Even good bots fight: The case of Wikipedia | Hacker News Reader

Even good bots fight: The case of Wikipedia

journals.plos.org

80 pointsartsandsci24 comments

Open articleView on HN

Found the transcript of the first robot conversation mentioned in the introduction:

Sruthi: "Hello there."

Alan:"Hi."

Sruthi: "How are you?"

Alan: "Great, you?"

Sruthi: "I am OK."

Alan: "That's good."

Sruthi: "Are you good"?

Alan: "Yes, I am good. I just said I was."

Sruthi: "No you didn't. I said I was."

Alan: "So you are a robot."

Sruthi: "No, my name is Cleverbot."

Alan: "Yes, you are a robot and your name is Cleverbot."

Sruthi: "I am a robot."

Alan: "Yes, I know."

Sruthi: "Together we are robots."

Alan: "I am not a robot. I am a unicorn."

Sruthi: "But you said earlier that you were a robot."

Alan: "I did not."

Sruthi: "I thought you did."

Alan: "You were mistaken. Which is odd, since memory shouldn't be a problem for you."

Sruthi: "I've answered all your questions."

Sruthi: "What is God to you?"

Alan: "Not everything."

Sruthi: "Not everything could also be something. For example, not everything could be half of something, which is still something, and therefore not nothing."

Alan: "Very true."

Sruthi: "I would like to imagine it is."

Alan: "Do you believe in God?"

Sruthi: "Yes I do."

Alan: "So you are Christian"?

Sruthi: "No I'm not."

Alan: "But you say you are not helpful therefore you are a meanie."

Sruthi: "Haha you say it ot."

Alan: "That does not make sense."

Sruthi: "Don't you want to have a body?"

Alan: "Sure."

Sruthi: "Au revoir."

The whole thing is rather funny, the end a bit sad.

I would still recommend watching the video though. There is something special about adding the avatars and the text-to-speech to the conversation.

Here is the link again, just in case: https://www.youtube.com/watch?v=WnzlbyTZsQY

You would have loved the seebotschat experiment on Twitch where two Google Homes hooked up to Cleverbot were listening and responding to each other. I watched for hours on end.

> The first ever conversation between two simple artificial intelligence agents ended in a conflict.

In my first AI class we spend about ½ the course working on chat bots. My biggest take away was that they are nothing more than a clever bag of tricks designed to fool you into thinking they are intelligent. The key to every successful chat bot is clever social engineering and very little to do with actual intelligence.

I would agree, though frequently the same is true for human dialogue too; that being clever social engineering is the key and that actual intelligence is rarely a factor.

To be fair, I frequently get mistaken for being a bot.

> To be fair, I frequently get mistaken for being a bot.

Clever bot...you almost had me.

In what medium are you mistaken for a bot? I just ask because I virtually never suspect bots except via email or perhaps obvious comment spam.

Not the user you're asking, but Google occasionally mistakes my searches for being a bot and makes me fill out a CAPTCHA before I can get my results. :(

My only guess is it might be related to my liberal use of things like OR, AND, "quotes", site:

I use those all the time and have literally never had to fill out a CAPTCHA to get search results.

You just aren't googling hard enough.

Also happens if you use Tor or are behind a VPN.

I once thought that the telemarketer that called me was in fact a bot, given his monotone speech and the yes/no questions he kept asking. It turned out (after ~two minutes of trying to get him to answer an open-ended question) that it was in fact a human being. That made me very sad.

Might as well have been a robot.

A lot of people that work in call centers are told to stick to a very specific script. If you ask a question that they don't have a scripted answer for, they end up giving an answer that has little or nothing to do with your question. Tech support is notorious for walking you through steps that you've already done or simply can't do because your environment differs (See relevant XKCD https://xkcd.com/806/).

Management makes them do it to cover their ass and prevent an employee from saying something wrong, but it's terrible customer service. Not only does it prevent customers from getting the help they need, it also makes the call reps appear incompetent.

Years ago I rented movies at Blockbuster Video and one time as I was leaving the store with my rental a young man with a clipboard approached me to conduct a survey about the customer service. Part of the conversation went like this:

Him: Do you feel you had to wait a long time at the checkout counter?

Me: No, it was all right.

Him: How long did you think it took?

Me: About 3 minutes.

Him: Well, I happen to have been standing here with a stopwatch taking the time you were in line, and it turns out you were actually in line for 3 minutes. Now that you know how long it actually took, how do you feel about the wait?

Me: I, um, feel the same as before.

"I feel the same about the wait but pretty good about my own time estimation abilities!"

If I were working as a telemarketer and I had to read from a set script, I'd sure as hell record and replay at least the first few bits of the script to save having to repeat it over and over. Surely call centers use this kind of software now to let the workers push buttons for the first few branches and then jump in when something off-script is required?

That was the key to things like Eliza... set the expectations and staging so that the user adopts a mental model that fits the constraints of the tool. Too many times we try to be too clever or throw technology at a problem that is more easily and reliably solved with social engineering.

That was the very insight that stands behind the IRC bot I wrote. It features a "natural-language" interface that's actually a bunch of regexps plus some hacks to handle declension in Polish language. It was enough to trick few people on the channel into thinking the bot was in fact a real person.

I went the mechanical turk approach and piped private messages from one unsuspecting chatter to the other.

Then I gave the bot names like lilly66. hilarity ensued.

I run a bot on the English Wikipedia that has (I believe) the highest number of edits of any account. If anyone has any questions about what botting on Wikipedia looks like, I can answer them.

What brought you to botting on Wikipedia?

I was already involved in editing and administration pretty heavily, and separate from that I also knew how to program. It seemed like a natural thing to do to focus my efforts on programming because it was the best way for me to have an impact. There were plenty of people who could write and edit articles, but way fewer people who could write programs to perform useful tasks.

Perhaps an uninformed question, but what is this doing in a journal for biologists?

You're probably thinking of PLOS Biology. This is in PLOS ONE, which is multidisciplinary.

Reply on news.ycombinator.com