Skip to content

Google in your Brain

2010-04-05 · Hank · watch on YouTube · Internet & YouTube · explainer learning

Hank mocks YouTube's bad auto-captions but argues Google has the data and incentive to master speech recognition soon; understanding meaning would need artificial intelligence we can't build, yet accurate recognition plus implants could change what it means to be a person.

What is said

00:00:29 — Greeting; auto-captioning is very bad but improving; Google's database, talent, money and ad incentives

  • ★★ Hank opens with a dramatic reading of YouTube's auto-captioning of John's video imploring him to teach John's son science, lest John explain the blue sky with pterodactyl poop. (playful) — 00:00:29
  • ★★ Hank says Google/YouTube's auto-captioning is currently very bad, as viewers can see by turning on captions for this video. — 00:00:29 · Google
  • ★★ Hank says Google is getting better at it and has analytical access to the world's largest database of recorded spoken words: YouTube. — 00:00:58 · Google, YouTube

00:01:27 — Contextual ads and voice phones; Google is creepy; recognizing words vs understanding meaning; that would be artificial intelligence, and we don't understand brains

  • ★★ Hank says Google has talent, money and strong incentives: understanding speech would let it run contextual ads on videos, and contextual ads are how Google makes all its money. — 00:01:27 · Google, online advertising
  • ★★ Hank says Google also wants people to be able to talk into their phones, which he can already do, but the broader implications of this technology kind of creep him out. — 00:01:27 · Google
  • ★★★ Hank says recognizing which words are said and understanding what they mean strung into a sentence are completely different tasks, and the second is even harder. — 00:01:27
  • ★★ Hank says Google has so far been unable to turn sentences into useful information; it needs cultural context and tone, and a program that could would pretty much be artificial intelligence. — 00:02:10 · Google
  • ★★★ Hank thinks we dramatically underestimate our brains' complexity: AI would require writing the software that runs our brains, and we have no idea how brains work. (hedged) — 00:02:10

00:02:39 — Implications of accurate speech recognition: live subtitles, machine translation, bionic implants and what it means to be a person; read Feed; it will come soon; sign-off

  • ★★ Hank says 100% accurate speech recognition in two languages would make subtitling conversations on handheld, in-ear or in-eye devices trivial, though machine translation would stay inaccurate, as Google Translate shows. — 00:02:39
  • ★★★ Hank thinks speech recognition combined with a bionic implant, which to him does not seem far off, would dramatically change what it means to be a person. (hedged) — 00:03:08
  • ★★ Hank says anyone who hasn't thought about that should read the novel Feed. — 00:03:08
  • ★★★ Hank says the mis-captioned snippets are funny now, but technology will probably soon get speech right, so it is worth thinking about what that will mean. (hedged) — 00:03:08

Humor (significant)

  • Hank solemnly reads YouTube's garbled auto-captions of John's sky-is-blue video, giant space birds and all. — 00:00:00 > «And that is why house are gray.»
  • Showing how badly the auto-captioning renders him, Hank says what it generally turns his speech into. — 00:00:29 > «Donald, report back with an elephant.»
  • After talking into his phone to show off voice input, Hank drops in another garbled, machine-transcription-style line. — 00:01:27 > «Robot Nita, I am a role model. What are your genes?»
  • Hank notes that recognizing which words are said is quite difficult. — 00:01:27 > «I have recently exciting sex if the Senate lasts a couple days.»

Threads in this video

Replies to

  • Fatherhood — «a video in which you implore me to teach your son about science»

BR2_6b8ecd0091