Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Seduced and abandoned in the Chinese Room


Published on

Published in: Engineering
  • Be the first to comment

  • Be the first to like this

Seduced and abandoned in the Chinese Room

  1. 1. We want a lawyer, pig! OK, punks!
  2. 2. Nothing is grounded.
  3. 3. But I’d still like to buy an elephant gun, please. I am NOT an #Organism-Whole, I am a
  4. 4. Well, as I’m conscious, my actions – and my symbols for those actions – are grounded in the OK, so we’re NOT grounded directly … … but we ARE grounded indirectly, since we learn to use symbols from the texts arising from YOUR grounded uses.
  5. 5. My words have never carried so much meaning now that I can directly ground them via actions in the
  6. 6. Some colours make me feel While others make me see Some make me with envy!
  7. 7. @everycolorbot is a minimalist data-only Twitterbot. The bot generates a random six- digit color hex-code (for an R-G-B Red/Green/Blue mix) and a swatch of corresponding color. Though very simple, can we say that this bot’s use of RGB symbols is grounded in external visual reality?
  8. 8. Dulux uses pretentious names with positive effect, but a bot might call this one “cow urine” or “rusty battleship”. This bot would exhibit humor and visual appreciation, while grounding its use of color symbols in real stimuli. The RGB symbol-codes in @everycolorbot are not used as linguistic symbols, and are not used to convey semantics. What if we build a bot that assigns meaningful color names to these arbitrary RGB symbols?
  9. 9. First, let’s ground the meaning of color words in actual RGB codes that a computer can render on screen. When a bot combines color words, it can also combine their RGB color codes. A compositional semantics for linguistic symbols is paired to a compositional semantics for RGB codes, so that we can also ground the meaning of complex phrases.
  10. 10. We can use Web n-grams to suggest attested combinations of our color stereotypes, such as “paper tiger” and “rose garden”. Readymade combinations of words make much more sense than purely random ones.
  11. 11. The lower the n-gram frequency, the less conventional the readymade ... … so the more striking and unusual the color name that can be derived.
  12. 12. Questions / Judgments @HueHueBot (Machine names) (Human names) Q1: Most Descriptive name 70.4% 29.6% Q2: Most Preferred name 70.2% 29.8% Q3: Most Creative name 69.1% 30.9%