Let me help expose the thinking here. Since I was an engineer who worked for Google over a decade and helped launch several health projects while there and wrote one of the docs that ended up convincing the leaders at Google Brain and Research to work on health problems (after the Google Health records debacle), the thinking is relatively straightforward:
Google spent heavily to build infrastructure and hire smart people. A small number of those people work on ads, which, while not "walled-off" from the rest of the company, does operate at arm's distance. A larger number of people work on other projects- many of them cloud, which is Google's other non-insignificant revenue source- doing all sorts of things, from building software and hardware infrastructure to do high performance computing (of many kinds) to optimizing models for growing products like Youtube through watchnext and Play Store through suggestions. Many of those people are ex-scientists who want to help use what they've learned to make life better for people- potentially through better medicines.
At the same time, most companies that work with health are incredibly low-tech, super-inefficient and mainly exist to waste patient's money. They fail to use ML and data engineering to maximize the value of their products, they run inefficient software and hardware, they can't explain why their products work or not.
So it stood to reason- at least in 2007, when I joined- that Google could use its technical prowess and ads $$$ to fund health research with the goal of providing a greater good for the world- while also making some revenue and profit from the products. While making the case for this I was extremely careful to repeatedly explain that Google had a serious reputational risk because people were skeptical that the company could both be a big tech advertiser, and a health research company. I said that we should always be up front and open about any deal we make, because eveyrthing is going to leak anyway and at least by being open you have some control over the narrative. And I also said that this work needed to be carried out ethically and in a way that was unrelated to existing consumer products (ads, search, android, etc).
Well, I was successful and managed to convince Google Research to get into health and Jeff Dean and Greg Corrado have been actively working in that area for some time. The work has had extremely modest success in terms of research output and products, but they're committed to it long-term. Later, as Google finally began to recognize Cloud as a business model, I began to engage with the cloud leadership, but with a slightly different pitch: Google, with its infrastructure, could be a great partner to pharma and EMR enterprises, as a service provider (servers and storage) as well as new ideas for data processing (Flume) and machine learning (this was pre-tensorflow; Sibyl was the system i wanted to sell).
This went on for some time and Google made a few products and published a few papers. Some are great and some are terrible, but I left because I came to the conclusion that Google simply can't provide the same level of innovation or success that Ads and Search had, when working in the health area. There are several reasons for this:
The data sources are not as rich as the web. The web provides copious data to make good Search and Ads systems. The data in health and research is ... terrible. Nobody gave any thought to featurization or any of the other important data engineering/ML engineering you need to be sucessful. A dataset like the one being discussed here is significant because it's homogenous.
Verily was getting going but at the time the focus was on medical devices, not machine learning for health research.
At the same time, Sundar has taken over and his approach... is not very enlightened. He was good at rapid growth in multiple products such as Android and Youtube, which were critical for Google to stay competitive when it looked like Facebook was going to own social. He has little to no understsanding of anything but growing consumer products (and when I say "consumer products", I mean the consumer is the product).
So my conclusion- after all of this- is that Google will ultimately fail to make the impact it could have, both because people don't trust google to handle this sort of data ethically, and because the upper leadership doesn't have the "DNA" (so to speak) to build products and do research in this area. Which is a real shame. But the good part is: there's nothing left inside of Google that makes them special in any way for health research. Everybody who wants to emulate the Google style has all the tools available in external open source repos and cloud systems (that includes TPU v4 which is a huge competitive edge for google).
Google spent heavily to build infrastructure and hire smart people. A small number of those people work on ads, which, while not "walled-off" from the rest of the company, does operate at arm's distance. A larger number of people work on other projects- many of them cloud, which is Google's other non-insignificant revenue source- doing all sorts of things, from building software and hardware infrastructure to do high performance computing (of many kinds) to optimizing models for growing products like Youtube through watchnext and Play Store through suggestions. Many of those people are ex-scientists who want to help use what they've learned to make life better for people- potentially through better medicines.
At the same time, most companies that work with health are incredibly low-tech, super-inefficient and mainly exist to waste patient's money. They fail to use ML and data engineering to maximize the value of their products, they run inefficient software and hardware, they can't explain why their products work or not.
So it stood to reason- at least in 2007, when I joined- that Google could use its technical prowess and ads $$$ to fund health research with the goal of providing a greater good for the world- while also making some revenue and profit from the products. While making the case for this I was extremely careful to repeatedly explain that Google had a serious reputational risk because people were skeptical that the company could both be a big tech advertiser, and a health research company. I said that we should always be up front and open about any deal we make, because eveyrthing is going to leak anyway and at least by being open you have some control over the narrative. And I also said that this work needed to be carried out ethically and in a way that was unrelated to existing consumer products (ads, search, android, etc).
Well, I was successful and managed to convince Google Research to get into health and Jeff Dean and Greg Corrado have been actively working in that area for some time. The work has had extremely modest success in terms of research output and products, but they're committed to it long-term. Later, as Google finally began to recognize Cloud as a business model, I began to engage with the cloud leadership, but with a slightly different pitch: Google, with its infrastructure, could be a great partner to pharma and EMR enterprises, as a service provider (servers and storage) as well as new ideas for data processing (Flume) and machine learning (this was pre-tensorflow; Sibyl was the system i wanted to sell).
This went on for some time and Google made a few products and published a few papers. Some are great and some are terrible, but I left because I came to the conclusion that Google simply can't provide the same level of innovation or success that Ads and Search had, when working in the health area. There are several reasons for this:
The data sources are not as rich as the web. The web provides copious data to make good Search and Ads systems. The data in health and research is ... terrible. Nobody gave any thought to featurization or any of the other important data engineering/ML engineering you need to be sucessful. A dataset like the one being discussed here is significant because it's homogenous.
Verily was getting going but at the time the focus was on medical devices, not machine learning for health research.
At the same time, Sundar has taken over and his approach... is not very enlightened. He was good at rapid growth in multiple products such as Android and Youtube, which were critical for Google to stay competitive when it looked like Facebook was going to own social. He has little to no understsanding of anything but growing consumer products (and when I say "consumer products", I mean the consumer is the product).
So my conclusion- after all of this- is that Google will ultimately fail to make the impact it could have, both because people don't trust google to handle this sort of data ethically, and because the upper leadership doesn't have the "DNA" (so to speak) to build products and do research in this area. Which is a real shame. But the good part is: there's nothing left inside of Google that makes them special in any way for health research. Everybody who wants to emulate the Google style has all the tools available in external open source repos and cloud systems (that includes TPU v4 which is a huge competitive edge for google).
Oh well.