Not-so-Stochastic Parrot
“Alexa, art.”
This is how a nearly two-and-a-half minute Instagram reel starts. These words are said by a sassy Gray parrot named Molly. She is reciting these words to the Amazon smart speaker in the room. The Alexa device produces a farting sound and then generates a description of the sound that was just produced (“That was a triumphant one”). After describing the sound, there’s the Alexa voice beckoning the parrot to continue interacting with the speaker (“Ask me for a gassy fart”). The video proceeds to feature the same pattern over and over again.
Bless the humans that made this all possible. That might not be your first thought, because there are no humans in the video. However, they’re the creative minds who are pulling all the strings and making this clip a reality. As AI becomes increasingly ubiquitous, I think it’s important to recognize the brilliance of humans, and the limitations of AI.
Too often, I see people responding to AI with amazement that gives credit and attributes to AI that don't feel appropriate. The technology is incredible and revolutionary. At the same time, I don't think we're ready for it: our brains haven't fully wrapped around the idea that non-humans can produce coherent text or auditory words. So, when we see language output carried by non-humans, we can't help but give some humanness to that device, software, or animal. Brains, you are SO cool, but slow down!
Also, let’s give credit where credit is due. The video clip I described earlier is AI use that I can get behind: robotic flatulence on command by a beady-eyed, modern-day tiny domestic dinosaur. When the parrot voices the command, “Alexa, fart”, the initial f-sound isn’t really there, so what I hear each time is “Alexa, art.” Yes, what the Alexa speaker produces each time is art, no question about that.
Some very sophisticated code has been produced, and then massive amounts of writing have been put into the model as training data. The program can analyze the language patterns in the writing, and then make a mathematical judgement about what's the mostly likely string of words to produce. It's a very difficult and applied math problem: it's not a poet, a writer, an assistant, a friend or anything else remotely human.
Also, let’s give credit where credit is due. The video clip I described earlier is AI use that I can get behind: robotic flatulence on command by a beady-eyed, modern-day tiny domestic dinosaur. When the parrot voices the command, “Alexa, fart”, the initial f-sound isn’t really there, so what I hear each time is “Alexa, art.” Yes, what the Alexa speaker produces each time is art, no question about that.
Chef’s kiss.
Perfection.
No notes.
And I'll bet you're curious about that fantastic video, so here it is (I sure hope I didn't hype it up too much for you):
And I'll bet you're curious about that fantastic video, so here it is (I sure hope I didn't hype it up too much for you):
Humans as Coders
The first set of humans I'm going to appreciate for making this moment possible would be the Amazon programmers. At some point, they decided that Alexa should respond to a “fart”/"art" command with fart sounds, and had to supply a variety of such sounds with descriptive comments about each one. They put each of these into the programming language that would allow the speaker to seamlessly offer these interactions on command.
But, is this really that special of a feature? I decided to do my own research and test it out with Siri. When I asked Siri to fart, the response I received was, “I’m software that’s made up of lines of code. So that’s not really part of my deal.”
I appreciate the implication that farting on command is part of anyone’s deal, so long as they’re not a computer program. Wouldn’t computer programs be the most capable of automatic flatulence? I digress...
I was not deterred by Siri's denial. I made further requests in attempt to get my iPhone to produce farting sounds. The response I received to the second command was a list of internet resources for fart sounds. My third attempt at heckling Siri to generate a fart sound resulted in a kind of farting-and-AI full circle moment: the suggestion I got was the song, “I Can’t Stop Farting Around my Wife” on Spotify, and the artist is “Beats by AI.”
I think that this rabbit hole is getting out of control: blame the childish part of my brain that didn’t move past finding farts funny. I’ll go explore those websites that Siri listed to get my fix.
One thing is clear from all of this: if the fart feature hadn’t been programmed into the Alexa code, then the speaker is not going to make those sounds, this video would never exist, and we’d be missing this important artifact from the AI discourse.
Humans as Bird Wranglers and Content Creators
Now, let’s get to the parrot, Molly. She's obviously very special, but her humans should really get a round of applause.A person had to know about the fart option from the Alexa speaker, and then teach the phrase “Alexa, fart” to Molly, likely over and over again. I'm guessing that there were moments of frustration and bribery, which make it all the more amazing that we get to see Molly voicing this demand dozens of times in the video in succession with only pauses for Alexa.
