WEBVTT
00:00:04.238 --> 00:00:10.409
¶ ¶
00:00:13.781 --> 00:00:16.415
NARRATOR: THERE'S
A QUESTION WE ALL ASK.
00:00:16.500 --> 00:00:19.785
WHAT'S GOING TO HAPPEN TOMORROW?
00:00:20.621 --> 00:00:21.704
NEXT WEEK?
00:00:21.789 --> 00:00:23.372
NEXT YEAR?
00:00:23.457 --> 00:00:27.760
WILL THE FUTURE BRING
US HAPPINESS OR SORROW?
00:00:28.679 --> 00:00:31.296
IN THE PAST, WE LOOKED
TO PSYCHICS, SEERS,
00:00:31.382 --> 00:00:33.882
AND ASTROLOGERS FOR ANSWERS.
00:00:33.968 --> 00:00:36.468
TODAY'S FORTUNETELLERS
ARE STATISTICIANS
00:00:36.554 --> 00:00:39.304
AND SOFTWARE ENGINEERS.
00:00:42.693 --> 00:00:45.844
BY FINDING THE HIDDEN PATTERNS
IN ENORMOUS AMOUNTS OF DATA,
00:00:45.946 --> 00:00:49.148
THEY CAN PREDICT THE FUTURE
WITH FAR GREATER ACCURACY
00:00:49.233 --> 00:00:51.784
THAN EVER BEFORE.
00:00:52.870 --> 00:00:55.654
THIS REVOLUTION IS
POWERED BY A NEW FORM
00:00:55.740 --> 00:00:58.657
OF ARTIFICIAL INTELLIGENCE.
00:00:59.827 --> 00:01:03.695
BUT BY SOLVING ONE PROBLEM,
HAVE WE CREATED ANOTHER?
00:01:03.798 --> 00:01:06.498
IF WE PEER INTO THE CRYSTAL
BALL, WILL WE SEE
00:01:06.584 --> 00:01:09.384
A WORLD WHERE MACHINES
SHAPE THE FUTURE,
00:01:09.503 --> 00:01:11.503
LEAVING THEIR
CREATORS FAR BEHIND?
00:01:45.923 --> 00:01:49.508
NARRATOR: IN THE 16TH CENTURY,
THE FRENCH SEER NOSTRADAMUS
00:01:49.593 --> 00:01:51.927
PUBLISHED "LES PROPHETIES,"
00:01:52.046 --> 00:01:56.515
A BOOK OF VAGUELY-WORDED
PREDICTIONS OF THINGS TO COME.
00:01:57.351 --> 00:02:00.352
HE CLAIMED HIS PROPHECIES WERE
BASED ON ASTROLOGICAL CHARTS
00:02:00.437 --> 00:02:03.939
AND VISIONS FROM BEYOND.
00:02:06.060 --> 00:02:08.777
NOSTRADAMUS FORESAW
TERRIBLE PLAGUES, WARS,
00:02:08.896 --> 00:02:13.098
EARTHQUAKES, FLOODS,
AND THE END OF THE WORLD.
00:02:13.200 --> 00:02:16.952
NOSTRADAMUS: NOT YET.
00:02:17.571 --> 00:02:21.490
NARRATOR: BUT HE
DIDN'T GIVE ANY DATES.
00:02:22.793 --> 00:02:26.328
MORE THAN 400 YEARS AFTER HE
UTTERED HIS LAST PROPHECY,
00:02:26.413 --> 00:02:29.748
THE APOCALYPSE
HAS YET TO ARRIVE.
00:02:29.834 --> 00:02:34.419
BUT OUR DESIRE TO PEER INTO
THE FUTURE STILL BURNS BRIGHT.
00:02:36.807 --> 00:02:39.624
ONCE, WE RELIED ON MYSTICISM.
00:02:39.727 --> 00:02:45.097
NOW, WE TURN TO ALGORITHMS.
00:02:47.601 --> 00:02:50.068
EVERY YEAR, THE FLOW
OF DIGITAL INFORMATION
00:02:50.154 --> 00:02:53.805
GETS BIGGER AND WILDER.
00:02:53.908 --> 00:03:00.662
AT LEAST 2.5 QUINTILLION BYTES
OF DATA IS PRODUCED EVERY DAY.
00:03:00.781 --> 00:03:03.866
COMPUTERS COMB THROUGH THIS
TORRENT OF INFORMATION AND
00:03:03.951 --> 00:03:08.320
FIND THE DIGITAL
TRAILS WE LEAVE BEHIND.
00:03:09.290 --> 00:03:14.593
THEY LET US LOOK INTO OUR
PAST TO SEE THE FUTURE.
00:03:14.678 --> 00:03:17.963
BUT FINDING THOSE TRAILS,
SEPARATING THE SIGNALS FROM
00:03:18.048 --> 00:03:21.166
ALL THAT NOISE IS
NOT ALWAYS EASY.
00:03:25.689 --> 00:03:29.024
THE MEN AND WOMEN WHO
SUCCEED ARE THE SEERS OF
00:03:29.143 --> 00:03:30.809
THE 21ST CENTURY.
00:03:30.895 --> 00:03:31.777
NATE SILVER: HI, I'M NATE
SILVER, NOT NOSTRADAMUS.
00:03:31.862 --> 00:03:35.113
WE'RE QUITE OPPOSITE IN FACT.
00:03:35.849 --> 00:03:37.950
I'M THE FOUNDER AND EDITOR IN
CHIEF OF FIVETHIRTYEIGHT AND
00:03:38.052 --> 00:03:40.619
AUTHOR OF THE SIGNAL
AND THE NOISE.
00:03:43.157 --> 00:03:44.790
THE SIGNAL IS THE TRUTH.
00:03:44.875 --> 00:03:47.376
IT'S WHAT THE REAL WORLD
IS, FOR BETTER OR WORSE.
00:03:47.494 --> 00:03:49.527
IT'S THINGS THAT LET YOU
UNDERSTAND HOW ONE THING
00:03:49.630 --> 00:03:51.630
RELATES TO ANOTHER.
00:03:53.584 --> 00:03:57.886
AND NOISE IS JUST USELESS
INFORMATION OR DISTRACTING
00:03:58.005 --> 00:04:04.176
INFORMATION THAT PREVENT YOU
FROM UNDERSTANDING THE TRUTH.
00:04:06.680 --> 00:04:09.381
THE AMOUNT OF INFORMATION
IN THE WORLD IS, LIKE,
00:04:09.483 --> 00:04:11.650
DOUBLING EVERY YEAR OR TWO.
00:04:11.752 --> 00:04:13.936
IT DOESN'T MEAN THE AMOUNT OF
USEFUL KNOWLEDGE IN THE WORLD
00:04:14.021 --> 00:04:17.022
IS INCREASING AT NEARLY
THAT RATE, RIGHT?
00:04:17.107 --> 00:04:19.691
A LOT OF IT IS A THOUSAND TEXT
MESSAGES THAT YOUR TEENAGE
00:04:19.777 --> 00:04:23.128
DAUGHTER SENDS EVERY MONTH OR
CAT PHOTOS ON THE INTERNET OR
00:04:23.230 --> 00:04:24.863
WHATEVER ELSE, RIGHT?
00:04:24.965 --> 00:04:27.165
AND SO IN THAT SENSE, IT'S
LESS AND LESS OF
00:04:27.251 --> 00:04:31.003
A SIGNAL-RICH ENVIRONMENT
AND THAT KIND OF CHANGES
00:04:31.088 --> 00:04:33.538
THE NATURE OF PREDICTION.
00:04:35.926 --> 00:04:39.761
NARRATOR: HUMANS ARE
PREDICTING MACHINES.
00:04:40.965 --> 00:04:43.598
WE TRY TO GUESS HOW OTHER
PEOPLE WILL ACT AND
00:04:43.717 --> 00:04:48.520
THE POSSIBLE OUTCOMES
OF OUR OWN ACTIONS.
00:04:48.605 --> 00:04:51.390
WE CALCULATE THE ODDS
OF SUCCESS OR FAILURE,
00:04:51.475 --> 00:04:54.059
ESPECIALLY WHEN
WE'RE TAKING A RISK.
00:04:54.144 --> 00:04:55.694
NATE SILVER: IT'S ACTUALLY
AMAZING IF YOU'RE OUT HERE ON
00:04:55.779 --> 00:04:57.646
THE STREETS OF NEW YORK AND
PEOPLE ARE KIND OF MAKING
00:04:57.731 --> 00:04:59.648
PREDICTIONS OR CALCULATIONS
ALL THE TIME ABOUT,
00:04:59.733 --> 00:05:02.634
"CAN I CROSS THE STREET?"
00:05:06.740 --> 00:05:08.373
BECAUSE WE REALLY DON'T PAY
ATTENTION TO TRAFFIC LIGHTS
00:05:08.459 --> 00:05:09.574
AND STUFF HERE.
00:05:09.660 --> 00:05:11.910
SO CAN I CROSS THE STREET
WITHOUT GETTING RUN OVER?
00:05:11.996 --> 00:05:13.912
HOW CAN I BE TALKING
ON MY CELL PHONE AND
00:05:13.998 --> 00:05:16.381
NOT RUN INTO MY
NEIGHBOR ON THE STREET?
00:05:16.467 --> 00:05:18.667
WAS IT FASTER TO
WALK OR TAKE A CAB?
00:05:18.752 --> 00:05:20.552
THOSE EVERYDAY
TYPES OF DISTINCTIONS,
00:05:20.637 --> 00:05:22.587
THEY'RE PREDICTIONS TOO.
00:05:28.095 --> 00:05:29.895
NARRATOR: HE MAY BE THE
MOST FAMOUS STATISTICIAN
00:05:29.980 --> 00:05:32.514
IN THE WORLD, BUT NATE
SILVER STARTED OUT
00:05:32.599 --> 00:05:35.100
APPLYING HIS GIFT FOR
CALCULATING THE ODDS
00:05:35.185 --> 00:05:39.021
TO A LUCRATIVE
CAREER PLAYING POKER.
00:05:39.106 --> 00:05:43.241
LATER, HE BEGAN HANDICAPPING
BASEBALL AND POLITICS.
00:05:43.327 --> 00:05:47.079
ALL THREE HAVE MORE IN
COMMON THAN YOU MAY THINK.
00:05:47.164 --> 00:05:49.147
NATE SILVER: I MEAN I THINK
THERE'S NOTHING QUITE LIKE
00:05:49.249 --> 00:05:52.701
POKER FOR REALLY LETTING YOU
EXPERIENCE WHAT PROBABILITY
00:05:52.786 --> 00:05:57.122
FEELS LIKE IN THE LONG RUN.
00:05:58.459 --> 00:06:00.258
'CAUSE YOU CAN SAY RIGHT NOW
AS WE'RE FILMING THIS,
00:06:00.344 --> 00:06:02.177
THIS WILL EITHER
LOOK RIDICULOUS OR BRILLIANT,
00:06:02.296 --> 00:06:04.296
BUT WE HAVE HILLARY CLINTON
WITH A 70% CHANCE OF WINNING
00:06:04.381 --> 00:06:08.333
THE PRESIDENCY, ROUGHLY, AND
DONALD TRUMP A 30% CHANCE.
