Microsoft’s obsession with artificial intelligence could be its downfall
Machine learning is enabled by disregard for security and privacy for little return so far
Twenty years ago as internet music piracy raged, I received a very unusual email from a primary school pupil in Kentucky. I have retained the original spelling: Downloading unlicensed MP3S was “leagal”, the child insisted, before offering a couple of reasons why acquiring the pirate content was morally justifiable.
“The bible dont say thou shall not download”, he argued (correctly). Then came the clincher: “Its not stealing its in the air.” If I can see it, the child argued, then I can take it.
But compared to the brazen heist Big Tech is perpetrating today, MP3 sharing was small beer. The generative artificial intelligence (AI) models scrape up every song, image and thought we’ve ever recorded just so they can burp it back up to us.
Last week Microsoft’s head of artificial intelligence echoed the child’s amoral argument.
Mustafa Suleyman is the Londoner who oversees Microsoft’s AI efforts across vast areas of the company’s consumer and search products.
This ingestion and reproduction has prompted a huge grassroots protest from digital creators of all kinds, with music companies joining the fray last week. It worries businesses, too. Give AI access to your company’s documents, and expect your confidential marketing plan to pop out in front of somebody else on the other side of the world.
Generative AI is a machine for erasing human originality – and replacing it with a bland derivative digital paste. It rewards the lazy and the dishonest while punishing the talented.
So Suleyman was asked: have the AI companies effectively stolen the world’s IP?
No, he replied, explaining that “with respect to content that is already on the open web, the social contract of that content since the Nineties is that it is fair use. Anyone can copy it, recreate with it, reproduce with it: that has been freeware, if you like.”
There you go: everything is in the air, and everything is free.
Not free for us but for Suleyman and his employer, a company worth $3 trillion (£2.4 trillion).
However the argument is as baseless as that offered to me from Kentucky. The doctrine of “fair use” is not what Suleyman thinks it is.
It’s a peculiarity of US law, an affirmative defence, rather than a formal copyright exception. It only applies in quite narrow circumstances, and it’s meaningless in most other countries.
Even more remarkable is the “social contract” that Suleyman evokes.
Judges are rarely impressed with defendants bringing their own, imaginary laws into the courtroom as they are obliged to interpret what’s on the books.
Microsoft employs a great many thoughtful people, who deliberate about the ethics of and consequences of their actions. Suleyman does not appear to be one of them.
The Londoner dropped out of Oxford where he was studying philosophy and theology because “it felt so abstract and impractical to me,” he once said.
Luck intervened: his brother was best friends with Demis Hassabis, the chess prodigy, former games entrepreneur and neuroscientist. Along with Kiwi developer Shane Legg, they founded Deepmind, with Suleyman providing the hustle. It was great timing as Deepmind was acquired by Google in 2014.
However, Deepmind was soon in trouble for an ethical violation: scraping millions of NHS health records it shouldn’t have used.
Suleyman himself was relieved of management duties in 2019 after allegations of bullying which led to him later issuing a public apology.
However, once again luck intervened as he was personally headhunted by Satya Nadella, Microsoft chief.
Two things annoy people intensely about performances like Suleyman’s.
One is the sense of entitlement. As the former AI developer and musician Ed Newton Rex explains: “Companies need three things for generative AI, and pay billions for the first two: AI talent and computing power. But they now expect to get the third, training data, for free.”
The other is the complete absence of boundaries. We need boundaries to respect human dignity, but Microsoft seems to have lost any sense of where these might be. For example, Microsoft recently introduced a feature into the pre-release version of Windows that surreptitiously captured your screen every few seconds.
It didn’t matter if what was on your display was your banking password or pornography, it was all caught and stashed away in a giant archive for Microsoft’s AI to use. It could be opened up and shared with the world with ease.
Microsoft took a long time to acknowledge the security concerns before removing the feature.
It appears that Microsoft has forgotten the primary rule of business, which is keeping your customers happy. Instead it has been sidetracked by stuffing AI into every corner of every product with a maniacal zeal.
If it paid more attention to its customers, it may notice that business seems to be losing faith in the utopian promises of AI.
Early trials are proving to be underwhelming. The Wall Street Journal last week reported first-hand from a range of large companies who had all found AI to be too error-prone to be useful.
A recent survey found that the financial returns from early test projects were “dismal”. Up to 42pc said it had no payback at all and fewer than 15pc of the trials were successfully completed.
For Microsoft to retain its customers’ trust, Mr Suleyman needs to go.
And Microsoft needs its AI to start showing some real results.