This is just my opinion and you’re invited to argue, agree, or simply to ignore it.  It’s not meant as an attack on any individual, but rather a comment on a process.  Bare with me while I first offer an analogy.

In the U.S., when election time rolls around, we see a plethora of political candidates “announce” they are running for office.  From the beginning, it’s obvious (even to the candidates themselves) that they don’t stand a chance of winning the election.  The speculation is that they are “running” for office for other reasons that largely have to do with personal recognition and monetary gain.  Often, they have a book to sell.  In other cases, keeping their name in the public eye allows them to charge higher speaking fees.  Often, they are people of little substance, and as H. L. Mencken once famously said, “Those who can—do. Those who can’t—teach.”  The result is that we have people with little or no real experience constantly lecturing us when, they, themselves have produced very little.

I can see that the “idea” of a Proof of Concept bot has its place in describing and displaying experimentation.  But it doesn’t seem to demonstrate very much about chatbots, or botmasters.  It seems more to be a way of settling arguments, or of boasting.

I enjoy seeing items related to the Proof of Concept, but what I find truly impressive is seeing a bot that can replicate, and maintain, a conversation.

The real work (and value) is achieved down on the ground, with the day-to-day grind of reviewing chat logs and meticulously making corrections and improvements to a chatbots working files.  I enjoy some of the fancy tricks botmasters are capable of doing, and I like being surprised when a bot does or says something… or performs in a way, I didn’t expect.  But, I’m disappointed when I see fabulous bots filled with known spelling errors, or who are incapable of maintaining a satisfying conversation.

The numerous additions of “Proof of Concept” bots are interesting, but they won’t win any contests, and they won’t ever get voted “Most Popular Bot”.  Show me bots who can chat.  That’s the true measure of a botmaster.

perhaps we should make a difference between a botmaster and a software developer. I know that for many, this is a mixed thing since the botmaster often wrote his own bot, or at least had to tinker with it enough to regard himself as somewhat of a programmer. But I have a suspicion that in the long run, these functions will tend to grow apart a little. Though, no doubt, the best will remain those who can do both.
Perhaps, for a person who leans more towards designing chatbots,  a ‘proof of concept’ could maybe be more about chat abilities, cool conversations,... While a developer might be more focused on the design concept, the UI, the pattern matcher,....

I know from my own experience that it took a long time of experimenting with different things before I finally started to figure out what works and what doesn’t. Proof of concepts, bad as they may be, tend to help with this.

Just saw the new edit button, looking good.

I agree with Jan. As a programmer still building their bot, for each module I build I want to test it out individually. This is a sort of proof of concept—it doesn’t necessarily speak to how the overall bot will perform, but it shows the module works as designed.

All this is very different than building a primitive complete bot and calling it a proof of concept. I think what bothers you about these endeavors, Thunder Walk, is that they seem too easy. As though their creators are implying that it is a simple step from their proof of concept to an award winning bot. I don’t know that this is what the creators truly intend. I think this type of bot mostly serves to highlight which parts of bot development are straight forward (the interface perhaps, or text to speech, or whatever they’ve included) and which parts require time and subtlety.

In my case, I’m not trying to create a new bot, but a new bot interface; one that you can actually talk to, rather than typing and reading. I’m not quite ready to show my work yet, but I’m getting close. Once I get it functioning, and have had the chance to test it in both XP and 7, then I’ll upload it to my web server, and post a link. It should be within the week, and maybe by tomorrow. I’ll keep you posted. Be warned, though, it won’t be a bot when I first post it. You’ll understand when you run it.

Thunder Walk - May 28, 2011:

This is just my opinion and you’re invited to argue, agree, or simply to ignore it.

I may friendly argue, with you Thunder Walk, regarding a POC. Yet, I also value and thank you for your opinion.  For proof of concept pilot projects, I find the K.I.S.S. Method from the NASA Jet Propulsion Laboratory  essential for attempting to prove newly written plans are feasible.

Thanks to everyone who’s responded so far.

I get that there is a difference between developers and mere mortals. Absolutely, the Proof of Concept demonstration bot has its place, if for no other reason than to create interest, and to resolve questions surrounding the possibility of a technique or method.

In a different forum, there were frequent questions concerning the feasibility of a “forum bot”. Was it possible? How was it possible? The question hung in the air for a long time. The forum was filled with chatbot and code experts, but no one had any idea.

Then, one day, someone joined the forum and demonstrated what no one else could do. But, while it functioned, it was totally useless. It worked, but it served no purpose. It was a bot without a mission, or an audience. I was impressed with the fact that someone was smart enough to accomplish what no one else seemed to be able to do, but I was disappointed to see how it was applied. It was a bot from a developer’s point of view, and without any consideration given to the user. It was like having Albert Einstein who couldn’t tie his shoes.

