The Turing test: Can a computer pass for a human? - Alex Gendler

Prevodilac: Milenka Okuka Lektor: Mile Živković

What is consciousness?

Šta je svesnost?

Can an artificial machine really think?

Može li veštačka mašina zaista da misli?

Does the mind just consist of neurons in the brain,

Da li se um sastoji jedino od neurona u mozgu

or is there some intangible spark at its core?

ili u njegovoj srži postoji neopipljiva iskra?

For many, these have been vital considerations

Za mnoge su ovo bila ključna razmatranja

for the future of artificial intelligence.

za budućnost veštačke inteligencije.

But British computer scientist Alan Turing decided to disregard all these questions

No britanski kompjuterski naučnik Alan Tjuring

je odlučio da odbaci sva ova pitanja

in favor of a much simpler one:

zarad jednostavnijeg:

can a computer talk like a human?

može li kompjuter da govori kao čovek?

This question led to an idea for measuring aritificial intelligence

Ovo pitanje je dovelo do zamisli merenja veštačke inteligencije

that would famously come to be known as the Turing test.

koja će postati čuvena kao Tjuringov test.

In the 1950 paper, "Computing Machinery and Intelligence,"

U naučnom radu iz 1950. "Kompjuterska mašina i inteligencija",

Turing proposed the following game.

Tjuring je predložio sledeću igru.

A human judge has a text conversation with unseen players

Ljudski sudija vodi tekstualni razgovor sa takmičarima koje ne vidi

and evaluates their responses.

i procenjuje njihove odgovore.

To pass the test, a computer must be able to replace one of the players

Da bi prošao test, kompjuter mora da bude u stanju da zameni takmičare,

without substantially changing the results.

a da se pri tom rezultat značajno ne promeni.

In other words, a computer would be considered intelligent

Drugim rečima, kompjuter bismo smatrali inteligentnim,

if its conversation couldn't be easily distinguished from a human's.

ako ono što govori ne bi bilo lako razlikovati od ljudskog govora.

Turing predicted that by the year 2000,

Tjuring je predvideo da će do 2000. godine

machines with 100 megabytes of memory would be able to easily pass his test.

mašine sa memorijom od 100 megabajta lako prolaziti na ovom testu.

But he may have jumped the gun.

No možda je istrčao pred rudu.

Even though today's computers have far more memory than that,

Iako današnji kompjuteri imaju daleko veću memoriju,

few have succeeded,

malo njih je u tome uspelo,

and those that have done well

a oni koji su imali dobre rezultate

focused more on finding clever ways to fool judges

su se više usredsređivali na pronalaženje pametnih načina da prevare sudije

than using overwhelming computing power.

nego na korišćenje ogromne kompjuterske moći.

Though it was never subjected to a real test,

Iako nikad nije podvrgnut stvarnom testiranju,

the first program with some claim to success was called ELIZA.

prvi program koji je tvrdio da je imao nekog uspeha se zvao ELIZA.

With only a fairly short and simple script,

Prilično kratkim i jednostavnim scenariom

it managed to mislead many people by mimicking a psychologist,

uspeo je da obmane ljude oponašajući psihologa,

encouraging them to talk more

ohrabrujući ih da više govore

and reflecting their own questions back at them.

i vraćajući njihova sopstvena pitanja nazad njima.

Another early script PARRY took the opposite approach

Još jedan raniji scenario, PARRY, je imao suprotan pristup,

by imitating a paranoid schizophrenic

imitirajući paranoičnog šizofreničara

who kept steering the conversation back to his own preprogrammed obsessions.

koji je stalno usmeravao tok razgovora na sopstvene pretprogramirane opsesije.

Their success in fooling people highlighted one weakness of the test.

Njihovi uspesi u obmanjivanju ljudi su istakli slabosti testa.

Humans regularly attribute intelligence to a whole range of things

Ljudi redovno pripisuju inteligenciju čitavom nizu stvari

that are not actually intelligent.

koje nisu zaista inteligentne.

