Placeholder Image

Subtitles section Play video

  • - Another big milestone for AI, a competition

  • between a former debating world champion and a computer.

  • - So many people are here working on this production.

  • - Many of the previous demonstrations that we have

  • seen were the realm of games.

  • And games have very well defined set of rules.

  • - You could do simple things like play games,

  • but you couldn't do complex things

  • like having a debate.

  • - There is so many things that are connected

  • together and so many things need to work smoothly.

  • I am actually a bit stressed.

  • I feel the team didn't have enough time to practice.

  • - So I have to ask you,

  • who do you thinks going to win tonight?

  • - Well I'm partial today of course.

  • What we really want is to create systems that

  • work with us so that we can make better decisions.

  • - Ladies and gentlemen, here we go, Project Debater.

  • - Greetings, Harish.

  • I suspect you've never debated a machine.

  • Welcome to the future.

  • - I started my amateur debating career back in high school.

  • The teacher for history made the mistake of returning

  • the exams and saying if you want to appeal

  • about your score, I'm willing to hear your thoughts.

  • And obviously nobody seriously considered to do

  • that except me.

  • Okay I understand if you think that this is my score.

  • I have a different view.

  • Let's argue about that, in writing.

  • Later on I started to think about writing,

  • you know comedy sketches.

  • I started to develop a sitcom called Puzzle.

  • Oddly enough, the final episode of that sitcom

  • is entirely focused on competitive debates.

  • They filmed 16 episodes

  • and it was a complete failure.

  • This is how I found myself doing computer science

  • and artificial intelligence.

  • - For centuries, people have always saw the games

  • as a great way to measure your intelligence.

  • When I play chess, I want to prove to myself I'm

  • smarter than the other side.

  • If you take the position that you have to hand code

  • a rule for every situation,

  • it is insanely hard.

  • What we now do is you give them examples,

  • and they'll learn from those examples

  • and make their own rules compared to having to craft

  • all the rules by hand.

  • - Games have been important for AI but games

  • differ from debate because the classical game is a set

  • of equations that's fully understood.

  • You can draw the rules of chance.

  • You cannot draw the

  • rules of conversation, that's impossible.

  • - Just a couple of days after the Jeopardy! event,

  • an email was sent to the entire research division

  • asking us what should be the next era grand challenge.

  • I was intrigued so I offered my office mate

  • at the time to spend an hour and brainstorm together.

  • And towards the end of the hour at some point

  • I suggested this notion of developing a machine

  • that will be able to debate humans.

  • So we submitted the proposal in a single slide.

  • They told us this is impossible and this is not

  • a very good plan.

  • - I think this is a characteristic of a grand challenge

  • because if it is very clear to you that this is doable,

  • then probably this is not a grand challenge.

  • And if it is very clear to you that this is not doable,

  • then probably you shouldn't start.

  • It should be somewhere in between but you need to take

  • into consideration that perhaps you will fail.

  • - I didn't hear anything for a few months.

  • Then, February 2012, Hayah Sofail

  • used this chat service in IBM to text me.

  • And she said, "Did you hear?"

  • And I said, "No, what?"

  • And she said, "It was selected, Debater was selected

  • as the next grand challenge."

  • I was very excited and I wrote her back,

  • "I want to thank you for all the support along the way."

  • And she said, "Don't thank me yet."

  • (laughs)

  • - I'm Harish Natarajan,

  • and my debating record is that I won the

  • European Championships in 2012,

  • the Oxford Intervarsity and the Cambridge Intervarsity.

  • I've twice won the largest open tournament in the world,

  • the LSE Open.

  • And certainly, even if I'm not necessarily the best

  • debater in the world, I'll certainly be very

  • high up most people's list.

  • What I think debating forces you to do is you just

  • have to defend and justify your own opinions,

  • but defend and justify often the very opposite

  • opinions as well.

  • And I think that process is so valuable because I think

  • it makes us interrogate the beliefs and the views

  • which we hold.

  • - One of the things that happens when you take a grand

  • challenge is that you have to slice it up

  • into questions that each of them is less grand.

  • From the beginning, one thing you would need

  • is a massive collection of newspaper and journal articles

  • on the order of 10 billion sentences.

  • - Then the Debater system needed capability of being

  • able to automatically identify the exact boundaries

  • of the claim within the billions of sentences.

  • - And figure out if those arguments are supporting your

  • side of the debate or the other side.

  • - This is something that is very easy for humans to do

  • based on our knowledge about concepts in the world.

  • And it is very complicated for machines to do.

  • Now you need to arrange a meaningful narrative.

  • - After the opening statement, now there's rebuttal speech.

  • - This is being able to listen to an opponent

  • and respond accordingly.

  • - Those were the main challenges that we had

  • in front of us.

  • - And if one of these steps is not working properly,

  • you will end up with a useless speech.

  • After we were selected, it was very hard for us

  • to make significant progress.

