Microsoft and OpenAI announced a strategic investment and partnership this week. We think the deal says a lot about the state of the AI industry. And we think Microsoft’s new role as the commercial gatekeeper for OpenAI’s inventions will likely benefit society as a whole.
What is OpenAI?
OpenAI was a non-profit founded and funded by Elon Musk, Sam Altman and some other PayPal mafia individuals. The original premise was that AI was so scary that the world needed a well-funded non-profit to save us all. We say “was” because Musk is no longer involved (citing conflicts of interest with Tesla) and OpenAI created a for profit subsidiary for which many of the key employees now work. It’s unclear how OpenAI will still be solely a force for good now that it is pursuing financial gains, since its founding premise was that avoiding the pursuit for financial gains was how it was going to be a force for good.
Are you confused? So are we. OpenAI says that it created a for-profit entity to attract investment (which, of course, negates the idea that it is non-profit). Either way, OpenAI has been successful in attracting investors, most recently, Microsoft.
What are the financial details of the deal?
Microsoft announced that it will invest $1 billion in OpenAI’s for-profit subsidiary over the next decade or more. At first glance, this seems like a lot of money, especially for an organization with only 100 employees. The scale of investment shows how expensive it is to develop and deploy large AI systems. However, $100 million per year isn’t a lot of money for Microsoft—in fact, that amount of money represents less than 0.1% of the company’s current operating expenses. From that perspective, $1 billion doesn’t seem that big and it illustrates the outsized financial advantage that the big tech platforms have in pursuing big AI, versus any other company of lesser financial means.
What are the business details of the deal?
OpenAI announced that it will shift its development to Microsoft’s Azure platform, giving Microsoft an important new customer. In some ways, this makes Microsoft’s “investment” seem like pre-funding of its own products. It’s also a way to to get an influential developer onto Microsoft’s platform that may lead to more purchases by other customers. This idea of investing in a partner to generate revenue echos back to Microsoft’s investment of $150 million in Apple in 1997, which helped assure Apple’s stability and protect Microsoft’s highly profitable Macintosh software division revenue.
OpenAI also announced it will license certain “pre-AGI” technologies to Microsoft. It’s unclear what “pre-AGI” means. AGI means artificial general intelligence—a vague, in-the-distant-future idea of an AI that is smarter and more capable than humans across many domains. Since there isn’t a defined finish line for AGI, it’s tough to know what OpenAI means by “pre-AGI.” A cynic might say that using the term “pre-AGI” is simply a way for OpenAI to say that it hasn’t completely given up on its non-profit charter of developing AGI to save the world.
Why does this deal matter?
OpenAI employs some of the best AI minds and has the potential to create truly life-altering technologies. OpenAI’s contribution to AI spans key areas of AI which are deemed critical for the development of AGI and which advance the state of the art. The frontier for AGI is AI that is more human-like: understands context, can learn from the world without being told, can learn fast with fewer examples, can be developed more economically. Some of the most notable include:
- generative adversarial networks, a type of AI that uses unsupervised learning to build its own knowledge about the world. GANs are especially powerful at generating content (say images or text). In very simple terms, they work by having one network generate fake content while another tries to identify it as fake. This makes learning very fast, which is a core goal of AI. By being at the leading edge of this technology, OpenAI has an edge in building AI which is forced to discover and internalize the essence of the data.
- language generation that is spookily accurate and eerily human-like. The researchers point to context in particular: “The model is chameleon-like—it adapts to the style and content of the conditioning text.” In fact, this work was so powerful that OpenAI initially chose not to release the model or the data, instead choosing to release a scaled-down version for others to examine.
- large-scale AI model training in a more predictable, and hence, economic way. The team developed simpler ways to anticipate and plan for training of models based on a simple statistical measure of how novel, expressed as signal-to-noise ratio, any given data set is to the AI. This work removes bottlenecks to even more powerful AI. The compute bottleneck is not insignificant: researchers observed that the compute being used to train the largest AI models is doubling every 3.5 months, a combination of economic viability and algorithmic performance.
OpenAI’s initial non-profit status provided some optimism that its technologies would be broadly beneficial for society. But the organization’s transition to a for profit entity and Altman’s new role as CEO exposes questions about the company’s governance. How will Altman harness the impressive power of OpenAI’s development team to generate a profit? What business model will the company pursue and what impact will they attempt to have on society as a whole?
We think this question is especially important for OpenAI. For instance, Altman believes that we, as a society, should be pursing a universal basic income (UBI). If OpenAI is able to create an AGI that is superior to humans, will that AGI push us down the path of a UBI? If OpenAI is correct that AGI is possible, who will decide how that AGI will act? If AGI relies heavily on unsupervised learning from data about the world, who will be responsible for steering it away from the most egregious bias in our digital corpus? Will there be some sort of democratic oversight or will society be at the mercy of a small group of billionaires who can afford to build the AGI?
We don’t doubt OpenAI’s or Altman’s good intentions. But we do think that these questions of governance are incredibly important, especially at this early stage, when decisions are made that can likely never be discovered, let alone rolled back.
If you are worried about AGI, be glad it’s Microsoft in charge
From what we see so far, Microsoft has a promising governance system and excellent people working on the hardest of ethical issues. They are not only pursuing the technology but pursuing the best use of the technology. Microsoft has a strong history of responsibility and its current management shows no indication of altering that legacy. And the company’s executives can be held accountable by its public shareholders, unlike founder-dominated companies like Alphabet and Facebook.
While this obviously isn’t a democratic answer to the development of the most powerful intelligence other than humans, it does provide some public involvement in how Microsoft deploys AI in the wild.
All of this, in our minds, makes Microsoft the best commercialization partner for OpenAI for society. If all goes well, OpenAI will be able to pursue far-reaching AI technology while Microsoft acts as the de facto gatekeeper for putting OpenAI’s inventions in the wild.
We believe that AI will have an extraordinary impact on society. Governing that impact will be complex and difficult. We are all just beginning to understand this emerging field, hopefully before it’s too late.