3 ms·

This is nothing new. This is a typical customizable chat bot with random Albert Einstein quotes. Albert Einstein: Everything is relative you know... You: Hi,

by unalone 18y ago

This is nothing new. This is a typical customizable chat bot with random Albert Einstein quotes. Albert Einstein: Everything is relative you know... You: Hi, there. Albert Einstein: Hi, I am Albert Einstein You: No, you're not. Albert Einstein: Write "Tell me something" and I will tell you something You: You are a computer. Albert Einstein: You are right. What gave me away? You: You are stupid. Albert Einstein: Oh yeah? Well, you are ugly, and I can learn. Clever, but it still needs a lot of work. Why haven't linguists ever approached this problem? I'm sure you could analyze how famous people talk, and really learn how to simulate their talking patterns, and make a bot that sounds like a specific human.

amichail 18y ago
I think what's different here is that submitting a chat rule is so simple that anyone can do it. Moreover, it's kind of fun to submit a chat rule and see how well it works out in actual chats.
- unalone 18y ago
  Yeah, but that's an old thing. A friend did this with Trillian for a group of us: it's neat at first, but it's simply preprogramming responses, even if the web site in question says it's teaching the computer to think.
  - amichail 18y ago
    Sometimes simple methods can work well. For example, Deep Blue brute-forced it with speed in chess. The idea here is to accumulate a large database of chat rules to provide convincing conversation.
    - unalone 18y ago
      Chess is far less complex than conversation, though. For conversation, the opening line alone will have trillions of possible permutations. Brute-forcing it won't work for something that incredibly complex. A more subtle solution is needed.
      - amichail 18y ago
        Humans are predictable though. That makes it tractable via brute-force.
        unalone 18y ago
        They're not so much predictable as anticipatable. You can't determine what they're going to to. You can only track the most likely conversation threads and attempt to meet them as readily as possible. The problem is, unlike in chess, there are no restrictions in a conversation. The trick with programming a really believable bot, I think, is not to brute-force it, but to create some sort of emotive simulator. Something that shifts the bot's mood as time goes on. That way, you're able to make the responses more believable, even if you don't anticipate responses to each line. The reason brute force doesn't work is because the human mind can't always anticipate things. However, we know why we don't anticipate things. We're able to tell when other people's responses are absurd and don't make sense; we can choose not to follow threads. Look at the snippet I posted above: the computer responds to everything as if it's logical conversation, which it's not. If a robot can't tell that, then the next step is to teach it how to tell. And you can't brute force that.

by unalone 18y ago

amichail 18y ago

I think what's different here is that submitting a chat rule is so simple that anyone can do it. Moreover, it's kind of fun to submit a chat rule and see how well it works out in actual chats.

unalone 18y ago
Yeah, but that's an old thing. A friend did this with Trillian for a group of us: it's neat at first, but it's simply preprogramming responses, even if the web site in question says it's teaching the computer to think.
- amichail 18y ago
  Sometimes simple methods can work well. For example, Deep Blue brute-forced it with speed in chess. The idea here is to accumulate a large database of chat rules to provide convincing conversation.
  - unalone 18y ago
    Chess is far less complex than conversation, though. For conversation, the opening line alone will have trillions of possible permutations. Brute-forcing it won't work for something that incredibly complex. A more subtle solution is needed.
    - amichail 18y ago
      Humans are predictable though. That makes it tractable via brute-force.
      - unalone 18y ago
        They're not so much predictable as anticipatable. You can't determine what they're going to to. You can only track the most likely conversation threads and attempt to meet them as readily as possible. The problem is, unlike in chess, there are no restrictions in a conversation. The trick with programming a really believable bot, I think, is not to brute-force it, but to create some sort of emotive simulator. Something that shifts the bot's mood as time goes on. That way, you're able to make the responses more believable, even if you don't anticipate responses to each line. The reason brute force doesn't work is because the human mind can't always anticipate things. However, we know why we don't anticipate things. We're able to tell when other people's responses are absurd and don't make sense; we can choose not to follow threads. Look at the snippet I posted above: the computer responds to everything as if it's logical conversation, which it's not. If a robot can't tell that, then the next step is to teach it how to tell. And you can't brute force that.