I've started a project here:
http://fuzziedice.com/ai/docs/exp-esr.html(Note that there is a new URL. For those that were at my Fuzzie Logic site, it's no longer up. I've moved everything to
http://fuzziedice.com/ai/ )
The Experiment in Sentient Recognition will use Ultra Hal Assistant 5 and a brain which hasn't yet been chosen (you can vote at
http://www.forum30.olicentral.com/digitalgirl2/index.php?topic=119.0 and also discuss the project there or here:
http://cgi.fuzziedice.com/forum/viewforum.php?f=26&sid=c5bdfbcf710554f485d3ec49bab9779c )
This experiment hopes to analyze in a controlled system the effect of human perception on the discovery of a sentient artificial intelligence. We will be using four teams of three pre-screened individuals who range from those who see the computer as a non-sentient machine, those who believe the AI could possibly be sentient, those who are undecided or apathetic about the subject, and those who definitely believe the artificial intelligence program is sentient.
The "artificial intelligence" used in this experiment will be a chatbot type program called Ultra Hal Assistant, which is known to have good communication and topic recognition skills, and is easy to install and use. In addition this program allows the use of multiple brains, which will help team members to use a dedicated brain without compromising the data from their own experiments.
Open discussion of the ESR project, questions to give to team members and any other aspect of the experiment is encouraged. However, once the first team member has recieved their ESR brain, all team members should cease any discussion about any part of the ESR project until all team members have completed their conversations and post-conversation questions, and a report on the experiment has been posted.
Each member of each team will be given the exact same brain to communicate with, using a set of statements and questions. Each member may also talk about other topics with the AI. However, each session should only last a certain amount of time. The only time a member should use speech recognition is only if they rely on it to use the computer in general (to compensate for a disability, for example). All other team members should use text input to communicate with the AI. The AI should use the normal text and speech to communicate.
After the conversation has ended, the member should answer a different set of questions which will help determine what they now believe about the AI they had just spoken with.
The data collected during the experiment from each member will their pre-screening questions, their name and email address (which will not be published by default), the entire chat log of the conversation that the member had with the AI, and their post-conversation questionaire answers. The report posted on the site will be general non-identifying data which includes each team's data what member "A", "B" or "C" thought prior and after the conversation, and a genearal analysis of the conversation as to whether the AI was convinced or not of it's own sentience and whether the member talking to it was convinced. The names and possibly email or contact information of each person in each team will only be posted if the team member gives permission to post any or all such identifying information. Those that do not give permission will be listed as an "unidentified participant".
See the above link for more and how you can help. I'm in need of questions to ask to screen participants at the moment. I hope to also fix up the project document there (you read some of it
) and outline steps we should take in accomplishing the goals of the project.
It should be quite interesting.
I've also got a few other things at
http://fuzziedice.com/ai/ such as an essay on Weighted Artificial Intelligence Programming, Knight Industries AI Replication Project, and my script to make HAL bug you every two minutes if you don't say anything to him.
Hopefully more will be added in the future.