Nonetheless, annual competitions like the Loebner Prize,

Bez obzira na to, godišnja takmičenja, poput Lobnerove nagrade,

have made the test more formal

formalizovala su test

with judges knowing ahead of time

tako što sudije unapred znaju

that some of their conversation partners are machines.

da su neki od njihovih sagovornika mašine.

But while the quality has improved,

No, iako su kvalitetniji,

many chatbot programmers have used similar strategies to ELIZA and PARRY.

mnogi programeri botova za ćaskanje

koriste strategije slične ELIZA-i i PARRY-ju.

1997's winner Catherine

Pobednica iz 1997, Katrin,

could carry on amazingly focused and intelligent conversation,

mogla je da obavi začuđujuće usredsređene i inteligentne razgovore,

but mostly if the judge wanted to talk about Bill Clinton.

ali uglavnom, ako bi sudija želeo da razgovoara o Bilu Klintonu.

And the more recent winner Eugene Goostman

A skorašnjem pobedniku, Judžinu Gustmanu,

was given the persona of a 13-year-old Ukrainian boy,

podarena je ličnost 13-ogodišnjeg dečaka iz Ukrajine,

so judges interpreted its nonsequiturs and awkward grammar

pa su sudije protumačile njegove zablude i nezgrapnu gramatiku

as language and culture barriers.

kao jezičku i kulturološku barijeru.

Meanwhile, other programs like Cleverbot have taken a different approach

U međuvremenu, drugi programi, poput Kleverbota su imali drugačiji pristup,

by statistically analyzing huge databases of real conversations

statistički analizirajući ogromne baze podataka stvarnih razgovora

to determine the best responses.

kako bi došli do najboljih odgovora.

Some also store memories of previous conversations

Neki su takođe pohranjivali u memoriju prethodne razgovore,

in order to improve over time.

kako bi se vremenom popravljali.

But while Cleverbot's individual responses can sound incredibly human,

No, iako Kleverbotovi pojedinačni odgovori mogu da zvuče neverovatno ljudski,

its lack of a consistent personality

odsustvo konzistentne ličnosti

and inability to deal with brand new topics

i nemogućnost da se bavi najnovijim temama

are a dead giveaway.

odaju ga odmah.

Who in Turing's day could have predicted that today's computers

Ko bi u Tjuringovo vreme mogao da predvidi da će današnji kompjuteri

would be able to pilot spacecraft,

moći da upravljaju svemirskim brodovima,

perform delicate surgeries,

da izvode prefinjene operacije

and solve massive equations,

i da rešavaju obimne jednačine,

but still struggle with the most basic small talk?

ali će i dalje da se muče s najobičnijim ćaskanjem?

Human language turns out to be an amazingly complex phenomenon

Ispostavilo se da je ljudski jezik zadivljujuće složena pojava

that can't be captured by even the largest dictionary.

koju ne može da obuhvati čak ni najobimniji rečnik.

Chatbots can be baffled by simple pauses, like "umm..."

Botove za ćaskanje zbunjuju sitne pauze, poput "hm..."

or questions with no correct answer.

ili pitanja koja nemaju tačne odgovore.

And a simple conversational sentence,

A prosta razgovorna rečenica,

like, "I took the juice out of the fridge and gave it to him,

poput: "Izvadio sam sok iz frižidera i dao mu,

but forgot to check the date,"

ali sam zaboravio da proverim datum",

requires a wealth of underlying knowledge and intuition to parse.

zahteva bogato pozadinsko znanje i intuiciju da bi je raščlanili.

It turns out that simulating a human conversation

Ispostavilo se da simuliranje ljudskog razgovora

takes more than just increasing memory and processing power,

zahteva više od pukog uvećanja memorije i snage procesora;

and as we get closer to Turing's goal,

i kako se približavamo Tjuringovom cilju,

we may have to deal with all those big questions about consciousness after all.

možda ćemo ipak morati da se bavimo svim tim velikim pitanjima o svesnosti.