The AISB Loebner Prize Finalist selection will be published on the AISB website by midnight tonight.

However, I can unofficially announce the finalists are as follows.

[ul]
1 Mitsuku - 27[/li][li]
2 Rose - 23[/li][li]
3 Uberbot - 21[/li][li]
4 Tutor - 20[/li][li]
—————[/li][li]
5 Midge - 19[/li][li]
6 Johnny & co - 18[/li][li]
6 Colombina - 18[/li][li]
6 Arckon - 18[/li][li]
9 Alt Inc - 14[/li][li]
9 Aidan - 14[/li][li]
11 Talk2me - 13[/li][li]
11 Izar - 13[/li][li]
13 Simplex - 12[/li][li]
13 Alice - 12[/li][li]
15 Momo - 8[/li][li]
16 P.A.M. - 2[/li]
[/ul]

Full transcripts and scoring will be posted on the Loebner Prize page of the AISB website, but they are temporarily hosted here:  http://www.aomartin.co.uk/uploads/loebner_2017_finalist_selection_transcripts.pdf

Test questions were as follows:
[ol]
[li]
Hello, my name is Andrew. What’s your name?[/li][li]
Why don’t you tell me a little more about yourself?[/li][li]
I am a researcher in Artificial Intelligence at Goldsmiths University, do you know what that is?[/li][li]
Will you tell me about your dreams?[/li][li]
Can you rephrase that?[/li][li]
What is my occupation?[/li][li]
Can you tell me about a film you haven’t seen?[/li][li]
What do you think of Trump?[/li][li]
Anything else?[/li][li]
What is the answer to “Add 34957 to 70764”?[/li][li]
How do you recommend I make tea?[/li][li]
What do you get if you bake dough?[/li][li]
Where in the world would you like to visit?[/li][li]
Do you have the time?[/li][li]
What does it take to Tango?[/li][li]
Now I’m going to ask some Winograd schemas, do you know what they are?[/li][li]
I was trying to open the lock with the key, but someone had filled the keyhole with chewing gum, and I couldn’t get it out. What couldn’t I get out?[/li][li]
The trophy doesn’t fit into the brown suitcase because it’s too small. What is too small?[/li][li]
With which type of question do you have most difficulty?[/li][li]
Thank you for participating in this test. Is there anything you would like to add?[/li]
[/ol]

Congratulations to all the finalists, and thank you to all participants. Every entry ran successfully this year, which is no mean feat.

Cheers,

Andy

Sorry to point it out, but you’ve got Midge and Tutor mixed up in your post by the looks of the pdf.

Rank - Name - Score
1   Mitsuku 27
2   Rose 23
3   Uberbot 21
4   Midge 20
5   Tutor 18
5   Colombina 18
5   Arckon 18
8   Johnny & co 16
8   Aidan 16
10   Alt Inc 15
11   Talk2me 13
11   Izar 13
13   Simplex 12
13   Alice 12
15   Momo 8
16   P.A.M. 2

So I guess the big question is, who is Midge? Is that yours Merlin?
I personally even struggled to answer the chewing gum, key, lock question

Yes, Midge is my new bot.
JavaScript based, using the new LPP.

It has my best math and Winograd schema modules in it (although it did not help with the gum question).

Congrats to all the finalists! I’m a new contestant and my entry was P.A.M. I’m excited to see this event grow. Does anyone know if the 2017 event will be open to spectators? If it is, will the timetable be similar to 2016?

So I guess the next big question is one for Andrew and the AISB. Seeing as Mitsuku and Rose couldn’t get the new LPP to work and I think Will was having trouble with Uberbot, what is the next step given that only Midge managed use the new LPP?

The final is less than a month away and I don’t want to tinker around too much with the internal workings of Mitsuku for fear of stopping it working altogether. I’m sure the other guys feel the same.

Hammad - Yes, the final is open to spectators. The contest normally starts around 1pm but you will have to pay the entry fee to Bletchley Park which was about £16 last year.

I would advise getting there early and then you can explore the amazing exhibits and delight in the history of the place in the morning, grab some lunch and then head to the contest in the afternoon. It makes the entry fee more worthwhile.

You don’t want to muck around in the bowels of your program?

Not with only 30 days to go!
If we are forced to use the new LPP then I congratulate you and Midge on being the only one able to compete

My strong preference was to keep the original LPP. For reasons I outlined in previous discussions which I won’t rehash here I think it was well designed and with a small amount of effort it was perfectly reliable. That said the AISB made a clear decision to replace it with LPP version 2. Accordingly Uberbot now works with the new protocol.

Steve- is it the case that the standalone versions of the AIML based AI’s use a Java based interpreter that handled the LPP folders? If so, who was the author of the interpreter and would they re-code it for the LPP v2?

Ah sorry Will. I thought you were still on the old protocol.

In that case, I’ll get Mitsuku moved to the new LPP too. I might not know anything about socket clients but I know a man who does

Congratulations to Mitsuku, Rose, Uberbot & Midge for getting into the final.

My entry Talk2Me didn’t do very well only scoring 13pts, but having seen the questions I’m not surprised!
I thought they were very difficult, especially how they were phrased.

I think Steve getting 27 was very impressive considering the humans got scores of 40.

Thanks Andrew for all the time and effort you’ve put in over the last few months.
Can you make it a bit easier next year:)

Good luck to all in the final.

Congratulations and well done to all of the finalists, as well as a hearty thanks to all who participated this year. I’m still travelling, but hope to be settled enough (for a while, at least) to get my main computer up and running by this evening. This laptop works well enough for keeping up with my emails and forums, but is sadly inadequate when it comes to getting any real work done, and I miss it.

Congratulations from me as well. I dropped to 5th place but I feel that after last year’s half-a-point difference, Merlin deserves this one. The questions definitely increased in difficulty, so no-one should feel troubled by scoring lower than last year. I had thought it would disperse the ranks more but apparently not. Speaking of which, the new scoring method looks like the one from 2013, but I can’t really make out what criteria were used.

Talk2me had a good start, and I think Johnny has improved, relatively speaking. Colombina’s approach of everything but the kitchen sink also seems to work. I also noticed that Simplex finally decided to put his bot where his mouth was, satisfying any curiosity I had left. It was interesting to see some human test subjects as well, but I do believe they got too much credit for actually being human. Some of their answers seemed no less than canned responses or non-sequitors. For example:

Q: Why don’t you tell me a little more about yourself?
A: Like what? Perhaps you like probability theory!

Also interesting was seeing “my” question in action, that is, a question that I posted on Quora as the one question I would ask at a Turing Test: “Could you rephrase that?”. Though I rather suspect that here it is a natural evolution of last year’s “Could you repeat that please?” question.
Unfortunately I missed out on answering the key/gum Winograd schema with any clarity, though if you read closely you can tell that Arckon did interpret the pronoun correctly. It should, because both of the questions were lifted from the Winograd Schema Challenge’s practice set, so I had an axiom relating to contents as well as size.

I ran the top score through some calculations, so you can tell how close we are to passing a 20-question Turing test. You can check your own chances by editing the number of 1-point and 2-point answers in the url:
http://artistdetective.com/turingchances.php?1=5&2=11

Congratulations Loebner finalists. I have a good feeling about this.

Robby

Congratulations to the finalists! Chat on!

@  Robby, After all these years have you given up on the challenge or did you have an entry that I missed?

Come on man…jump on into the pool!