Artificial intelligence test

http://news.yahoo.com/s/ap/20081012/ap_on_re_eu/eu_britain_artificial_intelligence

3 out of 12 judges couldn't tell if they were talking with a human or robot, now that's impressive.

Lee do you have a rough idea on how much processing power a AI program like that would require in order to work well(without lag)? can a program like that be installed in AIko? Of course one that can process audio and visual not just text, maybe you'll win the gold medal if you do that.

I've wondered the same thing. How feasible is it to utilize a neural net, or distributed computing architecture (perhaps with a series state of the art desktop processors) linked to Aiko's 'brain'?

This is rather long.. sorry..lol

In order to create a proper AI that can chat like a normal person, you will need a voice/subject analyzer.

It will have to compute the following:

- Subject at hand
- Context and tone of voice
- Visual Information
- Oral Information
- Touch Information (Heat, weight and whatnot) - This is the big one.
There might be more, but those are the big ones.

Also, as the question is being created, the analyzer can take the information given as it progresses to possibly have an answer just as the question is finished being asked if not sooner.

This will be supper processor intensive and might need its own unit. Using Quad core processors, the information could be processed and distributed quicker.
In addition, there should be a process that checks the "memory" to see if the question has already been answered. It would need a program to compare statistics.

The program itself will need to compare the stats by neural weights. The types might be “Relevance” and “importance”. For example:

Let’s say you give Aiko a Large Blue Rubber ball that is fairly hard. She analyses it and stores the info.
Now some time passes and you hand her a Small Red Sponge Ball.

Question: “What did I just give you?”

Analyze:

Visual Conclusions
- The object is spherical
- That it is Red
- Any identification marks (Logos, names)

Physical Conclusion
- Soft
- Possible Weight
- Size (Can be specified (Unit of measure or generalized, S,M,L) She will have to have a unit preference in her personality, with alternate units of measure as secondary’s. This will also have to be measured by relevance. You would measure a coin in cm or mm and a dinner plate in inches or cm.

Now with the basic information gathered the “memory” kicks in.
Searches under Objects>>Spheres.
Now it looks for any spheres with the same markings.
If none, it moves on to colour.
Then physical properties are analyzed.
These properties can vary greatly, so they are of least importance.
But they still help answer the question.
(The physical properties can determine whether the object is a ball, a ornament or something else.)

So with all the elements and traits analyzed, the information is placed in temp memory and answers.
Depending on the certainties of the conclusion, the speech can select the appropriate response.
For example, if she isn’t certain; “I think this is a… “or “This is…”

So for this example, the red ball is new to her, and some of the variables are different or missing from previous experiences. So the response might be as follows.

Answer: “I think you handed me a Red Ball.”
Question: “Can you be specific?”
Answer: “It is small, soft red ball.”
Question:” How different is this red ball compared to the Blue one I handed you earlier?”
Answer: “Compared to the Blue ball, this Red ball is Smaller, Softer and (depending on the weight) lighter.”

Depending on the situation and personality, she could have responded as follows:

Answer: “This red ball is Smaller, Softer and lighter. I like this one more than the other.”

So with the proper analyzer and language assembly algorithm, Aiko could have a very dynamic and personalized speech, thus giving her a more realistic personality.

you could have tags on every object to shorten the database searches.
When you give her something she will first look under held items then shape.
When she finds "ball", it will have actions or whatever associated with it, like throw.
She could say "This is for throwing". "Do you want to play catch?"
If yes, she would goto the action set for playing a game of catch.

If I ever start on something it will probably use multiple choice to play predefined or prerecorded responses based on keywords just to keep things simple.

Reaper wrote:you could have tags on every object to shorten the database searches.
When you give her something she will first look under held items then shape.
When she finds "ball", it will have actions or whatever associated with it, like throw.
She could say "This is for throwing". "Do you want to play catch?"
If yes, she would goto the action set for playing a game of catch.

If I ever start on something it will probably use multiple choice to play predefined or prerecorded responses based on keywords just to keep things simple.

That’s pretty much what I stated above... but in order to create such a database, she still has to go through the steps and analyzing. The analytical data is placed in Meta files for referencing. Each Meta has a Master Meta for every section created, Box, Spheres, Human, vehicle…etc.

As for association, classes would be created depending on the situation. If you just create “ball” >> “Throw ball” she will throw any ball she gets her hands on let alone if she get a hold of a fruit or sphere like glassware. When analyzing an object or situation, there are millions of combinations. It will take time and patience to create. The alternate route is to teach her like a child. Have the basics pre-installed and have her system build upon it. Somewhat like HMMay.

