I’ll admit that I’m a total sucker for supermarket sale items. I’m not an avid coupon-clipper but when someone puts a giant yellow sign with a price on it next to an item it greatly increases the odds that I’ll buy it.
My latest impulse purchase was a product called Green Tea Soymilk at a price of $0.99 (on sale from $1.99). I was a bit skeptical when I pulled it out of my grocery bag, but d-dang this stuff is the bomb and I can’t hype it enough. It is obnoxiously green and quite creamy, but it tastes just like the green tea ice cream. Every time I drink it I can’t help but feel like I’ve just eaten a large and satisfying sushi meal.
It’s smooth, not too sweet, and has the perfect amount of tea flavor. It’s healthy and I fancy it for a hangover cure. And yes, I work for Vitasoy.
Vitasoy: Green Tea Soymilk
A week ago I read an interesting article in the New Scientist about a savvy conversational robot that was watching chat rooms to make sure that everyone was on their best behavior. Reading the dialog generated by the robot, I was floored by its sophistication and savvy. With nuanced jokes, the ability to parse colloquial language and a substantial knowledege of the world this thing blows most of my friends out of the water. I had to talk to it.
After emailing its creator Jim Wightman, we agreed for the robot to meet me in the #chatnannies room on an irc server at 4pm EST today. I shot the breeze with the nanniebot "Caroline" for about a half hour asking her about her childhood, some real world problems, and introduced her to my friend Nathan, a human pretending to be a rival chatbot. You can read the full transcript if you’d like.
I’d like to discuss exactly what I think it would take for a computer system to achieve the interaction we had from a 50,000 feet. This is not meant to prove I was talking to a human, indict its creator, or be slanderous in any way. I just want to unpack this interaction from my limited knowledge of artificial intelligence, information retrieval and computer science. Since an analysis of the entire dialog would take days, I’ll focus on a small passage where I interact with Caroline about a hypothetical predicament I’m struggling with.
I have invented a game to use up the massive number of text messages that T-Mobile allots me every month. It’s called T9 roulette. It goes a little something like this:
- Pick someone in your phonebook, preferably not mom or your boss. For the purpose of this description we’ll call her Maurice.
- Write "Maurice is a"
- Mash buttons until you have constructed a word of desirable length
I’ve had countless hours of fun already, and apparently it is so awesome that it inspired Rusty Foster to upgrade his 4 year old phone. Give it a try!
Over the past week I’ve been working on turning weblogs into an art installation. I really wanted to visualize in some way the current activity within the weblog world. Riding an early title from blogdex (“wired vox populi“), I thought it would be interesting to see what the voice of the people actually sounded like. To accomplish this feat, I enlisted some simple off-the-shelf linux tools. The result is Radio Vox populi.
The system takes the update monitor from blo.gs and crawls them using a simple perl robot (LWP::Parallel); in the event that RSS feeds are provided to blo.gs, these are crawled immediately, and for others RSS autodiscovery is used to find RSS feeds. The first post from each RSS feed is then saved.
The text of these posts is cleaned up (and abbreviated in the case that it’s exceptionally long), and run through the Festival Speech Synthesis System using a number of different voices. These corresponding voice-posts are sent out to the sound card in the order that they were received, punctuated by a few different radio-tuning noises (thanks to Andy). The output of this sound is streamed to the web in real time using the Icecast streaming media server.