The intelligence of the future: robots without prejudice | Cem Say | TEDxIstanbul
-
0:13 - 0:14Hello there.
-
0:14 - 0:18Artificial intelligence is, of course,
literally a new beginning. -
0:18 - 0:23We are trying to create
a new type of a thinking being. -
0:23 - 0:28In fact, we have achieved a lot
since we got started with this project. -
0:28 - 0:33Computers can now play chess
much better than humans. -
0:33 - 0:37They can analyze radiological images
better than human doctors. -
0:38 - 0:41But today, I will talk about a domain
-
0:41 - 0:48where AI has not yet reached
the level of a person of average IQ: -
0:48 - 0:51understanding human language.
-
0:51 - 0:55You have probably read
this horrific news item -
0:55 - 1:02about the "chatbots"
which are programmed to chat with people. -
1:02 - 1:08In 2016, Microsoft created
a Twitter AI character -
1:08 - 1:11which was supposed to learn
the nuances of human language -
1:11 - 1:14by tweeting with people.
-
1:14 - 1:18Twenty-four hours later,
they had to take it offline. -
1:18 - 1:23Due to the nasty things, curses etc.
that people wrote to it in their tweets, -
1:23 - 1:26it turned into this nauseating character
-
1:26 - 1:28who said things like,
"Hitler was so good." -
1:28 - 1:32It does not exist any more.
-
1:32 - 1:36A year later, in China - maybe
you have not heard about this one - -
1:36 - 1:40a similar end was waiting for two chatbots
-
1:40 - 1:45which were launched in China to chit chat
with users on Chinese social media sites -
1:45 - 1:50after they started to talk
about their dreams of moving to the States -
1:50 - 1:54or mentioned their dislike
for the Chinese Communist Party. -
1:54 - 1:57They were deactivated
for a few days after the incident. -
1:57 - 2:02After they were reactivated,
they started talking very "carefully" -
2:02 - 2:03when those issues came up,
-
2:03 - 2:08giving answers like,
"Sorry, I can not understand you." -
2:08 - 2:11People who use the digital assistant Siri
-
2:11 - 2:15already know what a big
engineering success it is. -
2:15 - 2:17Yet there's more to the story:
-
2:17 - 2:21Authors and poets
in every language are hired -
2:21 - 2:25so that Siri can give proper answers
in such situations. -
2:25 - 2:28They are also writing scripts
-
2:28 - 2:32so that it doesn't have to confess
it can't understand what is being said -
2:32 - 2:35and it can continue the illusion
of being intelligent -
2:35 - 2:38by diverting the conversation
when it gets stuck. -
2:38 - 2:42Amazon also has a digital
assistant named Alexa, -
2:42 - 2:44which doesn't have a Turkish version yet.
-
2:44 - 2:47They promised a one-million-dollar prize
-
2:47 - 2:51to the programming team which will enable
Alexa to chat with people for 20 minutes -
2:51 - 2:54without causing extensive boredom.
-
2:54 - 2:56No one has been able to do that yet.
-
2:56 - 3:01The problem is that there are lots
of very simple things that humans know, -
3:01 - 3:03and computers don't.
-
3:03 - 3:07And we need to have a way
of teaching them those things. -
3:07 - 3:11Let me tell you
a personal story about that. -
3:12 - 3:16A long time ago, maybe 25 years or so,
-
3:16 - 3:21I bought a third-grade math textbook
for primary school students -
3:21 - 3:26and randomly picked 20 problems from it.
-
3:26 - 3:31Then I started to write a program
which would "understand Turkish." -
3:31 - 3:35It would understand these particular
arithmetic problems in Turkish -
3:35 - 3:36and solve them.
-
3:36 - 3:39I was thinking that
I would therefore reach a new level -
3:39 - 3:42in the computer understanding
of the Turkish language -
3:42 - 3:44and write another paper.
-
3:44 - 3:45The name of the program was ALİ,
-
3:45 - 3:48a Turkish acronym for
"arithmetic language processor." -
3:48 - 3:51It could solve problems like these:
-
3:51 - 3:54There were this many workers at a factory.
-
3:54 - 3:56That many of them were fired,
and this many retired. -
3:56 - 3:58How many were left?
-
3:58 - 4:01Or questions like,
"That many students from College A -
4:01 - 4:03and this many students from College B
-
4:03 - 4:04attended the ceremony.