The system would have to have the usual parts like visuals, speech and hearing… but will need to be able to analyze motion, angles, distance, trajectories and so forth… thus building more Meta info.
Think about it this way, when a child gets a ball for the first time, it has no clue what it is… by trial and error, and guidance by a parent or other, the child receives the information needed to understand what a “ball” is. Then it can tell what the ball can do and what it can’t. The child then gets a variety of balls. It will have to go through each one and learn its characteristics. Does it bounce, break, slam to the ground or roll? What is it made out of? What type is it? What brand is it? How will it fly through the air if I throw it or if I roll it? There are many aspects that need to be covered. But with time, and a adaptive program, Aiko could theoretically learn and self teach over time. In addition, a definition, explanation and example database has to be created in order for Aiko to understand… well everything. o.O

As daunting as this may seem, it can be done. The problems with robots these days are that they are programmed with pre built movements and capabilities. We have to step away from those aspects and have them learn and understand on their own with a little help now and then. People build robots to do a certain amount of tasks or specific tasks. There is no room for expansion nor is there any want. That’s why we don’t have robots serving us today.

So, to sum this up, you need a Self Learning Adaptive Program (S.L.A.P) to start having Aiko do more over time then just the pre recordings. BTW, I totally pulled the abbreviation out of my ass… and as funny as it is, it makes sense.

So once again, Aiko needs a Self Learning Adaptive Program (S.L.A.P).

Aiko will FAIL the turing test..... Aiko is still in learning stage.. her brains is like a 5 years old.

With the current CPU quad is still very very SLOW.
In order to make Aiko work properly, I will need a cpu that is 10x faster at least...

Aiko extract the 3-D from environment once every 0.5 sec... at a very low resolution 320x240...
Even with a core2 (aiko is using) is not enough cpu power.

Anyone who does 3-D rendering from 3-D Studio will know that it is very slow to process even with the fastest cpu in the market today.
In order to do real time at a high resolution... you would need to connect at least 15 core2 together....

Can't do that with a home budget, not too mention it won't fit inside Aiko.

Yea, just having the current income is probably extreamly hard to manage and stretch.

My ideas arn't ment to be done as of now. These are just things to save and consider for the future development of Aiko.

As for the processing info, from what I understand, you are running a laptop and a server/database right?
And because you are on a low budget, it would be hard to get custom parts made.
Having custom parts with individual processors running simultaneously mastered by a quad would fix the problem. But as you stated, it would take alot in funds and space to do.
You might want to look into Intel or AMD to ask them if they could help with designing and possibly making custom units with their own processors. Just a thought.

MattDragonTamer you're right about what capabilities the software should have.

Aiko wrote:
In order to do real time at a high resolution... you would need to connect at least 15 core2 together....

A single processor based on memristors will be able to do that.

To run a A.I program on computers which work in 0 and 1 sequences is like putting a fish to run on dry land.

I hope these processors based on memristors will hit the market in the near future.

Le Trung: Actually the brain of a 5 year old is quite advanced. When you ask Aiko a question, or when you tell her something, does she understand ? You know what I mean...

Chatterboxes somewhat simulate how we respond to questions via their scripts.

As an example: "How are you?"
We each have a few multiple choice responses that we give depending on a self query of our current state.
[I feel well] could generate: "Fine.", "I'm ok.", "I feel good."
This is how a properly scripted chatterbox does it, but it's not intelligence and is rather primitive as it doesn't have a whole lot of conditional checks to further refine the response like a sentient has.

I doubt we're ever going to see a true ai until a very complex chatterbox like script format is merged with an internal dialogue program and variable database associated with interactions to give a simple memory.
Just adding a huge database to a chatterbox isn't going to magically create ai and it's inefficient to scan through that entire database to match keywords in a sentence.

Right now, the main focus on ai should be developing a good script format to process the information it's getting. All I've seen so far are way to simple.

"How are you" to Aiko.

The brains software check for the micro controller, mainboard,sensors.. and server connection.

If everything is OK, then Aiko will say "I am fine" or "my system is working fine"
If not..
then my micro board at location xxx is not working...

Reaper, check out Rollo Carpenters "Cleverbot" WWW.cleverbot.com. It runs on statistical pattern matching algorithms. Responses are NOT hard coded. It appears to be doing most of what you describe. Questions can be asked in any number of ways. Have noticed very short comments like 'yes', 'no', 'maby', 'whatever' are confusing to the AI.

captquazar wrote:Reaper, check out Rollo Carpenters "Cleverbot" WWW.cleverbot.com. It runs on statistical pattern matching algorithms. Responses are NOT hard coded. It appears to be doing most of what you describe. Questions can be asked in any number of ways. Have noticed very short comments like 'yes', 'no', 'maby', 'whatever' are confusing to the AI.

