You don’t find needles in haystacks. Unless, the haystack is made of needles. From passive listen to active grabbing, platforms suck up your digital DNA like a Hoover™ Deluxe. They suck. Suck it up.
— — —
What is active grabbing?
Henry Shevlin is a doctor of philosophy. He teaches AI ethics at Cambridge University. Not far from where I was born. If you’re sensing a theme in these posts, you’re right. England!
Four years ago, he tweeted a pneumonic to remember the periods of the
Paleozoic era — “Catastrophic Overthrow Started Different Colder Period”. The tweet had one reply and seven likes. It was, by just about any standard, unnoticed by humanity.
Back then, OpenAI was actively grabbing tweets. A massive amount of data turned over – apparently for free – to train its large language models.
The other day, Dr. Shevlin asked OpenAI’s ChatGPT, what “Catastrophic Overthrow Started Different Colder Period” could mean. It “recognized” it as a “Clever pneumonic to remember the order of geological periods. There is no cognition. With the use of super-fast machines, OpenAI searched its trillions of inputs and matched it. It just didn’t cite its source. Because it can’t. Google found the same tweet and cited it. But, at this point, Google is grown kid that the family ignores where it does the same ‘ol remarkable stuff it’s always done. The new kids get lavish praise for saying, “Pisgetti.”
The good doctor tested other phrases to see if OpenAI could “learn.” Nope. Just repeat.
But, that’s not today’s story. This is. Everything we do, say, write, or choose to put on the world wide web in public spaces or our private ones is fodder for platforms. Not just to train AI, but – maybe – to sell us things.
Follow me down this rabbit hole. What if OpenAI got smart enough to share answers from that user. Or from like minded users? My friend, Matthew, who pointed me to this event called this confirmation bias. He’s WAY right.
A large-ish media conglomerate called Cox Media Group pitched clients that could target ads because they have devices in your hands and homes that passive listen to your conversations. You see, they snuck some innocuous language into their terms of services that lets them do it. If you’re talking about cars, you’re going to see car ads. Not because of what you like or who you follow. That’s public stuff. But because you mentioned it once to a friend offline or in the privacy of your own home.
This story didn’t benefit from Matthew pointing it out. Most people never saw it. Only one very-under-the-radar outlet reported it — 404 Media. That got Cox in enough hot water that they retracted their pitch. By then, everyone was talking about passive listening.
If an old school media company like Cox knows how to stealthly insert allowances into their ToS, it’s almost impossible to fathom how much data big platforms and tech companies have. I’m going guess haystacks full of needles.
The next time you react to a platform’s, “Uncanny ability to put the right ad in front of me,” think about this. Cox’s mistake was not passively listening or actively grabbing, it was boasting about it. Most big companies aren’t that boastful.