- Microsoft's investment in OpenAI may have been prompted by concerns about Google's AI advances.
- In a 2019 email, a Microsoft executive said he was “very concerned” about Google's AI capabilities.
- The email was released as part of the Justice Department's antitrust lawsuit against Google.
Newly unearthed emails show that Microsoft was “very concerned” about Google's capabilities in artificial intelligence in 2019, which may have spurred its investment in OpenAI.
In a lengthy email, Microsoft Chief Technology Officer Kevin Scott told CEOs Satya Nadella and Bill Gates that Google's AI-powered Gmail autocomplete has gotten “terribly better.” “I'm doing it,” he said.
He mentioned machine learning, adding that Microsoft is “several years behind our competitors in terms of the scale of machine learning.”
The email, with the subject line “Thoughts on OpenAI,” was made public on Tuesday as part of the Justice Department's antitrust lawsuit against Google. Most of Scott's emails were redacted.
In response, Nadella said the email emphasized “why we need to do this” and that he copied Microsoft's chief financial officer, Amy Hood, onto the chain. Ta.
Microsoft did not immediately respond to Business Insider's request for comment outside of normal business hours.
In 2019, Microsoft made the first $1 billion investment in a multi-billion dollar partnership with OpenAI.
Microsoft has benefited from well-timed investments ever since.
After increasing public and investor interest in AI after ChatGPT, Microsoft was able to act quickly and incorporate OpenAI's hot technology into existing products such as Bing and Microsoft 365. .
The speed at which Microsoft was releasing AI products even led some to wonder if its biggest rival, Google, was being left behind.
Google, a pioneer in AI technology, has since sought to counter narratives that it is lagging behind Microsoft. The company has released several products that compete with OpenAI's releases, including Bard, an AI-powered chatbot, and his AI model named Gemini.
Email exchanges from 2019 also show how Microsoft was keeping tabs on its rivals, with Scott calling the scale of OpenAI, DeepMind and Google Brain's AI ambitions “interesting.” In mentioning competitors' efforts, Scott mentioned his Google data center design and distributed systems architecture.
Scott touched on Microsoft's AI talent, saying the Bing, Vision and Voice teams have “very smart” people with machine learning expertise. He added that teams face constraints in scaling their ambitions, which suggests why Microsoft saw potential in partnering with OpenAI to realize its AI ambitions. added.
Scott said that when Open AI, Deep Mind, and Google Brain were competing to see who could pull off the “most impressive gameplay stunts,” he was “very dismissive of their efforts,” but that “that's not true. It was,” he added.
Please read the unedited portion of the email below. RL refers to reinforcement learning, NLP refers to natural language processing, and BERT refers to bidirectional encoder representation from transformers.
from: Kevin Scott
sent: Wednesday, June 12, 2019 7:16:11 AM
To: Satya Nadella; Bill Gates
subject: Re: Thoughts on OpenAI[Redacted]
What's interesting about what Open AI, Deep Mind, and Google Brain are doing is the scale of their ambitions, which range from data center design to computing silicon, networks, distributed system architectures, numerical optimizers, compilers, and programming frames. It's how we drive everything, right down to the work we do. , and high-level abstractions at the disposal of model developers. When all these programs were competing to see which RL system could accomplish the most impressive gameplay stunts, I had a huge disregard for their efforts. That was a mistake. When I leveraged all the infrastructure they built to build NLP models that we can't easily replicate, I started taking things more seriously. And as we tried to understand where the gaps were between Google and our capabilities when it came to training models, we became very concerned. As it turns out, just replicating BERT-large wasn't easy. Although we had a model template, it took up to six months to train the model because our infrastructure was not up to the task. Google had been using BERT for at least 6 months prior to that, so out of the time it took to jointly hack the ability to train a 340 million parameter model, we were able to get it into production and do the following: It took me a year to figure out how to move through the stages. A larger, more interesting model. We are already seeing the results of that effort in our competitive analysis of the company's products. One of the competitive metrics we look at for Q&A has jumped 10 percentage points in Google search thanks to models like BERT. Gmail's autocomplete is especially useful in mobile apps, but it's getting horribly better. [Redacted] We have some really smart ML people on Bind, the vision team, and the speech team. However, the core deep learning teams of each of these large teams are very small, and their ambitions are also constrained. So even if we start providing resources, we still have to go through a learning process to scale. And we are several years behind our competitors in terms of ML scale. [Redacted] from: Satya Nadella
To: Kevin Scott
CC: amy hood
sent: 2019/12/6 6:02:47 pm
subject: Re: Thoughts on OpenAIA very good email explaining why this needs to be done and why we will make sure our infrastructure people do it. Amy – FYI Send from Mail on Windows 10
Axel Springer, Business Insider's parent company, has a global deal that allows OpenAI to train models based on its media brands' reporting.
On February 28, Axel Springer, the parent company of Business Insider, joined 31 other media groups in filing a $2.3 billion lawsuit against Google in Dutch court, alleging losses caused by the company's advertising practices. I woke you up.
