Facebook M – The Anti-Turing Test • Hacker News Board

Re: Facebook M – The Anti-Turing Test

#71

Post by pinkrooftop » Tue, Nov 10, 2015, 2:41 PM UTC

Human or AI the level of service provided seems pretty amazing

Re: Facebook M – The Anti-Turing Test

#72

Post by username223 » Tue, Nov 10, 2015, 2:47 PM UTC

Can M wade through phone support menus and cancel Comcast subscriptions? I look forward to a darkly-humorous future in which we pit poorly-paid third-world citizens against each other in wars of call center attrition. Better, once we equip them with those sound-board UIs that play pre-recorded answers in native English-speaking voices (can't find the link), English can become a transmission protocol that few people deal with directly.

Re: Facebook M – The Anti-Turing Test

#73

Post by hodwik » Tue, Nov 10, 2015, 2:47 PM UTC

jhpriestley wrote: post #66

graeham wrote: post #11

"I'm AI but humans help train me" The implication to me is that the chat is with a human, who is using an AI tool with the intention of training that tool. What better way to train a new service than to launch it, then answer all the weird, unexpected questions with humans? Gradually more of the questions get answered, the AI gets better trained, and the human-AI becomes increasing more AI. Further, as the AI gets be…

AIs are already trained on huge corpuses of natural human communication, but they still aren't able to reliably analyze human language. Adding a few more gigabytes of chat logs isn't going to do much.

They weren't trained to read natural language, and then respond with some programmatic action filtered into natural language:

  Bot reads "What's the temperature like near me?"
  Person calls "$get user-local-temperature"
  API responds "{temperature:{f:77},{c:99}}"
  Human writes "It's 77 degrees outside!"

Training set now contains that relationship between that question, that API call, that response, and that natural language response (and probably the users location, age, gender, and so on, all captured in the meta-data about the response in the corpus).

  Bot reads "What's it like outside?"
  Person calls "$get user-local-weather"
  API responds "{weather:{now:Sunny},{today:Cold}}"
  Human writes "It's sunny now, but will be cold later today."

And so on. I think the goal here is training on standard API calls as the response, and taking their data return and converting it into grammatical sentences. It's a two step training process. Know which API to call, and know how to convert API response to natural language.

There's no serious corpus yet for that -- if this is real, it is important work.

Re: Facebook M – The Anti-Turing Test

#74

Post by gjm11 » Tue, Nov 10, 2015, 2:51 PM UTC

For something with a similar flavour, see http://dangermouse.brynmawr.edu/csem/coffeehouse.html; start reading where it says "Post Scriptum".

Re: Facebook M – The Anti-Turing Test

#75

Post by traverseda » Tue, Nov 10, 2015, 2:52 PM UTC

gjm11 wrote: post #74

For something with a similar flavour, see http://dangermouse.brynmawr.edu/csem/coffeehouse.html ; start reading where it says "Post Scriptum".

Didn't like that semicolon.

Re: Facebook M – The Anti-Turing Test

#76

Post by jasperry » Tue, Nov 10, 2015, 2:56 PM UTC

I wonder if you can ask M to use fewer exclamation points. From the conversations I saw in the article, it's a little too chirpy (or should I say "clippy") for my taste.

Re: Facebook M – The Anti-Turing Test

#77

Post by adrianb » Tue, Nov 10, 2015, 2:59 PM UTC

hellbanner wrote: post #37

So how well does this scale, if all of Facebook's users are using "M" like this?

Launch is limited to selected users in Silicon Valley.

Re: Facebook M – The Anti-Turing Test

#78

Post by thanatropism » Tue, Nov 10, 2015, 3:02 PM UTC

andreasvc wrote: post #43

Grue3 wrote: post #21

"Can you solve this CAPTCHA for me?" (provided the CAPTCHA is sufficiently OCR-resistant)

If you want to avoid giving information about whether you are an AI or human, you simply respond "No." Besides, the CAPTCHA's that are sufficiently hard to solve for computers are already hard for humans as well.

Question isn't "will you?" -- it's "can you?"

Re: Facebook M – The Anti-Turing Test

#79

Post by jhpriestley » Tue, Nov 10, 2015, 3:14 PM UTC

hodwik wrote: post #73

jhpriestley wrote: post #66

Earlier quoted context omitted.

AIs are already trained on huge corpuses of natural human communication, but they still aren't able to reliably analyze human language. Adding a few more gigabytes of chat logs isn't going to do much.

They weren't trained to read natural language, and then respond with some programmatic action filtered into natural language: Bot reads "What's the temperature like near me?" Person calls "$get user-local-temperature" API responds "{temperature:{f:77},{c:99}}" Human writes "It's 77 degrees outside!" Training set now contains that relationship between that question, that API call, that response, and that natural langu…

The translation of unrestricted human language into a restricted, "logical" language which could be understood mechanically, is another long-standing problem - going back as far as Liebniz - which is unlikely to be solved by a few gigabytes of chat logs.

Re: Facebook M – The Anti-Turing Test

#80

Post by drumdance » Tue, Nov 10, 2015, 3:26 PM UTC

andreasvc wrote: post #46

fab1an wrote: post #5

Facebook's strategy here is to build an AI brand before they have the actual technology, which could make a lot of sense. At the same time the interactions between M's team and its users will provide meaningful data to train the AI on.

I've never heard vaporware characterized as something that "could make a lot of sense"; how could it? Plus I'm doubtful whether the data would be very meaningful. A bunch of people adversarially trying to figure out whether the AI is real is not representative or generally useful data.

Microsoft Windows was vaporware for years. They famously did a "demo" that was just a manipulation of graphics. But Bill Gates correctly grasped that the future of the business rested on it and set about building the brand.