viewBox="0 0 51 48">Five Pointed Star Five Pointed Star

Can Machines Think?

The Tur­ing Test

Overview of Turing’s Theory

The Com­puter Ency­clo­pe­dia defines the Tur­ing test as:
The “acid test” of true arti­fi­cial intel­li­gence, as defined by the Eng­lish sci­en­tist Alan Tur­ing. In the 1940s, he said “a machine has arti­fi­cial intel­li­gence when there is no dis­cernible dif­fer­ence between the con­ver­sa­tion gen­er­ated by the machine and that of an intel­li­gent per­son” (2011).

Alan Tur­ing referred to his test as “The Imi­ta­tion Game” in his famous paper, where a human observer is sup­posed to dif­fer­en­ti­ate between a human being and a com­puter based on answers to type­writ­ten ques­tions (Tur­ing, 1950). He then basi­cally refuted the argu­ments of his time which asserted that machines can­not develop the abil­ity to think – the fol­low­ing objec­tions: the­o­log­i­cal, “The Heads in the Sand,” math­e­mat­i­cal, “Lady Lovelace’s”, con­scious­ness, “argu­ments from var­i­ous dis­abil­i­ties,” “argu­ments from infor­mal­ity of behav­ior,” evi­dence for Extrasen­sory Per­cep­tion, and “argu­ments from con­ti­nu­ity in the ner­vous sys­tem.”
Abram­son con­sid­ers the ques­tion of the pos­si­bil­ity of a machine being able to be built that passes the Tur­ing test to be partly a philo­soph­i­cal ques­tion of the mind as well as the phi­los­o­phy of com­puter sci­ence (2011, pg. 203–204, 217).

Arti­fi­cial Intelligence

A lot of the dis­pute over the pos­si­bil­ity of true arti­fi­cial intel­li­gence (AI) cen­ters around the def­i­n­i­tion of an algo­rithm (Hoff­man, 2010, pg. 205–206). Tur­ing refers to a learn­ing machine, start­ing with a child-like “mind” to achieve AI or a think­ing machine (Tur­ing, 1950). Hoff­man argues that the def­i­n­i­tion of algo­rithms are too general.

An infi­nite class of prob­lems for which it has been proven that there does not exist an algo­rithm, e.g. the deci­sion prob­lem of first order pred­i­cate logic. Actu­ally, there is no empir­i­cal evi­dence avail­able for the claims that humans do bet­ter than algo­rithms. Despite the fact, that this kind of objec­tions to the enter­prise of AI demand more from machines than from the human mind it appears also quite irrel­e­vant for the prac­ti­cal aspect of AI sys­tems whether a sys­tem is able to respond prop­erly to an infi­nite num­ber of con­ceiv­able requests with which the sys­tem will never be con­fronted. Since in prac­tice always only a lim­ited num­ber of requests will be addressed to a sys­tem it is suf­fi­cient if the sys­tem can respond to those as desired (2010).

Another objec­tion by Hoff­man involves think­ing cre­atively and the dif­fer­ence in the pat­tern of thought that can­not be reduced to an algo­rithm. Hoffman’s counter-argument is weak here, stat­ing the fal­lacy with this argu­ment is a par­tic­u­lar rule scheme is not given gov­ern­ing human thought.
Diane Proud­foot asserts the impor­tance and the still cur­rent rel­e­vance of the Tur­ing test in eval­u­at­ing AI, which has been dis­cred­ited by var­i­ous argu­ments of AI researchers (2011, pg. 951). She points out the ten­dency to anthro­po­mor­phize AI sys­tems skews researcher’s logic and that Tur­ing test is immune to such con­cep­tu­al­iza­tion (pg. 955). It has also been sug­gested that per­haps the gold stan­dard for intel­li­gence is not human intel­li­gence, obso­lesc­ing the Tur­ing test and IQ tests in favor of a “uni­ver­sal intel­li­gence test” (Biever, 2011, pg. 42–45). Biever makes a good point that per­haps an IQ test would not be applic­a­ble to a machine or even an alien life form.

Com­pletely Auto­matic Pub­lic Tur­ing test to tell Com­put­ers and Humans Apart (CAPTCHA)

Since my major is web devel­op­ment, CAPTCHA has great sig­nif­i­cance to me, as CAPTCHA is a bar­rier to auto­mated “bots” that exploit con­tact and com­ment forms as well as forums on web­sites. Most com­monly CAPTCHA is in the form of a dis­torted image that makes it dif­fi­cult for a com­puter to rec­og­nize the char­ac­ters. There are other meth­ods such as audio (speech recog­ni­tion), a puz­zle or math ques­tion, as well as cor­rectly iden­ti­fy­ing pic­tures with their cor­re­spond­ing labels (Lee & Hsu, 2011, pg. 81). Lee and Hsu con­ducted a usabil­ity study with var­i­ous meth­ods of text dis­tor­tion to deter­mine the dif­fi­culty for actual humans to dis­cern the characters.

Con­clu­sion

Today the sig­nif­i­cance and rel­e­vance of the Tur­ing test is under fire from many in AI research. Within the con­fines of human behav­ior mim­ic­ked by a com­puter, it is still very rel­e­vant and should still be con­sid­ered a stan­dard when con­sid­er­ing the “intel­li­gence” of a machine.

Ref­er­ences

Abram­son, D. (2011). Phi­los­o­phy of mind is (in part) phi­los­o­phy of Com­puter Sci­ence. Minds & Machines, 21(2), 203–219. doi:10.1007/s11023-011‑9236-0

Biever, Celeste. (2011). Ulti­mate IQ. New Sci­en­tist, 211(2829), 42–45.

Hoff­mann, A. (2010). Can machines think? An old ques­tion refor­mu­lated. Minds & Machines, 20(2), 203–212. doi:10.1007/s11023-010‑9193-z

Lee, Y., & Hsu, C. (2011). Usabil­ity study of text-based CAPTCHAs. Dis­plays, 32(2), 81–86. doi:10.1016/j.displa.2010.12.004

Proud­foot, D. (2011). Anthro­po­mor­phism and AI: Tur­ingʼs much mis­un­der­stood imi­ta­tion game. Arti­fi­cial Intel­li­gence, 175(5/6), 950–957. doi:10.1016/j.artint.2011.01.006

Skyt­tner, L. (2005). Gen­eral sys­tems the­ory: prob­lems, per­spec­tives, prac­tice. Hack­en­sack, NJ: World Sci­en­tific Pub­lish­ing Co. Pte. Ltd.

Tur­ing, A.M. (1950). Com­put­ing machin­ery and intel­li­gence. Mind, 59, 433–460, retrieved Sep­tem­ber 22, 2011 from http://loebner.net/Prizef/TuringArticle.html

Tur­ing test. (2011). Com­puter Desk­top Ency­clo­pe­dia, 1. Retrieved from EBSCOhost.


L. Ball
L. Ball

Father. Developer. Coffee Connoisseur. Amateur Guitarist.