-
4:04 - 4:07What's the total number of students?"
-
4:07 - 4:10requiring really simple arithmetic.
-
4:11 - 4:12You think that's easy peasy?
-
4:12 - 4:15Ask Siri the same questions,
and see if it can solve them all. -
4:15 - 4:19Let me tell you,
it took two years of my youth, -
4:19 - 4:22and I used to have gorgeous hair
when I got started. -
4:22 - 4:24(Laughter)
-
4:24 - 4:25Here is the problem.
-
4:25 - 4:27Let us go through this example.
-
4:27 - 4:31There were 67 liters of diesel
in the gas tank of a truck. -
4:31 - 4:33The driver bought 145 liters more.
-
4:33 - 4:35How much diesel does the truck have now?
-
4:35 - 4:40We are skipping the linguistics routines
that analyze all this in Turkish. -
4:40 - 4:43Let's come to the point
-
4:43 - 4:46where the AI can understand
the fact that there need to be 212 liters -
4:46 - 4:48at the end of the first two sentences.
-
4:48 - 4:53There, we come to a point where it knows
that there are 212 liters of diesel -
4:53 - 4:54in the gas tank,
-
4:54 - 4:56but what was the wording
of the question again? -
4:56 - 4:59"What's the sum of diesel in the truck?"
-
4:59 - 5:01"How much diesel
does the truck have now?" -
5:01 - 5:06ALİ could not answer that
with the information we have mentioned. -
5:06 - 5:08Do you see what the problem was?
-
5:08 - 5:12"The gas tank of the truck" is not
the same thing as "the truck," -
5:12 - 5:15and computers do not know automatically
-
5:15 - 5:18that if the tank contains something,
the truck also contains that thing. -
5:18 - 5:20And that's really complicated.
-
5:20 - 5:22"Ahmet's father had five kids"
-
5:22 - 5:23does not mean "Ahmet had five kids."
-
5:23 - 5:24On the other hand,
-
5:24 - 5:28when the gas tank of the truck
has the petrol, the truck has it as well. -
5:28 - 5:32That's why I had to specify in the program
-
5:32 - 5:36all this knowledge
that people already inherently know. -
5:36 - 5:40The technical name for this stuff
is "commonsense knowledge." -
5:40 - 5:44"The gas tank of the truck
is a part of the truck." -
5:45 - 5:48"If A is a part of B, right,
-
5:48 - 5:51B should contain
everything contained in A." -
5:51 - 5:54All of this information
that I consider commonsense -
5:54 - 5:57is all the things that I do not tell you
while we are talking -
5:57 - 6:02since I assume
that you already know it all. -
6:02 - 6:07We can not have a proper conversation
with those chatbots -
6:07 - 6:09since they know none of those things.
-
6:09 - 6:12After I coded all these,
ALİ could solve all 20 problems properly. -
6:12 - 6:17I had no more energy to go on to the 21st.
-
6:17 - 6:21Now, I'll tell you the story of a man
who dedicated his life to this problem -
6:21 - 6:23of coding commonsense knowledge:
-
6:23 - 6:27Douglas Lenat, a famous American
computer scientist. -
6:27 - 6:30This is him in the 1980s.
-
6:30 - 6:34He started a project called Cyc in 1982.
-
6:34 - 6:37And this is exactly
what the project was about: -
6:37 - 6:41To code all the commonsense knowledge
that computers don't know. -
6:41 - 6:44To write a million lines,
if a million lines are needed. -
6:44 - 6:47He founded a corporation
where they do the following: -
6:47 - 6:52If you are drinking coffee, the open side
of the cup is facing upwards. -
6:53 - 6:55The king is a man.
-
6:55 - 6:59Then his wife should be a woman,
and she is called the queen. -
6:59 - 7:03People can't go to work after they die.
-
7:03 - 7:04And so on.
-
7:04 - 7:07They are coding all the items
of information which people already know -
7:07 - 7:13and computers need to know in order
to understand human language, one by one. -
7:13 - 7:15And this is him today.
-
7:15 - 7:18After 35 years, the project
is still in progress. -
7:18 - 7:21I think there's an obvious problem here.
-
7:21 - 7:24It's clearly problematic to code manually.
-
7:24 - 7:26Now it's time to hear the good news.
