56
Fluidly representing the world: Way, way harder than you think

Fluidly representing the world: Way, way harder than you think

Embed Size (px)

Citation preview

Page 1: Fluidly representing the world: Way, way harder than you think

Fluidly representing the world:Way, way harder than you think

Page 2: Fluidly representing the world: Way, way harder than you think

The One Thing to Remembera year from now

Rules and statistics must DYNAMICALLY INTERACT to represent the world

appropriately in computer models of intelligence.

Rules are not enough

Statistics is not enough.

Page 3: Fluidly representing the world: Way, way harder than you think

Let’s start things off with a question that you will never forget...

Page 4: Fluidly representing the world: Way, way harder than you think

What do cows drink? FIRST ANSWER

COW MILK DRINK

Statistical, bottom-up representations

Bottom-up, Statistical(e.g., Connectionism)

conscious

unconscious

Page 5: Fluidly representing the world: Way, way harder than you think

These neurons are activated without ever have heard the

word “milk”

conscious

unconsciousCOW MILK DRINK

What do cows drink? FIRST ANSWER

Bottom-up, Statistical(e.g., Connectionism)

Page 6: Fluidly representing the world: Way, way harder than you think

What do cows drink? RIGHT ANSWER

Top-Down, Rule-Based(e.g., Symbolic AI)

Facts about the world: ISA(cow, mammal)

ISA(mammal, animal)

Rule 1: IF animal(X) AND thirsty(X) THEN lack_water(X)

Rule 2: IF lack_water(X) then drink_water(X)

Conclusion: COWS DRINK WATER

Page 7: Fluidly representing the world: Way, way harder than you think

What do cows drink? Context 1: Milk Context 2: Water

COW MILK DRINK

Statistical, bottom-up representations

Rule-based representations

Bottom-up + Top-Down

conscious

unconscious

Page 8: Fluidly representing the world: Way, way harder than you think

When a computer will be able to represent the world in such a way that, when it is asked, “What do cows drink?” it will be able to answer either “milk” or “water,” depending on the context of the question.

Artificial Intelligence:

Page 9: Fluidly representing the world: Way, way harder than you think

Rules are not enough

Page 10: Fluidly representing the world: Way, way harder than you think

A

Page 11: Fluidly representing the world: Way, way harder than you think

Rules to characterize an “A”

• Two oblique, approximately vertical lines, slanting inwards and coming together at the top of the figure

• An approximately horizontal crossbar• An enclosed area above an open area• Vertical lines longer than the crossbar

Page 12: Fluidly representing the world: Way, way harder than you think
Page 13: Fluidly representing the world: Way, way harder than you think

Characterize this word

Page 14: Fluidly representing the world: Way, way harder than you think

• It says “Gutenberg”

• It is written in a Gothic font.

• It looks like very old-fashioned script.

• Germans would have an easier time reading it than Italians.

• It has a very German feel to it.

• It is nine letters long.

• It contains the letter “e” twice, and the letter “u” once.

• It means “good mountain” in English

• It makes one think of the Vulgate Bible.

• The “t” is written in mirror-image.

• ……..

Page 15: Fluidly representing the world: Way, way harder than you think
Page 16: Fluidly representing the world: Way, way harder than you think
Page 17: Fluidly representing the world: Way, way harder than you think
Page 18: Fluidly representing the world: Way, way harder than you think
Page 19: Fluidly representing the world: Way, way harder than you think
Page 20: Fluidly representing the world: Way, way harder than you think
Page 21: Fluidly representing the world: Way, way harder than you think
Page 22: Fluidly representing the world: Way, way harder than you think

• And can also be read perfectly upside-down!…

Question: But is that really a property of “Gutenberg”?

Answer: Yes, in some contexts.

But does that mean that the representation of “Gutenberg” must always include: “Can be read upside-down when written in a VERY special way in pseudo-Gothic script”?

Page 23: Fluidly representing the world: Way, way harder than you think

Moral of the Story

In the real world, a fixed set of rules, however long, is never sufficient to cover all possible contexts in which an object or

a situation occurs.

Page 24: Fluidly representing the world: Way, way harder than you think

But the statistics of the environment are not enough, either.

Page 25: Fluidly representing the world: Way, way harder than you think

Without top-down conceptual knowledge we have no hope of understanding the following image

Page 26: Fluidly representing the world: Way, way harder than you think

Statistics is not enough

Page 27: Fluidly representing the world: Way, way harder than you think

Statistics is not enough

“A dark spot. Hmmm…. Doesn’t look like anything.”

Page 28: Fluidly representing the world: Way, way harder than you think

Statistics is not enough

“Pictures often have faces in them. Is that a face in the lower-left hand corner?”

Page 29: Fluidly representing the world: Way, way harder than you think

Statistics is not enough

“Nah, doesn’t join up with anything else.”

Page 30: Fluidly representing the world: Way, way harder than you think

Statistics is not enough

Page 31: Fluidly representing the world: Way, way harder than you think

Statistics is not enough

“Oh, THERE’s a face.”

