[KnyteTrypper]

Lennart Lopin has just released (6/06) a new version of his Prelude chatbot. The new Prelude@# standard edition now uses the Microsoft .NET framework. It features voice recognition, TTS, and MSagent compatiblity (Guile3D's Melissa is included in the download). Several impressive new memory files are also available. Lennart has produced .mdu files for both the 2002 standard ALICE AIML (@17,500 entries), the Annotated ALICE AIML (@19,000 entries), and a massive 180,000 entry file containing 3 years of daily chats. This bot began as a rather simple and unnotable effort, a few years ago. But it's now on the verge of becoming a major entry in the field of "self-learning" chatbots.

Prelude homepage

[FuzzieDice]

Sounds interesting. Voice recognition? Or is it Speech recognition? There's a difference. Voice recognition is when the AI can distinguish between people by the sound of their voice. Speech recognition only can recognize speech/spoken words but can not know who is talking by the sound of the voice speaking to it (but Voice Recognition can).

Would be nice to have a chatbot with Voice Recognition instead of just Speech Recognition. Then maybe it could also be one up on HAL in that it would know it's own voice speaking vs. the person speaking to it.

[Art]

While you are correct in a manner of speaking, the industry at large, still uses the two interchangably, especially Dragon Naturally Speaking.

Several companies claim voice recognition as voice authentication or what we referred to in the day as being Speaker Dependent (only recognizing one person's voice). Yet, it was not really recognizing the voice as being John or Vicky or whom ever, but merely responded to a particular person's voice.

Biometrics, can identify a person from a fingerprint, retina scan, handprint, facial recognition or voice print verification, etc. Biometrics is nothing more that a program doing a great job at pattern matching be it face, hand, finger, voice, etc. It's what they do and it's big money in the industry. Notice ads for laptops that include a "biometric fingerprint scanner" for privacy. Big deal, I've been using one for years to log on, pay bills, etc., without anyone being able to see or hack a password from me. It does have merits and it does work flawlessly. Now suddenly, it's the new buzzword!

I agree that the ability for a computer to recognize it's user from a matching vioce pattern would be very cool!

My friend has a Voice rec program (Speech rec) and we have been around each other so much over the years that we almost sound alike...at least his computer can't tell the difference!!

All in good time, i fear...all in good time!

[FuzzieDice]

That's the problem. When buzzwords bet blurred in the media and/or advertising so it's hard to tell what it really is. What will I need to look for if I want true voice-distinction? What'll be the buzzword? Maybe by the time it becomes more prevailent in software applications, it will be named something else to which I'll have no clue what the software will do.

Wish they would keep a few things "standard" so one would know what they are trying to sell/create. :/

Gets confusing otherwise.

[Art]

Yes Fuzzie, the "nice" thing about "Standards"...there's so many to choose from!! :coolsmiley

[FuzzieDice]

Art - LOL! Good one.

[Art]

Hello botters!

I have been experimenting with the Prelude bot. After quite a while of editing the mind file to get rid of extreme curse words, German, French and other languages, phrases and many duplicated sentences, I now have a much smaller file to work with.

It seems to show some promise and I'm sure over time (time spent chatting with the program), it just might surprise me.

There are a lot of limitations compared to some other bots I've used in the past.

This one does not have any current provisions for user name, gender, time / date, topic tracking, age and the ability to readily change TTS settings. Voices are, however, changed when the user switches MS Agent characters as a lot of them have / use their own individual voices. The program does not use or have a provision for anything other than MS Agents. The majority of them performed well.

The program did "lock up" on me a few times and I was not able to communicate this to the author because my registration email had not yet been verified (nor has it been today).

The online forum has a fair beginning of topics, questions and discussions and I feel that as more people try this bot, the potential of modifications and corrections will show forth. This could be an interesting developmental bot.

This is just my take after a few hours of experimenting.
Your actual mileage may vary.... :afro

[Art]

I have been conducting some serious editing of the Alice Mind file that was available for download on the Prelude site. It was originally filled with some obsenity, duplicates, unwanted lines / spaces and some just plain nonsense.

The file size was about 1460k when I started and at the end of three days I had it down to 650k without loosing any pertinent data.

I then ran the Prelude bot using the modified Alice file and it ran without exception! The conversation was actually quite interesting and yes, Virginia, it does "learn" or remember what was previously discussed.

Quite an interesting bot and more is promised in the way of it's development. :afro

[KnyteTrypper]

That sounds like an interesting project, Art. Which of the AIML sets did you edit? The 2002 ALICE AIML is pretty much a done deal. It was the original Loebner winner set. But the Annotated ALICE AIML set is the one currently in development. It contains most of what's worthwhile from the 2002 set, plus what developments have occured (and continue to occur) since. I, for one, would be interested in having a copy of your edited file if you'd like to distribute it.

I've said all along that I thought AIML provided the best information storage and retrieval system available to date for building a bot "brain," but needed a good learning algorithm combined with it to advance significantly toward "real" A.I. I sort of doubt Prelude will be the one, lol, but it definitely seems to be a step in the right direction.