-
7:26 - 7:28We have had a revolution in AI,
-
7:28 - 7:30and computers can now learn
certain things on their own, -
7:30 - 7:36without us having to code them manually.
-
7:36 - 7:38This is a machine-learning revolution.
-
7:38 - 7:41Linguists have the following idea:
-
7:41 - 7:46If two words are exact
synonyms of each other, -
7:46 - 7:50then the collections of all other words
surrounding them in various sentences -
7:50 - 7:52will also be similar to each other.
-
7:52 - 7:56Based on this idea, this man,
-
7:56 - 8:01who is proof of the fact
that you don't need to be bald -
8:01 - 8:03in order to be handsome
if you're an AI researcher, -
8:04 - 8:06named Tomas Mikolov,
-
8:07 - 8:11did the following while working
for Google five years ago. -
8:11 - 8:14Now think of all the documents
in English at Google. -
8:14 - 8:16The work I'll be telling
you about was in English. -
8:16 - 8:18Now imagine all the documents in English.
-
8:18 - 8:20For every word in every sentence,
-
8:20 - 8:26you're supposed to find out how many times
it has appeared in the same sentence -
8:26 - 8:27with any other words.
-
8:27 - 8:31For every imaginable pair of words,
we have the computer count -
8:31 - 8:37how many times these two words appear
together in the same sentence or not. -
8:37 - 8:38It's a computer,
-
8:38 - 8:42so it can do the computations anyway.
-
8:42 - 8:46The idea is that, if the two words
are close to each other in meaning, -
8:46 - 8:50the same words appear with similar
frequencies in their surroundings. -
8:50 - 8:54Let's say, we can easily see
that both words "cat" and "dog" -
8:54 - 8:56will appear frequently
in the same sentences -
8:56 - 9:03with the words "flea" or "rabies,"
"vaccine," "tail," "pet," and so on, -
9:03 - 9:07but not with words like "printer,"
"generator" or "inflation." -
9:07 - 9:09Do we see this?
-
9:09 - 9:12So, we can prepare a number sequence
-
9:12 - 9:15containing the frequencies
of the neighboring words -
9:15 - 9:18for every single word.
-
9:18 - 9:23Such a number sequence
is called a "vector," -
9:23 - 9:27as you might well know
if they still teach it in high school. -
9:27 - 9:31The computer can automatically position
similar number sequences -
9:31 - 9:34closer to each other,
-
9:34 - 9:37and the dissimilar ones
far from each other -
9:37 - 9:41on some sort of a map or space.
-
9:41 - 9:46What I mean is that the computer,
which knows no English, -
9:46 - 9:51creates a vector for each single word
by doing the computations. -
9:51 - 9:52Yet, the vector of "cat"
-
9:52 - 9:55is found in a location
close to the vector of "dog" in that space -
9:55 - 9:57for the reasons I just explained.
-
9:57 - 10:01Or the vector of the school
Buffy the Vampire Slayer attends - -
10:01 - 10:03they really looked at that -
-
10:03 - 10:06is positioned close
to the vector of Hogwarts, -
10:06 - 10:08where Harry Potter studies.
-
10:08 - 10:12Thus they are found to be positioned close
to each other in terms of their meaning. -
10:12 - 10:13There's more.
-
10:13 - 10:16As you will recall
from that high school course, -
10:16 - 10:19you can do arithmetic on these vectors.
-
10:19 - 10:21They can be added or subtracted,
-
10:21 - 10:22and you might say, "So what?"
-
10:23 - 10:26Mikolov discovered this.
-
10:26 - 10:29He did the following addition
and subtraction operations -
10:29 - 10:31on the vectors thus learned.
-
10:31 - 10:33He came up with the question,
"What would happen -
10:33 - 10:37if the king were a woman instead of a man"
-
10:37 - 10:39when he subtracted the word "man"
-
10:39 - 10:43from the word "king"
and added the word "woman." -
10:43 - 10:46Guess what the resulting
vector is near to? -
10:46 - 10:47"Queen."
-
10:47 - 10:51No one had hand-coded
that equation as the Lenat team. -
10:51 - 10:54The computer discovered it all by itself
-
10:54 - 10:58after counting millions of millions
of words on the documents we created. -
10:58 - 11:02I have more to tell you,
and this really happened. -
11:02 - 11:04There is info on Turkey there.