Don't forget: someone in Molly’s life had to film all of these moments, and edit them together in this fabulous mega-clip sequence. There is no repetition in Alexa's descriptions or suggestions. Molly has a variety of reactions and expressions.
I imagine that there are hours of footage that had to be deleted because the recordings were just repeating what had already been selected for the clip. Who knows what other efforts were made: storyboarding? Keeping track of the different categories of each of these unique instances? I can only imagine.
The final act suggests Molly's limits and where humans likely had to intervene in a way that couldn't be passed off as Molly's doing. Unprompted, the scene begins with the Alexa voice stating, "One hundred farts coming your way." Did they try to teach Molly to request 100 farts? If that had been successful, surely we would have seen it. The humans are clearly dedicated to their craft, but that was more than they could deliver, I suppose. Instead, a person likely had to prompt Alexa, and then stick around to capture Molly listening to the entire sequence. Bravo.
Humans as Engaged Audience
For me, the experience of watching this clip is elevated by the comment section. That's definitely not always the case on the internet. But, something about a parrot asking a robot to fart brings out the best in humanity.
Some comments that stood out to me:
Alexa is a little bit more into this than one would think
poor alexa can't even get a fart out before she's expected to produce anotherThe last bit where [Molly] was finally satisfied by 100 farts❤️❤️
From these, you can see people filling in the gaps and co-constructing a narrative. They're seeing intention that's not there, and that's incredible. Humans are meaning-making creatures whose observations paint stories without a basis in the actual facts of the situation. Of course, the creators of the video have done some work to lead people to those conclusions.
The following comment
Not Alexa encouraging this depravity 💀
The words "encouraging" and "depravity" are both doing some heavy lifting. First, "depravity" assumes that Molly is doing anything beyond just repeating what she's been taught by her people. When I see "encouraging", I see the writer has thought of Alexa having an intention beyond just carrying out what's in her code.
I love it. It's so much better than the stripped down elements of a bird and a speaker.
AI as a Robot
The namesake of this post is a reference to Bender et al's 2021 article, "On the Dangers of Stochastic Parrots: Can Language Models be too big? 🦜""Stochastic parrots" is the evocative label that the authors give to AI writing systems (aka, language models, large language models, or LLMs). This very deliberate choice is showing that the technology behind AI writing software doesn't have sentience or engagement with the language it's producing.
A little more of a breakdown here:
- Parrots produce what sounds like language, and usually don't know what it means. For instance, "parroting" as a verb that means repeating without engaging with the meaning of what's being uttered.
- "Stochastic" means probabilistic, and I've inserted an image of the definition below:
Molly is not making complex mathematical calculations to ask for farts from Alexa. I'm not 100% sure what's happening in that bird brain, but I'm guessing it's a soup of her human peppering her with the phrase "Alexa, fart", being a social creature that wants to please her human, and some behavioral training.
Any time you see something AI-produced that feels creative or inspiring, I can almost guarantee you that it's the result of a creative human, or your own amazing mind filling in the gaps to make some meaning.
In Conclusion
No magic here. Well, maybe some magic. But, it’s definitely not the magic of technology.
Instead, it’s the magic of sparky human brains making a million creative decisions and dedicating themselves to the time-consuming labor of coding, bird training, filming, and video editing. And then there's the magic in the eye of the beholder, who will see a story from the pieces that have been put forth.
And the result? A two-minute and 22-second video that tells a story of a flatulence-crazed parrot whose hunger is is only sated when she gets a sequence of 100 farts in a row.
Amanda! This piece turned out so great! I’m so glad I was able to be a part of it evolving over time. Nice work!
ReplyDeleteI love this perspective! "It's a very difficult and applied math problem: it's not a poet, a writer, an assistant, a friend or anything else remotely human."
ReplyDeleteI have sooo many thoughts, and most of them are just YES! AI is made by humans. it has inherited (or in many cases, plagiarized) the norms, biases, expectations, linguistic quirks, and art styles of humans. I think whoever decided to call ChatGPT "AI" really messed it up for us. Unsure what a better term is, but it's not intelligent!