Archived

This topic is now archived and is closed to further replies.

Nail

Original AI for desktop character

Recommended Posts

Hi. I need to create a desktop character that will have a cute, original AI and that will help it''s owner. I think that characters, realised in MS OFFICE 2000 are _VERY_ simple for the example. If you''ve seen Black&White game, then I think you can understand me. Your creature is learning in the game and it''s not annoying you. So... can any1 advise me any kind of AI for my desktop character?

Share this post


Link to post
Share on other sites
quote:
Original post by Nail
If you''ve seen Black&White game, then I think you can understand me. Your creature is learning in the game and it''s not annoying you. So... can any1 advise me any kind of AI for my desktop character?


Black And White has a very limited field of things that need to be learnt.
Same with the stupid office helpers.
What you need to consider is what your desktop dude REALLY needs to help you with. and what can be avoided.

You''ll need to find a way of monitoring user input without interfering too much with speed of the machine.

You''ll probably need to think about spiliting the AI over frames (or ticks), which is what they did with B&W and got a huge increase in speed.

You''ll probably be best off using some sort of state machine, rather than anything fancy.

a few thoughts, but I hope they''re helpfull




"Bad Day... F**K it!" -Stephen Baldwin (Usual Suspects)

Share this post


Link to post
Share on other sites
Can you explain what is splitting in frames (ticks)... Or give any links. I can imagine what is a state maching, but i still need some info on this... i''ll be glad if you can help.

Share this post


Link to post
Share on other sites

quote:
Original post by Nail
If you've seen Black&White game, then I think you can understand me. Your creature is learning in the game and it's not annoying you. So... can any1 advise me any kind of AI for my desktop character?



quote:
Original post by Nutter2000
Black And White has a very limited field of things that need to be learnt.
Same with the stupid office helpers.



No offence intended but it is VERY clear that neither of you know what is behind the little paper clip in Microsoft's Office software.

It's a very complex user model using a Bayesian Network to track user actions: mouse clicks, mouse moves, key strokes; and what each of these invoke. For instance, if a user tries to bold something and then undoes the action, underlines it and undoes it and then searches help on some topics relating to formating, the user model computes a set of beliefs about what the user might be trying to do (perhaps formating a title). It can then interact with the user to try and refine the information that it will present by asking a question. It's results are astounding in analysing complex sets of actions of the user and deciphering the users intent.

Unfortunately, the marketing morons at Microsoft had the paper clip added and this was the biggest drawback of the system. No one wanted to interact with an annoying piece of office stationary.

If you really want to understand the power of the Office Help system (or things like the new Printer troubleshooter) you should do a search for information on the Lumiere Project conducted by Microsoft Research.

Oh, and don't bag things until you understand them please.

Tim

Edited by - Timkin on June 24, 2001 10:03:39 PM

Share this post


Link to post
Share on other sites
quote:
Original post by Timkin
It's a very complex user model using a Bayesian Network to track user actions...


I'm sorry, but I didn't mean to step on anyone's toes there.

The reason I consider the Office Help to be simplistic is because it is a good example of what I consider to be AI for the lowest common denominator.

I, only a personal view mind you, find that it's probablity network has been trained for people who know nothing or little of M$ products or are fairly stupid.

I will admit I was harsh in calling it stupid, what I should of said was annoying and virtually useless to the majority of people once they pass a certain, relatively low , level of experience with the product.

Now you make a good point that the Bayesian network is fairly complex, however, the point I was trying to make (very badly I admit) is that the input set from the user is relatively simplistic compared to the input set generated from an overall desktop level helper. With, say word, the network has mouse clicks, drags, key presses and state changes and combinations thereof to contend with, which thereby constituted a rather nasty and complex probablity network back in 1993 when the Lumiere Project was concieved and executed for the '95 release of Office.
Also the probability computed of what the user is trying to do has too large a data set to cope with advanced usage. It is trained (I beleave anyway) on a research group consisting mostly of people who have little experience of Office.
Hence why it often gets it wrong for advanced users.

However, a Desktop helper would need to also take into account in it's set of current states which programs are currently running in the background, icons on the task bar, shortcuts on the desktop and so on, IN ADDITION to the mouse clicks, et al.
An immensely complicated task comparatively .
A Desktop helper would have to include a stable yet flexable AI to cope with the constant changes that monitoring a desktop entails.

Therefore, although the underlying AI in Office Help is complex, the input set and current-state set is not compared to that required for an Overall DESKTOP helper.

I have, however, not taken into account the vast NLP of the help search which truely enables it's power, implimented for the '97 release of office, however, for the job Nail is considering I can not foresee a reason (at least assuming no help file needed) that NLP need be implimented.
The NLP developed for the '97 release of Office is extremely good.