00:06:08.435 --> 00:06:09.601
BY THE TIME THIS IS AIRING,
00:06:09.686 --> 00:06:11.853
YOU GUYS WILL
KNOW WHAT HAPPENED.
00:06:11.972 --> 00:06:15.173
BUT MOST PEOPLE, WHEN YOU SAY
70-30, THEY DON'T KNOW QUITE
00:06:15.275 --> 00:06:16.191
WHAT THAT MEANS.
00:06:16.310 --> 00:06:18.276
THEY DON'T KNOW WHAT THE
30% REALLY FEELS LIKE.
00:06:20.814 --> 00:06:22.114
WHEN YOU PLAY POKER,
00:06:22.199 --> 00:06:24.015
THERE ARE CERTAIN DRAWS
YOU MIGHT HAVE, RIGHT?
00:06:24.118 --> 00:06:27.369
WHERE CERTAIN TYPES OF FLUSH
DRAWS MIGHT HAVE A 30% CHANCE
00:06:27.488 --> 00:06:28.820
OF COMING THROUGH.
00:06:28.906 --> 00:06:32.374
AND YOU EXPERIENCE THAT ON
BOTH SIDES HUNDREDS OF TIMES
00:06:32.493 --> 00:06:35.160
OVER THE COURSE
OF YOUR CAREER.
00:06:38.332 --> 00:06:41.566
AND SO YOU JUST KIND OF
GET A VISCERAL UNDERSTANDING
00:06:41.668 --> 00:06:44.753
OF WHAT PROBABILITY IS LIKE.
00:06:44.838 --> 00:06:47.889
NARRATOR: BY APPLYING
RIGID MATHEMATICAL PRECISION TO
00:06:48.008 --> 00:06:52.060
POLITICS, SILVER SUCCESSFULLY
CALLED 49 OUT OF 50 STATES IN
00:06:52.179 --> 00:06:57.983
THE 2008 US GENERAL ELECTION
AND ALL 50 STATES IN 2012.
00:06:59.353 --> 00:07:00.986
A KEY PART OF HIS APPROACH IS
00:07:01.071 --> 00:07:04.222
TO CONTINUALLY
UPDATE INFORMATION.
00:07:04.324 --> 00:07:07.275
HE DOESN'T MAKE ONE
PREDICTION AND STICK WITH IT.
00:07:07.361 --> 00:07:10.996
HIS FORECAST CHANGES AS
HE GATHERS MORE DATA.
00:07:11.081 --> 00:07:13.898
NATE SILVER: YOU ALWAYS START
OUT WITH A CERTAIN BELIEF
00:07:14.001 --> 00:07:15.617
ABOUT WHAT THE
WORLD LOOKS LIKE.
00:07:15.702 --> 00:07:18.904
AND YOU'RE ALWAYS
CHALLENGING THAT BELIEF.
00:07:19.973 --> 00:07:20.738
NARRATOR: ONE OF THE
BIGGEST CHALLENGES TO
00:07:20.841 --> 00:07:23.208
SILVER'S BELIEF CAME IN 2016.
00:07:23.293 --> 00:07:24.909
DONALD TRUMP: "YEAH-UH!"
00:07:25.012 --> 00:07:26.678
NARRATOR: WHEN DONALD
TRUMP FIRST ENTERED THE
00:07:26.763 --> 00:07:32.250
PRESIDENTIAL CAMPAIGN, SILVER
PUT HIS ODDS OF WINNING AT 50-1.
00:07:32.352 --> 00:07:35.120
BUT IN THE DAYS LEADING
UP TO THE ELECTION,
00:07:35.222 --> 00:07:38.523
HE DECLARED THE GAP BETWEEN
CANDIDATES WAS CLOSING AND
00:07:38.625 --> 00:07:43.778
THE ODDS OF A TRUMP VICTORY
WERE GROWING EVERY DAY.
00:07:44.398 --> 00:07:45.930
FOR THIS, HE WAS MOCKED
BY THOSE WHO CONFIDENTLY
00:07:46.033 --> 00:07:49.985
PREDICTED AN EASY CLINTON WIN.
00:07:57.494 --> 00:08:00.412
NATE SILVER: PEOPLE ARE VERY
STUBBORN ABOUT THEIR POINTS OF
00:08:00.497 --> 00:08:03.715
VIEW, ESPECIALLY IN REALMS
LIKE POLITICS, ESPECIALLY IN
00:08:03.800 --> 00:08:06.084
REALMS WHERE THAT INVOLVE
EXPERTISE, WHERE YOUR
00:08:06.170 --> 00:08:10.088
REPUTATION IS BANKED ON A
CERTAIN THEORY BEING CORRECT.
00:08:10.174 --> 00:08:13.258
THE SIGNAL AND NOISE IS ALL
ABOUT UNCERTAINTY AND
00:08:13.343 --> 00:08:15.627
HUMILITY AND THE
CHALLENGES OF FORECASTING.
00:08:15.729 --> 00:08:16.595
AND THEN PEOPLE ARE LIKE,
00:08:16.680 --> 00:08:17.512
"OH, THIS GUY CAN
PREDICT ANYTHING."
00:08:17.598 --> 00:08:19.397
AND SO THERE'S THAT CLASH.
00:08:19.499 --> 00:08:22.067
IT RESOLVES ITSELF ONCE YOU
START TO SCREW THINGS UP, RIGHT?
00:08:22.169 --> 00:08:24.269
ONCE YOU START TO MAKE SOME
BAD PREDICTIONS, IRONICALLY,
00:08:24.354 --> 00:08:26.905
THEN PEOPLE KIND OF UNDERSTAND
IT BETTER, IN A WEIRD WAY.
00:08:26.990 --> 00:08:29.807
BUT IT'S NOT ABOUT BEING
CLAIRVOYANT, IT'S NOT ABOUT
00:08:29.910 --> 00:08:31.993
BEING CERTAIN, IT'S
ABOUT PROBABILITY.
00:08:32.112 --> 00:08:34.812
YOU HOPE YOU'RE RIGHT
SLIGHTLY MORE OFTEN THAN
00:08:34.915 --> 00:08:36.314
THE AVERAGE PERSON.
00:08:36.416 --> 00:08:39.251
YOU'RE PROBABLY NOT GONNA BE
RIGHT MORE THAN SLIGHTLY MORE
00:08:39.336 --> 00:08:41.286
OFTEN BECAUSE THE WORLD'S
A COMPLICATED PLACE.
00:08:43.624 --> 00:08:44.656
BARISTA: WHAT'S ALL THIS?
00:08:44.758 --> 00:08:45.757
NOSTRADAMUS: EXCUSE ME?
00:08:45.842 --> 00:08:46.991
BARISTA: YOUR OUTFIT?
00:08:47.094 --> 00:08:49.177
NOSTRADAMUS: I AM
NOSTRADAMUS, THE SEER.
00:08:49.296 --> 00:08:52.347
BARISTA: NOSTRADAMUS,
MAY I ASK YOU A QUESTION?
00:08:52.466 --> 00:08:55.016
WHEN AM I GONNA DIE?
00:08:55.135 --> 00:08:56.768
NOSTRADAMUS: I DON'T
FIXATE ON MINUTIAE,
00:08:56.853 --> 00:08:58.353
JUST THE BIG PICTURE.
00:08:58.472 --> 00:09:00.355
BARISTA: ALL RIGHT.
00:09:01.525 --> 00:09:03.275
NARRATOR: IT'S SAID THAT
WHEN IT COMES TO MANKIND,
00:09:03.377 --> 00:09:10.148
ONLY ONE THING IS CERTAIN,
WE'RE ALL GOING TO DIE.
00:09:11.618 --> 00:09:16.154
WE DON'T KNOW EXACTLY WHEN,
BUT INSURANCE COMPANIES
00:09:16.256 --> 00:09:19.024
HAVE BEEN BETTING ON
IT FOR CENTURIES.
00:09:19.126 --> 00:09:22.077
DALE HALL: A LONG TIME
AGO, THE BEST ESTIMATOR OF
00:09:22.162 --> 00:09:25.297
MORTALITY THAT ANYONE COULD
DO WAS JUST ASK AN AGE.
00:09:25.382 --> 00:09:26.381
HOW OLD ARE YOU?
00:09:26.500 --> 00:09:28.032
AND THAT'S PROBABLY GONNA
BE YOUR BEST ESTIMATOR.
00:09:28.135 --> 00:09:31.035
THEN WE LEARNED OVER TIME
THAT THERE'S ALSO ADDITIONAL
00:09:31.138 --> 00:09:34.005
VARIABLES THAT
ARE VERY HELPFUL.
00:09:40.514 --> 00:09:44.549
AND YOU START COLLECTING
SOME PIECES OF IN,
00:09:44.651 --> 00:09:46.618
THAT INFORMATION THAT
COMPANIES CONTRIBUTE TO GET AT
00:09:46.720 --> 00:09:51.273
A BROAD SET AND A BROAD LOOK
FOR A FINAL MORTALITY TABLE.
00:09:54.411 --> 00:09:56.077
CHRISTINE HOFBECK: SO THE
SOCIETY OF ACTUARIES ALREADY
00:09:56.196 --> 00:09:59.664
HAS MORTALITY TABLES THAT
MIGHT SUGGEST THE PROBABILITY
00:09:59.750 --> 00:10:01.566
OF YOUR DEATH AT
DIFFERENT AGES.
00:10:01.668 --> 00:10:04.619
GENERALLY WHAT WE'RE DOING
WHEN WE DO UNDERWRITING OR
00:10:04.705 --> 00:10:07.205
WHEN WE GET INFORMATION ON
WHETHER TO WRITE A POLICY IS
00:10:07.291 --> 00:10:10.292
TO DECIDE IF YOU ARE MORE
OR LESS RISKY THAN
00:10:10.377 --> 00:10:11.743
THE AVERAGE PERSON.
00:10:11.845 --> 00:10:13.011
DALE HALL: GENDER,
MALE VS FEMALE.
00:10:13.096 --> 00:10:14.296
DO YOU SMOKE OR NOT?
00:10:14.381 --> 00:10:17.299
TOBACCO USER VERSUS
NON-TOBACCO USER.
00:10:17.384 --> 00:10:20.051
YOUR RELATIVE HEALTH
NOW GETS UNDERWRITTEN,
00:10:20.137 --> 00:10:21.553
SO WHAT'S YOUR
CHOLESTEROL LEVEL?
00:10:21.638 --> 00:10:23.138
WHAT'S YOUR BODY MASS INDEX?
00:10:23.223 --> 00:10:25.273
CHRISTINE HOFBECK: SO IF
YOU ARE LESS RISKY,
00:10:25.392 --> 00:10:28.860
THEN YOU ARE A RISK THAT
WE WOULD LIKE TO WRITE.
00:10:31.865 --> 00:10:34.766
NARRATOR: ACTUARIES ESTIMATE
THE RISK OF YOUR DYING SOONER
00:10:34.868 --> 00:10:36.951
OR LIVING LONGER THAN AVERAGE.
