Sulla
Robot Personality Language/AI Project
Hello every one I thought I would share something I am starting to work on and see if any one else was interested in working on the project.
(Starting point would be Creative Commons language with AI coming later if needed.)
The project is the Robot Personality Language/Ai Project. The idea behind the project is that a robot you live with daily would be better if the
AI was based around personality and emotion instead of a chatbot/assistant idea. And instead of communicating clear ideas would communicate in a nonsense language that suggested emotions.
There are several examples of this in movies. Examples would be R2D2 and other non speakind droids in star Wars. Minions from Despicable Me. wall-e and most of the other robots from wall-e the movie. And I am sure there are others.
-635631477427050781.png)
The AI would be mainly about creating a dynamic personality that you can both change in menu and affect by interaction over time. It would also be effected by other variables to make it more dynmaic just like real personalities. Examples could be anything from the weather to if your team wins to day of the week.
Whats needed:
Langauge>
Lets start with the language first. The langauge needs to be something consistant and well designed. Designing a language of this kind is harder than you would think. Ben Burtt created them for both R2D2 and Wall-E and has talked about what a challenge it was.
What the language has to have is a clear set of set of parameters and standards. Then we can have a large community of people creating and submitting audio files to it.
The language would all be in audio files not text to speech. You can not convey the needed emotion using text to speech.
Each sound file would need to be clearly labeled by suggested emotion, and sound. Something like (Dejection, Long Wistle). You would have both sounds and sentences. The sounds would be used to construct more complex sentences. Both of course would need clear labeling.
All counds would have a Creative Commons license.
AI>
The AI would be based on most of the same fundamentals as a chat bot but with a more complex emotion layer. And because the language is only conveying emotion is can be much much less precise. For example you can parse and analize speech input and conclude it is a question. BUt you do not have to know the exact question or much less the answer to it. The reply and be a something suggesting the robot does not know or something suggesting it is giving an answer. Or even something suggesting it is impatient with to many questions. Because of the un-exact natue of the robots response we will tend to anthropomorphize its response and give it a deeper meaning.
The robots reponse would of course alos be effected by its mood that day. Maybe it does not like cold weather and its cold that day. And maybe its a bit moody because you have not interacted with it in a while. And in general its personality has a limited patience. When you start asking it questions you are probably going to get a rude or annoyed response.
Ideally at some point it would be best if the AI had a GUI interface and a way to import and export personalities.
Any way this is a very brief overview while I am sitting here sick. Hopefully it will make sense
What I am working on know is the langauge
coming up with ideas and making tests etc. So far my tests suggest a Minion type language would be the easiest to make. You can record things and
then change them with pitch +8 and Timer Compression 60 to get roughly the same sound of voice. It also starts to sound a bit lik M-O from Wall-E
with some sounds. I am thinking something like this would work. A combination of nonsense words, beeps, bleeps and other sounds along with
grunts and laughs etc.
I would also be more than happy to work on a Star Wars droid style language but so far it seems much harder to pull off well.
Any way way what I would like to do if there is any interest is get a group together to work on the langauge and release the information on how to make and add pieces to it. Then worry about the AI later since you can use the langauge before the AI just having it be random.
So please let me know if you have any suggestions of feedback or would like to help etc. :)
Good idea, but someone is ahead of you on the AI part at least... see here EZ-AI
Well that is one reason I wanted to work on the language first. It could be used to any one using any AI or sound board etc. And if the AI is not needed that just saves work :)
You could use a NLP to analyse the text and then based on what comes back, you would make the speech a certain way, or if you have determined that the robot is in mood x, use the NLP to replace certain words with others
I am looking at developing an NLP now in EZ-AI to parse sentences for root words in a sentence, and then do research on these root words and other nouns. Also Freebase and and hopefully soon another API from Google will show associated words. There are also some API's that will do this too.
If you want too, you can also download all of the freebase dump and store it in a local database.
@Sulla You do have a good idea. Hope you didn't think I was knocking it... :)
Yes anything you can do to help the AI understand inputs will help figure out better responses. For example if the speech recognition could pick up inflections on the input it would help a lot. Same with picking up facial expressions.
No not at all Richard R
But I am a bit sick today so responses might or might not make sense both ways haha.
Hello,
I plan on building an EZ B minion. I have one of the toys which has a sound board built in. I had planned to just download the phrases and stick them in a folder and call on them.
Instead I will load them based on your chart topics and try to use key words to play back the correct file. This is a start for me to try to get a basic emotional response. I can then work on maybe grouping the words to better define the response. I will let you know how I make out.
I want to keep it simple because I am new at this.
Ron
Ahh very cool keep my informed how it works for you :)
Interesting related news. http://www.wired.co.uk/news/archive/2015-04/01/google-robot-personalities
Google patents customisable robot personalities
Programming personalities (sounds like uploading) "so that effectively you could get someone to live on after their death in robot form"
This is dated first of April, could it be an April Fools hoax? Sounds to me like it is.
A lot of what else is talked about in the article has already been done (prior art) so a patent would be difficult to obtain.
Tony
Hello Sulla,
I plan on using your basic idea in a Minion ( Stewart from the movie). I will be on a plane trip next week so I should have some time to go further with this idea. I want to also build a simple "conversation chart" like your emotion chart. Because the minions do not really speak a true language, I will try to come up with some "canned phrases" which will be used and combined for conversation. I have no idea as to if this will work, but will try it anyway. All and any suggestions will be valued.
Ron
I forgot to mention I have two sound board which I will pull and build mp3 files from, then fit them to the correct chart topic.
Tony the patent is online and it relates to making AI robot personalities and then storing those online etc. http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&u=%2Fnetahtml%2FPTO%2Fsearch-adv.htm&r=42&f=G&l=50&d=PTXT&s1=google.ASNM.&p=1&OS=AN/google&RS=AN/google
Interesting patent but seems rather general to get a patent.
Do not think its an aprils fools joke.
Ron are you going to pull sound files from the movies or try and make your own?
No, the minion dolls I bought have sound boards with phrases and the general known replies ( Hello, WHAAAT? Booboy, BEEDOO, laughter, etc). This way they are clean with no background sounds to filter. Yesterday I pulled the first group ( about 15 phrases ) from doll 1 using Audacity software, Today I hope to get doll 2 data. I will save them as wav files first. Once I collect everything, I will break them down to separate files using your chart and a conversation chart I will begin to develop. MP3 conversion will be last. I am going away next week so I am trying to gather all the raw stuff so I can work on it on the plane (I hope).
Once I get this far I will ask if anyone can help build our own voice sounds and phrases. My own voice kind of works, but I don't have the needed bass and French accent that helps define the minion sound.
Ron