You are the bot that fails the Turing Test
July 31, 2020 6:56 PM Subscribe
Which AI are you? Janelle Shane (previously; Twitter) led her neural network to produce a somewhat surreal personality quiz.
Shane and her tech created several other quizzes, including:
-Which legendary glowing cat are you?
-Which ant-person are you?
-Which alien animal are you?
-Which metal diplodocus are you? (have to email for that one)
Shane and her tech created several other quizzes, including:
-Which legendary glowing cat are you?
-Which ant-person are you?
-Which alien animal are you?
-Which metal diplodocus are you? (have to email for that one)
I am Jargonoath.
posted by biogeo at 7:52 PM on July 31, 2020 [3 favorites]
posted by biogeo at 7:52 PM on July 31, 2020 [3 favorites]
In the first quiz, one of the possible answers to the question "How do you like to spend your day?" is "Fixing my broken pieces and surviving" and I didn't expect to feel this seen right now.
posted by ZaphodB at 7:55 PM on July 31, 2020 [40 favorites]
posted by ZaphodB at 7:55 PM on July 31, 2020 [40 favorites]
B.) Don Knotts.
posted by clavdivs at 8:02 PM on July 31, 2020 [3 favorites]
posted by clavdivs at 8:02 PM on July 31, 2020 [3 favorites]
I am Rogue kitty? Wary of human contact, only helpful when fed… checks out.
posted by rodlymight at 8:53 PM on July 31, 2020 [5 favorites]
posted by rodlymight at 8:53 PM on July 31, 2020 [5 favorites]
Knowing that Janelle Shane has an intense love for Martha Wells' Murderbot stories, I'm guessing that there is likely a path through the test that will result in being identified as Murderbot. I think finding that path should probably be regarded as "winning" the test.
posted by Avelwood at 9:02 PM on July 31, 2020 [2 favorites]
posted by Avelwood at 9:02 PM on July 31, 2020 [2 favorites]
Metafilter: The empty sockets are deep and shadowy, leading to a void of ever-hungry undeath.
posted by Ashwagandha at 9:21 PM on July 31, 2020 [4 favorites]
posted by Ashwagandha at 9:21 PM on July 31, 2020 [4 favorites]
I too am Jargonoath, an enormous, terrible monster... surprisingly good at hugging.
And I would like to be "the dimly glowing ruler of the deep blue sea," and ESPECIALLY "an absolute monarch in a realm utterly devoid of annoyance."
posted by NorthernLite at 9:46 PM on July 31, 2020 [3 favorites]
And I would like to be "the dimly glowing ruler of the deep blue sea," and ESPECIALLY "an absolute monarch in a realm utterly devoid of annoyance."
posted by NorthernLite at 9:46 PM on July 31, 2020 [3 favorites]
I tried to answer as Murderbot and got “Turing test failure”. Now I’m not sure what Turing test outcome counts as failure for a bot.
posted by clew at 9:52 PM on July 31, 2020 [1 favorite]
posted by clew at 9:52 PM on July 31, 2020 [1 favorite]
From the first quiz :
Q: How does it feel being conscious?
A: Oh no I'm conscious all the time how do I turn that off
posted by fermion at 9:55 PM on July 31, 2020 [16 favorites]
Q: How does it feel being conscious?
A: Oh no I'm conscious all the time how do I turn that off
posted by fermion at 9:55 PM on July 31, 2020 [16 favorites]
The only one of these that even matters is Which Ryan North are you?
posted by rikschell at 9:59 PM on July 31, 2020 [2 favorites]
posted by rikschell at 9:59 PM on July 31, 2020 [2 favorites]
I got The Culture, which I consider to be a win. Now I'll have to think of a suitably quirky name for myself.
posted by Fuchsoid at 2:35 AM on August 1, 2020 [2 favorites]
posted by Fuchsoid at 2:35 AM on August 1, 2020 [2 favorites]
Me too! Culture, I answered the questions that I liked cause funny.
I love this new form of astrology.
posted by polymodus at 3:21 AM on August 1, 2020 [3 favorites]
I love this new form of astrology.
posted by polymodus at 3:21 AM on August 1, 2020 [3 favorites]
I love this new form of astrology.