00:10:37.070 --> 00:10:39.120
IF THEIR CALCULATIONS ARE
WRONG OR IF YOU DIE IN
00:10:39.239 --> 00:10:42.240
A FREAK ACCIDENT, THEY
WILL LOSE MONEY ON YOU.
00:10:42.326 --> 00:10:45.543
BUT ON AVERAGE,
THEIR BET PAYS OFF.
00:10:47.414 --> 00:10:52.167
IT'S A MATTER OF NUMBERS,
THE LAW OF LARGE NUMBERS.
00:10:54.304 --> 00:10:57.222
THE LAW STATES THAT THE MORE
MEMBERS ARE IN AN INSURED GROUP,
00:10:57.307 --> 00:10:59.391
THE MORE LIKELY IT IS
THAT THE NUMBER OF ACTUAL
00:10:59.476 --> 00:11:04.346
LOSSES WILL BE VERY CLOSE TO
THE NUMBER OF EXPECTED LOSSES.
00:11:04.431 --> 00:11:07.065
THE SAME MATH IS BEHIND
GAMBLING CASINOS.
00:11:07.150 --> 00:11:09.234
DALE HALL: IN BUILDING ANY
MODEL, YOU HAVE TO UNDERSTAND
00:11:09.319 --> 00:11:13.988
HOW THE MATH WORKS, BUT THEN
USING THAT IN ORDER TO HELP
00:11:14.107 --> 00:11:16.775
PREDICT BETTER FINANCIAL
OUTCOMES OR PUTTING BETTER
00:11:16.860 --> 00:11:19.110
ESTIMATES AROUND
FINANCIAL OUTCOMES.
00:11:19.196 --> 00:11:20.445
THAT'S REALLY
THE HEART AND SOUL OF
00:11:20.530 --> 00:11:22.347
THE ACTUARIAL PROFESSION.
00:11:23.333 --> 00:11:24.949
CHRISTINE HOFBECK:
I HAVE NOT PLOTTED OUT MY
00:11:25.035 --> 00:11:26.418
PERSONAL LIFE SPAN.
00:11:26.503 --> 00:11:28.787
HOWEVER, I DO OFTEN
GET THE QUESTION OF
00:11:28.872 --> 00:11:31.506
"OH, YOU'RE AN ACTUARY?
TELL ME WHEN I'M GOING TO DIE."
00:11:34.928 --> 00:11:35.993
AT THE END OF THE DAY,
00:11:36.096 --> 00:11:37.462
IT'S ALL PROBABILITIES.
00:11:37.547 --> 00:11:40.999
I COULD WALK OUTSIDE
AND GET HIT BY A CAR.
00:11:41.101 --> 00:11:43.268
I HOPE THAT DOESN'T HAPPEN.
00:11:44.721 --> 00:11:47.605
NARRATOR: THE MORTALITY MODELS
USED BY THE INSURANCE INDUSTRY
00:11:47.691 --> 00:11:50.225
PAVED THE WAY FOR A
COMBINATION OF DATA SCIENCE
00:11:50.310 --> 00:11:54.062
AND FORTUNETELLING CALLED
PREDICTIVE ANALYTICS.
00:11:54.147 --> 00:11:57.899
ERIC SIEGEL IS ONE OF
THE FIELD'S HIGH PRIESTS.
00:11:57.984 --> 00:12:00.402
ERIC SIEGEL: PREDICTIVE
ANALYTICS I WOULD SAY IS
00:12:00.487 --> 00:12:04.656
THE LATEST EVOLUTIONARY STEP
OF THE INFORMATION AGE.
00:12:04.741 --> 00:12:08.493
PREDICTIVE ANALYTICS IS BIG
DATA TECHNOLOGY THAT LEARNS
00:12:08.578 --> 00:12:12.046
FROM THAT DATA, WHICH SERVES
AS EXPERIENCE, HOW TO MAKE
00:12:12.165 --> 00:12:15.583
PREDICTIONS FOR EACH
INDIVIDUAL PERSON.
00:12:16.670 --> 00:12:18.703
NARRATOR: PREDICTIVE
ANALYTICS FINDS PATTERNS
00:12:18.805 --> 00:12:21.005
HUMANS DON'T SEE.
00:12:21.091 --> 00:12:24.175
FOR INSTANCE, AN ALGORITHM
DISCOVERED PEOPLE WHOSE
00:12:24.261 --> 00:12:27.095
FIRST NAMES ARE UNCOMMON OR
TYPICALLY ASSOCIATED WITH THE
00:12:27.180 --> 00:12:30.014
OPPOSITE GENDER, SAY,
GIRLS NAMED "MASON"
00:12:30.100 --> 00:12:32.100
OR BOYS NAMED "ASHLEY"
00:12:32.185 --> 00:12:35.653
ARE MORE LIKELY TO BE
DEMOCRATS THAN REPUBLICANS.
00:12:37.607 --> 00:12:39.691
AND EMPLOYEES WITH CRIMINAL
RECORDS ARE LIKELY TO BE MORE
00:12:39.776 --> 00:12:45.113
PRODUCTIVE ON THE JOB THAN
PEOPLE WITH NO PAST ARRESTS.
00:12:46.616 --> 00:12:49.000
ERIC SIEGEL: WELL, HERE'S THE
DEAL, YOU DON'T ACTUALLY NEED
00:12:49.085 --> 00:12:50.335
TO PREDICT ACCURATELY.
00:12:50.420 --> 00:12:52.069
PREDICTING BETTER
THAN GUESSING IS THE
00:12:52.172 --> 00:12:53.705
NAME OF THE GAME.
00:12:53.790 --> 00:12:57.175
ACROSS ALL OF THESE PROCESSES
THAT ARE IMPROVED BY
00:12:57.260 --> 00:13:00.762
PREDICTIVE ANALYTICS, IT'S
IMPROVING IT BY ADDING
00:13:00.881 --> 00:13:02.714
A LITTLE PREDICTION.
00:13:04.217 --> 00:13:06.250
NARRATOR: THE ABILITY TO
MAKE THESE STRANGE BUT USEFUL
00:13:06.353 --> 00:13:08.970
PREDICTIONS IS BEING
FUELED BY AN EVEN MORE
00:13:09.055 --> 00:13:12.423
SIGNIFICANT BREAKTHROUGH,
THE DEVELOPMENT OF
00:13:12.526 --> 00:13:16.294
A NEW FORM OF
ARTIFICIAL INTELLIGENCE.
00:14:00.574 --> 00:14:02.740
COMPUTERS HAVE BEEN BETTER
THAN HUMANS AT PLAYING CHESS
00:14:02.826 --> 00:14:06.778
SINCE THE 1990S.
00:14:07.948 --> 00:14:09.581
THAT'S BECAUSE THE MACHINES
GOT BETTER AND BETTER AT
00:14:09.666 --> 00:14:12.784
PREDICTING THEIR
OPPONENT'S NEXT MOVE.
00:14:13.620 --> 00:14:17.672
BUT IN 2016, GOOGLE DID
EVEN BETTER THAN THAT.
00:14:17.791 --> 00:14:20.625
THEY CREATED AN ARTIFICIAL
INTELLIGENCE THAT DEFEATED
00:14:20.710 --> 00:14:24.178
A HUMAN GRANDMASTER
AT THE GAME OF GO.
00:14:28.051 --> 00:14:30.501
GO IS IMMENSELY COMPLEX.
00:14:30.604 --> 00:14:32.270
AN AVERAGE TURN IN
CHESS OFFERS ABOUT
00:14:32.355 --> 00:14:34.222
30 POSSIBLE MOVES.
00:14:34.307 --> 00:14:37.609
A GO TURN OFFERS 250.
00:14:37.694 --> 00:14:39.227
THERE ARE TOO MANY
POTENTIAL MOVES FOR
00:14:39.312 --> 00:14:43.231
ANY COMPUTER TO PREDICT.
00:14:43.900 --> 00:14:46.868
SO THE GOOGLE TEAM DEVELOPED
A NEW FORM OF AI CALLED A
00:14:46.987 --> 00:14:50.622
DEEP LEARNING NEURAL NETWORK.
00:14:52.826 --> 00:14:55.543
DEEP LEARNING DRAWS ON
HARDWARE AND SOFTWARE THAT
00:14:55.662 --> 00:15:00.999
LOOSELY APPROXIMATE THE WEB
OF NEURONS IN THE HUMAN BRAIN.
00:15:01.534 --> 00:15:04.135
GOOGLE FED THE COMPUTER
MILLIONS OF GAMES,
00:15:04.220 --> 00:15:07.872
THEN HAD IT PLAY
NONSTOP AGAINST ITSELF.
00:15:07.974 --> 00:15:10.391
AND SOON, IT LEARNED.
00:15:10.510 --> 00:15:14.395
IT GAINED SOMETHING
LIKE INTUITION.
00:15:14.514 --> 00:15:17.899
THIS WAS A MAJOR BREAKTHROUGH
IN AI BUT WHAT DOES IT MEAN
00:15:18.018 --> 00:15:20.351
FOR HUMAN BEINGS?
00:15:21.021 --> 00:15:24.022
A SUPER POWERED AI COULD
CHANGE THE WORLD BUT WILL IT
00:15:24.107 --> 00:15:26.491
CHANGE IT FOR THE BETTER?
00:15:32.816 --> 00:15:34.916
BARISTA: HERE'S YOUR
DOUBLE LATTE, SIR.
00:15:35.269 --> 00:15:37.135
CHRIS: HEY, NOSTRADAMUS!
00:15:37.221 --> 00:15:38.887
I DIDN'T GET A CHANCE
TO SAY HI EARLIER.
00:15:38.972 --> 00:15:41.940
I FELT YOUR COLD LOOMING
PRESENCE BEHIND ME IN LINE.
00:15:42.059 --> 00:15:43.591
IT'S A REAL PLEASURE.
00:15:43.694 --> 00:15:45.560
I THOUGHT YOU WERE ONE OF
THOSE ANGRY GUYS THAT SITS
00:15:45.646 --> 00:15:47.279
ON STAGE DURING A
COLLEGE COMMENCEMENT.
00:15:47.398 --> 00:15:48.947
NOSTRADAMUS: DO YOU HAVE
QUESTIONS FOR NOSTRADAMUS?
00:15:49.066 --> 00:15:50.265
CHRIS: YEAH, YEAH, YEAH!
00:15:50.367 --> 00:15:52.617
SO LET ME GET THIS STRAIGHT,
HOW DID YOU PREDICT THE FUTURE?
00:15:52.736 --> 00:15:55.237
NOSTRADAMUS: WELL, I KIND
OF STOLE FROM SOME SCHOLARLY
00:15:55.322 --> 00:15:56.204
HISTORIANS, CLASSICAL
HISTORIANS, AND THE SCARIER
00:15:56.290 --> 00:15:59.574
PARTS OF THE BIBLE.
00:15:59.660 --> 00:16:01.827
COUPLE OF PEOPLE I JUST
WHAT YOU CALL RIPPED OFF.
00:16:01.912 --> 00:16:05.330
AND I WAS VERY VAGUE ON DATES
AND PLACES TOO OF COURSE.