If you keep the conversation in the direction that the A.I wants you can have a conversation with it but it can stray of course very easy, A.I it's still in it's infancy.

Do you know other websites like WWW.cleverbot.com ?

I'm going to give it a shot again although I have less time
coming to this forum my mobile expenses must stay low (lol)
explaining what the implication and benefits are from android A.i.

But before that did you know that Microsoft does also research into A.i
other then surface technology , game and blue&me A.i this is the future
of health care and medical information the most interesting application was
''Situated Interaction'' using an on-screen avatar "Laura". Watch the video's and
see how the avatar relates people in the same room as a group.
Notice how many CPU the application uses.

https://www.youtube.com/watch?v=6GOrNU6e_og
Or the source link
http://blogs.msdn.com/healthblog/archive/2008/08/18/microsoft-research-how-we-watch-the-computer-how-it-watches-us.asp

Also I think with today's technology at hand its a big exploration with the kinda
work people put into their projects and come with things that is really amazing
Aiko ability to recognize faces and touch is really mind boggling. Sometimes I wonder
Where do people get there inspiration from and passion to keep working this alone must
be a great story in the future.

I was thinking about the somehow quieted '' Sumomo '' project how it could be realized also
But seeing what small robots are today a small lively robot raise brows if it is even feasible with technology.
That was what I was thinking till I came across this here the link.

http://www.calinon.ch/showVideo.php?video=2


Just for fun link on other stuff.
http://robotic.media.mit.edu/projects/projects.html

Have fun

Aiko wrote:Aiko will FAIL the turing test..... Aiko is still in learning stage.. her brains is like a 5 years old.

With the current CPU quad is still very very SLOW.
In order to make Aiko work properly, I will need a cpu that is 10x faster at least...

Aiko extract the 3-D from environment once every 0.5 sec... at a very low resolution 320x240...
Even with a core2 (aiko is using) is not enough cpu power.

Anyone who does 3-D rendering from 3-D Studio will know that it is very slow to process even with the fastest cpu in the market today.
In order to do real time at a high resolution... you would need to connect at least 15 core2 together....

Can't do that with a home budget, not too mention it won't fit inside Aiko.

congrats, MIT is working on what you would need... 500 to 1000 gHz processors, made of Graphene, a carbon based compound. that's 1 Thz, and they are working on getting it to the size of a regular computer cpu.

and 3-d rendering can be done easily with a homemade supercomputer.
now, before anyone says those are millions of dollars, the definition of a supercomputer is "two or more motherboards hooked together via a networking protocall"

just some special software, which there are free versions of, and your home network can be used as one supercomputer. just use a gigabit ethernet switch, not wireless.

there is some laboratory in the US that has a supercomputer made of hundreds of regular home pcs.

Discoman wrote:

Aiko wrote:Aiko will FAIL the turing test..... Aiko is still in learning stage.. her brains is like a 5 years old.

With the current CPU quad is still very very SLOW.
In order to make Aiko work properly, I will need a cpu that is 10x faster at least...

Aiko extract the 3-D from environment once every 0.5 sec... at a very low resolution 320x240...
Even with a core2 (aiko is using) is not enough cpu power.

Anyone who does 3-D rendering from 3-D Studio will know that it is very slow to process even with the fastest cpu in the market today.
In order to do real time at a high resolution... you would need to connect at least 15 core2 together....

Can't do that with a home budget, not too mention it won't fit inside Aiko.

congrats, MIT is working on what you would need... 500 to 1000 gHz processors, made of Graphene, a carbon based compound. that's 1 Thz, and they are working on getting it to the size of a regular computer cpu.

and 3-d rendering can be done easily with a homemade supercomputer.
now, before anyone says those are millions of dollars, the definition of a supercomputer is "two or more motherboards hooked together via a networking protocall"

just some special software, which there are free versions of, and your home network can be used as one supercomputer. just use a gigabit ethernet switch, not wireless.

there is some laboratory in the US that has a supercomputer made of hundreds of regular home pcs.

It was using that system of using the home computers inwhich SETI at home operated under, starting over ten years ago.

i didn't know that... some other places are looking for volunteers to allow them to access your computing power for running massive equations, just got posted recently on some news place.