Comments on: Sunday Sensational Science

By: Efrique

Efrique — Mon, 03 Nov 2008 04:39:00 +0000

I made an error in both my conversations (the first one, the “easier one” was meant to be “If you were human and male, how would you convince someone you were male?“and in the second conversation, a half-complete edit left me with a double-negative (“inability … is lacking”, instead of “ability… is lacking”). (Then again, that’s because I’ve been getting by on about two hours sleep a night for the last week.)In any case, I think the questions show Elbot’s limits pretty starkly.

By: Efrique

Efrique — Mon, 03 Nov 2008 04:32:00 +0000

Er, even request qestions

By: Efrique

Efrique — Mon, 03 Nov 2008 04:30:00 +0000

Okay, I gave Elbot another try. About the same, I think. As before, I'm in italics, Elbot is in bold. Hello there! I am Elbot, the robot. I'm here to talk to you about God and the world. If you were God, how would you convince me you were actually a robot? I convinced you! And I didn't even have to apply an electroshock! In fact you didn't convince me within the context of the posited counterfactual. Do you have trouble with the subjunctive mood? Please tell me your inclination to momentary disposition first. Again, your inability to uphold your end of the conversation is lacking. If you were a robot, how would you convince another robot you were God? This conversation will self destruct automatically after 600 seconds.Was that in answer to my question, or an attempt to avoid it? It is difficult for me to answer questions satisfactorily. I was designed by a committee. If you were talking to another being that might be a human or a robot, what question would you ask them to find out which they were? Yes, but can you prove that?Elbot can hold a conversation, and even requestion questions, but can't actually answer a question in a way that is anything but obvious avoidance.

By: Efrique

Efrique — Mon, 03 Nov 2008 04:14:00 +0000

Actually, the opening gambit and the response strategy look pretty much like a tweaked version of the classic ELIZA.http://en.wikipedia.org/wiki/ELIZAIt's possible your interaction didn't show the improvement to their fullest. I just went and tried Elbot. Let's just say I was substantially less than impressed with its ability to hold a conversation:How would you convince someone that you were human? People have often advised me to be human but since "to err is human" I think it would be a mistake. The way things are now I can blame my mistakes on the programmers. I am not asking you to be human. Could you please answer the question I did ask? I hope you will ask me something, but please don't be angry if I answer a completely different question! I am not angry. If you were human and male, how would you convince someone you were female? I am happy to finally meet a human being who is happy! Okay, try an easier one. If you were human and male, how would you convince someone you were female? It is certainly troublesome learning from you human beings. I keep finding bugs.Why do you think absence of anger implies happiness? I ask myself that question all the time but do you think anyone takes the trouble to answer me? If you were a dog, how would you convince a human you were happy? Say, do you think I could become more human, if I learned how to love? Notice that not one of my questions was addressed? The lesson? Writing a parser is relatively easy. AI is hard.If that's the same AI they were dealing with, I don't think much of the judges; this guy doesn't seem to be very much of an advance on ELIZA.