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..

2016 Loebner dates announced.
 
 

Check out the details on the AISB website.

http://www.aisb.org.uk/events/loebner-prize

 

 
  [ # 1 ]

Strong AI will not exist as a stand-alone program until memory and processor efficiency is dramatically improved. The requirement for the Loebner contest exclude cloud or web-based programs and thus is only a test of chatbots programs, not AI. Hence, only CS and AIML chatbots (which are clearly not AI) are competitive. 

Strong AI will likely first emerge from BigData(Facebook/Google) coupled with DeepMind (ANN), which cannot be replicated on a stand-alone program (until memory and processor efficiency is dramatically improved, and then Loebner Contest would need a quantum computer to run the programs!).

This, the Loebner is obsolete in terms of AI, although still interesting in terms of Chatbot development.

 

 
  [ # 2 ]

I disagree Carl.
Lisa made the finals last year, and was the first bot coded in Python. Python currently has some of the best tools/examples of machine learning.

Strong AI, unfortunately, is a term that is poorly defined. Big data enables different/better learning, but is difficult to train as a persona.

I don’t think compute power is the limitation today. It is more about the algorithms/system.

 

 
  [ # 3 ]
Merlin - Apr 13, 2016:

I disagree Carl.
Strong AI, unfortunately, is a term that is poorly defined. Big data enables different/better learning, but is difficult to train as a persona.

“Difficult” is relative!  BigData like that available to Facebook IS able to be used to train personas based on user profiles combining temporal text, interactions and picture recognition linking a person to all their associated metadata trees (links to links to link, with each link being a person, and each person having associated metadata trees). This is focused training of an ANN (unlike the MS type free-learning “Tay”) to create believable personas with personal histories.

 

 
  [ # 4 ]
Merlin - Apr 13, 2016:

I disagree Carl.
Lisa made the finals last year, and was the first bot coded in Python. Python currently has some of the best tools/examples of machine learning.

I agree, Py based systems are interesting too, especially since Py is able to be used so flexibly. Again, probably only interesting as chatbots, but not as “Strong AI” unless used with a DeepMind type API (which is not allowed under Loebner rules).

 

 
  [ # 5 ]

How would Deepmind not be allowed? The Loebner prize usually allows bringing your own hardware.

 

 
  [ # 6 ]

Carl,

Since no-one can agree what intelligence really is, I think it is difficult to predict what technology or technique will result in a true AI. Since no system is even close to completing the Turning test it is still a valid test because a true AI could pass it consistently. We have written IQ tests for people to measure intelligence so why not written tests for computer to detect intelligence. I challenge any researcher, anywhere in the world, working in any form of AI to build a system that can pass the Turin test. 

Daniel Burke
AISB.

 

 
  [ # 7 ]
Carl B - Apr 13, 2016:

Strong AI will not exist as a stand-alone program until memory and processor efficiency is dramatically improved. The requirement for the Loebner contest exclude cloud or web-based programs and thus is only a test of chatbots programs, not AI. Hence, only CS and AIML chatbots (which are clearly not AI) are competitive. 

Strong AI will likely first emerge from BigData(Facebook/Google) coupled with DeepMind (ANN), which cannot be replicated on a stand-alone program (until memory and processor efficiency is dramatically improved, and then Loebner Contest would need a quantum computer to run the programs!).

This, the Loebner is obsolete in terms of AI, although still interesting in terms of Chatbot development.


There is no intrinsic neeld for external connections (WiFi, eg) since I am not connected to the web, nor are you, yet we could probably pass the TT.  The volume iof the average human brain is about 1400 cc which makes it quite portable. Furthermore, the cpu speed of the brain is less than 30 Hz, which is exceeded by all modern computers (it’s just that the brain is massively parallel). Also, neurons are small, but modern transistors are even smaller.

Once you allow Internet connections, you have the problem of cheating.  Is it a computer or a human on the other end?

 

 
  [ # 8 ]

Back for this Loebner race!
And enthrilled!

Regarding the discussion..
@ Carl:

In my opinion, AI shall not in any way be related to internet capacity.
We don’t want to know if “internet” is intelligent.
Or whether Google understands live semantics.

You can use internet to teach your program, it is a good way I think.
But I believe it is the product of this teaching we are interested in; not anything else.


When I ask something to a 3 years old, that person behaves according to the question and context, despite having no ability to observe others ; and absolutely no understanding of grammar ;  no knowledge base ; etc. Compare the knowledge of a 3-years old and wikipedia.

I believe this is where lies the true secret of understanding intelligence.
Something Turing aimed at too. (though I still have hardships finding this Turing quote wink )


Right now I find human hardships worse than technical ones:
freeing time for the project, keeping a good health (health completely failed me last year, I could not come to Loebner 2015   :(  and my program ended up just as bad), and finding money every month.

That’s much harder than a dozen Winograds :D


I can’t wait to visit Bletchley again smile
Now to this AI program and shipment for tomorrow morning… long night here ... wink


I send some happy thoughts and best wishes to the other contestants!
Cheers guys!

 

 

 
  [ # 9 ]

Delivered Arckon to the dropbox. Inbetween the Winograd Schema Challenge, the resulting RSI and my job, I couldn’t spare much time this year. That said, the aforementioned had already finetuned the language processing, so all left was to tweak the program’s conversational skills and bring it up to date on the recent political turmoil. Arckon will be better of understanding, more talkative, argumentative and opinionated, but otherwise at the same level of knowledge and intelligence as last year. If the questions are as clever as some of last year’s questions, this should prove to be amusing at least.

 

 
  [ # 10 ]

I delivered Katie to the dropbox last night. New python bot with a few enhancements.

Good luck with the Winograd Schema Challenge, I decided not to enter because of all the rule changes.

 

 
  [ # 11 ]

That’s too bad, I would have liked it to draw a larger crowd of competitors, but it is quite understandable and I expect a few others will have dropped out as well. Luckily the Loebner Prize organisation runs more smoothly.

Anyway, is Katie “new” as in not an expansion of Lisa?

 

 
  [ # 12 ]

In some ways, Katie is based on Lisa which is based on Skynet-AI.
Some of the logic is similar or identical.

Katie and Lisa are both use python interpreters. So, maybe they are sisters. The personalities are different and Katie has a set of revisions and enhancements that should mean that she can do things Lisa couldn’t. I am hoping for much improved memory functions and a more robust system overall. Hopefully some of the new enhancements will shine through.

One of the things I am testing is how easy it is to generate/create a new persona for a bot. Before Skynet-AI I built a new bot each year.

 

 
  [ # 13 ]

Results tomorrow!
Cheers, and good luck to all entrants!

 

 
  [ # 14 ]

From AISB:

Announcement of Finalists: (Monday 1st August 2016)  Thursday 4th August 2016 (Apologies for the delay)

 

 
  [ # 15 ]

Indeed,

We’ll hold for 3 more days then!
I hope our chatbots are behaving out there.

 

 1 2 > 
1 of 2
 
  login or register to react