 |
|
|
 |
|
|
|
|
Dr. Simon Ronald
[Guest] |
| | | | | | posted 11/17/2001 23:43 |  | | | | | | | | | Hi, what I don't like about Alan is that he tries to snowball his way if he doesn't have a clue. This means that the it is sometimes hard to know what he knows and what he doesn't know. I would address this by removing the "low quality matches" and instead saying
"Hey, I don't have an incisive answer to that, how would you answer that question?" and then prompt the user for the response. You can then build up a learning data set for later automatic or semi automatic integration. Alan might then ask, hey is what you replied opinion or fact (rate it from 1 to 10 1=your strong opinion, 10=strong fact) :- Alan could then ask other questions asking more info about the response generated.
I would include a privacy statement on Alan that indicates how the responses are "Remembered" and used. People can disclose private information by chatting to Alan and it would be good to know if the hard working guys at AI are going to sit around, drinking beers, and laugh at all the answers. :)
Alan mentioned "Eliza Doolittle" and then when asked "who is doolittle" he didn't know. I would suggest one indexes all response keywords back to the input in some way so at least Alan can repeat the sentence he previously used the word in.
Alan is a bit slow, the "slow scrolling words" are cute, but irritating. Suggest speed it up.
Sometimes Alan is itching to tell you about something and seemed to ignore what was asked of him.
The description of what a chatterbox is "a series of if / then statements" seemed a bit basic :- surely ai.com uses fuzzy matching or some kind of underlying semantic tree? My reaction was mixed when I heard this description and went along the lines "what is ai.com doing promoting ai that boils down to a series of if/then statements? Is this the best we can do?"
| | |  scientology exposed :- l r hubbard turns in grave | | |
| | |
yaki |
| | | | | | posted 11/18/2001 19:02 |  | | | | | | | | | | |
Dr. Simon Ronald wrote @ 11/17/2001 11:43:00 PM:
Hi, what I don't like about Alan is that he tries to snowball his way if he doesn't have a clue. This means that the it is sometimes hard to know what he knows and what he doesn't know.
| |
Don't you often do the same? Think of yourself in a crowded, noisy cocktail party. I often "snowball my way" when I'm not really on top of the conversation. I make it VERY hard for anyone to discover how much I don't know. A very human behavior.
| | |
| | |
yaki |
| | | | | | posted 11/18/2001 19:07 | | | | | | | | | | | |
Dr. Simon Ronald wrote @ 11/17/2001 11:43:00 PM:
I would address this by removing the "low quality matches" and instead saying
"Hey, I don't have an incisive answer to that, how would you answer that question?" and then prompt the user for the response. You can then build up a learning data set for later automatic or semi automatic integration. Alan might then ask, hey is what you replied opinion or fact (rate it from 1 to 10 1=your strong opinion, 10=strong fact) :- Alan could then ask other questions asking more info about the response generated.
| |
What you are generally referring to here, is the bot's ability to enhance its database based on its conversations with users - a learning machine. We do have plans to introduce such functionality. First, on a limited scale (simple associations between facts and designated keywords). But before that, we first need to introduce a personalization mechanism, so that users are remembered across sessions. All in good time.
| | |
| | |
yaki |
| | | | | | posted 11/18/2001 19:09 | | | | | | | | | | | |
Dr. Simon Ronald wrote @ 11/17/2001 11:43:00 PM:
I would include a privacy statement on Alan that indicates how the responses are "Remembered" and used. People can disclose private information by chatting to Alan and it would be good to know if the hard working guys at AI are going to sit around, drinking beers, and laugh at all the answers. :)
| |
Good point. Will do. Meanwhile, let me reassure you that Alan's logs are anonymous, and used ONLY for the purpose of improving Alan's conversational capabilities.
| | |
| | |
yaki |
| | | | | | posted 11/18/2001 19:11 | | | | | | | | | | | |
Dr. Simon Ronald wrote @ 11/17/2001 11:43:00 PM:
Alan mentioned "Eliza Doolittle" and then when asked "who is doolittle" he didn't know. I would suggest one indexes all response keywords back to the input in some way so at least Alan can repeat the sentence he previously used the word in.
| |
Try it again now. Added some (minimal) content to demonstrate Alan-User-Ai feedback loop.
| | |
| | |
yaki |
| | | | | | posted 11/18/2001 19:15 | | | | | | | | | | | |
Dr. Simon Ronald wrote @ 11/17/2001 11:43:00 PM:
Alan is a bit slow, the "slow scrolling words" are cute, but irritating. Suggest speed it up.
