Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Zanzetkuken The Great]

Just came across this interesting article.  I recognize that as far as we can tell is pretty much only a very advanced chatbot, but it does make me wonder two things.  Just what is the current leading edge of Artificial Intelligence that is publically known, and if this chatbot program was released, could someone be able to import the recent posts of a bunch of threads to create one with the personality of the Bay 12 Forums itself?  Maybe see if we could put it through a Myers-Briggs and a political spectrum test to see what we get.

[Sergarr]

I think there was a thread here where people created chatbots based on forumites, actually.

[Zanzetkuken The Great]

I think there was a thread here where people created chatbots based on forumites, actually.

I seem to recall this, but this one seems to be considered as making more accurate representations, and I'm not referring to one forumite.  I'm talking about every post by everyone over a day or two being imported into the thing and seeing what happens.

[wierd]

That was a markov chain generator, which is not terribly advanced as an AI.

It basically just looks for statistical connections between what word follows other words, and strings them together. Sometimes the output makes strange sense. Most of the time it doesn't.

Most chatbots are just modifications on such systems with some additional rules to help constrain the output as something more cogent.


more advanced AIs 'try' to process inputs through a 'model', so that the program can determine some kind of 'meaning'. (instead of just playing 'chinese room')

These are things like google's DeepMind ai, which takes an input, and runs it through a trained neural net 'model', where it attempts to find 'meaning' in the source image, and produce an 'enhanced' output.

It would be interesting to build a chatbot on many hundreds of layers deep of such neural net models, so that the AI understands what verbs, nouns, etc are, what a proper sentence is, etc, and tries to build a meaningful sentence as best it can. It would be quite computationally expensive though.

[Sergarr]

AI with coherent meaningful speech requires a certain rewarding agenda that said AI wants to achieve with its speech. Like, for example, getting maximum number of likes on its facebook posts.

[Starver]

Can you elaborate on that?

Why do you think there was a thread here where people created chatbots based on forumites, actually?

Do you feel happy talking about every post by everyone over a day or two being imported into the thing and seeing what happens?

Tell me more about the markov chain generator, which is not terribly advanced as an AI.

Why do you like, for example, getting maximum number of likes on its facebook posts?

[alway]

If you want to generate an algorithm that generates statements that would, say, get the most likes on Facebook, you'd best start by analyzing current things on Facebook. At which point you will run into issues. Namely, that someone's fame makes a significant difference as to the number of likes they get. And that someone's fame changes over time. You're dealing with too many variables at once in your samples alone that it would be nearly impossible to develop a meaningful sample of high-like-inducing Facebook posts on the levels needed to actually get a proper AI going. You can't simply normalize over all of the user's posts; you need to take into account time, popularity, etc. Too many variables that lie squarely outside of Facebook.

You say that, but this account with nearly a quarter million followers would beg to differ. :V
Warning: nsfw from time to time; images chosen by AI.
https://twitter.com/archillect

[RoseHeart]

AI with coherent meaningful speech requires a certain rewarding agenda that said AI wants to achieve with its speech. Like, for example, getting maximum number of likes on its facebook posts.

Well put.

[Starver]

Is it because of its posts, or the novelty of an AI on Twitter?

Facebook is 13 years old¹. I imagine that it's been done many, many times.

If not from scratch, then as an adaption of the multiple prior AI agents. My last post was parodying Eliza, which I once had running on a BBC Microcomputer, but also saw actually fooling people² on IRC in the early '90s. A quarter of a century after its first teletype-based 'fooling' of people was done, and two decades after the famous first Eliza/Parry³ interchange over the fledgling proto-Internet. Can't find a transcript of that, but this one's from slightly later. (I've seen worse Human-to-Human conversations!)


¹ Pretty much to the day! Although who knows when the first official/unofficial API connection was established to automate account updates without any further human input/conscious guidance at all.

² Who hadn't seen the innermost workings and thus not been so jaded by the 'obvious tricks' such as subject/object swapping, tense morphing and changing a statement clause into a viable questioning form without having to do much work on parsing the precise semmantics being used.

³ Artificially 'Intelligent' Paranoid Schizophrenic simulator.

[alway]

Is it because of its posts, or the novelty of an AI on Twitter?

Judging by the fact that it's somewhat non-obvious that it's an AI aside from the post timestamps... and the frequency with which people actually respond to it (often asking a question), as well as the fact that I wouldn't pull it up any more or less if it were a human, I would say the former. The reposted material has a certain style to it, and if anything, it's the novelty of that style which is interesting.

[Starver]

Honestly, I was going to ask if you were intentionally trying to sound like a chatbot.

Why do you say that?

[Starver]

Something just felt... off about the way you were speaking.

What makes you believe something just felt... off about the way I was speaking?

[Reelya]

If you want to generate an algorithm that generates statements that would, say, get the most likes on Facebook, you'd best start by analyzing current things on Facebook. At which point you will run into issues. Namely, that someone's fame makes a significant difference as to the number of likes they get. And that someone's fame changes over time. You're dealing with too many variables at once in your samples alone that it would be nearly impossible to develop a meaningful sample of high-like-inducing Facebook posts on the levels needed to actually get a proper AI going. You can't simply normalize over all of the user's posts; you need to take into account time, popularity, etc. Too many variables that lie squarely outside of Facebook.

Actually, there's more you can glean from just Facebook. e.g. the bot can look at patterns on engagement and the type of posts, amount of likes, exactly who is doing the liking etc. A famous person in Facebook who posts things will actually have measurable differences in their network compared to the non-famous.

Basically by analyzing everyone's friend networks and the types of posts that get liked, and how that gets spread through the network etc. e.g. the bot might discover that reposting news articles gets likes, and could in fact glean what sort of news articles it's existing friends also like already, etc, and as well as that, it could look at post frequency. You might get less likes if you posted e.g. every 10 seconds. But the bot could in fact do something really clever: maintain lists of friends, work out which posts they might like the most, then use view filters such that only subsets of it's friends are seeing each post it makes. It could post much more volume then without risking overload of individual users, thus gathering maximum data.

[Reelya]

No it's not. Social media researchers can glean a whole lot about connections without having access to Facebooks private data. You just need a few "follows" and before long you have a huge graph of connection data to keep track of.

[Reelya]

Fake posts to get likes, friends, follows etc.