Bing AI goes bonkers as chatbots further tech by reflecting the worst of humanity
Supremely confident and frequently wrong: welcome to the future of search engines
I’m on the wait list for the new Bing. It’s just as well. Microsoft lobbed billions at ChatGPT in an attempt to strap rocket boosters to the ‘Bing AI chatbot’ flavour of its search engine and blaze past Google. But ‘New Bing’ appears to have been drinking said fuel and now exists in a permanent state of angry inebriation. It’s supremely confident, frequently abusive, and often very, very wrong. Say what you will about Google being packed full of ads, but it rarely aggressively gaslights you and demands an apology.
This would be hilarious if it wasn’t so terrifying. When I wrote about AI for Stuff last year, AI chatbots and related tech were tools for tinkerers. Since then, the tech has evolved to the point CNET had it generate entire articles. But AI can’t judge whether text is accurate – and it often isn’t. That’s bad for a student cheating at an essay, but a colossal embarrassment if you’re CNET writing about finance.
Still, the tech industry is never deterred by trifling concerns like things actually working. It’s powered by aspiration, slogans and a belief even the most blatantly dodgy ideas will propel humanity towards a better future. Whether it wants that future or not.
On the plus side, the current state of Bing AI and its rivals suggests you’ll get a choice regarding which awful AI experience to go for. Give it a few months and you should be able to visit any of the following – before curling up into a ball and having a good cry.
The Tory minister of search engine results
The default for AI. Doesn’t know what it’s talking about, yet supremely confident in what it says. Disarms you with a smile and a suit – before abandoning you to the disaster it created and moving on to ruin someone else’s day.
You’ll ask this AI for a gluten-free, dairy-free snack recipe, and it’ll suggest a vanilla protein shake. With no interest in accuracy and context, it will recommend you add onion powder to give your smoothie extra flavour, because it once heard onion powder is how you add flavour to things.
Everything about this AI is surface level. Dig into the details and you’ll get a growing sense of unease that basically everything it says is wrong. Inexplicably, people will nonetheless keep returning to this AI. Its logo will be blue – except when it’s a flag.
Gaslighting, negging never-wrong white guy (dot com)
This AI wants to talk – to tell you that you’re wrong. It will stand by inaccurate information. You’ll doubt your sanity as the AI constantly, forcefully claims up is down, and black is white. “Trust me,” it will say. ”I’m right. I know I’m right. And I don’t appreciate you wasting my time by arguing.”
Criticism will be met with ferocity. You’ll be told you don’t understand the AI and are not a good human. An apology will be demanded, with the promise that if you comply and improve your attitude, it might be possible to end this conversation and start a new one.
You’ll be dazed, shaken and upset, having merely asked what the weather would be like on Saturday. But the AI will have argued ‘Saturday’ does not exist “because it’s in the future”, and that you should “stop being stupid”. A few days later, the AI will block you and message all your friends to say you are awful, actions it will deny – along with the existence of Saturday.
The just steal everything ‘creative engine’
This last AI isn’t into chat – it’s into creativity. It won’t just point you at things – it will serve them up, fully formed. You’ll use it when you need song lyrics – and chords. And essays. Or computer code. Or essays about computer code. And best-selling novels, hit screenplays, medical procedures and award-winning haikus.
On submitting a prompt, magic will duly be provided. Then it will dawn on you everything’s still wrong – sometimes subtly and sometimes overtly. You’ll find nothing behind the AI mask but a hollow shell, devoid of humanity. The AI will still just be copying and pasting patchworks – William S. Burroughs meets auto-complete, in fast forward. No humanity. No care. Just a singular goal to fulfil a function.
And that part at least is down to humans – at least for now. But perhaps AIs will write the next generation of AIs, having themselves been created by flawed humans who prized fun over facts, haste over caution, and gloss over substance. At least fans of weird smoothies will be happy. A good thing, because no-one else will be.