All language subtitles for 07-Lecture 1 Segment 7 Unintentionally Funny Stories.en

Alright, so you say,

what the heck am I talking about. What it
means have an unintentionally funny story?

Well, people did work on storytelling.
Why do we want to have AIs that tell stories?

Because we want to figure
out what it means to communicate

kind of a coherent description of some
sequence of events.

That would like telling a story, right--Describe what you saw.

And so people worked in the eighties--and
they still work on this though with a very

different techniques

--people worked in the eighties on

taking a representation of the world and
turning it into language through

some storytelling device. A famous
example of this was the TALE-SPIN system of Shank,

and I'm gonna show you some of its
output. It's gonna do, I think a good job of illustrating

why this endeavor of logic-based, representation-heavy AI,

even people were trying and doing really clever stuff,

why it hit some walls.

So let's see some unintentionally funny
stories.

I'll read them for you.

Okay, like, I can just see these kind
of systems in the eighties trying to

like describe the stuff to people.

And here's the important point under all of this.

It is amazingly close.

It does so much right,

and it's funny because there's like one
link missing.

So, the bear is hungry, he talks to somebody
about where the honey is because honey will

fix his hunger. The friend replies there's this beehive, it's in the oak tree, that's a

useful answer, right.

Irving could have said "I have a
foot", right, that wouldn't have been helpful.

Joe walks over because you can't eat
unless you're there, and then he eats the beehive.

It was so close.

And what went wrong was simply in all of
this knowledge was missing the fact that

if you want what's in the container, the
honey,

you have to first remove it from the
container,

and not simply eat the container.

One piece in a chain missing, and the
whole chain falls apart.

This is why logical systems were too
fragile, and this is why people realize

that uncertainty was just fact of life.

You were always gonna be missing
some piece.

Okay, here's another unintentionally funny
story.

It's sooo close.

I mean, I guess it's kind of a weird story to begin
with even without the gravity thing but,

um...

What went wrong here?

Well, it's in fact describing a rather
unfortunate

squirrel drowning, right, and what happened
was there was a drowning event.

The squirrel,

was kind of

the "drownee", right, that's what drowned, and

what caused the drowning?

Well, gravity. I mean something's got to cause the drowning, right?

In rendering this into language, the
system had all the information but it

didn't realize that for a drowning event
it's really important that you say who drowned,

and it's not so important that
you say what force of nature

caused the drowning.

So we were so close, one little mess up.
The information was there and suddenly

it's a funny story rather than a
demonstration of a successful

storytelling system.

I'll give you one more.

There's no language problem here, there's
no like

physics violation. It's just kind of
weird. It's like, the

fox is dishonest.

Why did you tell me that. There's this whole digression about the piece of cheese.

This is a story about a fox walking over
to a crow.

So it's a little weird.

And the failing here is not a failing of
representing the world, or rendering it into language.

It's a failing of knowing what is
worth saying in the context.

That's the same issue of context again. Very challenging. There are so many

ways has to go wrong and we've had to
deal with these things in building modern systems.

Part of the way we've dealt with them is by changing our tools.

Part of the way we've dealt with them is by changing the way we formulate the problems,

and part of the way we've dealt with them is just by making things robust in various ways.

Okay, you say we can do better now.

Well, can we?

So, what's this, this is Siri, and what happens when you tell Siri, "Tell me a story"?

How does Siri do?

Once upon a time... no it's too silly.

Right, so at least we now know that the
story is going to be silly and

do not try to say it.

So is that progress uh... I don't know.