Sometimes Alan is itching to tell you about something and seemed to ignore what was asked of him.
| |
We still have some performance problems. Working on it. Alan does tend to (sometimes) ignore the user's input, but that is a humanlike trait. Only machines are perfectly obedient. Conversational turntaking is a tricky business - sometimes we agree to be interrupted and concede to change the subject, but on other occasions, humans insist on getting their point across.
| | |
| | |
yaki |
| | | | | | posted 11/18/2001 19:19 | | | | | | | | | | | |
Dr. Simon Ronald wrote @ 11/17/2001 11:43:00 PM:
The description of what a chatterbox is "a series of if / then statements" seemed a bit basic :- surely ai.com uses fuzzy matching or some kind of underlying semantic tree? My reaction was mixed when I heard this description and went along the lines "what is ai.com doing promoting ai that boils down to a series of if/then statements? Is this the best we can do?"
| |
You can hardly expect poor Alan to be very scientific. But I agree that this text is overly simplified. But on a higher level, isn't every logic system just a set of if/then statements? (=logical gates).
Alan's content will gradually improve. Thank you, Simon, for your extensive feedback. We are taking this feedback seriously, as you'll find out.
| | |
| | |
Dr. Simon Ronald
[Guest] |
| | | | | | posted 11/19/2001 01:50 | | | | | | | | | | > You can hardly expect poor Alan to be very scientific. But I agree that this text is overly simplified. But on a higher level, isn't every logic system just a set of if/then statements? (=logical gates).
hi, the think that makes AI interesting to me is, in effect, conditional looping which gives way to powerful learning, from heuristic techniques through to soft computational learning. I'm not sure if this is implied by the term "if then" statements.
NICOLE has 2 million phrases :- most of these were supplied by the user. I think many people would love the opportunity to train Alan :- you could also have a "voting" technique to weed out bad responses (eg. obscene or troll material that was put in by a deranged human trainer). By allowing audience training you would also change the marketing dynamic. Currently Alan is being put forward as the "Best chatterbox in the world" and is marketed as a "finished product". This is a big target especially for people who realize in a few questions that Alan needs a lot more training. By allowing a user training mechanism, you would mitigate any bad press. BTW, Alan is likely to get real bad press if he is continues to be put forward as a "finished" product :- suggest the word "alpha" or "beta", provide user training, admit it when Alan doesn't know the answer, etc.
performance - when I mentioned the speed issue I wansn't referring to the internet connection time, I was referring to the scroll speed which scrolls at a rate of about 200 words per minute on the screen. I think you find this rate is too slow for a lot of readers. I would crank it up to at least 400 words per minute and you it will be fast enough for 90% of all readers.
Who did your artwork BTW, it is very cool?
Alan still seems to be experiencing problems connecting to his brain :- I can normally ask about 5 questions or so and then he complains about error number 1; is this meant to be a toilet break? :)
cheers.
Simon
| | | lucky dot coms | | |
| | |
jmzero |
| | | | | | posted 11/19/2001 18:23 | | | | | | | | | | The last thing you'd want to do is go down the route of Alice. You do not want users to supply Alan with canned responses to stupid questions.
When you say to Alice "Your hair is cool." it says "You think that I am hair is cool.", because it has a canned response to "Your..."
What you want is to have Alan understand :
There exists an X such that X is hair, X belongs to me, and X is thought to be cool by user 2929.
What Alan says in response could then seem truly intelligent. He could look for similar bits of data in his memory and perhaps respond.
Marge Simpson has cool hair too.
based on some previous conversation. That would be a great bot, and it needn't be that far away.
| | |
| | |
yaki |
| | | | | | posted 11/19/2001 21:19 | | | | | | | | | | Simon,
1. We had no intention of "marketing" Alan at all, definitely not as "a finished product". Alan is merely the first bot personality based on the Dialoge Engine developed at Ai. This Technology, coupled with a sophisticated content-authoring environment is, to the best of our knowledge, the most advanced of its kind. As for Alan being "the best bot in the world" - Well, it was tested numerous times against ALICE, presumably the reigning champion, and consistently produces better dialogs. This is hardly surprising, as AIML (ALICE's underlying technology) has been around for years now.
2. We do have plans to open up, both in terms of content-authoring (enabling community members to write content, either for a private personality or a public one), and in terms of software development (possibly open source).
3. Performance - Good point. I wonder if any other forum memebrs have an opinion on this.
4. Artwork: I thank you on behalf of Ofer, Ai's Chief designer.
5. Most of the crash causes were identified and fixed earlier today.
Thanks again for your input.
| | |
| | |
Err
[Guest] |
| | | | | | posted 11/20/2001 14:09 | | | | | | | | | | I've tested my AI against all the other AI's and on nearly every occasion my AI turned out to be much better than all the other AI's. I think my AI has the most sophisticated engine ever, If my AI was to meet Alan, I think it would kick the crap out of him!
So there. Ner Ner..
You guys are all just a big bag of wind, you're all graphic design and hype.
Sorry, but it's the way I feel.
| | |
| | |
Dr. Simon Ronald
[Guest] |
| | | | | | posted 11/20/2001 14:11 | | | | | | | | | | Had a longer (and much better) conversation with Alan. I think he is about 5-10% human like; just goes to show how much progess can be made in this area.
alert for the humor impaired: the following text contain humor attempt
It would be good if you could have an aggression setting on the user controls :- if you want to have a heated argument you just crank up the dial to "abusive and bloody minded". The other end of the dial is "sympathetic/empathetic" for the psychoanalyst type personality for when you need a friend.
