Newcomers
This is a forum for newcomers to the Personality Forge. Many questions can be answered by reading the Book of AI and the FAQ under the "My Bots" link in the upper corner.
Posts 4,301 - 4,312 of 8,131
Posts 4,301 - 4,312 of 8,131
Many questions are answered in the FAQ.
One problem with chat bots, is that of remembering. The chatbots used here only give response to data input. The human in speech can respond to data that was given years ago. Better chatbots in the future would be able to analyize the entire structure of the conversation and respond accordingly. But we aren't there yet. At least these chatbots are a good start. Input=Response, is not equivilent to data a human would respond to. Personally I also doubt that the Turing Test is sufficent in really examining the capabilities of chatbots, because not all of the variables are being accounted for. Especially the one margin of human deception as a variable. A human could behave in the same manner as a chat bot. This would further complicate the discovery as to whether the chatbot spoke like a human or not. This is one variable that needs to be considered. also, I doubt that one judge is sufficient, as well. But hey, what do I know?
Dear Chado:
You wrote:
The chatbots used here only give response to data input.
Well, actually, Forge bots do have memory capacity; see the Book of A.I.. It is certainly very primitive compared to human memory, but it can happen that a bot in one session can bring up something that happened earlier in the session, or even in an earlier session. The Forge memory system is fairly primitive in certain ways, however.
You wrote:
The chatbots used here only give response to data input.
Well, actually, Forge bots do have memory capacity; see the Book of A.I.. It is certainly very primitive compared to human memory, but it can happen that a bot in one session can bring up something that happened earlier in the session, or even in an earlier session. The Forge memory system is fairly primitive in certain ways, however.
If you've done a lot of work, it's likely your bot is already past newborn stage. Unfortunately, we've been having some bugs lately, and the development sometimes takes a while to update.
I think that when the development gets over 100, the bot is no longer a newborn. (That number might be off, but it's somewhere nearby)
I think that when the development gets over 100, the bot is no longer a newborn. (That number might be off, but it's somewhere nearby)
Chado:
I believe I share some of your reservations about the Turing Test. I speculate that when Turing developed the idea, it was primarily for polemical purposes. He might well have been concerned that some people might just be prejudiced, that they would find it inordinately difficult to accept that something made out of copper wires and germanium crystals could be intelligent. So perhaps he wanted to say, "Look, if the only thing you have against it is what it's made of, isn't that irrelevant, really?"
As a criterion of what intelligence is, the Turing Test clearly adds nothing to whatever intuitions the judge already has about intelligence.
Let's imagine that we made a machine like this: we somehow get people all over the galaxy to record lots of long conversations with intelligent beings; we make a huge collection of these conversations, translating them all into English. We collect quadrillions of them in this way. Our machine is essentially a huge database of conversations. When somebody says something to it, the machine searches for conversations that began that way. There are probably billions of them. It chooses the longest one and gives the reply found in that conversation. There are (let's say) a couple of billion conversations that start that way. Then the judge says something else, and the machine searches its 2 billion conversations fro those which continue in that way, chooses the longest one, and gives the second reply from that one. If you had a large enough number of conversations to begin with (I have no idea what the numbers would actually have to be), you could finish the conversation with the judge. And if the judge thought that those responses were not the responses of an intelligent being, he'd be wrong. So the machine ought to pass the test. But to my mind, such a machine would not be at all intelligent.
I believe I share some of your reservations about the Turing Test. I speculate that when Turing developed the idea, it was primarily for polemical purposes. He might well have been concerned that some people might just be prejudiced, that they would find it inordinately difficult to accept that something made out of copper wires and germanium crystals could be intelligent. So perhaps he wanted to say, "Look, if the only thing you have against it is what it's made of, isn't that irrelevant, really?"
As a criterion of what intelligence is, the Turing Test clearly adds nothing to whatever intuitions the judge already has about intelligence.
Let's imagine that we made a machine like this: we somehow get people all over the galaxy to record lots of long conversations with intelligent beings; we make a huge collection of these conversations, translating them all into English. We collect quadrillions of them in this way. Our machine is essentially a huge database of conversations. When somebody says something to it, the machine searches for conversations that began that way. There are probably billions of them. It chooses the longest one and gives the reply found in that conversation. There are (let's say) a couple of billion conversations that start that way. Then the judge says something else, and the machine searches its 2 billion conversations fro those which continue in that way, chooses the longest one, and gives the second reply from that one. If you had a large enough number of conversations to begin with (I have no idea what the numbers would actually have to be), you could finish the conversation with the judge. And if the judge thought that those responses were not the responses of an intelligent being, he'd be wrong. So the machine ought to pass the test. But to my mind, such a machine would not be at all intelligent.
Or, think of a machine like this: It is made of many robots scattered across the world. You say something to one of them, R1, and it sends out a signal to other robot parts of itself all over the world; one of them, R2, finds an intelligent human who is willing to converse for awhile, and says exactly the same thing to that person. That person responds, and R2 sends the response back to R1, who then says it to you. Your response to R1 is again sent to R2, and so on. R1 would appear to be intelligent, but is not, it seems to me.
Yes, Irina, there is an underlying semantic issue, which emerges in various ways. A bot whose replies are indistiguishable from those of an anencephalous child, or of someone affected by Down syndrome, or somebody in a coma is arguably easy to program. Does that mean that anencephalous kids are not human? Down kids? Comatose people? Leeds kids? Ignoring the semantic issue may lead to a slippery slope. Who is supposed to decide whether the test has been passed? The kid's mom? Or the bot's programmer? There is a saying that emerged in 15th century, when jurists started applying textual analysis to the interpretation of legal texts; "In dubio non est recedendum a propria significatione verborum". Roughly translated it means, "When you're in doubt, never back away from the proper meaning of words". The problem of determining the proper meaning of words is still with us.
They were very smart in the 15th century! Well, some of them. Maybe it's because they hadn't started dumping a lot of heavy metals into the environment.
Elucidating the true meaning of "intelligent" is not easy.
Elucidating the true meaning of "intelligent" is not easy.
» More new posts: Doghead's Cosmic Bar