It seems to me, that if you wanted to build a new car, you’d want to test drive it. You’d want to know how it performed under different conditions, with different drivers, and you wouldn’t merely be satisfied to see if the engine started when you turned the key once or twice. And you’d want to see how it performed with real, average consumers, not professional test drivers on a closed course.

Show me one or more bots (provide the URLs) I can actually chat with. Mitsuku, Morti, and ALICE bots don’t count.

Thunder Walk - May 29, 2011:

I get that there is a difference between developers and mere mortals. Absolutely, the Proof of Concept demonstration bot has its place, if for no other reason than to create interest, and to resolve questions surrounding the possibility of a technique or method.

It seems to me, that if you wanted to build a new car, you’d want to test drive it. You’d want to know how it performed under different conditions, with different drivers, and you wouldn’t merely be satisfied to see if the engine started when you turned the key once or twice. And you’d want to see how it performed with real, average consumers, not professional test drivers on a closed course.

If you were building a new car, most of the components would be tested separately on test rigs before they were integrated into a new car. Then, you may very well use professional drivers to test if the whole system works as predicted. Only after lots of testing would you let it into the hands of an unprofessional driver not on a closed course.

With bots, it is sometimes easier to test just 1 function at a time without worrying about interactions with the rest of the system.

Thunder Walk - May 29, 2011:

Show me one or more bots (provide the URLs) I can actually chat with. Mitsuku, Morti, and ALICE bots don’t count.

You can always try Skynet-AI. If you use Internet Explorer (and have MS Agent Installed), Skynet has the ability to speak the responses to you.

In the past I have used a proof of concept to test things like memory systems, learning and debug functions.

Off-line I currently have a proof of concept to try to answer general word math problems. I collected a variety of math problems and have been exploring techniques to be able to create a general solver. I will add that capability to the next on-line version of Skynet-AI. Right now the bot can recognize a variety of problem types (some it can solve and some it can’t). There are actually 3 factors that need to be worked out.
1. How should the input be entered into a chat dialog? Special math mode or free form speech?
2. How does the AI recognize a math problem and then identify the problem type.
3. How does the AI then solve the specific problem type.
If I were to try this on a full bot I think it would be much more difficult to test and verify results.

I did add a couple of the simple cases to the current on-line version.
Skynet currently handles problems like the following example:
If you have two apples and add two more how many do you have total?

I think there is a point to be made here.

Proof of Concept samples usually demonstrate a certain ability of a bot. They show how the bot figured out what you said.  They demonstrate that the bot can do math story problems.  Or they prove that conversing with a bot can be natural as talking to somebody in the room. And so on.

Here’s a comment from another discussion group about a good chat bot in competition:

they know where they live and how they travelled to the contest site; they have an occupation of some sort; they have opinions about other people and current events; and they’re interested in finding out about the judge as a person. What they *don’t* have is an encyclopaedic knowledge of the world.

I have yet to hear of any proof of concept showing these kinds of features.  No one portrays typical, ordinary life experiences, the common things which find a way into good conversations.  Instead they want to prove how smart their creation has become.  Or how well it diverts obscene language. Or how well it can interpret garbage inputs and stay on topic.  Or how it can join an on-line forum. Or a bunch of other little steps which seem like window dressing, like chrome to make the chat bot look cool.

Just because a bot remembers you told it you have a dog, it doesn’t mean the bot has a life! Indeed, unless it is supposed to be as obscure as a Rogerian doctor, that tactic makes a very poor chat.  And you can only listen to a sales pitch for so long before that diatribe becomes old.  In fact, personality is drab without a plot.

I wonder if even a good story-teller bot would be a worthwhile proof of concept. Is creativity the same as chatting? Can a bot “create” a virtual life full of opinions and experiences and daily participation in the local culture?  That would be some proof of concept (depending on how interactive it is, of course.)

For the sake of discussion, I’d like to narrow this down a bit.

What I sometimes see… what’s difficult to endure… is any developer stuck in the development stage.  They produce lots of new bots that can perform a “trick” or bots that go about their botting business in some unique way.  At the same time, they can’t point to a single bot that can participate in a conversation.  We all know that a bot is never finished, and working on a chatbot is probably a lifetime experience.  But, the kind of developer I’m talking about has never been able offer up a single bot that can go beyond the two or three lines they’ve been programmed to answer.

When you acquire an ALICE/Pandorabot, off the shelf, it needs some work.  There are spelling corrections needed, there’s a lot of dated material, there are some patterns that are just wrong because of their code or content.  But, off the shelf, you can chat with it.  I think if you design a new car it’s meant to be driven, if you create a new airplane it’s meant to fly, if you developed a new chat bot… well, you get the idea.

