Once there was a theory, that a [sufficiently advanced] artificial intelligence can learn all the knowledge of humanity by reading all the books that humans have ever written.
Lately, computer scientists agree, that it would not be enough, because we need to connect symbols to what they represent (It is called the symbol grounding problem).
AnyEyeDo's mission is to create an infrastructure, that will allow AIs of today and tomorrow to receieve accurate, real-time information about the physical world.
Our initial product, YaScout, uses human agents to collect visual information at, or around their location.
The goal of the project was to create a beautiful, simple, and intelligent photo sharing app.
In the world, where users are drowning in the photos that they take with their phones, the percentage of shared photos remains very small.
We've addressed that with AI analysis of user's gallery. WITH scanned user's photos and automagically suggested the best photos to share, and the groups of people to share them with.
Behind this simple and elegant UI ran the state-of-art image classifiers, face detection and recognition
How often do people take a group photo and then forget to send it to the group?
Have you seen people asking someone to take a photo with multiple phones, just to be sure that everyone has a copy?
We've developed a super-human facial recognition technology in order to help people get photos of themselves
Imagine being able to automatically, yet privately and securely, share all the photos of your spouse with them on an ongoing basis. That's the Face.R app
Helping users find photos of people, places, and things in an immensity of their photo collection is no small task.
Consider this simple query: "me in a hat". In order to fulfill it, a system needs to know what a hat is, and have an ability to recognize it in a photo. It also has to be able to recognize all the people in my photo collection, and determine which one of those is myself. Finally it needs to index all that in an efficient manner and make available for an immediate retrieval.
Our team was the first one ever to apply the Google Brain technology, now known as TensorFlow to the problem of image recornition, demonstraing the superiority of the Deep Neural Nets that are so popular today.
Hundreds of millions of people use this product all over the world.
In a close collaboration with the Android team we have given the eyes to the Google Assistant.
Using the technology from Google Goggles we are able to respond to voice queries such as "What is this?" or "How much does this cost?"
This project gave rise to the Google Photos application.
Google is a data-driven company. Analyzing the requests made to Google Goggles it became clear to me that the most popular category of visual search queries, is foreign language text. I approached the Google Translate team, and this product launch was a result of fruitful collaboration.
Together, we've conducted user studies, and developed an innovative UI in order to allow the user to specify what they want translated, and for displaying the results in a manner that makes sense to them.
This product launch resulted in a hockey stick growth of the feature, and contributed to the popularity of the Google Translate app.
Now people all over the world can undersdand signs and menus in languages they can not even type.
Google Goggles was a revolutionary application. Internally called "Visual Search", it strived to answer every possible visual query.
We do not often encounter unfamiliar things, that we can not describe with words, in the world around us. When we do, however, there is no better way to describe them, than to take a picture.
Incorporating truly revolutionaly machine vision technology stack behind the scenes, it was the first ever application to solve image recognition problem at scale.
It provided technology stack for many future products of Google, from Google Glass to Google Translate, to Google Seach by Image, and Google Photos.
As a product manager, responsible for quality of Google Products and services in Russia, I led several launches, the most significant of which was Google Transit Directions.
Launching a product in a new country is not as simple, as translating all the strings into a foreign language, althought that is a part of it. Local knowledge is necessary. Local data needs to be collected. Local laws need to be obeyed. Local peculiarities need to be take care of.
For example, in Russia many transit routes overlap, and they don't follow any particular schedule. The Google Engineering team in Zurich, where all public transit runs precisely on schedule, had to modify their system in order to allow user instructions to be displayed in the following form: "Take bus number 5 or tram number 7, whichever comes first".
Chrome Developer Tools was built into Chrome from the very beginning, in order to help developers build a better web.
© 2020 Alexei Masterov