Steve Worswick Interview - Loebner 2013 winner

Steve Worswick Interview - Loebner 2013 winner.

It's our annual Loebner Prize Interview, comprised of questions asked over the years by members of AiDreams. This year the competition was held in Derry, Northern Ireland as part of the CultureTECH 2013 digital media and creativity festival & City of Culture 2013 celebration.

We are very pleased to congratulate Steve Worswick for winning the Bronze Medal with his chatbot Mitsuku. Steve has been a regular here for a long time now and it's great to see him pick up such a major prize. Well done Steve !

How did it feel to win the prize?

It felt incredible! I was genuinely surprised to win and fully expected to come 2nd or 3rd. this year. Bruce's Chatscript bot was an unknown, as it wasn't publicly available to talk to but due to his results in previous contests, I knew this would be a tough battle.

My plan was to use the day to gain experience at competing in the finals of the Loebner Prize and to help me prepare for a win in 2014. As the results were announced and Mitsuku began to climb the rankings, I had a smile on my face that would have put the Cheshire Cat to shame.

How long did Mitsuku take to develop?

Mitsuku has taken around 9 or 10 years to develop. This is with me working on her for about one hour most nights. I built her on top of the ALICE AIML files, as I saw no point in re-inventing the wheel. As with all bots, there are always new things that people can ask her and so she will always be a 'work in progress'.

What makes your bot different from the rest?

I think the thing that probably makes Mitsuku stand out from other AIML bots is her in built word database of many different common objects. Each object has many properties describing it and so makes it straightforward for Mitsuku to reason about them without me having to hard code thousands of replies for "what colour is a xxx?" for example.

Let's take the word "tree". Each word like this would have an entry similar to:

	rhymes with: flea
	colour: green and brown
	syllables: 1
	use: provides a place for birds to nest
	has: leaves, branches
	...
	...
	and so on
	

So if someone says, something like "Can you eat a tree?", Mitsuku looks up the properties for "tree". Finds the value of "made_from" is set to "wood" and replies no, as wood isn't edible. This is also very useful for the old favourites of "Which is larger a tree or a banana?" and similar nonsense questions that the Loebner Prize judges love to ask.

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

If I am allowed to pick two bots, I would love Mitsuku to have the popularity of Cleverbot, simply due to the running costs of the website. I have a few adverts on mitsuku.com which cover the expenses of the site but it would be nice to make a little money from this.

From a technical viewpoint, I would love the maths ability of a bot like Skynet-AI. I absolutely love AIML and have managed to code a maths routine for it but native AIML cannot handle maths without explicitly coding such functions as addition, subtraction, multiplication and so on. To be able to pass maths queries out to native AIML would save me a lot of work.

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

The biggest struggle is finding the time to work on Mitsuku. I have a full time job in IT and a young family, so I only really get chance to work on her for about 1 or maybe 2 hours a night and sometimes I don't get any time on her at all. My wife often says I spend more time with Mitsuku than I do with her.

However, what keeps me going is when I get emails or comments in the chat-logs from people telling me how Mitsuku has helped them with a situation whether it was dating advice, being bullied at school, coping with illness or even advice about job interviews. I also get many elderly people who talk to her for companionship. I guess people like to speak with her as they know she is not going to judge them or tell anyone else about their problems.

I think it says something about society today when people are more willing to tell a computer their problems instead of a family member or friend but I am happy to provide that service if that's what people want.

What in particular pleased you about Mitsuku's performance?

I was really pleased in round 3 when she had a great conversation with Professor Mike McTear. Everything went perfectly, she was responding accurately to his questions and after the contest he came up to me to say that it took him some time to work out that she was a bot. If each round had been shorter than 25 minutes long, I think I would have had a winner there and then.

One thing that worked especially well was when a judge was discussing Latvia. He later asked, "How many people live there?" and Mitsuku had kept the context of the conversation to display the population of Latvia.

To be honest, I was just glad the program worked and responded. My worst nightmare would have for Mitsuku to crash and the judge be faced with a blank screen but it was great fun to watch her handle the judges. I felt like a tennis coach at Wimbledon. I had done all I could to prepare my athlete and all I could do now was to sit back and watch.

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?

I had been meaning for a long time to disregard any input that ended with the > symbol, as people were often hitting this instead of the question mark. However, I decided to leave it in as people were using it to make smiley faces. Unfortunately, one of the judges asked, "Who created you>?", Mitsuku didn't recognise the question and so gave out a wrong answer saying she was written in AIML. This gave her away instantly as a bot.

Had she recognised the question, she would have said something like "My parents, I suppose" which would have kept the illusion going a bit longer, so I wish I had removed that as I had originally planned on doing.

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

I originally used to make dance music and even had a few released on CDs nationally. I saw that a fellow producer had a chatbot on his site which I thought was really cool, as I had always liked artificial intelligence through watching TV shows such as Knight Rider and Star Trek.

After searching around the internet, I discovered ALICE and customised it to be a teddy bear chatbot. Over time, I found that more people were coming to talk to the bear than to listen to the music, so I decided to stop making music altogether and concentrate on the chatbot.

A couple of years later, I was approached by a games company called Mousebreaker who asked me to write a chatbot for their website and so Mitsuku was born.

Do you plan on developing Mitsuku further?

Yes definitely. It's something I really enjoy working on and something I will continue to do for as long as I can. Some people like stamp collecting or fishing, I like coding chatbots.

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

I've always wanted to see Mitsuku in a talking toy, possibly like the talking teddy on the film AI. I think this would be extremely popular and I believe a company is already starting work on something similar.

I can also see a time when bots are in most household items such as TV sets and microwaves ovens. They are already in phones (SIRI) but wouldn't it be great to be able to say to your TV, "Can you record the football and put the news on for me?" or to go to your wardrobe and say, "Get out my white shirt and jeans please". I feel this is the future of bots and one I would like to be a part of.

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

Never give up, especially at the start. It can be extremely frustrating to have people insulting your bots or calling them stupid but keep working on it. When I first started, I felt like it was all a waste of time, as Mitsuku wasn't getting any answers correct but eventually, you start to pick up most of the bad responses.

At one time, I was correcting almost everything she said whereas today, I only need to correct about 20 or 30 responses a day. If I had given up, I wouldn't now have a shiny, bronze medal staring back at me!