00:16:14.558 --> 00:16:18.310
NARRATOR: TODAY, PEOPLE DEMAND
A HIGHER DEGREE OF ACCURACY.
00:16:18.429 --> 00:16:21.930
SO THEY TURN TO DEEP
LEARNING NEURAL NETWORKS.
00:16:23.934 --> 00:16:26.868
DEEP LEARNING NEURAL NETWORKS
CAN PROCESS VAST AMOUNTS OF
00:16:26.970 --> 00:16:31.490
INFORMATION AND FAR GREATER
DEPTH THAN EVER BEFORE.
00:16:31.608 --> 00:16:36.178
LIKE HUMAN BRAINS, THEY'RE VERY
GOOD AT PATTERN RECOGNITION.
00:16:36.280 --> 00:16:39.197
WHEN A DEEP LEARNING NEURAL
NET SCANS A PICTURE, IT ASKS
00:16:39.283 --> 00:16:43.702
ITSELF, WHAT EXACTLY
AM I LOOKING AT?
00:16:43.787 --> 00:16:46.254
AND WHAT DO I DO
WHEN I SEE IT?
00:17:36.089 --> 00:17:38.039
BOB BOND: MY WIFE IS A GARDENER.
00:17:38.141 --> 00:17:42.344
AND SHE WAS HAVING TROUBLE
WITH NEIGHBOR CATS USING OUR
00:17:42.429 --> 00:17:44.229
YARD AS A LITTER BOX,
PARTICULARLY IN THE NIGHT WHEN
00:17:44.348 --> 00:17:46.565
WE COULDN'T KEEP AN EYE ON 'EM.
00:18:05.085 --> 00:18:09.788
I HAPPENED TO BE PLAYING WITH
THESE DEEP LEARNING SYSTEMS
00:18:09.873 --> 00:18:12.791
AND IT OCCURRED TO ME THAT THE
THING TO DO WOULD BE TO PUT A
00:18:12.876 --> 00:18:16.244
CAMERA IN THE FRONT YARD AND
SEE IF WE COULD TRAIN THE
00:18:16.346 --> 00:18:20.382
CAMERA TO RECOGNIZE CATS.
00:18:23.520 --> 00:18:28.089
BASICALLY, THE SYSTEM IS A
STANDARD SURVEILLANCE CAMERA
00:18:28.191 --> 00:18:30.892
IN THE YARD.
00:18:31.595 --> 00:18:33.695
THE SURVEILLANCE CAMERA DOESN'T
KNOW ANYTHING ABOUT CATS,
00:18:33.780 --> 00:18:36.982
WHAT IT DOES IS
IT DETECTS MOVEMENT.
00:18:37.651 --> 00:18:39.901
WHERE THIS SYSTEM IS
DIFFERENT, THOSE IMAGES ARE
00:18:39.987 --> 00:18:43.738
FED DIRECTLY INTO THIS
DEEP LEARNING SYSTEM.
00:18:43.824 --> 00:18:46.207
AND THE DEEP LEARNING
SYSTEM UNDERSTANDS HOW
00:18:46.293 --> 00:18:50.161
TO RECOGNIZE CATS.
00:18:50.247 --> 00:18:53.048
AND IF IT SEES A CAT IN
THE IMAGE, THEN IT TURNS ON
00:18:53.133 --> 00:18:55.050
THE SPRINKLING SYSTEM.
00:18:59.256 --> 00:19:03.508
I'M A PROGRAMMER WITH NVIDIA.
00:19:04.227 --> 00:19:07.795
NVIDIA DESIGNED THESE GPU
CHIPS TO BE GOOD AT IMAGE
00:19:07.898 --> 00:19:11.016
PROCESSING AND IT TURNS OUT
THAT THEY'RE VERY GOOD
00:19:11.101 --> 00:19:15.153
AT ARTIFICIAL INTELLIGENCE.
00:19:17.858 --> 00:19:20.808
THE REASON THEY'RE GOOD AT
ARTIFICIAL INTELLIGENCE IS
00:19:20.911 --> 00:19:24.579
THERE ARE MANY, MANY
PARALLEL PROCESSORS.
00:19:24.665 --> 00:19:27.866
AND EACH PARALLEL PROCESSOR
CAN BE PROGRAMMED TO TAKE A
00:19:27.951 --> 00:19:31.119
SMALL PART OF THE IMAGE OR A
SMALL PART OF THE ARTIFICIAL
00:19:31.204 --> 00:19:35.023
INTELLIGENCE NETWORK AND WORK
ON JUST THAT PIECE AND THEN
00:19:35.125 --> 00:19:39.761
THE PIECES OF ALL THESE
PARALLEL UNITS ARE AGGREGATED
00:19:39.846 --> 00:19:42.297
TOGETHER AND THE END
RESULT COMES OUT.
00:19:43.133 --> 00:19:44.966
NARRATOR: TO TRAIN HIS DEEP
LEARNING SYSTEM, BOB FED IT
00:19:45.052 --> 00:19:48.770
MILLIONS OF PICTURES OF
CATS ON THE INTERNET.
00:19:48.855 --> 00:19:50.972
WITH A LITTLE POSITIVE
RE-ENFORCEMENT, THE COMPUTER
00:19:51.058 --> 00:19:55.560
LEARNED HOW TO DIFFERENTIATE
CATS FROM OTHER ANIMALS.
00:19:58.815 --> 00:20:00.899
BOB BOND: WHAT YOU SEE HERE
IS THE CAT AS CAPTURED BY THE
00:20:00.984 --> 00:20:04.703
CAMERA AND THEN YOU CAN SEE
HOW THE DEEP LEARNING SYSTEM
00:20:04.821 --> 00:20:07.422
HAS CATEGORIZED THE
PIXELS IN THE CAT.
00:20:07.524 --> 00:20:09.574
NOW AT THE POINT
THAT I DEPLOYED IT,
00:20:09.660 --> 00:20:12.394
THE DEEP LEARNING SYSTEM
DIDN'T RECOGNIZE THEM VERY WELL.
00:20:12.496 --> 00:20:14.996
THE CATS ON THE INTERNET
ARE TYPICALLY, YOU KNOW,
00:20:15.098 --> 00:20:17.065
CUTE LITTLE GUYS LAYING
ON THE OWNER'S LAP AND
00:20:17.167 --> 00:20:20.335
A FRONTAL IMAGE OF THEM.
00:20:20.437 --> 00:20:21.586
AND THEY DON'T LOOK
LIKE THAT AT ALL.
00:20:21.672 --> 00:20:23.755
MY CATS ARE
GENERALLY SLINKERS.
00:20:23.840 --> 00:20:26.174
THEY'RE SLINKING
AROUND MY YARD.
00:20:28.729 --> 00:20:31.646
SO WHAT I HAD TO DO WAS TAKE
SOME OF THE IMAGES I CAPTURED
00:20:31.732 --> 00:20:35.233
OF THE CATS AND THEN ADD
THEM TO THE TRAINING.
00:21:04.798 --> 00:21:07.082
AND THAT WORKED
OUT REALLY WELL.
00:21:08.485 --> 00:21:11.519
ORIGINALLY, SOMETHING LIKE 30%
OF THE CATS WERE RECOGNIZED
00:21:11.605 --> 00:21:13.188
IN THE DEEP LEARNING SYSTEM.
00:21:13.273 --> 00:21:17.759
OVER THE COURSE OF THE NEXT
THREE MONTHS, I'VE RAMPED THAT
00:21:17.861 --> 00:21:21.062
UP GRADUALLY FROM THIRTY
PERCENT TO SIXTY PERCENT.
00:21:21.148 --> 00:21:22.981
THE CAMERA'S CURRENTLY
RECOGNIZING SOMETHING LIKE
00:21:23.066 --> 00:21:27.152
90% OF THE CATS THAT
COME INTO THE YARD.
00:21:28.438 --> 00:21:31.956
SO IT REALLY DOES
A VERY GOOD JOB.
00:21:32.793 --> 00:21:33.742
NARRATOR: SOME DEEP LEARNING
SYSTEMS ARE NOW BETTER AT
00:21:33.827 --> 00:21:38.379
RECOGNIZING IMAGES
THAN HUMAN BEINGS ARE.
00:21:38.465 --> 00:21:41.166
AND THAT INCREASE IN SYNTHETIC
BRAINPOWER IS HAVING A BIG
00:21:41.251 --> 00:21:46.254
EFFECT ON THE ACCURACY
OF OUR PREDICTIONS.
00:21:46.339 --> 00:21:49.340
FOR INSTANCE, PREDICTING
THE WEATHER MAY BE THE MOST
00:21:49.426 --> 00:21:52.960
CHALLENGING PROBLEM HUMANS
TACKLE ON A DAILY BASIS.
00:21:53.063 --> 00:21:56.347
THE WEATHER SYSTEM IS
UNIMAGINABLY COMPLEX.
00:21:56.433 --> 00:22:00.018
BUT BY COUPLING MACHINE
LEARNING TO PREDICTIVE MODELING,
00:22:00.103 --> 00:22:02.687
HUMANS HAVE GOTTEN MUCH
BETTER AT PULLING SIGNALS
00:22:02.773 --> 00:22:05.840
FROM VERY NOISY DATA.
00:22:07.911 --> 00:22:09.778
NATE SILVER: YOU PROBABLY
HEARD THE PHRASE THAT
00:22:09.863 --> 00:22:11.780
A BUTTERFLY FLAPS
ITS WINGS IN BEIJING AND
00:22:11.865 --> 00:22:16.584
THERE'S A TORNADO IN TEXAS,
WHERE ONE EVENT POTENTIALLY
00:22:16.670 --> 00:22:18.653
AFFECTS EVERY OTHER EVENT.
00:22:18.755 --> 00:22:20.088
IN WEATHER FORECASTING
FOR MANY THINGS,
00:22:20.173 --> 00:22:22.624
THEY'VE FIGURED OUT
EVENTUALLY YOU'RE KIND OF
00:22:22.709 --> 00:22:24.592
SIMULATING THE BUTTERFLY, RIGHT?
00:22:24.678 --> 00:22:27.161
BY HAVING A VERY, VERY
POWERFUL COMPUTER AND A VERY
00:22:27.264 --> 00:22:29.964
GOOD THEORY ABOUT HOW
THESE INTERACTIONS WORK.
00:22:33.353 --> 00:22:35.603
A GENERATION AGO, WEATHER
FORECASTING WAS KIND OF THE
00:22:35.689 --> 00:22:37.806
BUTT OF JOKES REALLY
AND IT REALLY WASN'T THAT
00:22:37.891 --> 00:22:40.608
MUCH BETTER THAN RANDOM.
00:22:40.694 --> 00:22:42.777
BUT IT'S IMPROVED
BY LEAPS AND BOUNDS.
00:22:42.863 --> 00:22:44.896
BOTH FOR PREDICTING EVERYDAY
EVENTS LIKE TEMPERATURE AND
00:22:44.981 --> 00:22:47.899
PRECIPITATION AND FOR MAJOR
CATASTROPHIC EVENTS LIKE
00:22:47.984 --> 00:22:50.618
HURRICANES WHERE THE ACCURACY
OF HURRICANE PREDICTIONS HAS
00:22:50.704 --> 00:22:55.523
INCREASED BY ABOUT 300%
OVER THE PAST 25 YEARS.
