Rob Medeksza Interview - Loebner 2007 Winner

UltraHal - Third Time Lucky !

ultrahal It's that time of year again, when after many monthsand even years of hard work, programmers finallyoffer their chatbots up forjudgement intheannual Loebner Prize competition. Although there were only three entrants this year, competition was still stiff, especially considering one of the entrants was the two-time winner Rollo Carpenter. However, by the final reckoning of the competition that challenges programmers to create the most human chatbot possible, this year itwaswon by none other than Robert Medeksza of Zabaware with his bot 'UltraHal'. I am very glad to say that Robert kindly responded tosome questions I presented him with shortly after hissuccess. So without further ado here is the interview....

Firstly congratualtions on winning this years Loebner Prize. Last year AIDreams interviewed Rollo Carpenter and I am hoping you can spare a few minutes responding to a similar exercise. So here are the questions....

Thanks for the congratulations. I'd be happy to answer your questions.

How did it feel to win the prize ?

It felt quite good. My first time competing in the Loebner competition was in 1997 after I first founded Zabaware. At that time Hal placed 5th place out of 5 programs. I was disappointed but not discouraged and kept on working on improving Hal. It was not until 2006 that I entered Hal again, this time winning 2nd place. And now finally in 2007, Hal has placed 1st. So it did feel like quite a good accomplishment after 10 years of working on it.

Was your UltraHal the same system as you offer for purchase and if the winning system was different, what (if any) enhancements did you make to it ?

It was based on the Ultra Hal Assistant 6.1 brain with some important modifications and enhancements. In 2006, I pretty much entered the default Hal 6.0 brain, so I knew from that experience that wouldn't work to win. Hal, by default, does not try to pretend he is a human with any kind of life history and experiences. The purpose of the competition is to fool the judges into believing the program is a human, so Hal had to be modified to do this. I taught Hal over 3000 facts about himself including where he works, where he lives, descriptions of his family, his religious beliefs, etc. The Hal in the competition was called Steve and is 31 years old (born 5/12/1976), lives in New York City, works as a network technician for NBC. In addition I had Hal read wikipedia entries on New York City, Hugh Loebner, Alan Turing, and the Turing Test. All of this information I put at a higher priority in Hals brain to make it more likely he would use these responses rather than his default responses. This modification seemed to help Hal out in 2 of his conversations with judges, the 2 judges that scored Hal the highest.

Were you disappointed by the lack of competition ?

I would definitely like to see more bots entered into the competition. There seem to be many good bots around that aren't being entered. I hear Dr. Wallace may enter ALICE next year if he modifies it to work with the rather unique Loebner protocol. However, the 2 other competitors that entered were very worthy competitors. Rollo Carpenter's Jaberwacky for instance has a brain database about 150 times larger than Hal's. This actually seemed to be a disadvantage for him during this competition as it sometimes took his bot a bit long to answer a judge as the computer probably wasn't powerful enough to handle that amount of data smoothly.

Are there any special moments that you experienced in thecompetition ?

It is nice meeting with like-minded individuals interested in AI technology and chatterbots. Overall, it was a positive experience.

What in particular pleased you about your bot's performance ?

It was good seeing Hal responding correctly with one of the 3000 responses he was taught specifically for this contest. I was also pleasantly surprised at times when Hal answered a difficult question correctly. For example, at one point a judge asked Hal what color Kermit the frog is and Hal knew the answer.

Were there any moments where you were disappointed by UltraHal's responses ?

At times, Hal gives seemingly random answers to the judges queries. The judges may either find this as being spontaneous or think it is a completely off the wall response. Also, even though Hal knew what color Kermit was, he didn't know Miss Piggy was his girlfriend.

Did you learn anything that will help you with further development of UltraHal ?

Yes, the conversation transcripts from all the bots will be very helpful for further development.

Last year we asked Rollo Carpenter if he had any advice to pass on to budding bot makers - is there anything you would like to suggest ?

It takes a lot of time, patience, experimentation and hard work. But if its what you like to do, there is still a lot of room for new technology.

Finally, how do you see the 'world of bots' progressing in the coming years - care to share any insights ?

I see this technology exploding and becoming mainstream in the coming few years. I know many large companies are working on this type of technology waiting for the right time to enter the market with a bang. For example, Zabaware is working on exciting technology with some rather large corporations. I can't give any details because of non disclosure agreements, but I can tell you it has great potential.

Many thanks to Robert for findng the time to take part in this interview and once again - congratulations for being this years winner.

  • For a transript of UltraHal's conversations with each judgeplease follow this link...
  • To view video taken at the prize please follow this link...