Inside Microsoft’s push for winning AI formula
Software giant is trying to hold on to a crucial edge over rivals Google and Amazon, finds Gareth Corfield in Redmond, Washington
Artificial intelligence’s “big bang” moment is here – and it will bring a lot of money to those who get it right. As scientists and ethicists argue over whether the advent of advanced chatbots will usher in a new industrial revolution or the end of humanity itself, the businesses developing AI are rushing to put it in everything they can.
Microsoft is at the front of the race. The Pc-pioneer was quick to back Openai, the start-up behind transformational software CHATGPT. The Bill Gatesfounded company signed a deal to use Openai’s technology in its Bing search engine and reportedly invested $10bn (£8bn) in the company.
CHATGPT offers Microsoft a chance to leapfrog Google in search, but the technology brings far more than just that. Satya Nadella, its chief executive, has said AI will shape everything the company does in future.
During an unexpectedly sunny visit to Microsoft’s usually rain-soaked campus in Redmond, Washington, executives demonstrate ways in which they hope AI will boost its business.
One example: the technology has been plugged into its Powerpoint software to help office workers complete hours of work on presentations in minutes.
A series of bullet points fed to Copilot – Microsoft’s name for its AI chatbot integrations – result in the creation of an entire slide deck, complete with stock images, corporate branding and even the right colour scheme. Full sentences extrapolated from the terse bullet points garnish each slide.
Walter Sun, the company’s vicepresident of AI for business applications, believes Copilot could even become an “autopilot” by 2028. Rather than tasking a human to oversee the creation of a slide deck with the help of AI, an executive could simply ask software to do all the legwork. “We start by shipping features where there’s a human in the loop to make sure everything works,” says Sun. “Maybe in five years, these things will become more self-service. I think it’s possible”.
Copilot was first trialled commercially on Github, a Microsoftowned collaboration website for software developers. Instead of humans writing code and then posting it to Github for others to critique (and borrow from), Microsoft used the GPT programme to create a virtual assistant capable of writing software code all by itself in response to user prompts.
AI assistants within Microsoft’s products will become ubiquitous, writing code, emails and Powerpoint presentations for people while they deal with less automatable tasks such as meetings and presentations. However, limiting the role of
‘Future AI systems will have superhuman labour abilities and be smarter than humans’
humans in day-to-day business raises knotty questions.
“The question is, are we designing Powerpoint in a way that takes away my agency, my humanity?” asks Frank Shaw, the company’s marketing chief. “Or are we empowering them to spend more time on the parts that matter?”
Arvind Krishna, IBM’S chief executive, said that thousands of back office staff are likely to be “replaced by AI and automation over the next five years”. The company is stopping hiring in some areas and investing in software.
Connor Axiotes, a researcher at think tank the Adam Smith Institute, believes mass adoption of AI will lead to higher levels of unemployment in a way that no other past technologies have.
“Future AI systems will have superhuman labour abilities, be smarter than humans and encroach on a domain that no transformative tech ever has – cognitive labour,” he says.
“That part of working ability, which has been humanity’s comparative advantage, will no longer be so.
“This is why I think, more than ever, we are likely to see these stickier kinds of unemployment.”
Jon Friedman, Microsoft’s corporate vice president of design and research, is sanguine. “In every age in the history of technology, the thing that we thought was going to replace human jobs created more human jobs,” he says.
Microsoft’s representatives are keen to stress that the company wants to “augment” what people can do with AI instead of replacing them outright. Yet it is difficult to ignore the potential for AI to destroy jobs. Even as tech companies plough ahead, there’s the question of whether AI is good enough to do the job in the first place.
During the Copilot trial on Github, one software developer complained that the AI sometimes “suggests things that are just wrong”. Another said using the product was akin to “having to deal with a trainee programmer who knows how to search on Google” – not a glowing endorsement.
CHATGPT has a well documented tendency to “hallucinate”: produce convincing-sounding answers to questions that are in fact wrong. Plagiarism is also an issue when it comes to creating material.
Rashik Parmar, chief executive of the British Computer Society, says that AI programs such as CHATGPT are “only as good as the source of data” used to train them. He warns: “If somebody starts to poison some of the data sources then that can create all kinds of ramifications, right? We’re all super excited about what can be done without really thinking through the potential risks.”
Some people are thinking about the risks: hundreds of tech leaders including Elon Musk, Tesla’s chief executive, penned a recent letter calling for a moratorium on AI development until the risks were properly understood.
Meanwhile, Geoffrey Hinton, known as the “Godfather of AI”, recently quit Google with a warning that AI was on the verge of becoming dangerously powerful.
Ciaran Martin, the former head of the National Cyber Security Centre, has warned the technology “risks undermining the fabric of our society” by making the spread of convincing fake news so easy.
At Microsoft, writing Powerpoint presentations appears a long way off from the downfall of humanity.
The company is keen to press ahead with its advantage in AI, hoping to gain a crucial edge over rivals including Google and Amazon. If it does not, it risks falling behind. The cost of losing ground should not be underestimated. Google, whose experiments with AI have been lower profile, had $120bn wiped off its shares after a botched demo of its own chatbot, Bard, showed it giving a wrong answer. Prof Mike Wooldridge, of the Alan Turing Institute, says: “What we’re seeing is a massive pivot in ‘big tech’ towards this technology, because this finally delivered a mass market product for AI.
“They’re all broadly in the same sort of ballpark. But Openai and Microsoft have a year’s head start.”
Openai ultimately aims to create “artificial general intelligence” (AGI), regarded by most researchers as the field’s “holy grail”: the creation of true machine consciousness. AGI would allow machines to essentially think for themselves and this breakthrough moment could be reached within years, some in the field believe.
It is AGI that pessimists are most worried about. Once machines can be set off to complete tasks with relative autonomy, how can we control what they do? Thought experiments abound about how relatively simple commands could lead to the end of humanity.
Would Microsoft adopt AGI if Openai creates it?
“Honestly, I don’t know,” Eric Boyd, the company’s corporate vice president of AI platform technology, says in Redmond. It is difficult to speculate “on this thing that doesn’t exist”, he says.
However, Boyd adds: “I can’t imagine we wouldn’t take advantage of some really advanced technology.”