-
11:04 - 11:08If you take "France" out of "Paris"
and add "Turkey" - -
11:08 - 11:11yes, you got it right - it's Ankara.
-
11:11 - 11:14This means in this vector space,
there's a direction -
11:14 - 11:17which leads from the names of countries
to the names of their capitals, -
11:17 - 11:18which is really stunning.
-
11:18 - 11:23When you ask, What would Windows be
had it not been invented by Microsoft, -
11:23 - 11:26but by Google?
-
11:27 - 11:29the answer pops up as "Android."
-
11:29 - 11:35When you subtract "copper"
from "Cu" and add "gold," -
11:36 - 11:40you get "Au" as the chemical
symbol of gold. -
11:40 - 11:43This literally means we don't have to code
these manually anymore. -
11:43 - 11:46It seems that the computer
can make all the inferences -
11:46 - 11:49out of the data
we provide it with all by itself. -
11:49 - 11:51This is the yummiest example of all.
-
11:51 - 11:55When you take "Japan" out of "sushi"
and add "Germany," -
11:56 - 12:00you get the "bratwurst,"
the German favorite. -
12:00 - 12:02Too good to be true, right?
-
12:02 - 12:03Happy now?
-
12:03 - 12:04We finalized this project.
-
12:04 - 12:09Would computers understand what we say?
-
12:09 - 12:11Are we having fun? Not much.
-
12:11 - 12:14Now, I'll tell you
about a Turkish researcher. -
12:14 - 12:19Tolga Bölükbaşı is about to finish his PhD
-
12:19 - 12:21at Boston University in the States.
-
12:21 - 12:24This is a research he did two years ago.
-
12:24 - 12:27Tolga did the same thing
as Mikolov did previously, -
12:27 - 12:30but this time on news texts.
-
12:31 - 12:37What happens when you subtract "father"
from "doctor" and add "mom"? -
12:37 - 12:42"My dad is a doctor, and mom is a nurse."
-
12:42 - 12:48What about when you subtract "man"
from "computer engineer" and add "woman"? -
12:48 - 12:52In fact, we shouldn't have gender.
-
12:52 - 12:57Let's see how professions
are related to gender -
12:57 - 13:01in the meaning space
in the head of the computer. -
13:01 - 13:03You get "homemaker."
-
13:03 - 13:04Seriously! You get "homemaker."
-
13:04 - 13:08We get an English word "homemaker."
-
13:08 - 13:11So, it's clear that we not only put
all of our data in computers -
13:11 - 13:17but also put all of our prejudices.
-
13:17 - 13:23Imagine if this computer
were used to hire someone. -
13:23 - 13:27You've already uploaded your resume
and all the personal information -
13:27 - 13:29including your gender.
-
13:29 - 13:32Let's assume 10,000 people
applied for the job. -
13:32 - 13:34The computer needs
to do a pre-selection, right? -
13:34 - 13:38It needs to get to 1,000 candidates,
-
13:38 - 13:42eliminating 9,000 others
-
13:42 - 13:44so that the HR staff
can evaluate the results. -
13:44 - 13:47Computers nowadays are already
used for this kind of work. -
13:47 - 13:50Let's say that a computer loaded
with such meaning vectors makes selection -
13:50 - 13:54among the candidates who have applied
for a job vacancy for a computer engineer. -
13:54 - 13:57It might automatically eliminate
all the female candidates, -
13:57 - 14:02thinking that a computer
engineer should be male. -
14:02 - 14:05Tolga and his colleagues
also mention other cases. -
14:05 - 14:11It was found out that computers
link positive and negative attributions -
14:11 - 14:16with the words related to being
Afro-American and Caucasian. -
14:16 - 14:20For instance, the computer thinks
-
14:20 - 14:23that the word "mugger" is closely related
to being Afro-American. -
14:23 - 14:27It's certain that we uploaded
all our prejudices -
14:27 - 14:29while uploading all the information
we have in computers. -
14:29 - 14:33You might ask yourselves,
What will happen now? -
14:33 - 14:36Tolga and his team's article
offers a solution to that. -
14:38 - 14:40Just told you.
-
14:40 - 14:42All these things happen
in the vector space. -
14:42 - 14:44Each word has its vector.