Page 32: Fluidly representing the world: Way, way harder than you think

Statistics is not enough

“But, again, it doesn’t make sense, just an isolated face…”

Page 33: Fluidly representing the world: Way, way harder than you think

Statistics is not enough

“Let’s look at that dark spot again. A shadow?”

Page 34: Fluidly representing the world: Way, way harder than you think

Statistics is not enough

“Hey, TREES produce shadows. Is there a tree around? THAT could be a tree!”

Page 35: Fluidly representing the world: Way, way harder than you think

Statistics is not enough

“If that’s a tree, that could be a metal grating.”

Page 36: Fluidly representing the world: Way, way harder than you think

Statistics is not enough

“But trees with metal gratings like that are on sidewalks. So where’s the kerb?”

Page 37: Fluidly representing the world: Way, way harder than you think

Statistics is not enough

“If this is a kerb, it should go on straight. Does it? Yes, sort of. Now what’s this thing in the middle?”

Page 38: Fluidly representing the world: Way, way harder than you think

Statistics is not enough

“That sort of looks like a dog’s head. That could make sense.”

Page 39: Fluidly representing the world: Way, way harder than you think

Statistics is not enough

“But heads are attached to bodies, so there should be a body. Hey, that looks like it could be a front leg.”

Page 40: Fluidly representing the world: Way, way harder than you think

Statistics is not enough

“The rest fits pretty well with that interpretation, too. Why so spotty?A dalmatian, yeah, sure, drinking from a road gutter under a tree.”

Page 41: Fluidly representing the world: Way, way harder than you think

Dynamically converging on the appropriate representation

Bottom-up

Top-down

Context-dependentrepresentation

Page 42: Fluidly representing the world: Way, way harder than you think

N.B. The representation must allow a machine to understand the object in essentially the same way we do.

Consider representing an ordinary, everyday object

Page 43: Fluidly representing the world: Way, way harder than you think

The object:A cobblestone

Page 44: Fluidly representing the world: Way, way harder than you think

A “cobblestone” is like….

• a brick

• asphalt

• the hand-brake of a car

• a paper-weight

• a sharpening steel

• a ruler

• a weapon

• a brain

• a weight in balance pans

• a nutcracker

• Carrera marble

• a bottle opener

• a turnip in French

• an anchor for a (little) boat

• a tent peg

• a hammer

• etc….

• a voter’s ballot...

Page 45: Fluidly representing the world: Way, way harder than you think

May 1968, Paris

UNDER 21, this is your ballot.

Page 46: Fluidly representing the world: Way, way harder than you think

So, how could a machine ever produce context-dependent

representations?

Page 47: Fluidly representing the world: Way, way harder than you think

Searching for an analogy

Context

- I was late in reviewing a paper.- The editor, a friend, said, “Please get it done now.”- I put the paper on my cluttered desk, in plain view, so that it simply COULD NOT ESCAPE MY ATTENTION…it would keep making me feel guilty till I got it done.- I wanted to find a nice way of saying I had devised a way so that the paper would continue to bug me till I had gotten it done

Page 48: Fluidly representing the world: Way, way harder than you think

Bottom-up: statistics-based “sub-cognitive”, unconscious par of network

Top-down: semantic, rule-based, conscious part of network

Swat mosquito

Do the review

“Something that won’t go away until you’ve taken care of it”

hungry child

No, too harsh to relate a paper to review to a hungry child

an itch

No, too easy to just scratch it; papers are hard

to review

dandelions

You can’t make them go away, ever

a mosquito

Until you get up and swat it it NEVER

stops buzzing!

Page 49: Fluidly representing the world: Way, way harder than you think

Representations of an object/situation must always be tailored to the context in which they are found. A

machine must be able to do this automatically.

Page 50: Fluidly representing the world: Way, way harder than you think
Page 51: Fluidly representing the world: Way, way harder than you think
Page 52: Fluidly representing the world: Way, way harder than you think
Page 53: Fluidly representing the world: Way, way harder than you think

Solar System Representation used in BACON (Langley et al., 1980) derived from Kepler’s data.

?

Page 54: Fluidly representing the world: Way, way harder than you think

Is the appropriate representation of this figure ….

Page 55: Fluidly representing the world: Way, way harder than you think

….. just the juxtaposition of 64 of these?

And if so, just how is the flickering that we humans see represented?

Page 56: Fluidly representing the world: Way, way harder than you think

Fluidly Representing the World: the Hard Problem of Artificial Intelligence

Discovering how a system could develop dynamically evolving, context-sensitive representations is almost certainly The Hard Problem of artificial intelligence. Moreover, it always has been; it just hasn’t been recognized as such.

We argue for a continual interaction between bottom-up and top-down processing, thereby allowing the system, on its own, to dynamically converge to a the context-dependent representations it needs.

This ability is at the heart of our own intelligence and AI MUST face the challenge of learning how to fluidly represent the world if there is ever to be any chance of developing autonomous agents with even a glimmer of real human intelligence.