Bing’s AI chatbot doesn’t think like a human. Neither does your dog.

Unlike Microsoft’s New Bing (a topical bot integrated with OpenAI technology), my spaniel Hobbes cannot speak natural language sentences. But that didn’t stop it from feeling like Hobbes was actually giving a series of verbal responses to various inputs.

Enter: I will be home in 5 hours. output: OMG, a long-lost member of the pack is back from the wilderness! Let’s party! Input: Postman approaching. output: caveat! murderer! Watch them drop their bags and run away! Enter: Go outside, come back in 5 minutes. output: OMG, a long-lost member of the pack is back from the wilds! Let’s party!

What I’m doing is classical anthropomorphism: “attributing human traits, emotions and behaviors to non-human beings”, as ChatGPT said when asking for a definition (this is Suspiciously similar to the first sentence on the website Anthropomorphic Wikipedia page(opens in new tab); Call me, ChatGPT!)

Hobbes’ brain processes visual, auditory, and (especially) olfactory vortices in ways that I, as a non-dog, don’t really understand. But my ability to anticipate his reactions and fine-tune them in training leads me to imagine a cartoon speech bubble over that dog’s adorable head.

Millions of people are now doing something very similar with AI chatbots. See the bizarre reactions elicited by journalists and other early Beta his testers to pinpoint human motives. Entertainment has trained us to find villains in sci-fi movies. Watch a sci-fi movie villain.Newbing is trying to gaslight us(opens in new tab)! that is liar manipulating emotions(opens in new tab)! Evil!this is Spooky(opens in new tab)! that Tried to dissolve the reporter’s marriage(opens in new tab)!

Hmm, no.of OpenAI magic(opens in new tab) It’s certainly a smart algorithm, but it’s just a bunch of machine learning algorithms.These are the smartest things we’ve ever made, and they’re also silly enough to make Hobbes’ cotton-candy brain look like Einstein. is explained Machine learning behavior similar to pigeons(opens in new tab)We are literally talking about bird brains.

Think of New Bing as a parrot if you didn’t have so many animals in one article. Because it is essentially quoting us to us. This can get very annoying very quickly or feel strangely comfortable.

A brief history of artificial intelligence

Humans are anthropomorphic engines.As babies we look for faces in our environment as soon as you can concentrate(opens in new tab)and we will not stop (many “Faces in Things” Instagram account(opens in new tab) An example of how this engine works in the brain). Anthropomorphize animals, rocks, clouds, the sun, the moon, and the stars. That’s what led to almost all myths, not to mention astrology.

For thousands of generations, we’ve learned that dogs are about 1,000 times more likely to drop food when they’re sitting right in front of us with sad eyes on their faces. That’s why they came to flock to us.Cats, meanwhile, were honed at the pitch of human baby cries(opens in new tab) and actually I put it in my throat when I’m hungryThis is intelligence. )

Anthropomorphism is also why the famous Mechanical Turk — an 18th-century fake chess-playing robot secretly commanded by a chess player in a nearby cabinet — was a successful trickster for 80 years. Napoleon played against his 40-year-old machine and cheated. So did Benjamin Franklin.Mary Shelley didn’t even start science fiction frankensteinThe word “robot” was coined two centuries ago, and we were ready to believe in human-like beings made by humans.

So why should we humans in the internet age need to be any different? We are not. Or have you ever anthropomorphized Facebook as if it were a living thing in its own right? Zuckerberg’s meme monster. We were blown away by the artificial intelligence talk.From his high-concept HAL 2001 Star Wars droids become Cylons Battlestar Galactica, Star Trek data. We expect AI (true AI, what experts call intelligence in general) to be something we see in more lives than aliens. (never an alien.)

AI: Internet Autocomplete

Here comes OpenAI’s ChatGPT and, thanks to significant investment by Microsoft, New Bing, which takes advantage of our natural personification. Not exactly a mechanical Turk, but powered by a human. Many humans, aka the Mechanical Turk, like Amazon’s global peaceworker marketplace. Humans this time are not even partially paid. Humans are all online.

One of the most compelling explanations for AI chatbot technology is thatinternet autocomplete(opens in new tab)OpenAI is just scraping all the knowledge and creativity we have out of our heads and putting it into a huge data set that has almost everything in it. ChatGPT was trained on data sets including Reddit. New Bing clearly likes to munch on news, which might explain why it’s so easy to write a story about yourself.

Either way, AI chatbots are directing that information at us in a format that the algorithm believes will be useful. (I tried other verbs that didn’t sound perceptual to the algorithm, but they were all anthropomorphic: to judge, to predict, to suggest, to hope. See, even in our language, here I fail.)

It’s an anticipation of what could be useful from this incredibly vast repository of information that’s been buzzing around in these viral threads.Apparently it is because Those are threads.

too many questions, human

What is this Microsoft says in their strangely unsigned blog:(opens in new tab): “In lengthy chat sessions of 15+ questions, Bing can become repetitive, not always helpful, or provoked or induced to answer in a tone that is designed for us. There is a chance…a very long chat session can confuse the model as to which question they are answering.”

Did a time-saving AI chatbot write this? I saw. New Bing “will attempt to respond or reflect when asked to provide a response that may lead to unintended style. This is an important scenario that requires many prompts.”

In other words, if you start repeatedly prompting the AI ​​to think about its “shadow self,” new york times Reporter Kevin Roose did in a huge, multi-part overtime-inducing conversation with New Bing. Like my dog, it does what you trained it to and just wants to please you!

However, AI cannot actually display dog-level intelligence. This includes lots of real emotions and a pretty miraculous nose. ChatGPT and its ilk are rather insect-level intelligences, inputs, very basic instincts and outputs. And if it’s too long, there will be many irregular buzzing sounds. Build a reliable search engine out of it.

This does not detract from OpenAI’s work to date.We have the incredible ability to automate a lot of low-level white-collar tedious work, like writing the first draft of an email explosion or a document that should sound like it was written by a committee anyway. We’re talking about tools. Supercharge your coding process(opens in new tab)expect software updates to be cheaper and faster.

And yes, it can conveniently answer many search-like queries. The trick is that the bot always determines if he has one of the frequent factual ‘hallucinations’. Perhaps you need to build a fact-checking bot to maintain your chatbot. I can’t.

But when it comes to recreating human intellect and emotion, the funhouse mirror versions of the weird conversations and internet chatter born out of chaos just don’t cut it. When we’re talking to a chatbot, in a very Turing-test-like way, we still know, even if it’s surprising us.

Ask a few questions and you’ll see how it robotically constructs answers. You will smell it as surely as Hobbes smells the postman. Moving from New Bing to General Intelligence? Convincing Hobbes not to party with Puck is easy.