From helping position Azure with the developer crowd, to building ethics and trust into artificial intelligence, Microsoft’s US$1 billion investment in OpenAI shows the Redmond giant has its sights set on being ready for the complex world of artificial general intelligence.
But first, a quick refresher course. In recent years, artificial intelligence (AI) has moved from concept to reality, and beyond that to an umbrella term encapsulating numerous technologies: with machine learning and deep learning techniques in particular prospering as a result.
While these have been defined as early, ‘narrow’ examples of AI – technologies designed for a specific and programmed purpose, like recognising objects, faces, language or finding patterns in data – at the top of the tree is what is commonly referred to as ‘artificial general intelligence’ (AGI), otherwise known as ‘true’ AI and something which is still restricted to the confines of science fiction.
In short, it is the belief that AGI will come to combine human-type, flexible thinking and rationale with computational advantages, like near-instant recall or the ability to crunch, analyse and make sense of huge data sets.
This has naturally led to some concerns, especially around the technology’s long-term societal impact. Tesla (and OpenAI) founder Elon Musk has previously called AGI the ‘biggest existential threat’ to humanity, while the late Cambridge University Professor Stephen Hawking told the BBC ‘the development of full artificial intelligence could spell the end of the human race’.
In the more immediate future, Google’s director of engineering, Ray Kurzweil, has predicted an AGI capable of passing the Turing Test will exist 10 years from now, and that by the 2040s, affordable computers will be able to perform the same number of calculations per second as the combined brains of the human race.
For Joanna Bryson, associate professor in computer science at the University of Bath, the definition of AGI has come to mean different things to different people.
“AGI isn’t really a thing," Bryson said. "When people think AGI will become like us, that’s a crazy 17th-century idea that humans are at the apex of evolution, and that something faster will soon pass us.”
Instead, she breaks it down to four definitions and says for most, AGI has ‘come to mean two contradictory things – a system that can be all-knowing, and a system that is human-like’.
Yet for all the mystery, AGI is one of the reasons that Microsoft has ploughed $1 billion into OpenAI, the AI research lab based in Silicon Valley.
What Is OpenAI?
Founded in December 2015, OpenAI serves as a research lab to explore the future of AGI. In its short tenure, it has accrued a number of high-profile backers, including Tesla CEO Elon Musk, who cofounded the institute, former Y Combinator chairman and current OpenAI CEO Sam Altman, and LinkedIn cofounder Reid Hoffman. Its explicit mission is to ‘ensure that artificial general intelligence benefits all of humanity’.
“We want advanced AI systems to work with people to solve currently intractable multi-disciplinary problems, including global challenges such as climate change, affordable and high-quality healthcare, and personalised education,” an OpenAI spokesperson told Computerworld.
Technically, OpenAI has used deep learning – a technique based on neural networks, which seek to emulate how the human brain works – to produce some early results.
To date, such work has been limited to standalone experiments in things such as automatic text generation, manipulating a robot hand, and game-playing, with the latter seeing OpenAI’s software beat an e-sports team in the Dota 2 video game.
Interestingly, while OpenAI started out as a non-profit entity, it has operated under the business license of OpenAI LLP since the turn of the year with a new ‘capped profit’ business model.
This has allowed the company to “raise and return investment capital, just like a normal startup”, but it has to adhere to the company’s original charter of developing AGI for the benefit of humankind.
In this capped profit model, profit generated from any investment in OpenAI LLP will be passed onto the overarching non-profit group, which will then utilise these funds as they see fit.
This profit is capped at 100x the investment for first-round investors though, meaning if you invested $10 million today the profit cap only truly comes into effect after that initial investment has generated some $1 billion in returns.
That total cap remains unclear and will “change as we raise investment capital, but we’ll always keep it well below the market cap of today’s largest technology companies,” so says the OpenAI spokesperson.
“The cap is designed to be relevant if we succeed in creating AGI, which we think will generate unprecedented economic benefit."
