FFAI Turing Test

1. Loebner Prize

1.1. Basically, a Turing test.

1.2. Humans and chat bots.

1.2.1. https://www.youtube.com/watch?v=6OYXJNBjzsc

1.3. 2.5 minutes each for computer and human.

1.4. Transcript (2008)

1.5. 2008 Loebner Results

2. issues

2.1. unrestricted or restricted subject matter?

2.2. knowledge / background of the judges?

2.3. "Artificial Stupidity" - humans are often characterized by their limitations, not their abilities; simulating those is not a sign of intelligence.

2.4. is it relevant to AI research?

2.5. What is the motivation of the human subjects? Are they colluding to make the machine look better? Do they have secret signals?

3. basic setup

3.1. two agents both communicating via "text chat" (a teletype), one is a computer, the other one a human

3.2. judges can communicate with both agents and may ask questions

3.3. after some time communicating, they have to decide which of the agents is human and which is the computer

4. motivation

4.1. a test for weak AI

4.2. see whether machines can "act intelligently"

4.3. use a double blind paradigm with human judges

4.4. quite analogous to how we test medicines

5. proposed by Alan Turing (and others)

5.1. we want to build machines that are "intelligent", how can we tell whether we succeeded?

5.2. example