by the new feature. There are no people secretly reading your e-mail and coming up with replies, the company said. As with its ads in Gmail, software is scanning the content of your messages, not Google employees. "In developing Smart Reply we adhered to the same rigorous user privacy standards we’ve always held -- in other words, no humans reading your email," the company said.
One hurdle Google's engineers faced in developing the software was making the suggested replies less, well, touchy-feely. Google Senior Research Scientist Greg Corrado explained:
"Some analysis revealed that the system was doing exactly what we’d trained it to do, generate likely responses -- and it turns out that responses like “Thanks", "Sounds good", and “I love you” are super common -- so the system would lean on them as a safe bet if it was unsure.
For the real version of the program, which will roll out to iOS and Android Inbox users over the next week, engineers have toned down the affection to make the software a bit more work-appropriate.