In 2019, Microsoft executives on the highest stage had an anxious electronic mail trade about AI that might in the end kickstart its AI funding.
Again then, chief know-how officer Kevin Scott despatched a four-page electronic mail to CEO Satya Nadella and cofounder Invoice Gates with the topic line “ideas on OpenAI,” which outlined his fears that Microsoft was falling drastically behind within the AI race. On the time, ChatGPT was nonetheless greater than 4 years away from being launched to the general public. However even then, Scott realized OpenAI and Google had made extraordinary steps ahead of their work on AI.
“The factor that’s fascinating about what OpenAI and DeepMInd and Google Mind are doing is the size of their ambition,” Scott wrote.
The e-mail, which is closely redacted, got here to gentle as a part of the Justice Division’s antitrust investigation into Google and was first reported by Enterprise Insider.
That very same 12 months, Microsoft would make investments $1 billion in OpenAI, the very firm Scott cited in his electronic mail. Ultimately the tech big would go on to speculate a minimum of one other $10 billion into the startup, which is credited for popularizing AI chatbots for on a regular basis use with the launch of ChatGPT. The 2 corporations at the moment are intertwined. Microsoft brings its huge assets, the necessity for which Scott outlines in his electronic mail, whereas OpenAI brings its cutting-edge AI experience that had so preoccupied the Microsoft government.
In his electronic mail, Scott says he miscalculated what precisely Google and OpenAI have been attempting to perform with their AI work. On the time, DeepMind, a startup owned by Google, was attempting to construct an AI system that would play the Chinese language board recreation Go, which Scott appears to be referencing.
“I used to be extremely dismissive of that,” Scott wrote. “That was a mistake.”
Scott marveled at how Google and OpenAI had constructed a whole infrastructure round their AI push. Within the electronic mail, he stated he was stunned Google and OpenAI had designed information facilities, sourced silicon chips, and constructed programming frameworks to allow all their work, based on the e-mail.
“After they took all the infrastructure that they’d constructed to construct NLP [natural language processing] fashions that we couldn’t simply replicate, I began to take issues extra severely,” Scott instructed Gates and Nadella.
Microsoft didn’t reply to a request for remark.
The arrival of AI has made lots of these assets in extraordinarily excessive demand. AI requires extraordinary quantities of computing energy to each run and prepare the fashions behind chatbots like Google’s Bard and OpenAI’s ChatGPT. Knowledge facilities have change into scorching properties each as tech investments and as real-estate property, with among the largest funding companies racing to nook the market. Silicon chips, or semiconductors, went via an unprecedented scarcity as increasingly corporations tried to hoard them, fearing they’d run out. Their significance is greatest exemplified by Nvidia’s legendary inventory surge over the previous 12 months. AI fashions additionally necessitated huge quantities of cloud computing, which corporations like Google and Microsoft invested closely in over the past decade.
However again in 2019, when he despatched the e-mail, Scott realized how vital all these infrastructure upgrades have been to creating the kind of AI that’s now commonplace. “As I dug in to attempt to perceive the place all the functionality gaps have been between Google and us for mannequin coaching, I obtained very, very nervous,” Scott stated.
Scott says on the time, it took Microsoft about six months to coach one in every of its AI fashions as a result of “our infrastructure wasn’t as much as the duty.”
Microsoft additionally realized it was lagging its opponents when it comes to the personnel it had devoted towards machine studying and AI analysis. Since then, workers well-versed in AI have discovered no scarcity of job provides (some with million-dollar pay packages) from corporations keen to rent them. The push to rent AI expertise would ultimately unfold past the tech sector to just about each trade within the company world.
Microsoft, based on Scott, had some “very good” machine-learning consultants, however they lacked the proper assets and headcounts to make a notable dent in deep studying, the complicated coaching mechanisms used to develop AI fashions. That meant their work took longer than it ought to have, a worrying prospect amid the approaching AI arms race. “We’re a number of years behind the competitors when it comes to [machine learning] scale,” Scott stated.
In the meantime, Nadella appeared to take Scott’s intensive issues to coronary heart. Nadella cc’ed chief monetary officer Amy Hood and replied: That is “why I would like us to do that.”