I’d like to see a newly developed Proof of Concept bot that could chat with me.  It doesn’t even have to be as good as ALICE.  But, I don’t understand building a house, and then not being able to proceed beyond the front door.  I think developers are good at what they do, and I’d like to see more.  I’d also like to see what else they can offer.  As far as I know, you can chat with any bot that’s entered into the Chatterbox Challenge, or the Loebner Prize Competition.  Why would a developer want to keep their bot a secret… if one really exists?

Thunder Walk - May 29, 2011:

Show me one or more bots (provide the URLs) I can actually chat with. Mitsuku, Morti, and ALICE bots don’t count.

@ http://79b.org I made a bot menu to four bots you can chat with.

So you think all bot demonstrations should be loaded with tens of thousands of records of data by default?  Well, maybe they should.  If a bot can handle a dozen test data records, why not just plug in 50,000 while you at it?  Food for thought.  In any case, that is good feedback.  Thanks for the teamwork, Thunder.

The question of keeping a bot secret is quite obvious, if one really thinks about it.  Being criticized doesn’t even have to play into that scenario. Just observe how many here demand to see such a thing.

I don’t hear a bunch on this forum bragging about their best friends.  That’s just not so cool.

“Bots are things, like slaves, to be paraded around like shiny new cars.  Show off your bot!  It won’t care.”

Only what you really want to see is a bot that does care (not a monster Frankenstein, that is)...

Not all bots are actors showing off on the internet or just proof their authors are great.  A bot can be more like a private diary than like a published novel.

Thunder Walk - May 28, 2011:

The real work (and value) is achieved down on the ground, with the day-to-day grind of reviewing chat logs and meticulously making corrections and improvements to a chatbots working files.

The real work is making it easy to do that, as you’re talking to the bot, and eventually to get the bot to do it by itself. Then each user can have his own customized bot and “most popular” becomes irrelevant :)

8PLA • NET - May 29, 2011:

Thunder Walk - May 29, 2011:

Show me one or more bots (provide the URLs) I can actually chat with. Mitsuku, Morti, and ALICE bots don’t count.

@ http://79b.org I made a bot menu to four bots you can chat with.

So you think all bot demonstrations should be loaded with tens of thousands of records of data by default?  Well, maybe they should.  If a bot can handle a dozen test data records, why not just plug in 50,000 while you at it?  Food for thought.  In any case, that is good feedback.  Thanks for the teamwork, Thunder.

None of those bots seem like the excellent work you’re known for.  They seem more like the kind of things you can quickly acquire from a web site.  Geez, I expected more.  Mizuki is just an ALICE/Pandorabot.

As for how much data a Proof of Concept bot should have, I’ve never really thought of an amount, just that it might resemble a chatbot, and that replies might say something other than, “I have no answer for that.”

By the way, I would never be a part of any team who would have me as a member. 

@ Gary Dubuque and Robert Mitchell.

The question of keeping a bot secret is quite obvious… if any bot actually exists.

There are a couple of members of this forum I truly respect. I’ve known them (as well as you can know someone online) for a long while. I know them best through their work, and their involvement with other forums where they answer questions, give advice, and help people like me. They’re confident in what they know and do, and they know who they are. You can probably guess their names.

Occasionally, I’ve also come to know some “pretenders”. Not many, just a handful. They’re extremely intelligent, and they can solve the chatbot riddles that puzzle even those who really know this stuff. One way you might recognise such people is that they are name-droppers, or they quote obscure documents from MIT or NASA. But, somehow they can never show you their chatbot other than something like an ALICE clone. They can’t point to a chatbot they’ve worked on for years, a bot they’ve put a lot of sweat, effort, and thought into, they way you can tell someone has with Mitsuku, Morti, or ALICE.

Thunder Walk - May 30, 2011:

Mizuki is just an ALICE/Pandorabot.

Mizuki, is a RiveScript clone based on the world famous Mitsuku.  Imitation is the sincerest form of flattery.  You know, I may want to go shopping at the Mitsuku store: http://www.cafepress.co.uk/mitsuku now that the weather is getting nice.

Mizuki, I don’t think can be classified as a Pandorabot, she is running Perl/CGI on 79b.org servers.  I think ALICE/Pandorabot runs Lisp on their servers. Awesome!  You just suggested that Misuki has brought RiveScript to the level of Pandorabot.  Thanks for the compliment.

Beyond that, building a bot from scratch for the sheer enjoyment of it, is fun.  There are several ways to do it.  So, it is fun to try to do the same thing in different ways.  One problem with building advanced chatbots, is what do you do when they are finished?