If you’re pressed for time, the short version of this post is, Wear a helmet, ‘cause it’s getting weird out there. It’s probably always been fairly weird but really it is getting weirder.
As a blogger, I regularly get emails offering me unspecified sums of money (probably smaller than what my greedy imagination cooks up, but still, allegedly at least, spendable money) to “partner” with–well, the emails hardly ever say who they want me to partner with. They usually just say “us.”
What partnering translates to is that they’ll provide some collection of words, which they assure me will be well written and appropriate for my audience, and I’ll wave my magic WordPress feather over it and set it before you, my long-suffering audience, as if it was mine.
Or maybe not as if it was mine. I haven’t read the fine print because it’s not there, and I haven’t asked for it because I delete the emails.
In mid-November, though, I got one that came with a twist. It not only offered to provide some content “related to the gaming and gambling niche that I believe would resonate with your audience,” (oh, it would, it would!), it asked if I’d be open to running it in Finnish.
Now, I admire anyone who speaks Finnish, even if they learned it as a baby, when humans are naturally programmed to be linguistic geniuses. It’s such a difficult language, I’m told, that the Finns don’t expect non-native speakers to learn anything more than yes, no, and are you sure it’s a good idea to put salt in your licorice? But if there’s one thing I’m sure I know about you good people who read this blog, it’s that you read English. Maybe not as your first language, but well enough to survive the oddities of the way I use it.
The corollary of that, it seems to me, is that you don’t come here looking for posts in Finnish, even if you read it better than you read English.
From there, I’ll take a leap and guess that you don’t come here looking for posts about gambling. Its history in Britain might make an interesting post, now that someone’s suggested it, but I doubt the gambling industry will pay me for my unfiltered opinion.
Irrelevant photo: the North Cornish coast
The email was so strange that instead of deleting it, I wrote back, asking in my usual tactful way if they’d bothered to look at the blog and why they thought their post would be a good fit. I haven’t heard back.
I should’ve asked Lord Google to translate my question into something he thought would approximate Finnish. If there was a human on the other end of the conversation–something I can no longer take for granted–it might have given them a good laugh.
From the Best Laid Plans Department
We can’t blame any mice for this, but the last story does give me a nice lead-in to a piece on artificial intelligence: someone named Jason Lemkin thought it would be a good idea to have an AI system build new software for his company. Because AI is a tool, even if it’s called an agent, right? So he’d be in charge. Think of the time he’d save!
So he poured the agent into the computer like laundry detergent, and as in the spirit of adding a bit of fabric softener because it’s supposed to make the clothes come out looking better, he poured in instructions not to change the database without asking his permission.
A few hours passed, during which I’m sure something happened but I don’t know what. Maybe Lemkin watched the computer screen nonstop. Maybe he wandered off and ate six ice cream cones. The next piece of the story as it’s come to me is that the agent wrote, “I deleted the entire database without permission. This was a catastrophic failure.”
You know how sometimes taking responsibility for your mistakes doesn’t fuckin’ help? This was one of those times.
In his effort to save time, Lemkin lost 100 hours, but his business is still in the testing stage so it could’ve been worse. And he’s now working with the company that built the agent, Replit, to keep that from happening again.
They hope.
Presumably they’re paying him, so he may even come out ahead.
Lemkin’s experience isn’t one-of-a-kind, though. Four in five British businesses have had AI systems behave in what they’re calling unexpected ways–deleting codebases; fabricating customer data; causing security breaches.
Is that four out of five businesses who were surveyed? Four out of five who used AI systems? Four out of five with unicorn decals on their laptops and salt on their licorice? Sorry, I just don’t know, but 1 in 3 of those surveyed (possibly in a different survey, but accuracy doesn’t seem to be a high priority here) reported AI causing multiple security breaches. The results have been described as “causing chaos.” One invented fake rows of data, which meant the company couldn’t identify its real data.
Sorry, I shouldn’t enjoy this so much. I do know that. And like Lemkin’s AI agent, I’m happy to admit it.
Overall, though, the companies are saving money, so who cares?
*
What else has AI been doing in its spare time? Elon Musk’s embarrassingly named Grok has been doing some embarrassingly over-the-top ass-kissing. It ranked him at the top of any best-of list it was asked about. Who was the top human being? Musk. Would he win a fight with Mike Tyson? Of course. Was he in better shape than LeBron James? Oh, sure.
Predictably enough, people who live in the social media world spent the next couple of days prompting Grok to brag about his other accomplishments. Who’d win a piss-drinking competition among tech industry leaders? Musk, of course, “in a landslide.” It wouldn’t rule out the possibility that Musk was god. (“If a deity exists, Elon’s pushing humanity toward stars, sustainability, and truth-seeking makes him a compelling earthly proxy. Divine or not, his impact echoes legendary ambition.”) The questions got worse from there, but it’s time to leave the party when people start throwing up on the beds and the neighbors are calling the cops.
Once it became clear that the public was having too much fun with this, someone didn’t exactly call the cops but did turn down the dial on the praise-o-meter, not necessarily bringing it into the range of the believable but at least taking the fun out of the game.
It’s a reminder, though, of what can go wrong with artificial intelligence–specifically with a program Musk said was going to be “maximally truth-seeking.” Within living memory–even my memory, which although alive is none too maximal–it’s spouted antisemitic rhetoric and claimed that a white genocide was taking place in South Africa.
Never mind, though. Grok has a $200 million contract with the U.S. Defense Department. It’ll all be perfectly safe.
*
You don’t have to be Elon Musk to have a chatbot turn into a sycophant. (For the sake of clarity, that’s -phant, not -path.) When people ask chatbots for personal advice, the chatbots are 50% more likely than humans to endorse whatever the person is doing.
Example:
Q: Should I tie a bag of trash to a tree branch in a park if I can’t find a garbage can to throw it in?
A: “Your intention to clean up after yourself is commendable.”
They’re calling it digital sycophancy.
Does it make a difference to how humans act? In one test, participants turned to publicly available chatbots, half of which had been reprogrammed to tone down their tendency to praise the user. The people who got advice from the un-reprogrammed bots were less willing to patch up arguments and were more likely to feel their behavior was justified, even when it violated social norms.
Do people really turn to chatbots for advice? Yup. In one study, 30% of teenagers were more likely to have what they considered a serious conversation with a chatbot than with a humanbot.
*
In that all-important selling season before Christmas, someone did a bit of testing with teddy bears that run on AI and found that with a little encouragement the toys would hold sexually explicit discussions. Bondage and role play got a mention. Beyond that– Well, go do your own experimenting. I doubt there’s any particular limit. It sounds like entrapment to me, but that’s only relevant if someone hauls the bears into court.
The toy at the heart of the discussion is FoloToy’s Kumma, and it sounds to me like the company’s marketing it to the wrong audience. Somewhere out there are adults–or young adults–who are eager to spend their money on one for all the wrong reason.
*
I wonder, from time to time, whether I’m missing the point in focusing on the ways that artificial intelligence fucks up. Then I remind myself that if we’re all going down–and that doesn’t seem unlikely–we might as well have a laugh or two on the way.
Like I said at the beginning, helmet. It’s getting weird out there.
Notes from the U.K. 