AI Zone Admin Forum Add your forum

NEWS: Chatbots.org survey on 3000 US and UK consumers shows it is time for chatbot integration in customer service!read more..

Steve Worswick wins Loebner Prize 2017!
 
 

I know this news was already announced in another thread, but it deserves more prominent placement here on Chatbots.org.

Steve Worswick and his bot Mitsuku (http://mitsuku.com) were awarded the 2017 Loebner Prize bronze medal for “Most Human Computer” in the annual Turing Test contest held at Bletchley Park, UK.

After 4 rounds of judging chat sessions with bots and human confederates, Mitsuku received the top ranking among the 4 competitors.

The award was announced by the small robot in the attached picture (I hope my attachments work).  Because once again, no competing bot fooled any judge into believing it was a human, the Silver Medal remains unawarded. 

I am proud of the fact that AIML is the only bot language where more than one bot personality using the same technology has taken the prize.  Between me and Steve, ALICE and Mitsuku, AIML has won the contest 6 times.

We can thank Nir Oren for developing a new, simpler, message-based protocol.  Because this was the first year using the new protocol, there were bound to be glitches.  When Steve asked me to modify our LoebnerAB AIML program for the new protocol, I booked a trip to be on hand at the contest in case of the inevitable technical problems.  In my experience it really makes a difference if the contestants can be present to provide tech support for their entries.  I was pleased to meet Will Rayer in person and have the chance to discuss his extensions to AIML.

Many thanks also to Bertie Muller of the AISB for taking on the Hurculean effort of organizing the Loebner Prize contest.  This was of course the first year the contest was held after Dr. Loebner passed away last December, and he was missed greatly.  Prof. Muller did a great job taking on the task, which always requires more work than you might expect.  As the past winners said in our memorial letter (http://www.aisb.org.uk/publications/aisbq/AISBQ145.pdf), the past winners are willing to do whatever we can to help make the contest live on.

Thanks also to the other contest organizers, volunteers, judges, and human confederates, as well as to Bletchley Park for hosting the event. 

Image Attachments
steve_loebner_2017.jpg
 

 
  [ # 1 ]

Great sentiment Richard.
Good work on your tech and congrats again to Steve.

 

 
  [ # 2 ]
Richard Wallace - Sep 18, 2017:

I know this news was already announced in another thread, but it deserves more prominent placement here on Chatbots.org.

Steve Worswick and his bot Mitsuku (http://mitsuku.com) were awarded the 2017 Loebner Prize bronze medal for “Most Human Computer” in the annual Turing Test contest held at Bletchley Park, UK.

After 4 rounds of judging chat sessions with bots and human confederates, Mitsuku received the top ranking among the 4 competitors.

The award was announced by the small robot in the attached picture (I hope my attachments work).  Because once again, no competing bot fooled any judge into believing it was a human, the Silver Medal remains unawarded. 

I am proud of the fact that AIML is the only bot language where more than one bot personality using the same technology has taken the prize.  Between me and Steve, ALICE and Mitsuku, AIML has won the contest 6 times.

We can thank Nir Oren for developing a new, simpler, message-based protocol.  Because this was the first year using the new protocol, there were bound to be glitches.  When Steve asked me to modify our LoebnerAB AIML program for the new protocol, I booked a trip to be on hand at the contest in case of the inevitable technical problems.  In my experience it really makes a difference if the contestants can be present to provide tech support for their entries.  I was pleased to meet Will Rayer in person and have the chance to discuss his extensions to AIML.

Yes, it can’t be stressed enough that there were bound to be some glitches the first time through. I think it is important though to keep moving forward with the new interface protocol, because one starts getting ahead of the stated goal if your contest is about AI and then base the outcome on “fake typing,” or the selection. Time to abandon LPP and work with LPP2 so that everybody can take advantage of the improved bandwidth left over for actually getting the good chat. That’s shiny.

 

 

 
  [ # 3 ]

Nice work Steve and Mitsuku!

 

 
  [ # 4 ]

I wonder if the the prize was lost the most times by the same technology as well? mad

Why didn’t you book a trip to be on hand at the contest for your own contest entry? cool smirk

One of the contest entries is like a drunk old lady in a casino showing up for the free booze.  kiss

 

 
  [ # 5 ]

Also guys, top tip with the new protocol. Don’t make your bot write out a lot of text instantly. Mitsuku delayed her answer depending on how long her response was which one of the judge’s said made her more humanlike. Some of the bots wrote 2 or 3 lines of text in less than a second after the judge had pressed enter.

 

 
  [ # 6 ]

I put that same kind of response length-based delay into the UI that I wrote for Bruce, though he thought the delay was too long (I had based on 200CPM, which is high average typing speed). I think he opted for 400CPM, which is faster than the most people can type, but is still slower than the 750CPM for the world record for sustained typing (typing speeds over short bursts were recorded at over 1,000CPM).

 

 
  [ # 7 ]

I think 200CPM is perhaps a more natural response. That’s roughly 40 words per minute, an average typist. I used to work for an attorney whose secretary could do 100WPM with very few errors. My own estimate would be C + 200CPM, where C = time to realize that a response has arrived + time to read response + time to formulate a reply.

There is another reason to use a slower typing estimate. The slower you reply, the fewer chances the judge has to trip you up. smile

 

 
  [ # 8 ]

It just depends on whether you think your bot stands a good chance of passing for a human. The upside to fast responses is fewer crossing messages and more opportunity to display what your bot is made of. Judges who base their decision on response speed are the same as those who base it on whether the bot makes typos. It would make a poor victory if those were the main criteria.

 

 
  [ # 9 ]

Don said, “It just depends on whether you think your bot stands a good chance of passing for a human. “

Of course I do. And they have.

 

 
  [ # 10 ]

I suppose the context was lost that I was addressing Steve’s advice. I meant “you” as in general “one who participates in this contest”.

 

 
  [ # 11 ]

I apologize to Steve for answering his question. Don, humanity has a lot to offer.

 

 
  [ # 12 ]

No problem. I’m staying out if it as there seems to be a bit of sour grapes directed towards the Loebner Prize and my entry this year for some reason.

 

 
  [ # 13 ]

Yeah, sorry about that. I thought Mitsuku’s performance was good, it’s just that everything that bothers me about Turing Tests seemed to be back with a vengeance this year, including a judge who considered the making of typos important. I think I’ll take a break from this stuff.

 

 
  login or register to react
‹‹ ALexa      Loebner Prize Report 2017 ››