IT Essentials: AI sharks are circling
Welcome to the feeding frenzy
It's a classic dilemma illustrated famously by Mayor Vaughn in Jaws who had to weigh up whether or not to close Amity Island's beach. Think of the money and run the risk of a couple more randoms getting chewed, or prioritise the common good and lose out on the lucre?
We all know which path he chose and how that went.
In the era of AI, Microsoft, Meta, Adobe, Slack, Dropbox and other tech giants are faced with a similar dilemma. Obey the rules, build on brand values and adhere to previously stated principles, or tweak the Ts&Cs, hoover up as much data as possible in the shortest possible time and to hell with the consequences.
They are choosing as Mayor Vaughn did. Of course they are.
On 26th June, new Facebook and Instagram privacy terms will see users' data - years of personal posts, images, chats and online tracking information information - made available for use for unspecified "artificial intelligence" uses and to be shared with unspecified third-parties. Users can opt out before the deadline, but this is a cumbersome process that involves filling out a form and explaining why they might object to such as reasonable proposal. Isn't that against laws such as the GDPR? You betcha. Does Meta care? Hell no. In the age of AI grabbing as much data as possible is "legitimate interest" don't you know? Got a problem with that? See you in court (again).
Then there's Adobe. Like many software companies, Adobe moved to the subscription-only model many moons ago in order to guarantee a steady income stream. A knock on benefit is Adobe can now block users who don't agree with changes in its terms of use. Adobe's new Ts&Cs give the graphics giant the right to "access your content through both automated and manual methods." Adobe has now started locking applications like Photoshop and Substance 3D that customers have already paid for until they consent to the new terms. Has this gone down well with customers? As well as might be expected, but alienating your (now captive) user-base is just the price you have to pay when you have your eyes on the big AI prize.
Which brings us to Microsoft. Chastened after a drubbing from the US authorities over its lax cloud security, Microsoft turned on its PR charm for a while with talk of transparency, responsibility, privacy and security. Then came Recall. Described as a "privacy nightmare" by campaigners, Recall, integrated into Windows 11, takes regular screenshots of all your activities and stores them locally in an encrypted database. Encrypted, that is, until you log in, at which point anyone else lurking on your endpoint can also see what you've been up to.
"This database file has a record of everything you've ever viewed on your PC in plain text," said security expert Kevin Beaumont. Until Friday, when Microsoft backed down, Recall also had opt-out terms that were not entirely clear. It is now opt-in, but why not make it that way in the first place? Beaumont described Recall as a strategic mistake and an "act of self harm," but maybe Microsoft sees things differently. Imagine what you could do with all that data.
It takes trillions of tokens to train the latest AI models, and there are only so many articles, pictures, videos and messages out there to tokenise. The big fish seem determined to gobble them all up before anyone else can have them, and in advance of regulators getting into gear. AI regulation is complex, nuanced and slow and it seems they are betting on a winner-takes-all strategy. This is, I believe, a bad thing for many reasons.
Silicon Valley's guiding maxims grow at all costs, move fast and break things, and ask for forgiveness not permission did a lot of damage to the nascent web, the consequences of which we see now in terms of poor security, pervasive surveillance, lock in and a massive centralisation of power. As employees at AI firms complain about being unable to speak out about potential risks, previous promises of openness and transparency are being chucked overboard, ethics teams disbanded. Lessons, it seems, have not been learned even as the tech grows exponentially more powerful and ubiquitous. We're gonna need a bigger world.
Recommended reads
Penny Horwood looks at how the Bat Conservation Trust is using machine learning to explore the nocturnal acoustic landscape of bat communications.
And Samara Lynn, editor of our US sister site MES Computing, digs into the best tools for detecting text written by generative AI