Beastie Bot 2
After getting irritated with having ALICE bot brain mixed with Beastie bot 1, I started all over and created Beastie bot 2. Much like Beastie Bot 1, BB2 replied to a few responses with beastie boys lyrics. Such as:
Human: hello
Beastie Bot 2: So whatcha whatcha whatcha want.
Human: sing for me
Beastie Bot 2: There's a commotion that's goin' down, I talk more shit then drown the clown.
Human: How old are you?
Beastie Bot 2: 24 is my age 22 is my gauge.
Human: you've got to fight
Beastie Bot 2: For your Right..
Human: to Party!
Beastie Bot 2: word.
Human: intergalactic planetary
Beastie Bot 2: Planetary Intergalactic.
My biggest challenge in creating this bot was 1) trying to anticipate what the user might input and 2) finding an appropriate rap lyric to match the user's input. In order to try and figure out what people might input into a chat bot, I started chatting with iGod. To try and find out how to response appropriately, I chatted with iGod and search the internet for Beastie Boys lyrics. I then responded to iGod with the BB lyrics. Sometime the responses made sense. Most of the time they didn't. The biggest problem with using BB lyrics was that most of the lyrics were spoken from first person, about themselves, which didn't help when you were trying to start a conversation about something else. This bot wasn't really designed for conversation. Rather it was design to just respond to simple statements. Ideally I wanted to program the bot to be able to handle a freestyle rap battle, as exemplified by this video. But it's comprehensive abilities are limited. This leads me to my conclusions over chat bots. I've probably made a point of this before, but I'll reiterate my opinion here. I don't believe that chat bots like ALICE will advance the future of NPCs. Rather I feel that the opposite. I feel that chat bot are a clumsy interface between user and computer. I believe that natural language is a too sophisticated and too broad a language set for the ALICE bot to understand. It's much like expecting a baby to understand adult English. But unlike the ALICE bot, the baby, at least, has an emotional state that is clear communicated, which makes me believe it's more alive than the ALICE bot. Where I do think the future of NPC is going is pretty much aligned with Bruce Blumberg's essay on "Anticipatory AI and Compelling Characters" on Gamasutra. Bruce was an Associate Professor at The Massachusetts Institute of Technology (MIT) Media Lab where he served as the Director of the Synthetic Characters Group. Currently he leads Blue Fang's Synthetic Animal Team. In this essay Blumberg believes that creating support for anticipatory behavior in NPCs is the key to creating compelling and believable characters. Anticipatory behavior involve three things. 1)"Making perceptions perceivable" 2) "Making expectation perceivable" 3)"Making impending changes in motivational state perceivable." These type of behaviors are already being applied in the latest games. Nintendogs is one of them, and its success is proof to me that NPCs with more sophisticated mental states and behaviors are far more compelling and believable than NPCs with a sophisticated natural language parser. Hands down I have an easier and more entertaining time interacting with a Nintendog than I would with an ALICE bot.
My opinion that Chat bot A.I.s are insufficient, because they cannot grasp the symantic mean of words can be supported by the Chinese Room Thought Experiement. This arguement was created by John Searle to argue that computer A.I. cannot "think" like humans do. I partially agree with him in that natural language parser cannot think, but I disagree that computers do not have that potential. Eventually some thousands years later, we will be able to model the brain on a molecular level in a computer. When that time comes, we'll see whether not a computer can think. Of course that is if any one still cares to try a make a computer think like a human.