Bruce Wilcox Interview - Loebner 2014 winner

Bruce Wilcox Interview - Loebner 2014 winner

It was that time of year when once again some of the world's best chatbots went head to head in the annual Loebner Prize Competition.

Making it through to the final four this year were Mitsuku, Uberbot, Rose and Izar. The event itself took place at Bletchley Park in the UK on Saturday 15th November.

There were four judges including James May the television presenter, well know for his role in the hit motoring show Top Gear.

Despite some technical glitches that have attracted a lot of discussion; the judges arrived at a winner. So congratulations go to Bruce Wilcox for his third win. Here follows our annual interview and many thanks to Bruce for taking part once again.

How did it feel to win the prize again ?

It would have felt good, except for the random chaos of the contest, where lines of judge input were broken up into fragments. So it was hardly a real conversation. But the Loebner's is always somewhat random and I figure we have at least a 50/50 chance each year, so winning this year after losing 2 before that and winning 2 before that is more or less par.

How long did Rose take to develop ?

Rose has been developed for 2.5 years (part time obviously). The first year was a scramble because we had to abandon Suzette, who wasn't our property, so only had time to start on Rose in the 2nd half of that year. Rose was then shelved the next year for Angela. So Rose was picked up again and dusted off since we didn't own Angela, and continued again into this year.

What makes your bot different from the rest ?

Rose has a more competent/comprehensive natural language engine which makes her more likely to accurately recognize the user's input meaning and provide an appropriate output. She has a stronger ability to handle conversational idiom (like tag questions, multi-sentence input in response to a question like "Do you have pets" --> user says "yes. I have a dog"). Rose's answers are more likely to be informative of her life rather than mere quibbling/stalling. And Rose can take charge of a conversation in addition to merely responding to an input.

What feature of another bot do you wish your bot had ?

I would like to have web search (e.g. Google access) like Skynet or such. Not that it could be used  during the Loebner's. Suzette originally was designed more like Cleverbot and a had Google search capability to find personal answers to questions about her, but then Google starting blocking our queries and I changed technologies to what eventually led to ChatScript.

How much of a struggle is it and what is the incentive to keep on improving ?

People are always suggesting new uses that require new features and I keep exploring new ideas. I created document mode to be able to read Wikipedia and data mine information using a specially designed chatbot, for example. My incentive generally is that I love applied research, so I create things that I or someone else can use. The most recent significant capability addition this year was "theater mode", where multiple chatbots could interact with each other and the audience/user could also address either or both when invited.

What in particular pleased you about Rose's performance ?

She had interesting things to say and could stay on topic despite the breaks in user input.

Knowing what you know now, if there was one thing that you could have changed about your bot to be better prepared for this contest, what would it be ?

If we'd known James May was going to be one of the judges, we would have put in topic on shirts and Top Gear and asked questions to seek him out.

What was the thing that got you started working on chatterbots ?

I initially planned to work on body language and gesture recognition for Blue Mars, but delays in their development caused me to start a separate chatbot project while waiting for their avatars to be usable.

Did you use any A.I. techniques like semantic analysis, emotion analysis, learning or inference, and how much use were they in this contest ?

I use all of these, but they were of no real use in this year's Loebner.  

Do you plan on developing Rose further ?

The capabilities underlying her certainly. Whether or not Rose herself has her story extended or my wife and I switch to an entirely new persona remains to be seen. We have done Suzette, Angela, and Rose. Maybe it's time for someone new.

What do you see for the future of Rose or bots in general ?

Bots continue to take over the world. The latest, Amazon Echo, is a permanent household bot that can get information for you or provide entertain. Currently not a personality, but maybe in the future.

Is there any advice you would like to pass on to budding bot makers out there ?

Get lots of testing- smoothness pays off as much as general capability. Develop a viable personality. Have a lot of persistence if you intend to compete. Your competitors will have been developing their bots for years (often a decade). Use ChatScript (Loebner winner Suzette was 2 years to develop, the first commercial release of Angela was 5 person months, Rose was 3 years).

Other resources

Society for the Study of Artificial Intelligence and Simulation of Behaviour - More details and transcripts.

Videos available on YouTube, filmed by Daniel Burke...