The Turing test’s insight into humanness.

Illustration from Dean Burnett's Guardian spoof of the June 2014 reissued 'Turing Test Passed' Story
Illustration from Dean Burnett’s spoof of the June 2014 reissued ‘Turing Test Passed’ Story

I’ve heard people reacting in two ways to the hyped announcement about Eugene passing the Turing Test. Some claim the test should be harder: longer term and  more complex, others that it doesn’t show machines doing thinking. I disagree with both complaints. I think the test is a brilliant one, and very insightful and informative about what it means to be a language machine.

Turing (1950) wrote:

“I believe that in about fifty years’ time it will be possible, to programme computers… [to] play the imitation game so well that an average interrogator will not have more than 70 per cent chance of making the right identification after five minutes of questioning”.

So however hyped, the basic facts of the story are more or less correct, and I find it quite amazing (given that the Eugene chat bot was first written in 2001) that Turing got the timing spot on. However, I do agree that the news story and most interpretations of the meaning of the Turing Test are nonsensical from a scientific standpoint.

It seems likely to me that since 2001 many 13 year olds along with a great many other humans would fail the test as described, and equally likely that many more advanced chatbots would be able to pass it quite easily. This wasn’t the case in 1950 when the social meaning of computing would have been unrecognisable to contemporary judges, and vice versa.

Given that Turing’s computing challenge was passed, quite trivially, some time ago, the research challenge posed by the test as a socio-historical milestone, and the challenge for cognitive science in general since then is figuring out how, when and in what ways humanness is an ascribable quality.

There is a nice discussion of exactly this problem in QM’s very own CS4FUN – although I’m not sure who (or what) wrote it.


Identifying Emotions on the Basis of Manual Activation

Arash Eshghi started what turned out to be a very productive fight on our CogSci listserv with this press release: Carnegie Mellon Researchers Identify Emotions Based on Brain Activity and its attendant paper: Identifying Emotions on the Basis of Neural Activation.

I came up with a press release of my own, I might at some point get round to doing the Atlantic Salmon paper on the subject.

Press Release: University Researchers Identify Emotions Based on Finger Activity

New Study Extends “Palm Reading” Research to Feelings by Applying Machine Learning Techniques to Keyboard Data

For the first time, scientists at a university have identified which emotion a person is experiencing based on finger activity.


 :)      happy

 :(      sad

-----fig 1--------

The study combines keyboards and machine learning to measure finger signals to accurately read emotions in individuals. The findings illustrate how the finger categorizes feelings, giving researchers the first reliable methods to evaluate them.

“Our big breakthrough was the idea of testing typists, who are
experienced at expressing emotional states digitally. We were fortunate,
in that respect, that EECS has so many superb typists”

said a professor.

For the study, typists were shown the words for 9 emotions: anger, disgust, envy, fear, happiness, lust, pride, sadness and shame. and were recorded typing them multiple times in random order.

   :§      8(>_<)8
x-(    ;-b...
 _ _ 
( " )   :-c

 DX        >:(
:0=     :-)

  :(     *:-}

  !-}       (-_-)
-----fig 2--------

The computer model, using statistical information to analyse keyboard activation patterns for 18 emotional words was able to guess the emotional content of photos being viewed using only the finger activity of the viewers.

“Despite manifest differences between people’s psychology, different
people tend to manually encode emotions in remarkably similar ways”

noted a graduate student.

A surprising finding from the research was that almost equivalent accuracy levels could be achieved even when the computer model made use of activation patterns in only one of a number of different subsections of the keyboard.

“This suggests that emotion signatures aren’t limited to specific
regions such as the qwerty parentheses cluster, but produce
characteristic patterns throughout a number of keyboard regions”

said a senior research programmer.