Stochastic popinjay and Perso-Arabic art / adab

« previous post | next post »

‘Stochastic Parrot’: A Name for AI That Sounds a Bit Less Intelligent

An ancient Greek word for guesswork fuels a term that suggests supersmart computer programs are just mimicking whatever they see

Ben Zimmer, WSJ, Word on the Street (January 18, 2024)

In his capacity as chair of the American Dialect Society's 2023 Word of the Year competition new words committee, our Language Log colleague Ben Zimmer oversaw the selection of candidates from the "special ad-hoc category related to one of the most buzzed-about stories of 2023: artificial intelligence."

Our new category included an array of AI heavy hitters. There was “ChatGPT,” the name for OpenAI’s chatbot, which is so successful it often gets used generically for any generative AI system. There was “LLM,” short for “large language model,” the machine-learning algorithm trained on mountains of text that powers AI programs. And there was “hallucination,” for AI-generated responses that are untethered from reality.

Triumphing over all of these was “stochastic parrot,” used by AI skeptics to point out that large language models are not capable of understanding language but instead have been programmed to parrot back plausible-sounding synthetic text.

The term was coined by University of Washington computational linguist Emily M. Bender in a widely cited 2021 paper. As it happened, Bender was in attendance at the Word of the Year event and spoke about how her coinage of “stochastic parrot” took off.

The first part of the phrase, “stochastic,” means “randomly determined,” and is derived from the ancient Greek word “stokhastikos” meaning “based on guesswork.” A “stokhastes” was someone aiming to predict the future through divination. The word is recorded in English from 1662, when the theologian John Owen wrote of the “stocastick faculty,” meaning the power of conjecture.

It would take until the 20th century for the term to get applied to modern statistical modeling. The Russian statistician A.A. Chuprov used the word “stochastical” in a 1923 paper to mean “based on the theory of probability,” and the shorter form “stochastic” became a standard way to refer to probabilistic distributions and patterns.

Bender put a new spin on the word when she and her co-authors published their paper on the risks of “stochastic parrots,” or “large language models that are impressive in their ability to generate realistic-sounding language but ultimately do not truly understand the meaning of the language they are processing.”

“Parrot” has referred to the mimicking bird since the 16th century, and soon after it entered the lexicon, it came to be used for someone who repeats words without understanding, as well as a verb meaning “imitate like a parrot.”

That clever locution, "stochastic parrot",  swiftly grew in popularity in the AI community, particularly among those who wished to minimize what ChatGPT and similar enterprises could accomplish.

As “stochastic parrot” snowballed in popularity, Bender witnessed how her creation swiftly spun out of her control. She notes bemusedly how the term often gets attributed to other (male) researchers, and how it has been interpreted as a slur against AI. People who are invested in the idea that LLMs can achieve artificial general intelligence, Bender told me via email, “take ‘stochastic parrots’ as an insult to the machines that they have anthropomorphized (and maybe identified with).”

Before we dismiss what AI can achieve as mere mimicking, perhaps we should take a look at what parrots stood for in other cultural traditions.

Sanskrit fables associated with parrots were regularly compiled and translated as part of the Perso-Arabic traditions of adab, or moral wisdom, most notably the famed Tuti-nama (Book of the Parrot), in which a domestic parrot regales his mistress nightly with entertaining tales to prevent her from leaving the house for romantic dalliances while her husband, a trader, is traveling. "A sort of Scheherazade thing in reverse," [Rajeev] Kinra muses. "But the main point," he adds, "is that the parrot loomed large in the medieval and early modern Indo-Persian literary imagination as a symbol of Indian lore, and Indian storytelling…".

[Source:  "A 17th-Century Indian Parrot: Interpreting the Art of the Deccan", Shatha Almutawa, Perspectives on History (Jul 1, 2015)]

Parrots, stochastic or otherwise, can live to be old and wise.  Don't sell them short.

 

Selected readings

[Thanks to June Teufel Dreyer]



4 Comments

  1. AntC said,

    January 20, 2024 @ 11:56 pm

    New Zealand's realnational bird, for example.

  2. ardj said,

    January 21, 2024 @ 7:35 am

    One should not overlook that parrotts can be quite original in their speking, and sometimes even offer nice and accurate prophecies. For instance Skelton's popagay replies to Galathee, who asked about 'frantycknes and folyshnes whyche ys the great state':
    "Wilfulnes and brayneles now rule all the raye…
    He triumfythe, he trumpythe, he turnythe all up and downe"

  3. Chris Barts said,

    January 21, 2024 @ 9:18 am

    The interesting question is, how much of human behavior is a stochastic parrot?

  4. Jarek Weckwerth said,

    January 22, 2024 @ 1:36 am

    @Chris Barts Spot on! Many foreign language courses and exams actually work on this principle. You give the student some kind of content, because you must not assume they have anything to say on their own, and the content must be standardized. Then what you're interested in is only the quality of the language they, erm…, parrot.

RSS feed for comments on this post