00:22:55.625 --> 00:22:57.909
AND THAT SAVES A LOT OF LIVES.
00:23:03.667 --> 00:23:05.216
MARY BETH GERHARDT: THE
DIFFICULTY IS DEFINITELY
00:23:05.335 --> 00:23:06.835
FINDING THE TRUTH.
00:23:06.920 --> 00:23:08.753
IT'S NOT ABOUT MAKING A
FORECAST A LOT OF THE TIMES.
00:23:08.872 --> 00:23:11.639
A LOT OF TIMES, IT'S FINDING
OUT HOW CONFIDENT YOU ARE IN
00:23:11.725 --> 00:23:14.375
THAT FORECAST AND
DETERMINING WHAT YOU KNOW AND
00:23:14.477 --> 00:23:15.727
WHAT YOU DON'T KNOW.
00:23:15.846 --> 00:23:17.545
BUT THANKFULLY, THE MODELS
HAVE GOTTEN REALLY GOOD.
00:23:17.647 --> 00:23:21.399
SO WE'VE GOTTEN BETTER
AT PREDICTING EVENTS.
00:23:24.237 --> 00:23:25.737
JOE SIENKIEWICZ: SO I
STARTED OUT 28 YEARS AGO.
00:23:25.856 --> 00:23:28.723
JUST IMAGINE
FORECAST INFORMATION CAME
00:23:28.825 --> 00:23:30.325
IN THE FORM OF PAPER.
00:23:30.427 --> 00:23:31.860
PILES OF PAPER.
00:23:31.962 --> 00:23:33.695
IT LIMITED THE
AMOUNT OF INFORMATION
00:23:33.780 --> 00:23:35.797
THAT WE COULD LOOK AT.
00:23:35.899 --> 00:23:38.933
WE SEE THINGS NOW IN THE
MODELS THAT WE'RE ACTUALLY
00:23:39.035 --> 00:23:41.920
IN SOME WAYS LEARNING AND
CONFIRMING USING OTHER
00:23:42.038 --> 00:23:44.572
INFORMATION, OBSERVATIONS,
SATELLITE DATA.
00:23:44.674 --> 00:23:47.375
SO THINGS HAVE CHANGED
ENORMOUSLY IN MY CAREER.
00:23:47.460 --> 00:23:50.011
NARRATOR: TODAY, FORECASTERS
CAN RELY ON A VAST ARRAY OF
00:23:50.096 --> 00:23:54.632
WEATHER SENSORS ON THE GROUND,
AT SEA, IN THE AIR,
00:23:54.718 --> 00:23:57.084
AND EVEN SPACE.
00:23:57.187 --> 00:23:59.921
THIS INFORMATION IS FED INTO
COMPUTER MODELS THAT BUILD
00:24:00.023 --> 00:24:03.725
ON OUR DEEP KNOWLEDGE OF
ENVIRONMENTAL PHYSICS.
00:24:04.861 --> 00:24:05.643
DR WILLIAM LAPENTA: THE
ATMOSPHERE IS BASICALLY
00:24:05.729 --> 00:24:08.396
A FLUID, JUST LIKE
WATER, FISH TANK.
00:24:08.481 --> 00:24:11.933
AND FLUIDS ARE DEFINED BY
MATHEMATICAL EQUATIONS IN
00:24:12.035 --> 00:24:14.101
TERMS OF THEIR STRUCTURE
AND HOW THEY WOULD
00:24:14.204 --> 00:24:16.955
EVOLVE WITH TIME.
00:24:20.126 --> 00:24:23.044
SO WE GO FOR THE MAJOR TERMS
IN AN EQUATION AND THEN WE PUT
00:24:23.129 --> 00:24:25.797
THEM INTO A COMPUTER MODEL AND
WE TRY TO GET A SOLUTION OF
00:24:25.916 --> 00:24:29.250
HOW THE ATMOSPHERE WILL EVOLVE.
00:24:31.421 --> 00:24:33.922
NARRATOR: DR WILLIAM LAPENTA
IS THE DIRECTOR OF NOAA'S NINE
00:24:34.007 --> 00:24:37.225
WEATHER PREDICTION CENTERS.
00:24:37.310 --> 00:24:40.511
WHEN A TORNADO IS FORMING OR
A HURRICANE IS BREWING,
00:24:40.597 --> 00:24:45.433
THE ACCURACY OF HIS MODELS CAN
BE A MATTER OF LIFE AND DEATH.
00:24:45.518 --> 00:24:48.820
BUT BECAUSE THERE ARE SO MANY
VARIABLES IN THE ATMOSPHERE,
00:24:48.939 --> 00:24:52.523
NO MODEL SPITS OUT THE
RIGHT PREDICTION EVERY TIME.
00:24:52.609 --> 00:24:56.644
NOAA COMPENSATES BY RUNNING
THEIR MODELS DOZENS OF TIMES,
00:24:56.746 --> 00:25:00.865
INTRODUCING RANDOM
VARIATIONS IN THE DATA.
00:25:01.918 --> 00:25:04.535
THE RESULT IS A CLUSTER
OF POSSIBLE FUTURES
00:25:04.621 --> 00:25:07.038
CALLED AN ENSEMBLE.
00:25:07.123 --> 00:25:08.957
DR WILLIAM LAPENTA: SO LET'S
JUST SAY THAT MY PUTT
00:25:09.042 --> 00:25:11.542
WAS A HURRICANE TRACK, OKAY?
00:25:11.628 --> 00:25:13.995
SO I PUTT ONCE,
YOU SEE THE BALL ROLL,
00:25:14.097 --> 00:25:15.129
YOU SEE IT ROLL A CERTAIN WAY.
00:25:15.215 --> 00:25:19.050
OKAY, THAT'S ONE
PIECE OF INFORMATION.
00:25:20.553 --> 00:25:23.721
SO AN ENSEMBLE MEANS YOU
WOULD DO THAT MANY TIMES OVER.
00:25:23.807 --> 00:25:29.694
MAYBE 20 TIMES,
MAYBE 40 TIMES.
00:25:30.697 --> 00:25:32.680
AND THE THING IS EVERY TIME
YOU PUT IT DOWN AND STROKE IT,
00:25:32.782 --> 00:25:34.983
SOMETHING CHANGES.
00:25:35.068 --> 00:25:37.452
SO THEN IF THE TRAJECTORIES
OF THE BALLS ARE VERY CLOSE TO
00:25:37.554 --> 00:25:41.322
EACH OTHER, THAT MEANS YOU
EITHER HAVE A VERY GOOD MODEL,
00:25:41.408 --> 00:25:43.691
OR THERE'S A HIGH LEVEL
OF PREDICTABILITY IN THAT
00:25:43.793 --> 00:25:46.995
HURRICANE TRACK, OR BOTH.
00:25:48.331 --> 00:25:50.031
MARY BETH GERHARDT: THESE ARE
CALLED SPAGHETTI PLOTS,
00:25:50.133 --> 00:25:52.400
AND IT'S ESSENTIALLY JUST
DETERMINISTIC GUIDANCE
00:25:52.502 --> 00:25:54.886
OVERLAID WITH A BUNCH OF
ENSEMBLE GUIDANCE, SO YOU CAN
00:25:55.005 --> 00:25:58.205
REALLY SEE A LOT OF
INFORMATION ON ONE SCREEN.
00:25:58.308 --> 00:26:00.375
WHEN THEY'RE CLOSER TOGETHER,
YOU KNOW, YOU HAVE HIGHER
00:26:00.477 --> 00:26:01.943
CONFIDENCE IN YOUR FORECAST.
00:26:02.045 --> 00:26:03.978
YOU KNOW THERE'S MORE
CERTAINTY IN YOUR FORECAST.
00:26:04.080 --> 00:26:07.382
AND THEN AS YOU GO OUT IN TIME,
SOLUTIONS BEGIN TO DIVERGE.
00:26:07.484 --> 00:26:09.851
WE BECOME MORE UNCERTAIN
ABOUT THE FORECAST,
00:26:09.953 --> 00:26:12.487
AND REALLY JUST THE
SLIGHTEST DIFFERENCE CAN MAKE A
00:26:12.572 --> 00:26:15.189
HUGE CHANGE IN THE FORECAST.
00:26:16.860 --> 00:26:17.859
DR WILLIAM LAPENTA:
WE'RE NEVER GONNA GET
00:26:17.944 --> 00:26:19.660
A PERFECT NUMERICAL FORECAST.
00:26:19.746 --> 00:26:21.229
YOU CAN'T HAVE AN
OBSERVATION EVERYWHERE.
00:26:21.331 --> 00:26:23.698
YOU CANNOT KNOW WHERE THE
TEMPERATURE IS AT EVERY SINGLE
00:26:23.783 --> 00:26:26.834
POINT IN THE ATMOSPHERE.
00:26:27.537 --> 00:26:29.370
NARRATOR: WEATHER PREDICTION
MAY NEVER BE PERFECT,
00:26:29.456 --> 00:26:32.623
BUT THE STEADY GAINS WE MAKE IN
ACCURACY HAVE SAVED THOUSANDS,
00:26:32.709 --> 00:26:36.677
PERHAPS MILLIONS OF LIVES.
00:26:36.763 --> 00:26:43.084
NOW, SCIENTISTS ARE TACKLING
AN EVEN BIGGER PROBLEM.
00:26:43.186 --> 00:26:46.304
CAN PREDICTIVE TECHNOLOGY
PREVENT THE APOCALYPSE?
00:26:52.804 --> 00:26:55.621
NARRATOR: TODAY, PREDICTIVE
ANALYTICS IS BEING USED TO
00:26:55.723 --> 00:26:58.957
TACKLE QUESTIONS THAT ONCE,
ONLY MYSTICS AND MADMEN WOULD
00:26:59.060 --> 00:27:02.895
HAVE DARED TO ANSWER.
00:27:02.980 --> 00:27:06.165
MADISON: SO, HOW
DOES THE WORLD END?
00:27:06.267 --> 00:27:07.800
NOSTRADAMUS: A GREAT FIRE
WILL COME FROM THE SKY AND
00:27:07.902 --> 00:27:09.985
LAY WASTE TO THE WORLD.
00:27:10.104 --> 00:27:13.989
MADISON: WHEN?
00:27:15.243 --> 00:27:16.692
NARRATOR: SIXTY-FIVE
MILLION YEARS AGO,
00:27:16.778 --> 00:27:21.831
AN ASTEROID WIPED OUT
THE EARTH'S DOMINANT SPECIES.
00:27:21.949 --> 00:27:24.083
IT COULD HAPPEN AGAIN.
00:27:24.168 --> 00:27:26.702
BUT IF WE COULD SEE THE NEXT
KILLER ASTEROID COMING AND
00:27:26.788 --> 00:27:30.623
PREDICT WHEN IT WILL STRIKE,
HUMANITY MIGHT ESCAPE
00:27:30.708 --> 00:27:33.542
THE FATE OF THE DINOSAURS.
