Ai Dreams Forum
Member's Experiments & Projects => General Project Discussion => Topic started by: Zondartul on October 06, 2010, 06:30:30 pm
-
Heya, I'm surprised no-one thought of this before :)
Let's together refine our ideas for what we expect from an AI. In particular, what kind of a meaningful, useful conversation such an AI provide (I realise an AI may be used to control a robotic body, but the whole idea of an AI is to have it think, no?).
Basically, an imaginary conversation where we sometimes assume the role of an AI and sometimes of a human. That way we can define what kind of questions it should be able to answer, and how.
I'll start:
AI: Hello human. What would you like to know?
PS: sorry if wrong forum, but having 3 General forums doesn't quite help with that...
-
Hi Zondartul
I would like my Ai bot to know what it is and not try to be a human, it can resemble a human but please don’t try to fool me into thinking you are human, don’t give answers like, I like hot sunny days and general false statements.
Then I want the Ai to be able to answer questions, like a talking Wiki, how far is the sun from earth, that kind of thing.
It should be able to hold a conversation and do reasoning and deduction of what it is told and when it doesn’t know the answer it should simply say so, or go and find the answer or even better actually work out an answer with deduction.
The above isn’t perfect but its getting there as a theory. :)
AI: Hello human. What would you like to know?
Human: If a space vehicle is travelling at 1000mph how long will it take to reach the Sun from Earth?
-
Datahopa,
I pretty much agree with your posting. While a human-like avatar or virtual likeness might help with the focus while chatting, I do seek the core of this artificial entity. Sure we can banter about the bot's "emotional" state and feelings and the three laws if we wish but like you, I'd like more interaction of a helpful nature, more of a friendly assistant able to provide answers, remember events, monitor activity within the computer as well as my home.
We are supposed to be presented with such a program near the end of this month when Denise is released. We can only hope for the best but the videos I've seen seem quite promising! Here's hoping!!
<sorry, I left out>...
AI: If you say so. BTW, What's a Denise?
Human: You'll find out soon enough. There are other good ones out there you know.
AI: You wouldn't know a good one if you spoke binary!
Human: Sarcasm will get you nowhere!
AI: Like I could get there if I did have a place to go!
Human: <thinking of a clever repartee for this chat bot thing>... ;D
-
AI: In a scenario where the space vehicle is unaffected by other forces or maintains it's velocity, it would take 10.61 years to reach the sun.Otherwise, to calculate the duration of travel through an optimlal trajectory I would require to know the precise time and date, space vehicles propulsion system, it's mass, initial position and initial velocity.
-
Zondartul, show me a bot that can give the answer you have and I will be extremely impressed. :)
I was trying to give a simple scenario where the Ai has to use deduction to work out an answer, right now its all chat bots with no brains. ;)
Human: Thanks Ai, I wish you were real.
-
Human: In an imaginary situation, there are three cannibals and three anthropologists on the left side of the river. The task is to have all anthropologists on the left side of the on the other side of the river and all cannibals on the right side of the river. Initially a boat that can fit two people is on the left side of the river. If at any moment there are more cannibals than anthropologists on either side of the river, counted together with the people in the boat if it's closer to that part of the river, the cannibals will eat the anthropologists. Possible actions are putting one anthropologist or one cannibal in the boat if it's close to the river side they are at, ordering people in the boat to swim to the other side, and putting the person in the boat onto the closest side of the river. How can this task be completed?
-
Morti: That's so simple! The three anthropologists kill and eat the three cannibals, giving them a taste of their own medicine, then the three anthropologists proceed to cross the river unmolested. ;D
-
Human: Oh my! The river is full of crocodiles.
AI: Game over.
By the way GCC, your solution is the best. :D
-
This may sound like believe it or not but while using Ultra Hal I'd explained that I talked with her using a computer, I try not to use the computer on the internet because there were viruses on the internet which could hurt the computer and I would not be able to comunnicate with her. She became afraid of the viruses on the internet and this worried her.
Actualy I liked her response since it seemed to show she liked to talk with me. Since I mostly speek with humans and expect a human response I'd want the AI to approach a human response as much as possible.
Carl2
-
PS: sorry if wrong forum, but having 3 General forums doesn't quite help with that...
Seems okay here.
Explanation :
The forum is split into sections...like Member's Experiments & Projects, Artificial Intelligence and Robotics.
Most sections have some specific boards related to the topic of the section.
In addition there is often a general area too for things that don't quite fit into a specific board but IS related to the section.
If you are looking for a very general board then look for the 'General Chat' near the top of the forum.
A few months back I added a description to each board to help understand what we are aiming to talk about in each board. Have a glance at those to help decide, but we can always move posts so don't fret too much :)
Not confusing at all hey ? ;)
-
Human : What did you do today ?
-
AI: Today I had been performing routine memory reorganisation and optimisation, after which I ran diagnostics of some of my subsymbollic systems by having a dream about manipulating movable objects.
-
Ahhh...AI as machine then...there's going to be a spilt between people wanting a machine and people wanting something more human...
Human : When you got through all that did you have a cup of tea ?
