I’m an AI person rather than a chatbot person, and I joined this forum partly because outside of this chatbots forum, it’s “AI Winter” out there, so there are very few intelligent discussions going on about AI. So if I start posting original threads in the “AI Thoughts” section, what kind of posts should I make? What is appropriate for that section or for this forum regarding AI topics? Should I lean my topics toward natural language topics that chatbot people would want to read, or should I just post anything about AI? What would you all like to talk about regarding AI? Should I start a thread on my own AI work that I update when I have interesting things to report? Have other people started their own threads on their own chatbot work? I can get pretty spacy, in a Kurzweil/Penrose/Moravec/Hawkins sort of way, talking about artificial life, the singularity, and so on, so I don’t want to go too overboard… unless people enjoy that kind of thing as a kind of humor or something.

Mark, please feel free to post just about anything that’s near to your heart, with regard to AI. Just because our primary focus is on chatbots, doesn’t mean that we don’t also appreciate other forms of Artificial Intelligence, as well. If you want to give your thoughts on NLP, or Neural Nets, or the coming of the “singularity”, or even AI in SciFi, the field of discussion is wide open. We have a rather eclectic group of individuals here, with an exceedingly wide range of tastes, so don’t feel as if you need to hold back in any way, shape, form or fashion. Just be prepared to back up your opinions is all I ask.

As to starting a thread to discuss your current AI project(s), whether they’re chatbots or not, I say Go for it! I’m 100% certain that I’m not the only one interested, and you MAY find someone with the skills, experience and interest to be able to help move your project forward in ways you may not have even imagined. I know, because it’s happened here before.

And listen, don’t fret about “spacey” here, either. Most of us could use a chuckle now and again (some of us more than others! ) so as is said in “The Good Book”, Let not your heart be troubled. And to paraphrase myself, “Let ‘er RIP!”

it’s “AI Winter” out there

There are lots of competitions happening over at Kaggle‘s. But it is usually focused round something very specific. It’s great for testing out your own machine learning algorithms. Though most people appear to be using something like SPSS or python with the scikit package.
There are also a lot of AI related tweets, but you need to filter a lot and find the right people.

Mark Atkins - Oct 13, 2012:

[...] Should I start a thread on my own AI work that I update when I have interesting things to report? [...]

Yes indeed, please do give us an initial description of your work, and updates on its progress. Your profile on this prestigious Chatbots.org website says that you have knowledge of the “VPython, Java, C++, C, LISP, PROLOG, CLIPS, JESS, JavaScript, PHP, HTML, MATLAB, Scilab, R, Ada, Pascal, BASIC, FORTRAN, ALGOL, assembly language, Smalltalk, ATLAS, SQL, MySQL” programming languages—most impressive! Since you include JavaScript, please be advised that

http://www.scn.org/~mentifex/Dushka.html

in JavaScript for MSIE is a Russian artificial intelligence that I have been working on for almost a year now. If chatbotters are going to post about their AI-related projects, it would be especially interesting to read discussions of their AI progress over time. The Dushka Russian AI is pertinent for demonstrations of AI performance because, even if you do not know Russian, you can click with Internet Explorer on the link and and see a complex panopticon of how the intelligent machine thinks both “super rosam” and “sub rosa.” That is, the Russian AI provides interlinear status-reports on what thought-processes are going on behind the scenes, such as “fetching an auditory engram” or “activating a known Russian verb-form” or using the VerbGen() module to generate a needed Russian verb-form, or running the ReJuvenate mind-module to clear out the oldest memories and make way for new memories. This Russian AI is getting visits not only from Russia and Ukraine, but from all over the world. Indeed, the much-vaunted “Technological Singularity” may come a’swirlin’ out of Mother Russsia now that the Dushka Russian AI has gotten loose in the world. Best of luck!  -Artur Arturovich

Dave:

Thanks. I’m bursting with things to discuss. I’ll probably start with the topic of Microworlds, which I’ve needed to think about a lot, lately, and I’d like to hear some opinions about those, especially on a certain Microworld I created myself.

Jan:

That’s an interesting link, thanks. Later I saw your post in the contest section about the Kaggle link. The Kaggle contest that interested me most was the digit visual recognition task, but that doesn’t pay anything and my system is not ready for that level of detail yet, so I’m going to have to skip that. I had an idea for the chess rating algorithm competition, which does pay, but I don’t currently have the time to do the needed research and development on that. All the others look impossibly hard to do well, based on prediction within extremely complex phenomena like human preferences, the economy, molecular behavior, human health, etc. I saw there are some technical discussions associated with each contest, which made me aware of some technology I barely know anything about, like random forests. I didn’t read enough about joining a competition to know what the rules are, especially if a person must submit their code/algorithm at the end, and how they prevent cheating on the digits contest when people could just write down and submit their answers that they figured out by eye. If I’m using an entire, complicated, unpublished system to recognize handwritten digits, I’m not going to want to submit my “code”, and they probably couldn’t understand it even if I did.

