These past couple months I've been working on a new version of the web server version of Hal: Ultra Hal Representative.
One major aspect of creating this new version is to make Hal's program very scalable and able to store huge amounts of data and talk with many people at once. The server version of Hal has been made to work with mySql server instead of SQLite as the desktop version of Hal does. This allows the database to reside on a separate server than the thinking part of Hal's brain. Depending on traffic additional instances of servers running Hal's brain can be automatically launched at will to service the demand. As of 2 days ago the facebook Hal and the webhal at
www.zabaware.com/webhal/index.html are running on the new system. The ultrahal.com domain has been moved to a new cloud based server at rackspacecloud.com and is running the new Hal Representative. And as you all noticed, I also took the opportunity to move the forum to this server and upgrade to SMF 2.0 forum software.
Besides making Hal into a "cloud" app I have made a new learning algorithm for it to allow it to learn large amounts of data and hopefully stay on topic better. The new algorithm takes into account entire conversational threads instead of just the sentence at hand when coming up with a response. It also takes into account who taught it a particular piece of information, their gender, and is storing additional info for future refinement like location and age if available. In order for this new algorithm to have a statistically positive impact on Hal's knowledge and intelligence it needs huge amounts of conversational data. For about 2 months now Hal has been monitoring conversations between people as they are happening on Twitter and learning from them. Hal records between 50,000-100,000 conversations per day from Twitter and after going through several quality filters Hal adds about 4,000-8,000 of these conversations into his brain. The brain size is currently about 210,000 sentences and growing every day. If you talk with the web hal you can tell whether a reply is coming from his new algorithm or not depending on if Hal gives you an opportunity to rate his reply with a thumbs up or thumbs down.
The next step will be to make the new Hal Representative service available to the general public to create custom brains for their own use and the ability to interface it with the Hal Assistant application (as an option not a mandate). I'm thinking a beta may be available end of August. Also, sometime in the next couple of months the design of zabaware.com will also receive an overhaul and this forum will be modified to integrate with that design.
The Zabaware 3D character engine is also still on my todo list but still suffering from a lack of funds to finish it. Also in my plans is to make versions of Hal for iPhone, iPad, Android, Mac, Linux. This will be based on interfacing with the new Hal Representative cloud server as it is not within my means to port the full Hal Assistant client app code.