Microsoft deal explained
With shades of Google’s acquisition of DeepMind, the investment has made Microsoft the “exclusive” provider of cloud computing services to OpenAI, and the two companies will combine to experiment, develop and potentially commercialise new AGI technologies, as well as building ethics and trust into these new technologies.
In the short-term, the companies will build a computational platform in the Azure cloud, where they will be able to train and run AI models, as well as building hardware on top of Microsoft’s supercomputing technology.
OpenAI will license some of its technology to Microsoft to commercialise, while Microsoft becomes OpenAI’s ‘preferred commercial’ partner.
The group also becomes an exclusive Microsoft cloud partner, having worked with ‘multiple cloud providers for its compute needs’ in the past. OpenAI didn’t address our questions as to how much skills and resources crossover there will be between the two organisations.
“We’re partnering to develop a hardware and software platform within Microsoft Azure which will scale to AGI,” says the OpenAI spokesperson.
“OpenAI is producing a sequence of increasingly powerful AI technologies, which requires a lot of capital for computational power. The most obvious way to cover costs is to build a product, but that would mean changing our focus. Instead, we intend to license some of our pre-AGI technologies, with Microsoft becoming our preferred partner for commercialising them.”
Structurally, OpenAI’s board will continue to take all decisions about the research agenda, and how its work is used, with one Microsoft director to be appointed to the board.
Why has Microsoft invested? For talent, developer buy-in and AI consolidation
For independent industry analyst Jack Gold, the partnership is as much about market consolidation as it is about Microsoft positioning for the future, given the small amount of AI workloads that currently require cloud-based compute.
“OpenAI is trying to consolidate the various frameworks and platforms that are currently available for AI, to make it less of a many component ecosystem,” says Gold.
“Currently, there are lots of frameworks and tools, and that’s typical of a new market so there will be consolidation. The question is, who will lead the consolidation? Microsoft will compete with Google and AWS among others to do so, so any partnerships it can participate in that can offer more potential customers is a good thing.”
Nicholas McQuire, vice president of enterprise research at analyst house CCS Insight, admits that the OpenAI investment announcement in July “was a bit of a surprise”, predominantly because AGI remains something of a “research ivory tower – super advanced and, to an extent, a slightly narrow discipline of research”.
Indeed, he points to Microsoft’s existing AI research areas being in more immediate, narrow application areas of AI, such as Natural Language Processing (NLP), cybersecurity and quantum computing.
Yet he still sees a few clear reasons for the investment: “On some level, it looks like Microsoft would naturally want to be affiliated with some form of research entity, which can drive thought leadership, push the boundaries of AGI, and then – in a third way – be able to take advantage of research breakthroughs in the Azure business as well, which is obviously similar to what Google does with DeepMind – albeit at arms lengths.”
Diving deeper, he suggests the investment is about positioning Azure on the roadmap to be ready for AGI, which is a ‘supercomputing set of workloads’, especially among developers who are experimenting with machine learning.
“Having a capability around AGI is a way to showcase to the developer community – particularly those now going into machine learning – the advanced use cases that are coming down the line,” said McQuire.
For all of that, there is a general awareness that Google is leading the AI arms race today, based on its raw capabilities and talent pool.
This is why Joanna Bryson believes that this deal comes back to a talent attraction initiative for Microsoft as much as anything else: “They have a real problem with looking cool. Clearly, they are the adults in the room, and clearly, they want to be the adults in room, but they want to look cool so people want to work with them.
“Like DeepMind, it helps to attract more talent. In the best case, OpenAI is like how Google looked in 1998; they have a cool product, they claim they don’t care about money, they have lots of good talent, and everybody knows they are creating the best stuff.”
But she says this PR and talent attraction initiative could ultimately lead to more: “In the best case, it’s a project that will change how people live. But there’s a big piece around commercialising it.”
Microsoft did not comment at the time of writing.