Arthur:

Your program’s monitoring/perception of its own activity sounds interesting. As for my having learned all those languages, a serious drawback is that I’ve spread myself out too thin instead of having learned one language particularly well.

The Kaggle contest that interested me most was the digit visual recognition task, but that doesn’t pay anything

Yep. There are a couple of ‘educational’ competitions. Digit recognition is one of them. Predicting who lived and died on the Titanic is another one (which has a great video that explains random forests in laymen’s terms).
I have found them very useful for testing and comparing my own algorithms against what R or sikuli can do.

I saw there are some technical discussions associated with each contest, which made me aware of some technology I barely know anything about, like random forests.

After doing some competitions, you sort of start to get to see which algorithms perform well. Random forests is usually amongst the winners. Various regression algorithms, n-grams and decision trees are also very popular. A good algorithm by itself isn’t enough though, you need to know how to apply it (like random forests) and usually, you need a mixture of things, like n-grams combined with logistic regression.

I didn’t read enough about joining a competition to know what the rules are, especially if a person must submit their code/algorithm at the end, and how they prevent cheating on the digits contest when people could just write down and submit their answers that they figured out by eye. If I’m using an entire, complicated, unpublished system to recognize handwritten digits, I’m not going to want to submit my “code”, and they probably couldn’t understand it even if I did.

Rules vary depending on the competition. In the case of the educational ones, there really isn’t such a big need to catch cheaters: it’s not going to get you anywhere. But sometimes, the winning model actually gets used and the organiser can determine what has to happen with the code: only open source allowed, source has to be included with the model submission (in a password protected zip), no source requirements at all. Everything is possible. In general though, there is always an open source version of the algorithm available (sikuli is open source, some java libs, I’m open sourcing everything,...).

Hi Mark,

after you came your way from A.I. to the chatbot-place,
can you tell us, how one could go back?
What goods do you find here and what is missing jet
to build a chatbot-based A.I.?
There should be a way to do so:
Prof. Dietrich Dörner once showed us the way to build an A.I.
on the basis of a steam-engine: )
Regards,

Andreas

Andreas Drescher - Oct 18, 2012:

What goods do you find here and what is missing jet
to build a chatbot-based A.I.?

Sorry for my long delay to respond, Andreas. Recently I was just forced to move and to change jobs (in some ways for the worse), plus I’ve been trying to read all the A.I. posts in this forum to get “the lay of the land” before making generalizations or statements of opinion, so my time has been pretty tight lately.

In my opinion, those of you who have realized that one of the secrets to AI is the ability to represent the real world are on the right track. I was surprised to find so many people here already realized that, which many people in the AI community still seem not to realize. Relying on language alone will not produce AI. I think it’s important for chatbot enthusiasts to realize that if they want to achieve AI rather than just produce impressive demos with language that ultimately will not generalize very far. If one critical thing is “missing” for intelligent chatbots, that’s it. I could complain about Turing Tests, but my suggestions for intelligence testing would be too far out of the realm of chatbots, so I’ll just skip that topic. My posts elsewhere in this forum about the definition of intelligence imply my beliefs for such intelligence testing.

From the very beginning, mobile creatures whose excitable cells were capable of conveying information about conditions outside the body had a survival advantage over those whose movements were independent of whatever was going on outside. Obviously, the organism that flees in the absence of predators and feeds willy-nilly is doomed to be prey for those more lucky organisms fitted out with cells coordinating representations of the world with movement in the world. With increased complexity of behavioral repertoire comes increased capacity for representing the environment.
(“Neurophilosophy: Toward a Unified Science of the Mind-Brain”, Patricia Smith Churchland, 1989)

It is now deeply puzzling how the robot might be instructed so as not to be a fool, a problem that in AI research is called the frame problem (McCarthy and Hayes 1969). How do humans manage not to be fools? What does our “common sense” or “intelligence” consist in? The more we try to solve the robot’s problem of sensible behavior, the more it becomes clear that our behavior is not guided by explicit sentential instructions in our store of knowledge (Dennett 1984a). Specifying the knowledge store in sentences is a losing strategy. We have knowledge, all right, but it does not consist in sets of sentences. We know about moving babies away from hazards without having detailed lists of what counts as a hazard and how far to move the baby. Our “relevant-access mechanism” is imperfect, since we are tripped up from time to time, and tort law is full of instances of such imperfection. The right things do not always occur to us at the right times. Nevertheless, we manage on the whole to survive, reproduce, and do a whole lot more.
(“Neurophilosophy: Toward a Unified Science of the Mind-Brain”, Patricia Smith Churchland, 1989)

As far as the “goods” I find here, let’s just say that professionalism and maturity go a long way. Also, the post about robots forced me to bring my knowledge of the state-of-the-art on robots a lot more up to date. Every new topic that’s interesting forces me to learn, which I like, especially for topics like robots and quantum computers. Somewhere along the way I hope to increase my knowledge of parsing and grammar.