Google I/O has been a bit low-key in recent years — no surprise with the ongoing global pandemic. This year’s Google I/O still wasn’t the personal soiree we’ve grown accustomed to, but there were a surprising number of hardware announcements like the Pixel 6a and Pixel Watch that grabbed the most attention. One announcement could turn out to be the most important in the future, though: The auto-summary announcement, which debuts in Google Docs, leverages Google’s years of AI advances to potentially save us all a lot of time. It’s also fascinating from a technical perspective.
Attention to detail
Google has relied heavily on machine learning for years, but unfortunately this stuff is deviously complex and opaque to users. Google and other companies regularly report huge breakthroughs in the way AI understands the world, but it’s hard to understand the importance of these advances until they seep into features we use every day. But if they do, hoo boy.
The AI magic of Google Photos comes to mind: it is incredible at classifying images† On numerous occasions, I’ve gone from vaguely remembering a photo to actually finding the exact image in Google Photos in just a few seconds. This is a demonstrable example of AI making technology more useful, and automated summaries in Google products could be equally innovative as it moves from a tech demo and esoteric AI theory to a real product.
The line of Google AI summaries goes back to the “transformermodel conceived by his researchers in 2017, leading to language processing and generational systems such as GPT-2 capable of self-supervised pre-training. That means much of early AI training takes place without human oversight, so it’s faster and more adaptable than older methods like recurrent neural networks. More recently, Google has refined these techniques for the Pegasus text summary model in 2020.
Transformers have been instrumental in natural language understanding (NLU) and natural language generation (NLG) because they can mimic something you and I take for granted: attention. Our brains have the ability to focus on various tasks, filter out irrelevant information…or at least what your brain does. thinks is irrelevant. This allows you to focus on a difficult problem, but it can come back to bite you, such as when you’re tinkering with your radio or (god forbid) texting while you’re supposed to be driving a motor vehicle. Getting a machine to do this too (or hopefully better) is no small feat, but Google has created a model with Pegasus that allegedly focuses on the right data.
To train Pegasus, Google provided the AI with news articles and other lengthy web documents, but it deleted entire sentences and challenged the machine to recover them. Yes, that sounds like something even us humans would struggle with, but a computer can browse data 24/7 until it starts to understand language. While Pegasus is capable of self-supervised learning, Google says the model still needs some fine-tuning. However, only about 1,000 supervised examples were needed to modify the model, while older transformers may require 10,000 or more to do a worse job. With some refinement to make Pegasus smaller and more efficient, we end up with: a hybrid AI who can digest and summarize long documents.
Giving meaning to your digital life
Google CEO Sundar Pichai talked about the upcoming summary feature at the start of the keynote, suggesting to me that Google sees this as an important initiative. Automated summaries come first in Docs, but that’s just the beginning. Google also showed an early example of automated digests for Google Chat, with short summaries for missed conversations. I’d kill for something like that in Slack, making it nearly impossible to catch up on multiple channels in the morning or after a long meeting. Google even plans to generate meeting summaries in Google Meet, presumably with speech transcription.
I’m not a developer or AI expert, but I do spend a lot of time editing and analyzing long documents. Making a cohesive summary is difficult and time consuming, and the fact that a machine can do it instantly is amazing. If it is indeed can do that. We don’t know how good these summaries will be or if there are any ways to fool the AI. I can imagine people plugging in text with the intention of getting humorous or offensive summaries. This still feels like the start of something important. Would it even be smart to change the way we write long documents so that an AI can more easily summarize them for us? Could be!
Even if the summaries aren’t perfect, Google gives you a chance to address that. AI summaries are optional and you can edit them before anyone else sees them. Perhaps they will become more accurate over time. The Chat preview shows up and down buttons to rate the summaries, similar to how human volunteers evaluated Pegasus during testing. Reportedly, testers didn’t rate human-generated summaries higher than AI-generated summaries, so we might be surprised at how accurate the AI already is. We are all bombarded with information all day long, and an AI that can understand your personal data and summarize it all could be life-changing.
Sony’s upcoming LinkBuds S looks like one of the smallest ANC earbuds on the market
About the author