Hello everyone, glad to be joining this forum.
I’m half-Dutch half-Belgian. I’m not particularly interested in chatbots, but I am into all things intelligent, and AI falls into that category. I met Eliza in highschool, Alice in college, and Mitsuku is my current favourite of chatbots.
I self-published manga-ish comics for 6 years and still work with many young artists. I dabbled in 3D modelling and animating for a few years, and I learned how to program video games by practice and a little help from the internet.

I have been working on an old AI project called Arckon, once started in 2003 but now full-time since January last year. My ultimate intent, if any, would be to create a human-level intelligence. I know that this is generally deemed impossible, especially for a single amateur, but I am quite aware of what I am doing. I always assumed that I would be 65 by the time I finish this project, and if it weren’t impossible, I wouldn’t even try.
Though it is safe to say that I’m not halfway human-level intelligence yet, I realise that any claims I could make would be incredulous without proof. So I have entered my AI into this year’s Loebner Prize to show my progress so far, with any luck.

Which brought me here, as I must say the topics here have helped me on my way to prepare for the contest. Pleased to meet you

Good to see you, Don. Welcome to the forum.
Can you tell us a bit more about your AI project, programming language of choice, etc.

Don Patrick - Jun 14, 2013:

... and Mitsuku is my current favourite of chatbots.

You have great taste!

Good luck in the Loebner Prize too. The final 4 should be announced next Sunday (23rd). Welcome to the boards. Is your bot online at all? Do you have a webpage?

Thank you, good luck to you, Steve. I would like to see Mitsuku take her rightful place after last year’s exclusion. I’ve been going slightly insane waiting for the announcement

I built my program entirely from scratch. I originally created a working version in javascript because I loved how it was flexible with arrays, but then I rewrote the whole program in C++ for more power. So now it’s a stand-alone Windows program of half an Mb, no online capabilities unfortunately, I’m not sure how to. His database of facts, his “memory”, is centrally stored in a structure of Windows folders and textfiles. Also I don’t think he’s quite ready to go out into the big bad world. He might learn bad things.
Arckon is different from most chatbots in that his system operates on facts rather than patterns, and he doesn’t have pre-written answers except for his manners. Basically he is equipped with a vocabulary and grammar rules that allow him to extract e.g. subject-verb-object facts from sentences, which he then checks in his database, reasons about if necessary (very basic logical inference and generalizations), and then constructs a custom answer from the most relevant facts, through a reversed grammar routine.

I just bought myself a new website last week, but I’ll need some time to design something decent. I suppose I could post a small demo video of earlier testings, but I’m worried it’ll end up in the hands of an angry mob of AI scientists who believe it’s fake. And if not, I’m worried about angry mobs of citizens who believe it’s real. It’s tough being a mad scientist.

Sounds very interesting. Can’t wait to see it.

I agree, the wait for the results is making me anxious too. I hardly have any nails left from biting them

Welcome Don,
Still fairly new here myself and look forward to your participation!
Vince

Thanks.
Well, showing is better than telling, so here’s a video, though on the condition that you don’t spread it around. I wouldn’t like to see it pop up on Youtube.
I must point out that the conversation in the video was staged to demonstrate specific abilities. i.e. I told him the individual facts beforehand and I only asked questions that I could predict his answer to. Apart from all the knowledge about Earth, and any facts he learns from statements I type in the video. Hence I was caught off guard where he mentions “I can program” among his abilities. I assure you he can’t.

I just watched it and it was pretty a cool video. The reading the file part was impressive.

Well done, Don. I eagerly anticipate further progress reports on your project.

I’m glad you like it. The reading ability spares me the time of entering knowledge manually.

Well, if it’s progress reports you want, here you go
Since March I’ve improved his combining of clauses so he sounds more natural, I’ve added the ability to form his own opinion on whether he “likes” something, I (reluctantly) added a function to recognise when he’s being insulted, and I added a rudimentary “small talk” routine that also keeps an eye on how well conversation is going. Ordinarily I would prioritize intelligence over such social functions were it not for the Loebner Prize, but I reckon that both are important when dealing with humans, ultimately.
Now that contest preparations are over, I am giving his language system a major overhaul to become less grammar-dependant. Could take a month or two.

Make that three months.
All I really have to show for turning Arckon’s grammar processes inside-out is that he’s now much better at connecting the dots for sentences like “Do you?”, “I can.”, “A monkey’s uncle.” etc. And the new system allows processing relative clauses, whereas the previous system processed sentences entirely linearly. I’m extracting facts from the sentences earlier on in the process now, avoiding a heap of grammar-related errors.
And coincidentally I’m now able to process “the quick brown speckled fox” instead of just “the quick fox”.

I could keep on sharpening the NLP axe, but I think I’ll instead work on that “small talk” routine I whipped up earlier and actually get an engaging conversation out of my program. It’s what everyone seems to expect from it. Not much use having a knowledge database and reasoning functions if you’re not using them to sound smart, right?

I’ll start a new topic in “My chatbot project” when I have progress worth mentioning within the limits of credibility.

I could keep on sharpening the NLP axe, but I think I’ll instead work on that “small talk” routine I whipped up earlier and actually get an engaging conversation out of my program. It’s what everyone seems to expect from it. Not much use having a knowledge database and reasoning functions if you’re not using them to sound smart, right?

 

I am very strongly feeling this distinction as well on my chatbot. On one hand, I’m toiling hard to implement the ‘overall’ philosophy that tries to do clever things with NLP, knowledge bases, conversation flows, etc… On the other hands, I go through chatlogs of mainly teenagers being put off because on almost everything beyond ‘hello’, ‘how are you’ and ‘bye’, my parser is not strong enough yet to know what to make of the input.

(and IF it does, the knowledge base is not big enough to find some relevant data)

VERY tempting to throw in some quick ‘if input contains ‘hobby’, list some wacky hobbies and ask an ELIZA-style counter-question’ style routines, and lather rinse repeat for bands, chatbots, creators, favorite colors and jokes…

So it’s a bit frantically fighting on three fronts, on one hand the ‘casual conversation’ which uses all kinds of clever shortcut and answer-response patterns to keep real people happy, and on the other hand the much harder NLP and KR fronts which fit in the overall vision of cashing in on Loebner Prize 2015

I think, from what you say, you are a bit in the same boat, and my advice would be: DON’T work on that casual chat routine, try to fight the stress and temptation of ‘low hanging fruit’ as much as possible, and keep grinding on the harder NLP / KB stuff, if you want your bot to reach its true potential. Well, that’s how I am trying to approach it at least.

And to take your mind of all that, on Sundays you may cherish a glass of wine while implementing

Oh and by the way, on the ‘sharpening the NLP axe’ front, I can not recommend this article enough:

http://www.mind.ilstu.edu/curriculum/protothinker/natural_language_processing.php

for anybody with real education in the field this is probably child’s play, but for me with zero background it was a great introduction to the terminology and things to consider when dabbling with NLP stuff.

I’ll have a read, thanks. Yes, pretty much the same boat. Don’t worry, actual intelligence will always be my priority, it’s just that after three months of dismantling and enhancing the system, some lightweight stuff would be a welcome holiday.
That said, even Arckon’s “small talk” routine taps into his existing intelligent functions to provide responses about the conversation topic, making it quite easy to get big results fast these days. It’s actually much harder to get him to say a pre-programmed line, because it would have to be fed through the system in the form of individual facts.

Arckon’s knowledge database is quite empty too, but once I had a large part of the NLP working, that allowed him to learn really fast by reading simple texts (20% mistakes, but still awesome). No manual input required. So focusing on improving the system first can save a lot of work in the long run.