A proposal to beat below-the-belt selfies: Crowdsourced machine learning using victims' image stashes

Empty your inboxes, people, we can duck this problem together


Column It’s often said that the second most important job in Australia – behind Prime Minister – is captain of the Australian men’s cricket team.

The last holder of that office recently resigned in disgrace after it emerged a flurry of texts he exchanged with a staffer crossed the line from racy to harassment as one message included an unsolicited dick pic.

After his resignation, a friend who is both a cricket fan and an AI expert messaged me.

"This can’t go on," she said.

"Well, he’s gone," I replied. "And there’s an end to it."

"I don’t mean that. Of course he’s gone,” she responded. “This happens all the time. Every day women are getting photos they never asked for, of bits that they’d rather not see."

"True dat. But what can you do about it? ‘Penis erectus conscientum non habet.’"

She rolled her eyes at my Latin. "It’s not as though I’d expect men to do anything about it. Apparently you all think your bits are suitable for framing."

"Well, if you can’t stop men being stupid – and god knows many have tried – what else is there?"

"Oh, I reckon this is a perfect problem for machine learning."

"What? How?"

"All these photos of all of these bits, they’re flying through systems that route them from >ahem< sender to an unfortunate receiver. I reckon it would be dead simple to build a filter on the receiver."

"A filter?"

"Something that could match those photos against an abstract, prototypical, possibly even Platonic bit, and intervene."

"So if it looks enough like someone’s bits…"

"…you’d stop it right there. And replace it."

"You’d replace a photo of someone’s bits with someone else’s?"

For that I got a look that telegraphed how can you be this dim? "No. You’d transmit another photo altogether. Of a duck."

"Which duck? Whose duck? Why a duck?"

"Everyone loves a good photo of a nice, friendly duck. So much less… surprising. And yet…"

"And yet, if someone received – apropos of nothing in particular – a photo of a duck…"

"…They’d know they’d been spared the sight of some bloke’s bits."

"Oh…"

"You see it now."

"It’s genius."

She giggled and said, "I know."

"There’s only one problem," I replied. "Bits come in all shapes and contours and colours. That’d mean quite a large… erm…training set."

"And that’s the second half of my plan. To train the model to detect the bits, we ask for contributions."

"From blokes?" I shuddered.

"No," she said, in a tone you’d use when addressing a poorly trained dog. "From all the women who have over the last twenty years been subjected to all those photos of all those blokes’ bits."

"Oh."

"Yes. Every time another unsolicited photo arrives, women can… deposit?… those bits in the appropriate receptacle. Improving the model."

"And, at the rate unsolicited bits are flying around these days…"

"Yeah," she sighed. "I reckon it should only take about a week."

"You know, I now see seventh graders using web-based tools to train TensorFlow-based models. With pretty remarkable results. What you’ve described is basically a secondary-school project. The sort of thing that might win a science fair."

"It’s dead easy. We’ve got all this great tech, and it’s time to make it really useful, by adding some c**k-blocking to our messaging apps. Of course, you’d need to get Apple and Google and Facebook onboard..."

"But they need this. Everyone is tired of online harassment and looking for a way to be seen to be doing the right thing. This is just that sort of feature. It’s a natural. By the end of next year, I reckon it’ll be a tickbox feature on most of your sharing services. Just click on ‘Ducks for d**ks’, and you’re protected."

"That’s my hope. It truly feels like it’s an idea whose time has come. And then, every time some bloke over-enthusiastically shares their bits, well, they’ll be gone for a duck." ®

PS: Yes, we know this has been done before – see here, here, and here, for example – we're saying it should be more widespread and work well.

Broader topics


Other stories you might like

Biting the hand that feeds IT © 1998–2022