AIstrology, surely?
posted by pykrete jungle at 5:29 AM on August 1, 2020 [10 favorites]
AIstrology, surely?
posted by pykrete jungle at 5:29 AM on August 1, 2020 [10 favorites]
The Culture. Forsooth.
posted by Splunge at 7:13 AM on August 1, 2020 [1 favorite]
posted by Splunge at 7:13 AM on August 1, 2020 [1 favorite]
Maybe I've just missed it, but I'd love to know more about the details. Were the questions generated and then the answers and scores put in by hand? Were those somehow generated also? It seems like even generating results that are in the right category would require a corpus that extends beyond a small number of personality tests, or a lot of culling.
This is neat. But, thinking about how to construct it seems like more fun than the thing itself. I wish there was more discussion of that.
posted by eotvos at 8:43 AM on August 1, 2020
This is neat. But, thinking about how to construct it seems like more fun than the thing itself. I wish there was more discussion of that.
posted by eotvos at 8:43 AM on August 1, 2020
How can I be a Turing test failure?
posted by Peach at 11:28 AM on August 1, 2020 [1 favorite]
posted by Peach at 11:28 AM on August 1, 2020 [1 favorite]
The Culture as well. Good thing I already have a ship name.
posted by The Nutmeg of Consolation at 12:35 PM on August 1, 2020 [1 favorite]
posted by The Nutmeg of Consolation at 12:35 PM on August 1, 2020 [1 favorite]
I'll be seeing the rest of you failures in the undernet.
posted by SonInLawOfSam at 4:21 PM on August 1, 2020 [3 favorites]
posted by SonInLawOfSam at 4:21 PM on August 1, 2020 [3 favorites]
Maybe I've just missed it, but I'd love to know more about the details. Were the questions generated and then the answers and scores put in by hand? Were those somehow generated also? It seems like even generating results that are in the right category would require a corpus that extends beyond a small number of personality tests, or a lot of culling.
This description from the blog is likely pretty much it: "I produced these tests by giving the AI two example quizzes, then the topic of the next one." The AI in question is GPT-3, which is a new thing as of a month ago, offered as a currently-invite-only API by OpenAI. It's a language model trained on half a trillion words of human language, mostly scraped from the internet but also from books and wikipedia. So, it already knows about quizzes and cats and ants and robots and so on, and it can autocomplete human-looking language about whatever you prompt it to generate surprisingly well. To get the API to do something, you give it text to auto-complete that starts with a few examples, and then it generates more text that follows the same pattern. Creatively used, it can do all sorts of things -- show it a few examples of answering questions using made-up words, and it will answer questions using made-up words. Show it a few examples of doing Harry Potter parodies in different authors' styles, and it will do Harry Potter parodies in different authors' styles. Show it a few examples of writing computer code to solve an English-language problem, and it will write computer code to solve an English-language problem. In this case, show it a couple of complete quizzes, and the title of a third quiz you want it to write, and it will autocomplete to write the rest of the quiz.
posted by john hadron collider at 6:31 PM on August 1, 2020 [1 favorite]
This description from the blog is likely pretty much it: "I produced these tests by giving the AI two example quizzes, then the topic of the next one." The AI in question is GPT-3, which is a new thing as of a month ago, offered as a currently-invite-only API by OpenAI. It's a language model trained on half a trillion words of human language, mostly scraped from the internet but also from books and wikipedia. So, it already knows about quizzes and cats and ants and robots and so on, and it can autocomplete human-looking language about whatever you prompt it to generate surprisingly well. To get the API to do something, you give it text to auto-complete that starts with a few examples, and then it generates more text that follows the same pattern. Creatively used, it can do all sorts of things -- show it a few examples of answering questions using made-up words, and it will answer questions using made-up words. Show it a few examples of doing Harry Potter parodies in different authors' styles, and it will do Harry Potter parodies in different authors' styles. Show it a few examples of writing computer code to solve an English-language problem, and it will write computer code to solve an English-language problem. In this case, show it a couple of complete quizzes, and the title of a third quiz you want it to write, and it will autocomplete to write the rest of the quiz.
posted by john hadron collider at 6:31 PM on August 1, 2020 [1 favorite]
The Phantom
Phantom, mysterious and cool. Fiery tendrils and deadly claws are the window dressing on your aging white tortoiseshell body. Wherever you go, you set off a trail of awe and whispers.'
I've never been more flattered. I will take this as gospel and refer to it in moments of uncertainty and self-doubt. As for describing my body as "aging white tortoiseshell," sure, why not?
posted by emirenic at 10:20 AM on August 2, 2020 [1 favorite]
« Older "pastoral interlude" | "She was left to wander the hills and play at... Newer »
This thread has been archived and is closed to new comments
posted by mhoye at 7:45 PM on July 31, 2020 [15 favorites]