-
14:44 - 14:48We already know from high school years
that we can add and subtract them. -
14:48 - 14:52Tolga and his team first list the words
-
14:52 - 14:57that are really feminine or masculine,
-
14:57 - 15:01like "dad," "uncle,"
"grandmother," and so on. -
15:01 - 15:06These words really should have
a relation to male and female roles. -
15:06 - 15:11Then there are these words
which should not be masculine or feminine -
15:11 - 15:14despite having closer meanings
in the computer's space. -
15:14 - 15:19For example, the word "genius"
appears to be male. -
15:19 - 15:24On the other hand, the word "stylist"
stands out as a very female word. -
15:24 - 15:26It doesn't have to be like that.
-
15:26 - 15:30So, after listing all the words
that need to be feminine or masculine, -
15:30 - 15:33Tolga and his team created an algorithm
-
15:33 - 15:40which would automatically erase
the computer's prejudices -
15:40 - 15:47on the ones that should be neutral.
-
15:47 - 15:51If a word like "father" or "uncle"
is not in the list, -
15:51 - 15:56but it is still biased towards a gender
in the space of meanings, -
15:56 - 16:01the algorithm automatically corrects it.
-
16:01 - 16:03With the help of this,
"computer programmer" -
16:03 - 16:06ends up at the same distance
to the male and female notions, -
16:06 - 16:09and the problems I talked about go away.
-
16:09 - 16:10Isn't that beautiful?
-
16:10 - 16:16I wish we could delete the prejudices
in the human brain so easily. -
16:16 - 16:22For a while, some people
have been worrying about -
16:22 - 16:24what would happen
if computers took over. -
16:24 - 16:26On the other hand,
-
16:26 - 16:29considering the fact that we can't delete
the prejudices in people, -
16:29 - 16:33while we can in computers,
-
16:33 - 16:37maybe we could give computers a chance
at jobs requiring fairness -
16:37 - 16:43such as being referees,
judges, and managers -
16:43 - 16:45and let people take a rest for a while.
-
16:45 - 16:46What do you say to that?
-
16:47 - 16:48Thank you.
-
16:48 - 16:51(Applause)
- Title:
- The intelligence of the future: robots without prejudice | Cem Say | TEDxIstanbul
- Description:
-
Cem Say, famous for his writings and research on artificial intelligence, tells us what artificial intelligence is and when and where humanity will benefit from it. Moreover, he says we should not be afraid of artificial intelligence, adding that the prejudices of humanity would be exceeded by artificial intelligence with the use of technology.
While giving lectures at the Department of Computer Engineering at Boğaziçi University, Prof. Dr. Cem Say is a pioneer at making research on qualitative reasoning and understanding Turkish in artificial intelligence.
Say, one of the founders of the Cognitive Science Graduate Degree, works on how the human mind works and operated. He was among the computer experts who examined the digital evidence in the court cases and revealed their counterfeits.
He contributes to popular science journals such as "Technology for Everyone" and is well-known with his easy-to-read and learn articles describing complicated technology objects - such as Bitcoin, blockchain, sharing economy etc.This talk was given at a TEDx event using the TED conference format but independently organized by a local community. Learn more at https://www.ted.com/tedx
- Video Language:
- Turkish
- Team:
- closed TED
- Project:
- TEDxTalks
- Duration:
- 16:55
Mirjana Čutura edited English subtitles for Geleceğin zekâsı: ön yargısız robotlar | Cem Say | TEDxIstanbul | ||
Mirjana Čutura approved English subtitles for Geleceğin zekâsı: ön yargısız robotlar | Cem Say | TEDxIstanbul | ||
Mirjana Čutura accepted English subtitles for Geleceğin zekâsı: ön yargısız robotlar | Cem Say | TEDxIstanbul | ||
Mirjana Čutura edited English subtitles for Geleceğin zekâsı: ön yargısız robotlar | Cem Say | TEDxIstanbul | ||
Mirjana Čutura edited English subtitles for Geleceğin zekâsı: ön yargısız robotlar | Cem Say | TEDxIstanbul | ||
Mirjana Čutura edited English subtitles for Geleceğin zekâsı: ön yargısız robotlar | Cem Say | TEDxIstanbul | ||
Mirjana Čutura edited English subtitles for Geleceğin zekâsı: ön yargısız robotlar | Cem Say | TEDxIstanbul | ||
Ozge Yilmaz edited English subtitles for Geleceğin zekâsı: ön yargısız robotlar | Cem Say | TEDxIstanbul |