00:27:35.513 --> 00:27:40.933
PAVLOS PROTOPAPAS AND HIS TEAM
AT HARVARD ARE WORKING ON IT.
00:27:41.018 --> 00:27:43.519
PAVLOS STARTED OUT AS A
THEORETICAL PHYSICIST.
00:27:43.638 --> 00:27:46.605
THEN HE BEGAN WORKING IN BIG
DATA AND IMMEDIATELY SAW ITS
00:27:46.691 --> 00:27:49.475
POTENTIAL FOR ASTROPHYSICS.
00:27:50.645 --> 00:27:52.077
PAVLOS PROTOPAPAS: MY MENTOR
USED TO CALL ME A PARASITE
00:27:52.180 --> 00:27:54.613
BECAUSE I USED TO GO AND TAKE
DATA FROM OTHER PEOPLE AND
00:27:54.699 --> 00:27:57.316
DO ANALYSIS BECAUSE THEY DIDN'T,
THEY WOULD JUST USE THE DATA
00:27:57.401 --> 00:28:00.035
FOR THEIR SPECIFIC
PROBLEM, BUT I SAID,
00:28:00.154 --> 00:28:01.687
"LOOK, THE DATA THERE,
THERE ARE STUFF THERE,
00:28:01.789 --> 00:28:03.038
WE CAN FIND THINGS."
00:28:03.157 --> 00:28:04.957
SO I STARTED GETTING
THE DATA FROM EVERYBODY AND
00:28:05.042 --> 00:28:07.526
STARTED DOING ANALYSIS.
00:28:07.628 --> 00:28:10.129
NARRATOR: TO PAVLOS, THE
ENORMOUS POOL OF DATA
00:28:10.214 --> 00:28:13.499
COLLECTED BY SATELLITES AND
TELESCOPES IS LIKE TREASURE
00:28:13.584 --> 00:28:15.501
BURIED IN THE OCEAN.
00:28:15.586 --> 00:28:16.585
PAVLOS PROTOPAPAS: WELL FIRST
OF ALL, ASTRONOMY HAS BEEN
00:28:16.671 --> 00:28:20.306
COLLECTING DATA SINCE
THE BEGINNING OF TIME.
00:28:20.391 --> 00:28:24.844
PEOPLE LOOK AT THE SKY, THEY
START WRITING DOWN THINGS,
00:28:24.946 --> 00:28:26.729
THEY COLLECT DATA
ABOUT WHERE THE STARS ARE,
00:28:26.848 --> 00:28:29.849
WHERE THE PLANETS ARE, WHEN THE
SUN RISES, WHEN THE SUN SETS,
00:28:29.934 --> 00:28:32.318
AND ALL THIS THING.
00:28:33.437 --> 00:28:35.404
NARRATOR: AS THE GURU OF
"EXTREME COMPUTING,"
00:28:35.523 --> 00:28:37.773
PAVLOS RUNS PREDICTIVE
SIMULATIONS ON ALMOST
00:28:37.859 --> 00:28:41.393
UNIMAGINABLY LARGE SCALES
LOOKING FOR SOLUTIONS
00:28:41.495 --> 00:28:45.114
TO THE WORLD'S
MOST COMPLEX PROBLEMS.
00:28:46.417 --> 00:28:49.084
AND FEW CHALLENGES ARE GREATER
THAN HIS CURRENT MISSION,
00:28:49.203 --> 00:28:51.604
TO FIND AND PREDICT
THE PATHS OF EVERY POTENTIAL
00:28:51.706 --> 00:28:54.473
PLANET-KILLER IN SPACE.
00:28:54.575 --> 00:28:58.377
PAVLOS PROTOPAPAS: NASA HAD A
MANDATE TO FIND ALL ASTEROIDS
00:28:58.462 --> 00:29:03.048
THAT ARE ABOUT 1093 YARDS
BIG, WHICH HAS BEEN MET NOW.
00:29:04.685 --> 00:29:09.438
NOW THEY HAVE A NEW MANDATE TO
FIND ALL OF THEM AT 153 YARDS.
00:29:09.557 --> 00:29:12.892
AND 153 YARDS IS, IF IT HITS
EARTH, IT'S BIG ENOUGH TO
00:29:12.994 --> 00:29:16.111
DESTROY A CITY LIKE BOSTON.
00:29:19.817 --> 00:29:25.404
SO RIGHT NOW, THE EFFORT IS
TO FIND ALL 153 YARD ASTEROIDS
00:29:25.489 --> 00:29:28.707
OUT THERE THAT
POSSIBLY CAN HIT EARTH.
00:29:30.211 --> 00:29:31.660
NARRATOR: THE MILKY
WAY ALONE CONTAINS AT LEAST
00:29:31.746 --> 00:29:34.580
A HUNDRED BILLION STARS.
00:29:34.665 --> 00:29:37.666
WE HAVE ONLY RECENTLY BEEN
ABLE TO SPOT PLANETS AROUND
00:29:37.752 --> 00:29:40.052
THE NEAREST STARS.
00:29:40.137 --> 00:29:42.838
SO, SPOTTING A
460-FOOT-LONG CHUNK OF ROCK
00:29:42.924 --> 00:29:44.957
WOULD SEEM NEARLY IMPOSSIBLE.
00:29:45.059 --> 00:29:47.092
BUT PHYSICS GIVES US A WAY.
00:29:47.178 --> 00:29:49.228
PAVLOS PROTOPAPAS:
STARS DON'T MOVE.
00:29:49.313 --> 00:29:51.430
I MEAN, THEY MOVE IN THE SKY
BECAUSE WE MOVE AS EARTH,
00:29:51.515 --> 00:29:55.768
BUT THE RELATIVE
POSITIONS DON'T MOVE.
00:30:00.157 --> 00:30:04.143
PLANETS, ASTEROIDS,
AND EVERYTHING, MOVE.
00:30:05.746 --> 00:30:09.949
SO IF YOU LOOK AT THE PLANET
AT NIGHT, AT 11:00 PM.
00:30:10.034 --> 00:30:13.652
AT 12:00, YOU SEE IT HAS MOVED
RELATIVE TO THE STARS, RIGHT?
00:30:15.790 --> 00:30:17.873
NARRATOR: TO FIND MOVING
OBJECTS IN SPACE, YOU COMPARE
00:30:17.959 --> 00:30:21.860
A SERIES OF PICTURES TAKEN
OVER TIME, SAY, TWO IMAGES
00:30:21.963 --> 00:30:24.897
A NIGHT OVER MONTHS OR YEARS.
00:30:25.800 --> 00:30:28.500
THEN YOU REMOVE ALL THE
SOURCES THAT DON'T MOVE.
00:30:28.602 --> 00:30:30.970
YOU'RE STILL LEFT WITH
A MOSAIC OF LIGHT, BUT
00:30:31.055 --> 00:30:34.707
MACHINE-LEARNING SOFTWARE CAN
DETECT THE TINY CHANGES IN
00:30:34.809 --> 00:30:37.109
THESE MILLIONS OF FLARES.
00:30:38.145 --> 00:30:39.278
PAVLOS PROTOPAPAS: SO
YOU HAVE TO LINK THEM AND
00:30:39.363 --> 00:30:40.362
YOU HAVE TO CONNECT THEM.
00:30:40.481 --> 00:30:41.680
IT'S LIKE CONNECTING DOTS.
00:30:41.782 --> 00:30:46.568
AND YOU CONNECT DOTS SO THEY
GO ACCORDING TO AN ORBIT
00:30:46.654 --> 00:30:49.021
IN THE SOLAR SYSTEM.
00:30:49.123 --> 00:30:50.489
AND THIS IS
A CLASSIC PROBLEM OF
00:30:50.574 --> 00:30:52.708
SEARCHING
THREE-DIMENSIONAL SPACE.
00:30:52.827 --> 00:30:56.161
THESE ASTEROIDS MOVE BASED ON
HOW THEY FEEL GRAVITY FROM THE
00:30:56.247 --> 00:30:58.414
SUN AND THE OTHER PLANETS.
00:30:58.499 --> 00:31:01.383
SO, THE MOMENT THAT YOU
KNOW WHERE THEY ARE, YOU CAN
00:31:01.502 --> 00:31:05.004
ACTUALLY MAKE THE ORBIT
VERY, VERY WELL, RIGHT?
00:31:05.089 --> 00:31:08.724
BUT MORE OBSERVATIONS WE HAVE,
MORE TIMES WE SEE THIS OBJECT
00:31:08.843 --> 00:31:12.377
AND FURTHER AWAY IN TIME WE
HAVE THEM, MORE CONSTRAINED WE
00:31:12.480 --> 00:31:15.514
CAN MAKE THE ORBIT TO THE
POINT EVENTUALLY YOU CAN
00:31:15.599 --> 00:31:18.901
CONSTRAIN IT ENOUGH TO KNOW
IT'S GONNA COME CLOSE OR
00:31:19.020 --> 00:31:21.320
IT'S GONNA HIT EARTH.
00:31:22.606 --> 00:31:24.556
NARRATOR: IF WE CAN PREDICT AN
ASTEROID STRIKE IS COMING,
00:31:24.658 --> 00:31:27.326
WE MAY HAVE A FIGHTING
CHANCE TO STOP IT.
00:31:27.411 --> 00:31:30.362
BECAUSE IT'S NOT
ENOUGH TO SEE THE FUTURE,
00:31:30.448 --> 00:31:33.332
WE WANT TO CHANGE THE FUTURE.
00:31:42.710 --> 00:31:44.493
BRENNA BERMAN: THE GOAL OF THE
PREDICTIVE ANALYTICS PROGRAM
00:31:44.578 --> 00:31:48.413
IN CHICAGO IS TO FIND PATTERNS
IN DATA AND TO USE THOSE
00:31:48.516 --> 00:31:52.451
PATTERNS TO MAKE OUR CITY
SERVICES MORE EFFECTIVE.
00:31:53.821 --> 00:31:56.288
TOM SCHENK: SO IT WAS A
PROJECT WHERE WE WERE LOOKING
00:31:56.390 --> 00:31:58.190
AT ALL THESE DIFFERENT
SORT OF REQUESTS
00:31:58.275 --> 00:31:59.391
YOU CAN MAKE OF THE CITY.
00:31:59.493 --> 00:32:00.809
WE CALL THEM SERVICE REQUESTS.
00:32:00.895 --> 00:32:03.595
WHEN YOU CALL 311 ON THE
PHONE, YOU MAKE A REQUEST.
00:32:03.697 --> 00:32:05.447
AND WE'RE LOOKING FOR
CORRELATIONS AS WE'RE MINING
00:32:05.566 --> 00:32:07.933
THROUGH A BUNCH OF DATA
00:32:12.139 --> 00:32:14.456
TO TRY AND FIGURE OUT, YOU KNOW,
IS THERE A RELATIONSHIP HERE?
00:32:14.575 --> 00:32:16.742
CAN WE PREDICT
SOMETHING FROM IT?