As for your statement that the paper clip was added by the marketing people at M$ all I can say is "welcome to the wonderfull world of Software Development, please leave your sanity by the door "

- ********************************************************* -

The comment I made about the AI in Black And White is also valid to a certain point of view, however, before I get flamed for THAT as well, I'd like to point out that the Black And White Creatures are amazingly good at what they do. The AI, in my opinion, far exceeds that of the Office Help, and in only 4 years dev. have produced an extremely realistic AI considering the limitations of the system, ie. with an excellent 3D engine as well as the AI.

I read somewhere Richard Evans (Lionhead - Head of AI) describe that the AI for the creature took only approx 3% CPU time, and the villages 5% CPU time, NOW THAT is something to boast about!

The way they did it was to split the AI over frames, ie do a bit of AI each frame so that the overall effect was the same but took less processor time used.
I personally beleave that this method is the way forward for a lot of AI, esp. in games.

- ********************************************************* -

In Summery - I beleave the Office Help to be a usefull and extremely good yet relatively basic AI implimentation that was state of the art 6-8 years ago. The load of the AI accross the CPU is well implimented and efficient.
The entire project is a worthwhile and well executed AI solution that has taken some extremely intelligent people a fair amount of time and money to solve. I would not in anyway like to belittle their acheivements.
IN MY OPINION , it is annoying to the user once a RELATIVELY low level of experience in the product has been reached by the user.
It is probably more a statement on the many problems of Human Computer Interaction than actual AI.
HOWEVER IN MY OPINION , when compared to the task of an overall desktop helper it is necessarily simple.

I AM NOT A GENIUS OF AI , nor do I consider myself to be.
I am merely a student of life who says things how he sees it,
and tries to help people where possible.
apologies if I step on anyones toes that is NOT my intention.




quote:
Original post by Nail
Can you explain what is splitting in frames (ticks)... Or give any links. I can imagine what is a state maching, but i still need some info on this... i'll be glad if you can help.


no offence but it sounds like you're starting out too complicated.


"Bad Day... F**K it!" -Stephen Baldwin (Usual Suspects)


Edited by - Nutter2000 on June 25, 2001 6:55:16 AM

Share this post


Link to post
Share on other sites
To the person who explained about the bayesian network, even though it was a complex form of AI back in 1993. It''s still no more than lots and lots of if(...)else(...) statements, looped and nested.

Share this post


Link to post
Share on other sites
I''m sorry to everyone who is going on abot all the complexity behind the damn paper clip but in the end, the bloody thing is one hell of a lot of wasted cash. You can get the same results with 1000 lines of code and a massive index file (which the paper clip has anyway). I have never found the paper clip to observe what i do. I ask it a question ill get the same set of suggestions each time, no matter what i have done before. I would be so much happier if i could simply press F1 and it would use my last few actions as guesses of what i might want. Just like when you place your cursor over a function in VC++ u get help on that function. That would be infinitely more helpful than fighting with the paper clip.

Plus the actual written help in Office sucks pretty bad. A lot of the stuff isn''t well/at all documented. It''s great for computer newbies, but grrrr, where''s my hardcore reference????

In short, Im glad the paperclip is gone, and i would mourn the lost money and time that went into it, but for Microsoft, it''s pocket change at most.

Resist Windows XP''s Invasive Production Activation Technology!

Share this post


Link to post
Share on other sites
I think the most important thing is the user interface.

So, do not put an annoying BIG paperclip on someone''s screen! Perhaps use part of the taskbar, or create a small floating line of text...

As for the text: I''d advise you to create an algorithm that can vary the things it can say... Otherwise, when it keeps repeating the same lines, people will get annoyed with it.

Also, make sure it can also do constructive things by it self. The user should be able to give it a list of commands it has to execute. In a smarter/better way then Sheduler (or whatever the MS thing is called).

Share this post


Link to post
Share on other sites
quote:
Original post by Nutter2000

Therefore, although the underlying AI in Office Help is complex, the input set and current-state set is not compared to that required for an Overall DESKTOP helper.




Taking that point of view for your original post then yes, I would certainly agree with you. 8^)

Tim

Share this post


Link to post
Share on other sites
quote:
Original post by Erlog
To the person who explained about the bayesian network, even though it was a complex form of AI back in 1993. It''s still no more than lots and lots of if(...)else(...) statements, looped and nested.


Perhaps you could enlight us as to HOW a Bayesian Network is exactly the same as a rule based system???

Tim

Share this post


Link to post
Share on other sites
quote:
Original post by Timkin
Taking that point of view for your original post then yes, I would certainly agree with you. 8^)



Thank you.
appologies for being extremely vague about what I ment though!





Edited by - Nutter2000 on June 27, 2001 10:29:21 AM

Share this post


Link to post
Share on other sites