  • Only four months after we started to work,

  • for the topic physical education should

  • be mandatory in high school,

  • the Debater system detects the following

  • claim in Wikipedia:

  • "Several studies have shown that lizards display

  • no benefit from exercise."

  • This is going to sound ridiculous, okay?

  • They're going to ask us about physical exercise,

  • whether this should be mandatory or not,

  • and how are we going to understand that these claims

  • are actually referring to lizards and not to

  • high school students?

  • It will be very hard to get even close to a meaningful

  • debate between this system and an expert human debater.

  • - First time I heard about Project Debater was

  • one of those small articles probably back in 2014

  • where they talked about this Debating Watson Project.

  • And at the time it certainly seemed like a pipe dream.

  • Humans are rarely just convinced by evidence

  • and logical rational argument.

  • It's about really connecting with them in one

  • of many different ways, some of which

  • aren't as rational as others.

  • - What I do is I coach debaters and I tell them

  • this is what you need to do in order to improve.

  • And usually a human debater, it just kind of,

  • their brain understands what I'm saying

  • and they can convert that into being a better debater.

  • Sometimes not, but usually.

  • And in this case, I am saying it to the programmers

  • and they're saying okay how do we take that learning

  • process that happens inside people's heads and convert

  • that into some sort of algorithm?

  • And that's really where the magic happens.

  • - It seems to me that this is a motion that basically

  • for the government it's impossible to defend.

  • - Well you can agree in general that you'd defend,

  • like any negative aspects to begin with.

  • - What I'm saying is that if you will debate me,

  • not as an expert debater, I will make this point.

  • Is there an answer to that?

  • - Being able to show to the world that you can

  • take a computer program and teach it not just to

  • play a game that has very specific set of rules

  • and very specific ways of winning but something

  • that is so,

  • so difficult for humans to learn and so subjective in how

  • you see it and how you hear arguments and how

  • you respond to arguments.

  • I think that that's, it's incredible.

  • When I joined the team, it was not at a level

  • where it would be competitive.

  • - My automatic analysis indicates that Yara

  • suggested that the abolition of the entire

  • fossil fuel field is going to create and the traumatic

  • effects of that and lastly,

  • however considering relevant data suggests some information

  • that may contradict this claim.

  • - I first heard about IBM Debater because they were

  • asking for volunteers to record speeches for them.

  • Basically, they wanted to build up this database

  • of speeches from people with a lot of debating

  • experience and then train the computer

  • using that database.

  • - I think at first, Project Debater would sometimes

  • give arguments that weren't necessarily for its side.

  • - Such mandatory minimum sentences

  • lead to overcrowded prisons and rightfully so.

  • These were the main points I wanted to raise.

  • - The types of arguments, the connections between arguments,

  • the way that we sometimes tell jokes

  • or sometimes we're more serious,

  • those are all things that they

  • pulled in for Project Debater.

  • - I often compare Project Debater to my kids and

  • their development because I guess my oldest

  • was born about the same time the program was born.

  • In terms of personality,

  • they'd be I guess on par.

  • My seven year old has a great personality,

  • they do develop that quite early.

  • It's scary but also nice to see that Project Debater

  • is developing a personality as well.

  • - I don't know whether to call my opponent naive,

  • old fashioned or just a romantic.

  • But this way or that, my opponent is wrong.

  • - I remember the Debater, she called him naive

  • and it was really hurt.

  • Suddenly he's laughed at not only by a machine,

  • but the audience are now bursting out laughing.

  • My name is Assaf Gavron,

  • I'm an author most of the days.

  • I'm in charge of writing what we call generic texts.

  • Little bits and little phrases that are put together

  • by the software.

  • Metaphors or quotes, opening sentences, jokes.

  • We had a lot of discussions about who is Debater?

  • - My friend, you are speaking at the extremely

  • fast rate of 203 words per minute.

  • Don't hurry, we have plenty of time.

  • Please slow down and present your arguments calmly.

  • - We never write the texts to a specific debate.

  • We don't know what the debate will be.

  • No, the magic of Debater is that it can argue on

  • any topic thrown at it under the sun.

  • So we write a general text that could fit

  • some topics and then the system has to identify them

  • and use the right ones in the right time.

  • - Climate change is what allows them to poison the minds

  • of young kids.

  • - Science, we know the earth is warming.

  • Will there be more droughts?

  • Yes.

  • - How definitive is the evidence?

  • Is there any room for debate?

  • - This is the Natasha Hall Show, on CJAD800.

  • - We are speaking with Dr. Aharanoff,

  • she is the global manager for IBM's

  • artificial intelligence Project Debater,

  • thanks for being here.

  • How nuts is that it could take artificial intelligence

  • to teach us to understand other human perspectives better?

  • - I think that's where AI should be going.

  • - Why?

  • Why do we need robots to teach us how to debate?

  • - Ultimately making a decision involves how you weigh

  • your different values.

  • It's not all about information and I think

  • that AI is not gonna do for you.

  • But it can give