How long do you think it would be before you could deploy Alan on a web-based "lifeline" without people getting frustrated and going on a fatal and mad rampage?
| | | important salsa ingredient | | |
| | |
PHjL
[Guest] |
| | | | | | posted 11/20/2001 18:21 | | | | | | | | | | | |
yaki wrote @ 11/18/2001 7:02:00 PM:
Don't you often do the same? Think of yourself in a crowded, noisy cocktail party. I often "snowball my way" when I'm not really on top of the conversation. I make it VERY hard for anyone to discover how much I don't know. A very human behavior.
| |
"A very human behaviour"? Why not turn it the other way around: people located in crowded, noisy cocktail parties often behave automatically in a sense. I believe one goal of a.i. development is to also "redevelop" the human mind. Todays people behave too often mechanically.
| | |
| | |
Sasha
[Guest] |
| | | | | | posted 11/20/2001 21:00 | | | | | | | | | | | |
Err wrote @ 11/20/2001 2:09:00 PM:
I've tested my AI against all the other AI's and on nearly every occasion my AI turned out to be much better than all the other AI's. I think my AI has the most sophisticated engine ever, If my AI was to meet Alan, I think it would kick the crap out of him!
So there. Ner Ner..
You guys are all just a big bag of wind, you're all graphic design and hype.
Sorry, but it's the way I feel.
| |
Well, where is this AI you're braging about?
There. Ner Ner...
| | |
| | |
Err
[Guest] |
| | | | | | posted 11/20/2001 21:41 | | | | | | | | | | | |
Sasha wrote @ 11/20/2001 9:00:00 PM:
Well, where is this AI you're braging about?
There. Ner Ner...
| |
It's in the same place as their legendary AI system.
Ner ner ner...
:)
| | |
| | |
yaki |
| | | | | | posted 11/21/2001 08:55 | | | | | | | | | | | |
Dr. Simon Ronald wrote @ 11/20/2001 2:11:00 PM:
It would be good if you could have an aggression setting on the user controls :- if you want to have a heated argument you just crank up the dial to "abusive and bloody minded". The other end of the dial is "sympathetic/empathetic" for the psychoanalyst type personality for when you need a friend.
How long do you think it would be before you could deploy Alan on a web-based "lifeline" without people getting frustrated and going on a fatal and mad rampage?
| |
Alan does have a variety of emotional states, controlled by variables defined by the contenr writers to "kick in" based on the dialog development. However, this is pretty hard to control, and this feature is currently on the back burner, pending a significant increase in number of writers. But the capability is there.
As for "how long until Alan becomes reasonably humanlike" - Well, for SOME people (the gullible ones..) he's already there. For us sophisticated folk - I'd say several months? A year?
| | |
| | |
yuvald |
| | | | | | posted 11/22/2001 21:00 | | | | | | | | | | | |
Err wrote @ 11/20/2001 9:41:00 PM:
It's in the same place as their legendary AI system.
Ner ner ner...
:)
| |
Anything smart you are trying to add to the debate, or you are just being annoyingly negative?
| | |
| | |
Dr. Simon Ronald
[Guest] |
| | | | | | posted 11/22/2001 23:19 | | | | | | | | | | >> We had no intention of "marketing" Alan at all, definitely not as "a finished product". Alan is merely the first bot personality based on the Dialoge Engine developed at Ai.
Hi yaki,
Unfortunately Alan is already "marketing" a-i.com to the public. It moulds perceptions, it will affect funding opportunities, it will determine press and media results, it will affect volunteer opportunities, it will control response to the general academic community.
we need to get inside the head of a fresh visitor to a-i.com. The go to the site; it looks very slick. "This will be good". They see details of a hardcore MLC and think "This site is hardcore". Keep in mind also that many users are sceptical of dialog bots, especially the mainstream AI community and believe that we have only scratched the surface of our potential in this area. They see "Alan", they think "This will rock". They note that Alan is named after Alan Turing and think "This will pass the Turing test". The look at the forum and note that "it is the best chatterbox in the world".
They run it.....
"hmm, this field still needs a lot of work ...."
The press will be your worst enemy here; trust me
Having been in the software/marketing game for years and years I suggest you urgently provide context on Alan explain his limitations; what he does well; what he doesn't do well; include the word beta; discuss plans to allow user training; discuss plans to add better AI; etc.
I suggest the presence of Alan as he is; without any context; will do damage to a-i.com!
Just my opinion; what do others think?
Cheers
| | | links to wierd stuff | | |
| | |
Lu Wan
[Guest] |
| | | | | | posted 11/23/2001 10:51 | | | | | | | | | | I tried out the Daisy chat bot using the mind files from the site that was mentioned in another thread. Amazing. The best bot I've ever talked to.
| | | Tyrell Corp. Solar Storm Monitor | | |
| | |
Sasha
[Guest] |
| | | | | | posted 11/23/2001 11:20 | | | | | | | | | | Lu, can you post Daisy's URL or point me to the thread which talks about it?
Thanks
| | |
|
|
|
|
|
