A Google AI model developed a skill it wasn’t expected to have

From Yahoo Finance:

Concerns about AI developing skills independently of its programmers’ wishes have long absorbed scientists, ethicists, and science fiction writers. A recent interview with Google’s executives may be adding to those worries.

In an interview on CBS’s 60 Minutes on April 16, James Manyika, Google’s SVP for technology and society, discussed how one of the company’s AI systems taught itself Bengali, even though it wasn’t trained to know the language. “We discovered that with very few amounts of prompting in Bengali, it can now translate all of Bengali,” he said.

Pichai confirmed that there are still elements of how AI systems learn and behave that still surprises experts: “There is an aspect of this which we call— all of us in the field call it as a ‘black box’. You don’t fully understand. And you can’t quite tell why it said this.” The CEO said the company has “some ideas” why this could be the case, but it needs more research to fully comprehend how it works.

CBS’s Scott Pelley then questioned the reasoning for opening to the public a system that its own developers don’t fully understand, but Pichai responded: “I don’t think we fully understand how a human mind works either.”

AI’s development has also come with glaring flaws that lead to fake news, deepfakes, and weaponization, sometimes with so much confidence, in what the industry calls “hallucinations.”

Asked if Google’s Bard is getting a lot of “hallucinations,” Pichai responded: “Yes, you know, which is expected. No one in the, in the field has yet solved the hallucination problems. All models do have this as an issue.” The cure, Pichai said, is around developing “more robust safety layers before we build, before we deploy more capable models.”

Link to the rest at Yahoo Finance

3 thoughts on “A Google AI model developed a skill it wasn’t expected to have”

  1. I hope that the Bengali translations are better than — well, to name languages I or a family member reads/writes with reasonable facility — Arabic, Farsi/Persian, German, Italian, Japanese, Portuguese (Brasiliero), Russian, and Spanish. Or English into any of those languages. I have no reason to be overoptimistic;† I won’t quote the horrific, laughter-spasm-inducing “translation” of the TSA “prohibited items in carry-ons” into Italian that turned a couple of items into borderline-profane street-Italian insults. (That may have all too accurately “translated” the TSA’s intent, but still…)

    † One nonfungible brownie point to the first person who correctly identifies the source and speaker of that line, from memory without a search of any kind (except perhaps pawing through one’s own record/CD collection).

  2. Not a surprise to anybody who undefstands the common roots of live languages:

    https://lingo-star.com/bengali-language/?v=4326ce96e26c

    Even basque, is a live language that LLMs should learn without much trouble.

    Here’s Bingchat’s take on basque:

    “The Basque language, also known as Euskara, is a **language isolate** spoken by the Basque people in the Basque Country, which straddles the westernmost Pyrenees in adjacent parts of northern Spain and southwestern France ¹. It is the last surviving Paleo-European language spoken indigenously in Europe, predating the Indo-European languages that invaded Europe during the Palaeolithic and now dominate the continent ¹.

    The origins of the Basque language are not well understood, but it is believed to have been present in and around the area of modern Basque Country before the arrival of the Indo-European languages in western Europe ¹². The Basque language is classified as a language isolate, meaning it has no known linguistic relatives ¹.

    Some scholars have suggested that Basque may be related to the Iberian language, which was spoken in the Iberian Peninsula in the pre-Roman era ². Others have proposed that Basque and Aquitanian, a language spoken in southwestern France, may have derived from the Pre-Indo-European languages spoken in Europe before the arrival of Indo-European speakers in the region ².

    The terms “vasco” and “basque” are inherited from the Latin ethnonym Vascones, which in turn goes back to the Greek term Οὐάσκωνες (ouaskōnes), an ethnonym used by Strabo in his Geographica (23 CE, Book III) ¹.

    The whole thing is right in the wheelhouse for (ahem) LARGE LANGUAGE MODELS!

Comments are closed.