The new con­ver­sa­tion­al­ists on the block

Wangaratta Chronicle - North East Regional Extra - - Tech and Science - WITH CHRIS FEBVRE,

WE hu­mans are a nar­cis­sis­tic bunch.

We cre­ate in our own im­age, an­thro­po­mor­phise an­i­mals and inan­i­mate ob­jects, and idolise those which best re­flect hu­man­ity’s traits back at us.

From the tale of Pyg­malion, the an­cient Greek sculp­tor who fell in love with his own sculp­ture, to the mod­ern day science fic­tion film ‘Ex Machina’ (a movie I highly rec­om­mend), in which an ar­ti­fi­cially in­tel­li­gent ro­bot dupes both her human cre­ator, and the man hired to test her, our fas­ci­na­tion with hu­manesque con­structs has stuck with us from the be­gin­ning of recorded his­tory.

In re­cent times, this ob­ses­sion has driven com­puter sci­en­tists to at­tempt to de­velop ar­ti­fi­cial in­tel­li­gences by mim­ick­ing the way the neu­ral net­works in our brains process in­for­ma­tion.

While not en­tirely new, in­ter­est in chat­bots (ar­ti­fi­cial in­tel­li­gences de­signed for the pur­poses of com­mu­ni­cat­ing with a human user in the most human-like way pos­si­ble), have cer­tainly in­creased markedly.

And with the re­newed in­ter­est, de­vel­op­ers have been hard at work in­fus­ing the ‘human fac­tor’ into their chatty apps.

The race to cre­ate re­al­is­tic chat­bots arose soon af­ter the first dig­i­tal com­put­ers were de­vel­oped, as did the tests for mea­sur­ing ma­chine in­tel­li­gence.

De­vel­oped by Alan Tur­ing, the ‘ Tur­ing Test’ in­volves a human eval­u­a­tor who is tasked with en­gag­ing in text based con­ver­sa­tions with both an ar­ti­fi­cial in­tel­li­gence, and an­other human who acts as a foil.

Af­ter five min­utes of con­ver­sa­tion with each en­tity, the eval­u­a­tor se­lects the one they think is human.

If the eval­u­a­tor can’t dis­tin­guish be­tween the two, or picks the in­cor­rect one, the pro­gram is deemed to have passed the test.

In 1990, Amer­i­can in­ven­tor Hugh Loeb­ner in­tro­duced the Loeb­ner Prize, adopt­ing the prin­ci­ples laid out by Tur­ing in 1950.

The an­nual com­pe­ti­tion grants prizes to chat­bots who man­age to fool the judges or present the most human-like con­ver­sa­tional tech­niques.

Just as in the ‘ Tur­ing Test’, the Loeb­ner prize also in­volves a human who con­verses with a judge at the same time as the A.I via a text mes­sag­ing in­ter­face, and the judge must dis­tin­guish be­tween the two.

Last year, the chat­bot Mit­suku took out the prize, which she had won in 2013 as well.

Mit­suku’s suc­cess, a chat­bot de­vel­oped by Steve Wor­swick, is largely due to a broad and in­ter­wo­ven data­base of com­mon ob­jects as well as their char­ac­ter­is­tics and re­la­tion­ships to each other.

Wor­swick looks at the bot’s con­ver­sa­tional logs to iden­tify top­ics users are ask­ing about and man­u­ally in­cludes them in Mit­suku’s vast knowl­edge data­base in or­der to im­prove rel­e­vance and ‘human-ness’.

For ex­am­ple, be­cause of Mit­suku’s Ja­panese name and anime-like avatar, many users ask her about Ja­panese cul­ture and anime shows, which Wor­swick then man­u­ally in­puts an­swers to, aid­ing her learn­ing, and im­prov­ing her re­sponses.

You can chat with Mit­suku at http://www.mit­suku.com/.

CHATTY: Ad­vances in tech­nol­ogy and a re­newed in­ter­est have given rise to some sur­pris­ingly human-like chat­bots.

Newspapers in English

Newspapers from Australia

© PressReader. All rights reserved.