00:32:17.845 --> 00:32:20.379
AND SOMETHING THAT KEPT COMING
UP OVER AND OVER WAS BEING
00:32:20.464 --> 00:32:22.965
ABLE TO PREDICT WHERE
RATS ARE IN THE CITY.
00:32:24.585 --> 00:32:26.051
JOSIE CRUZ: RODENT
CONTROL, CAN I HELP YOU?
00:32:26.137 --> 00:32:28.087
WE ARE THE BUREAU
OF RODENT CONTROL.
00:32:28.172 --> 00:32:30.455
WE WORK UNDER
STREETS & SANITATION.
00:32:30.558 --> 00:32:32.307
AND WHAT WE'RE DOING
TODAY IS WE'RE WORKING
00:32:32.426 --> 00:32:34.393
THE PREVENTIVE BAITING PROJECT.
00:32:34.462 --> 00:32:35.344
TOM SCHENK:
WE CALL IT RAT PATROL,
00:32:35.429 --> 00:32:36.979
IS THEIR NICKNAME FOR IT.
00:32:37.264 --> 00:32:38.063
JOSIE CRUZ: 4900 TO MAIN.
00:32:38.149 --> 00:32:40.482
FROM 4991, MR COGLAND COME UP.
00:32:40.601 --> 00:32:43.735
I'M TELLING YOU I WAS REALLY,
REALLY SHOCKED WHEN
00:32:43.821 --> 00:32:45.404
THEY CAME TO ME AND SAID,
00:32:45.489 --> 00:32:48.157
"WE'RE PREDICTING WHERE
THERE'S GOING TO BE RATS."
00:32:49.276 --> 00:32:51.026
TOM SCHENK: WE STARTED OFF WITH
ABOUT 350 DIFFERENT VARIABLES,
00:32:51.112 --> 00:32:53.779
350 DIFFERENT PIECES OF DATA TO
TRY TO UNDERSTAND AND
00:32:53.864 --> 00:32:56.832
PREDICT WHERE THESE RODENT
COMPLAINTS WERE GONNA HAPPEN
00:32:56.951 --> 00:32:58.584
ACROSS THE CITY OF CHICAGO.
00:32:58.669 --> 00:32:59.868
AND AFTER WE
DID SOME STATISTICS,
00:32:59.954 --> 00:33:02.004
WE KIND OF SETTLED ON
ABOUT 31 DIFFERENT FACTORS
00:33:02.123 --> 00:33:03.755
THAT SEEM TO
PREDICT REALLY WELL.
00:33:03.841 --> 00:33:06.508
FOR INSTANCE, THINGS LIKE
SANITATION CODE COMPLAINTS.
00:33:06.627 --> 00:33:09.044
GARBAGE IN THE ALLEYWAY
OR BROKEN TRASH BINS.
00:33:09.130 --> 00:33:11.130
AND SO WE CAN TAKE A LOOK AT
ALL THAT DATA AND THEN WE CAN
00:33:11.215 --> 00:33:15.000
ACTUALLY PREDICT WHERE THESE
BLUE DOTS ARE GOING TO HAPPEN.
00:33:15.970 --> 00:33:17.502
JOSIE CRUZ: THE ADDRESSES
THAT WE GET ARE LOCATIONS THAT
00:33:17.605 --> 00:33:20.973
THEY'VE PREDICTED THAT WE
HAVE NOT GOTTEN ANY 311 CALLS.
00:33:21.058 --> 00:33:24.643
THEY ARE PREDICTING TO ME HERE
THERE'S GONNA BE RODENT ISSUES.
00:33:24.745 --> 00:33:26.395
SO, THAT'S WHY
IT'S PREVENTIVE.
00:33:26.480 --> 00:33:29.231
WE'RE OUT THERE AHEAD OF IT TO
SEE IF IT'S TRUE, AND GET TO
00:33:29.316 --> 00:33:31.783
IT BEFORE IT GETS
TO BECOME A PROBLEM.
00:33:45.916 --> 00:33:48.800
WE HAD NOT HAD A COMPLAINT AT
ALL HERE, SO BY THEM GIVING US
00:33:48.886 --> 00:33:50.969
THESE ADDRESSES, IT'S
BASICALLY GETTING US A STEP
00:33:51.055 --> 00:33:53.005
AHEAD OF WHAT'S GOING ON HERE.
00:33:53.090 --> 00:33:55.390
AND SURE ENOUGH,
THEY WERE RIGHT.
00:33:56.727 --> 00:33:58.710
JUAN GONZALEZ: WE'VE SEEN
VARIOUS RAT HOLES THROUGHOUT
00:33:58.812 --> 00:34:02.514
THE ALLEY, ALONG WITH RAT
DROPPINGS EVERYWHERE, SO
00:34:02.600 --> 00:34:05.601
THERE'S A VARIETY OF DIFFERENT
THINGS THAT WILL TELL YOU THAT
00:34:05.686 --> 00:34:07.853
THERE'S RAT ACTIVITY GOING ON.
00:34:11.942 --> 00:34:14.226
JOSIE CRUZ: WITH OUR
RODENTICIDE, THE RAT WILL DIE
00:34:14.328 --> 00:34:15.894
WITHIN THREE TO FOUR DAYS.
00:34:15.996 --> 00:34:19.665
AND THEY WILL DIE WHEREVER
IT HITS THEM BASICALLY.
00:34:37.134 --> 00:34:38.684
I KIND OF THOUGHT,
00:34:38.769 --> 00:34:41.019
"REALLY? OFF A COMPUTER, YOU'RE
GONNA TELL ME, YOU KNOW, THAT
00:34:41.105 --> 00:34:44.890
THERE'S A POSSIBILITY THAT
THERE COULD BE RATS THERE?"
00:34:44.975 --> 00:34:47.392
BUT, YOU KNOW, I WAS
VERY, VERY SURPRISED,
00:34:47.478 --> 00:34:48.560
AND VERY PLEASED.
00:34:48.646 --> 00:34:49.861
I'M VERY HAPPY THAT
WE HAVE THIS GOING.
00:34:49.947 --> 00:34:50.896
I REALLY DO.
00:34:50.981 --> 00:34:52.648
IT HELPS ME TREMENDOUSLY.
00:35:02.076 --> 00:35:03.375
TOM SCHENK: IF WE CAN PREDICT
EVERYTHING PERFECTLY,
00:35:03.460 --> 00:35:06.161
THAT'S JUST LUCK.
00:35:06.997 --> 00:35:08.780
IT'S NOT ABOUT BEING
RIGHT ALL THE TIME.
00:35:08.882 --> 00:35:11.833
IT'S ABOUT BEING RIGHT
MOST OF THE TIME.
00:35:14.922 --> 00:35:16.638
NARRATOR: MOST PEOPLE DON'T
HAVE A PROBLEM WITH STOPPING
00:35:16.757 --> 00:35:20.976
RAT INFESTATIONS OR
ASTEROID STRIKES.
00:35:21.679 --> 00:35:23.262
BUT WHAT HAPPENS WHEN THIS
TECHNOLOGY DIGS INTO EVERY
00:35:23.364 --> 00:35:24.930
PART OF OUR LIVES?
00:35:26.066 --> 00:35:27.182
ERIC SIEGEL: SO WE'RE AT A
REALLY INTERESTING TIME IN
00:35:27.268 --> 00:35:30.902
HISTORY NOW WHERE IT'S ONLY
SLOWLY DAWNING ON MOST OF US
00:35:30.988 --> 00:35:33.605
AS CITIZENS JUST HOW MUCH
INFORMATION ORGANIZATIONS,
00:35:33.691 --> 00:35:35.440
BUSINESS AND GOVERNMENT
ORGANIZATIONS,
00:35:35.542 --> 00:35:38.076
HAVE ABOUT US AS INDIVIDUALS.
00:35:38.162 --> 00:35:41.496
BECAUSE FOR THE MOST PART,
THEY HAVE THE INCENTIVE TO
00:35:41.615 --> 00:35:44.166
NOT REALLY DIVULGE EXACTLY
HOW MUCH THEY HAVE.
00:35:44.285 --> 00:35:46.084
IT'S SO VALUED TO THEM.
00:35:46.170 --> 00:35:48.670
IT'S SO VALUABLE TO THEM.
00:35:49.657 --> 00:35:50.822
NARRATOR: RIGHT NOW,
CORPORATIONS CAN TRACK YOUR
00:35:50.924 --> 00:35:54.126
MOVEMENTS WITH CELLPHONES
AND SURVEILLANCE CAMERAS,
00:35:54.228 --> 00:35:57.346
MONITOR YOUR SPENDING, AND
RECORD EVERY KEYSTROKE
00:35:57.464 --> 00:35:59.631
YOU MAKE ON A COMPUTER.
00:35:59.717 --> 00:36:02.217
USUALLY, THEY USE THIS DATA
TO TRY TO SELL YOU SOMETHING.
00:36:02.303 --> 00:36:05.554
BUT THE POWER OF PREDICTION
GOES FAR BEYOND THAT.
00:36:05.639 --> 00:36:08.223
ERIC SIEGEL: THE THINGS THAT
MATTER MOST, QUITTING A JOB,
00:36:08.309 --> 00:36:10.675
HOW YOU'LL PERFORM IF YOU'RE
HIRED, WHETHER YOU'RE GONNA
00:36:10.778 --> 00:36:13.278
COMMIT A CRIME, WHETHER YOU'RE
AT RISK OF COMMITTING
00:36:13.364 --> 00:36:15.314
AN ACT OF TERROR,
WHETHER YOU'RE PREGNANT,
00:36:15.399 --> 00:36:16.982
THESE ARE ALL
VERY POTENT THINGS.
00:36:17.067 --> 00:36:20.736
THEREFORE, THERE'S A LOT OF
VALUE IN PREDICTING THEM,
00:36:20.821 --> 00:36:25.490
BUT THERE'S ALSO A REALLY
HIGH RISK OF MISUSE.
00:36:26.710 --> 00:36:27.859
NARRATOR: THE FEAR OF USING
COMPUTERS TO PREDICT THE
00:36:27.961 --> 00:36:32.914
FUTURE IS GREATEST WHEN IT
COMES TO LAW ENFORCEMENT.
00:36:33.000 --> 00:36:36.201
IS IT POSSIBLE TO KNOW WHERE
CRIMES WILL HAPPEN AND
00:36:36.303 --> 00:36:38.036
WHO WILL COMMIT THEM?
00:36:43.352 --> 00:36:45.820
NARRATOR: WHEN PHILIP K DICK
PUBLISHED "MINORITY REPORT" IN
00:36:45.905 --> 00:36:49.857
1956, IT SEEMED OUTLANDISH,
THAT IN THE FUTURE,
00:36:49.959 --> 00:36:53.527
WE WILL BE ABLE TO PREDICT
CRIMES BEFORE THEY HAPPEN.
00:36:53.629 --> 00:36:58.666
BUT THAT IS EXACTLY WHAT
WE ARE TRYING TO DO TODAY.