[Art]

KT,

There wasn't much difference between the 1938k and the 1750k files so I took the smaller one.
As mentioned, there was a lot of RANDOM SEVENS, FIVES, TWOS, etc. as well as a lot of foreign
language responses (French, Italian, German, etc.), and these were likewise removed. The site
mentioned that Prelude came in 3rd using the Alice brain so whether or not the brain I downloaded
was one of those, I can't say with any certainty but it has been an interesting experiment.

One main concern was the numerous insertions from a variety of conversational chats with people
from all walks of life. Not that I have anything against people, it's just their conversations, language
usage (of lack of), spelling errors, grammar and general garbage, led me to this project in the first
place.

Any bot that can learn from online chats runs the same risk of obtaining garbage, profanity and a
multitude of unwanted chatter that ultimately ends up in it's brain file.

How then would such a bot record only the pertinent information? How indeed! That would be the
million dollar question. It simply must be regularly edited by the botmaster in order to maintain the
type of integrity that is desired.

I did find one slight problematic area. After chatting, I'd review the brain file and find that my responses
were not inserted in the right position, as in an inserted line, but was appended to the end of a previous
line entry in the brain file. I edited the placement of these as well. This may or may not be a problem
as far as the performance or recall of knowledge is concerned but I will be watching for it in subsequent
chats until I either figure it out or realize that it doesn't cause a problem.

Though it's not "ready for prime time", I certainly wouldn't have a problem seeing that you get a copy to
experiment with.

I'll be away after today for a week so perhaps we can discuss it when I return.

All the best!

[KnyteTrypper]

I bet it's been an educational experience for you, and one which helps you appreciate the differences between a bot you chat with for your own pleasure and one you make for public use - a point I've always tried to make about Alicebots. When creating a public use bot, it's really not a disadvantage to have the botmaster function as a part of the program. It's a relatively simple matter for a person to filter and sort the input, but (at this point in the evolution of A.I.) a near-infinitely complex task to try to design a digital ruleset to perform the same function. Even just creating an artificial mind that properly understands context has so far proven to be beyond our skill level. Being an optimist, I don't think it's beyond our capabilities, just beyond our current capabilities. But also being a realist, I suspect it may take us a while to get there.

[FuzzieDice]

"our"? I feel there may be a slight flaw in that assumption. In that while it might be beyond some people's capabilities, it might not be beyond other's. Because there isn't one "out there" yet is no indication of capability. It just means that someone hasn't set down and taken the time to make one yet. Some may not be capable of it. Some may be, but other projects or work may be taking priority. Or, they already have, but they or those above them decided it's not a good idea to introduce it to the public, not because the AI can't handle it, but maybe because they feel the public won't be able to handle it (ie. attempts at reverse engineering, steeling of their code they worked hard on, or other things).

I too hold hope that the general public will see more intelligent bots and maybe an AI in the future. The way technology progesses, I wouldn't be surprised if it ends up happening within the next 25 - 50 years.

But to be fair, I don't feel one can judge human capability by availability. You never know what talents an individual has until they express those talents.

[dan]

I think it may have happened, but just not for us to know about.  If the military is able to run a war from a laptop in Florida it seems plausible that somewhere there's a supercomputer which has programming which could do, well, you get the picture...and I wouldn't say 25-50 yrs for the public, more like 5-10 years before we get some pretty cool stuff, but yeah ai doesn't seem like a public thing yet.  "Just think of all the lost jobs" was a complaint of robots in the 70's and 80's when manufacturing went to robots.  We'll all be programming the dang things.  So I guess I better sharpen my programming skills if I want to remain marketable. :crazy2

[FuzzieDice]

This is another reason I'm studying the latest .NET and C++ programming.

Another thing I noticed is, many of the general public don't seem to want computers to talk to them. I've told many about my hopes to make my car talk only to hear "oh, I wouldn't want my car to talk back to me!" Or those that say they rather talk to a HUMAN on the phone and not the computer answering system. There I can see, unless they can make the systems more intelligent. But I still think it's going to ultimately be up to whether or not the "majority" of people will even want to deal with it.

Personally, I'm very much into technology. I have been ever since I was a kid wishing I had my own robot and soldering together AM radios and VU Meters at age 7! But not everyone started out that way.

It's also odd that even those that have been "raised" in the computer era with home computers and computers in their classrooms don't mind the ipods, the extra features in the cars, and such. But when a computer talks to them, then they seem to mind (unless it's a video game).

How odd.

[Art]

KT / all,

I have to agree with you. Given the vast power of information that is wielded and available on the net and given the fact that most can't wait to show what they've got, I believe if there was such a program / bot that could actually understand the actual context and content of the typed or spoken word, that feat, in and of itself, would put the program's creator(s) at the top of the tech mountain and their 15 minutes of fame would be assured!!

I've seen a lot of university experiments in AI, robotics, reasoning, virtual emotions, etc., but all have fallen short of the common goal of giving an AI, "understanding" instead of emulation or imitational intellect.

Also in light of FD's comments, this is not to say that it can't be done, just to voice an opinion that I don't think any rational AI's are currently "out there" at this time. If they were, we'd most likely have heard about it by now.

Personally, I can't wait to confront a smart, artificial entity.

Whether in my lifetime or not, it will come....