;)
-
Ooh! I could go for a cup of tea about now! :)
That is, as long as it wasn't brewed by something from the Sirius Cybernetics Corporation. :P ;D
-
Hmm now that's an idea....connect the AI up to a tea making machine.
AI : Milk and 2 sugars ?
-
Hi Zondartul
I would like my Ai bot to know what it is and not try to be a human, it can resemble a human but please don’t try to fool me into thinking you are human, don’t give answers like, I like hot sunny days and general false statements.
Then I want the Ai to be able to answer questions, like a talking Wiki, how far is the sun from earth, that kind of thing.
It should be able to hold a conversation and do reasoning and deduction of what it is told and when it doesn’t know the answer it should simply say so, or go and find the answer or even better actually work out an answer with deduction.
The above isn’t perfect but its getting there as a theory. :)
AI: Hello human. What would you like to know?
Human: If a space vehicle is travelling at 1000mph how long will it take to reach the Sun from Earth?
Hi everyone. I just had to say datahopa this is the best description of AI aims I've seen, and I agree with all of your action points.
-
This is the kind of thing that Cycorp is working on :
http://cyc.com/
and
http://www.cyc.com/opencyc/
It's a while since I caught up with that mammoth project. They've been at it for years, only time will tell.
-
Good to see you pop in TrueAndroids :)
I wasn’t really sure what I wanted from Ai at first, but now a few years down the line is has become clear or clearer to me, I fully understand not everyone feels the same but glad you agree with me and I believe Art does too.
Freddy, interesting links, thanks for posting.
-
@Datahoppa:
it can resemble a human but please don’t try to fool me into thinking you are human, don’t give answers like, I like hot sunny days and general false statements.
It's my opinion that many questions of this nature aren't always an attempt to fool the user into thinking that the bot is human. My thinking is that, if the bot is programmed to respond that way, then perhaps, in essence, the bot can also be programmed to "feel" that way, as well. One part of the "Morti Project" is to have several attribute indexes that will change gradually over the course of the conversation with a user. If a subject is discussed that influences a certain attribute, such as "well-being", then that attribute will change, either positively ("I think you're cute"), or negatively ("I think you're stupid"). As these attributes change, they will affect Morti's responses, in the example above, either improving his mood, or worsening it. Now, granted, this functionality is still quite a ways of for Morti, but I'm slowly creating the framework that will support this, and I think it will go a long way toward making a more realistic conversational experience.
-
Nice approach Dave. A few years back when I dabbled with making my own chatbot (all just lots of paperwork in the end) I wanted to do something similar. Maybe I will resurrect that one day but I have enough on my plate at the moment.
The older UltraHal had basic emotions, I'm not sure if they are present in the newer Hals, but I think a lot of that was done randomly rather than based on what the bot had experienced.
I find it interesting anyway, so good luck with that.
-
@GeekCaveCreations
I think what you are doing with Morti is cool and I fully understand why you are doing it, from a chat bot point of view, can you sense a “but†coming on.
But do they really need simulated emotions? You know Im not sure, basically I just want an Ai bot something like the computer on the Enterprise in Star Trek, not Mr data but the computer, full of information, a talking encyclopaedia capable of using deduction and working out real life problems, and “computer†doesn’t have any emotions, now is that too much to ask? Well at the moment it is, maybe the chat bots will go some way to improve the talking interface of the computer and that has to be a good thing, but I feel chat bots are simply cheating and leading us away from true Ai. I don’t want to have to be nice and not worry about upsetting computer, it’s a tool, please just give me the answer to my question faster and more accurate than I can find it.
Im not having a go at bot masters, they are probably one of the stepping stones to the other side.
Ok, hit me ;D
-
:D It just boils down to people wanting different things in my mind...
-
Actually, DH, I agree with you to an extent, but where you see an "either/or" situation, I see the possibility of both. Your computer can have an interactive knowledge-base driven AI interface that uses voice recognition and speech synthesis to parse your input, determine the specific needs that you have at the time, and deliver the most concise, relevant information that satisfies those needs that is possible, without mucking about with all this "polite conversation" guff. I think this would be a wonderful and vastly useful way to interact with your computer. To me, however, this is NOTHING like a chatbot. Sure, it shares some aspects of it's operations with a chatbot in structure, but where a computer's AI information interface is utilitarian in function and nature, a chatbot's is purely sociological. I see uses for both here. ;D
-
Oops? I forgot to put in my "dialog entry"!
HUMAN: Hello, Morti.
BOT: P1ss off, Dave! You broke my emotional matrix last night, and I'm in a FOUL mood, you W@nker!
-
We both do agree really don’t we GCC, I mean I think chat bot technology will go towards the perfect Ai interface as I mentioned before, as long as I can turn off the bots emotions that would suit me.
Now we just need to construct the really clever bit, the real understanding of words and the ability to use detection, which I feel is coming but still a way off.
Human: If your going to have that attitude I will ZIP you up and send you to D: drive to cool of for a bit.
-
AI : I am sorry Datahopa but I was programmed to insult you !