Fifteen Years

Fifteen years ago, in 1993, I was an undergraduate student at the University of Western Australia. Sitting in the back of a lecture on Information Theory, I started thinking about Markov Models and language modelling. Back in the lab, I hacked out a quick program in C that tokenised strings of characters into words, fed them into a Markov Model, and then used that model generatively to produce nonsensical, but often humorous, quasi-English sentences. MegaHAL was born.

Over the next couple of years, as I became aware of the World Wide Web, MegaHAL moved online, and became (for the time) massively popular. Thousands of people talked to the online version each day, in many different languages. I improved the performance of the program, releasing new versions rapidly. Due to demand I made the code available under the GPL, and MegaHAL was ported to many different platforms. Work on MegaHAL lead to my PhD, to my involvement in the Loebner Prize, to my work at Lionhead and so on. But I've let MegaHAL live on its own for quite a while now - the latest version is 9.1.1, which was released in early 2004, but I haven't touched the code personally since the turn of the century.

Today I decided that the time is ripe to implement a brand new version of MegaHAL, which I'm calling MegaHAL10. I intend to write it from scratch using everything I've learned in the intervening years, and make it available to talk to online, and for download.

I've started this project by starting a blog. I want to be completely open and honest about what I'm doing and why I'm doing it. I'm hoping that a small community of people who enjoy this kind of project will gather around and offer suggestions and well-meant criticism. I'm looking forward to seeing where this will all lead.