In the decades I’ve been writing about technology, I’ve never seen a something take off as fast as generative AI – or seen anything so hyped.
The promises about how genAI will change our lives have come fast and furious. It will write the Great American Novel, compose symphonies, steal millions of jobs. It will do our work for us, help authoritarian regimes spread misinformation, and ultimately become an existential threat to humankind.
The claims have been outrageous. But unlike many overhyped technologies, genAI has proven it can bring in significant revenue. Sam Altman, CEO of OpenAI, the maker of ChatGPT, recently told his staff that the company was raking revenue in at a pace of $1.3 billion per year, roughly $100 million a month. Compare that to 2022 — before the launch of the generative AI chatbot that took the world by storm; OpenAI had a total revenue that year of just $28 million.
Microsoft has invested $13 billion in OpenAI, and that doesn’t account for how much money the company is spending on AI internally. Google, Facebook, Amazon, and others are investing many more billions. AI startups seemingly become valued at more than $1 billion overnight.
So, when I got my hands on Copilot for Windows when it was released late last month, I was expecting big things. After all, this was Microsoft’s chance to make a big splash, to show how genAI could improve the world’s most popular desktop and laptop operating system, installed on an estimated 1.4 billion PCs. Before the release, Microsoft promised Copilot for Windows "empowers you to create faster, complete tasks with ease and lessens your cognitive load – making once complicated tasks, simple."
Landing with a thud
To say I was underwhelmed by Copilot for Windows would vastly understate just how disappointing I found it. To begin with, it didn’t give Windows a single new capability. The things you can do with Windows that includes Copilot are exactly the same things you can do with Windows without Copilot.
That was bad enough. Worse was that it didn’t make Windows easier to use in any meaningful way. It can help you accomplish only the most basic of tasks. What can it do? If you want to use dark mode, you can tell it to “Turn on dark mode” and answer “Yes” a few seconds later when Copilot asks you to confirm that you want it turned on. That saves you from going to Settings > Personalization > Colors and then choosing dark from the dropdown menu in the “Choose your mode” setting.
Similarly, if you ask to change the image on your lock screen, it pops up the Setting page for personalizing your background, where you can select a different picture to use. It also asks if you want to perform several other similar tasks, such as changing your desktop background.
And that, I am sad to say, is pretty much the basics of what it can do. Overall, it helps you bypass having to hunt for the right settings for a limited number of often not-so-important tasks.
However, there are countless things it won’t help you with. Ask it to connect you to a Wi-Fi network, for example, and rather than popping up the right settings, it only offers a set of instructions, which are the exact same instructions you’d get by using Bing’s chatbot.
Another example: When I asked it to check for Windows updates, rather than checking for updates, it instead launched a troubleshooter that would help me fix any problems with Windows updating itself, even though I wasn’t having any troubles with that.
It also had the tendency to indulge in non sequiturs — asking, for example whether I’d like it to tell me a joke.
There’s even worse. It requested that I ask its name. I asked, and it answered, “My name is Bing. I am a chat mode of Microsoft Bing, a web search engine that can help you find information, images, news, and more.”
“I thought you were Copilot for Windows,” I replied.
“No, I am not Copilot for Windows,” came the response. “I am Bing, a chat mode of Microsoft Bing, a web search engine that can help you find information, images, news, and more.”
After $13 billion spent, Microsoft built an AI that doesn’t even know its own name — and then bull-headedly insists on being called by the wrong name, even when you point out it’s wrong. I’d say that’s not a great return on investment.
What does this mean for the future of genAI?
What does this high-profile failure mean? Has the technology been blown full of hot air, and will it eventually deflate? Will we see a re-run of the dotcom crash? A flashback to false metaverse promises?
Not likely. The dismal performance of Copilot for Windows is due more to a failure on Microsoft’s part than of the technology itself. AI truly can be a transformative technology when used properly. But Microsoft needs to tone down the AI hype, and only promise what it can deliver.
As the saying goes, to man with a hammer, the whole world looks like a nail. To that I add: To a company with AI hopes, every problem it sees can be solved with AI. (It's worth nothing that Copilot for Windows is just one such tool Microsoft is rolling out. There's also the coming Microsoft 365 Copilot, a security Copilot, and Copilot variations for other Microsoft services. They may well fare better than what's been added to Windows.)
Generative AI alone can’t make Windows simple to use. Windows is a nearly 40-year-old operating system with plenty of complexity, confusion, and illogic. Merely slapping a genAI bot on top of it doesn't do the trick.
Microsoft could have made a far more useful and powerful AI-based Copilot for Windows if it were willing to put in the time and work. As Computerworld Windows blogger Chris Hoffman points out, Microsoft never did the work to truly integrate Copilot with Windows itself. That would have required writing specific APIs that Copilot could hook into for many tasks. If Microsoft had done that, when you asked Copilot to do something for you, it would make a call to the API, and it would do what you wanted, rather than searching the Internet for an answer and spitting that answer back to you.
Microsoft decided not to put in the effort, and the results are an embarrassing failure. It won’t be the first, not just for Microsoft, but for others as well. Plenty of companies are piggybacking onto the genAI hype. They’ll promise you the sun, moon and stars and deliver barely anything at all.
So be very suspicious of genAI promises. Expect the worst. Maybe then, sometimes, you’ll be pleasantly surprised.