00:36:59.752 --> 00:37:01.869
COMPUTER PROGRAMS SEARCH
SOCIAL MEDIA POSTS,
00:37:01.971 --> 00:37:05.639
CENSUS DATA, POLICE REPORTS,
AND A VAST POOL OF PUBLIC AND
00:37:05.725 --> 00:37:08.375
PRIVATE RECORDS, LOOKING FOR
PATTERNS THAT COULD ADD UP
00:37:08.478 --> 00:37:11.846
TO CRIMES ABOUT TO HAPPEN.
00:37:11.931 --> 00:37:14.231
IT'S CALLED
"PREDICTIVE POLICING."
00:37:17.687 --> 00:37:20.688
CRITICS SAY PREDICTIVE POLICING
IS INHERENTLY FLAWED,
00:37:20.773 --> 00:37:23.991
BECAUSE IT RISKS LABELING
INNOCENT PEOPLE AS CRIMINALS .
00:37:24.076 --> 00:37:26.827
IT INADVERTENTLY
TARGETS MINORITIES AND
00:37:26.913 --> 00:37:30.865
IT JUDGES YOU ON YOUR
PAST, NOT WHO YOU ARE TODAY.
00:37:34.370 --> 00:37:36.837
BUT PROFESSOR JOEL CAPLAN
BELIEVES WE CAN GET THE
00:37:36.923 --> 00:37:40.958
BENEFITS OF PREDICTIVE
POLICING WITHOUT THE BIAS.
00:37:41.043 --> 00:37:42.343
PROF. JOEL CAPLAN: I THINK
I'VE ALWAYS BEEN FASCINATED
00:37:42.428 --> 00:37:44.762
BY SPACES, AND
THE CONCEPT OF SPACE,
00:37:44.881 --> 00:37:47.681
WHETHER IT'S BLACK HOLES
OR, YOU KNOW, THE SPACE
00:37:47.767 --> 00:37:49.683
OF A PERSON'S LOCAL ENVIRONMENT.
00:37:49.769 --> 00:37:52.186
AND I'VE ALSO BEEN INTERESTED
IN PUBLIC SAFETY, AND LAW
00:37:52.271 --> 00:37:55.222
ENFORCEMENT, AND POLICING, AND
THE FACT THAT PUBLIC SAFETY
00:37:55.308 --> 00:37:58.425
AFFECTS AND INFLUENCES OUR
LIVES ON A REGULAR BASIS,
00:37:58.528 --> 00:38:00.194
WHETHER WE KNOW IT OR NOT.
00:38:07.403 --> 00:38:09.537
CHIEF HENRY M WHITE JR.:
WELL, I CAN TELL YOU WHAT'S
00:38:09.622 --> 00:38:10.654
UNIQUE ABOUT ATLANTIC CITY.
00:38:10.740 --> 00:38:12.122
IT'S UNLIKE ANY OTHER CITY.
00:38:12.241 --> 00:38:15.659
THIS IS A TOWN THAT'S CONSTANTLY
CHANGING ON A DAILY BASIS.
00:38:18.948 --> 00:38:22.666
WE ONLY HAVE APPROXIMATELY
40,000 YEAR-ROUND RESIDENTS.
00:38:22.752 --> 00:38:26.787
BUT WE HAVE APPROXIMATELY 24
MILLION VISITORS EVERY YEAR.
00:38:26.889 --> 00:38:30.123
AND EACH VISITOR BRINGS
THEIR OWN SET OF ISSUES.
00:38:30.226 --> 00:38:33.594
ROBBERIES, SHOOTINGS,
AGGRESSIVE BEGGING, DRINKING
00:38:33.679 --> 00:38:36.981
IN PUBLIC, SLEEPING IN PUBLIC.
00:38:45.942 --> 00:38:48.025
SO EACH AND EVERY DAY,
ATLANTIC CITY IS CONSTANTLY
00:38:48.110 --> 00:38:51.912
EVOLVING WITH A TOTALLY
DIFFERENT POPULATION SEGMENT.
00:38:55.334 --> 00:38:57.701
WE'RE JUST NOW GETTING
INVOLVED WITH A SORT OF
00:38:57.787 --> 00:38:59.753
FORM OF PREDICTIVE
ANALYTICS TO PREVENT CRIME
00:38:59.839 --> 00:39:02.373
HERE IN ATLANTIC CITY.
00:39:02.458 --> 00:39:05.226
AND THEN THE PROGRAM IS
CALLED RISK TERRAIN MODELING.
00:39:10.466 --> 00:39:13.767
PROF. JOEL CAPLAN: RISK
TERRAIN MODELING WAS DEVELOPED
00:39:13.853 --> 00:39:16.937
BY DR. LES KENNEDY AND I AT
RUTGERS UNIVERSITY, AND
00:39:17.023 --> 00:39:20.307
IT'S A METHOD OF SPATIAL RISK
ANALYSIS THAT DIAGNOSES
00:39:20.393 --> 00:39:24.845
FEATURES OF THE LANDSCAPE AND
HOW THEY INTERACT AND OVERLAP
00:39:24.947 --> 00:39:27.197
TO CREATE UNIQUE BEHAVIOR
SETTINGS FOR CRIME.
00:39:29.201 --> 00:39:30.851
CAPT. JAMES SARKOS: SO, WHAT WE
LIKE A LOT ABOUT RISK TERRAIN
00:39:30.953 --> 00:39:32.352
MODELING IS THAT IT
FOCUSES ON PLACES AND
00:39:32.455 --> 00:39:33.654
NOT NECESSARILY PEOPLE.
00:39:33.739 --> 00:39:35.656
THE BEST WAY TO JUDGE, IF
WE'RE DOING A GOOD JOB AS
00:39:35.741 --> 00:39:38.158
POLICE OFFICERS, IS NOT BY
HOW MANY ARRESTS WE MAKE.
00:39:38.244 --> 00:39:40.160
ALTHOUGH, A LOT OF TIMES, IT
SEEMS LIKE PEOPLE DO JUDGE US
00:39:40.246 --> 00:39:42.663
BY THAT, BUT WE WANNA BE
JUDGED BY PREVENTING CRIME.
00:39:42.748 --> 00:39:45.082
THAT'S OUR ULTIMATE GOAL,
IS THE PREVENTION OF CRIME.
00:39:45.167 --> 00:39:46.467
IN THE END, WE DON'T
WANNA MAKE AN ARREST IF
00:39:46.552 --> 00:39:47.868
WE DON'T HAVE TO.
00:39:47.970 --> 00:39:49.970
WE'D RATHER THE CRIME NEVER
OCCURRED TO BEGIN WITH.
00:39:50.673 --> 00:39:51.722
PROF. JOEL CAPLAN:
CRIME ISN'T RANDOM.
00:39:51.841 --> 00:39:55.342
AND CRIME OCCURS VERY
FREQUENTLY AT CERTAIN PLACES.
00:39:55.428 --> 00:39:58.729
AND IF WE DON'T FOCUS ON THE
PLACES, WE'RE NOT GOING
00:39:58.848 --> 00:40:02.316
TO CHANGE THE ATTRACTIVE
QUALITIES OF THOSE LOCATIONS.
00:40:03.819 --> 00:40:06.287
IT DOESN'T MATTER HOW
MANY PEOPLE WE ARREST,
00:40:06.389 --> 00:40:08.739
IT DOESN'T MATTER HOW
MANY PEOPLE AGE OUT OF CRIME,
00:40:08.858 --> 00:40:10.691
AND IT DOESN'T MATTER HOW
MANY POTENTIAL VICTIMS
00:40:10.776 --> 00:40:12.943
WE WARN TO PROTECT THEMSELVES.
00:40:13.029 --> 00:40:17.731
IF THE PLACE ISN'T CHANGED,
IT WILL CONSTANTLY ATTRACT
00:40:17.833 --> 00:40:19.900
SIMILAR BEHAVIOR
AS IT DID BEFORE.
00:40:21.253 --> 00:40:23.070
NARRATOR: WORKING WITH
ATLANTIC CITY POLICE,
00:40:23.172 --> 00:40:25.372
CAPLAN GATHERED DATA
ABOUT THE PLACES WHERE
00:40:25.458 --> 00:40:28.676
MAJOR CRIMES HAD OCCURRED.
00:40:28.761 --> 00:40:31.345
THE DATA WAS MATCHED TO
GEOGRAPHICAL INFORMATION
00:40:31.430 --> 00:40:33.347
TAKEN FROM GOOGLE MAPS.
00:40:33.432 --> 00:40:37.417
THEN HIS ALGORITHM LOOKED FOR
PATTERNS, HOT SPOTS FOR CRIME.
00:40:37.520 --> 00:40:39.803
PROF. JOEL CAPLAN: THESE WERE
IDENTIFIED AS THE PHASE ONE
00:40:39.889 --> 00:40:42.473
TARGET AREAS FOR
THE INITIAL ROLLOUT OF THE
00:40:42.558 --> 00:40:44.124
RISK-BASED POLICING INITIATIVE.
00:40:44.226 --> 00:40:47.361
IF WE WERE TO FOCUS ON THE
TOP FIVE PERCENT OF THE
00:40:47.446 --> 00:40:51.065
HIGHEST-RISK PLACES IDENTIFIED
BY THE RISK TERRAIN MODEL,
00:40:51.150 --> 00:40:53.784
WE WOULD ACCOUNT
FOR 45% OF ALL
00:40:53.903 --> 00:40:56.904
THE SHOOTING INCIDENTS IN 2015.
00:40:58.124 --> 00:41:00.124
NARRATOR: SOMETIMES, THE
SYSTEM IDENTIFIED PLACES THE
00:41:00.242 --> 00:41:03.377
POLICE ALREADY KNEW ATTRACTED
CRIME, SUCH AS CONVENIENCE
00:41:03.462 --> 00:41:07.131
STORES AND VACANT PROPERTIES.
00:41:08.517 --> 00:41:12.052
BUT LAUNDROMATS ALSO
CAME UP AS PROBLEM ZONES.
00:41:13.172 --> 00:41:14.888
PROF. JOEL CAPLAN: WE'VE
TESTED ROBBERY FOR EXAMPLE
00:41:14.974 --> 00:41:17.558
IN NEWARK, KANSAS
CITY, AND CHICAGO.
00:41:17.643 --> 00:41:21.145
AND WHILE THE CRIME TYPE IS
THE SAME, AND WHILE THE CRIMES
00:41:21.263 --> 00:41:25.315
CLUSTER ACROSS JURISDICTIONS
CREATING HOT SPOTS OF CRIME,
00:41:25.434 --> 00:41:28.936
THE RISK FACTORS THAT CREATE
THE CONTEXTS FOR ROBBERY ARE
00:41:29.021 --> 00:41:32.189
VERY DIFFERENT IN EACH
ONE OF THESE SETTINGS.
00:41:32.274 --> 00:41:37.111
AND, FOR EXAMPLE, BARS ARE NOT
A SIGNIFICANT RISK FACTOR FOR
00:41:37.196 --> 00:41:40.814
ROBBERY IN KANSAS CITY,
WHILE THEY ARE A SIGNIFICANT
00:41:40.916 --> 00:41:44.084
RISK FACTOR IN
NEWARK AND CHICAGO.
00:41:44.170 --> 00:41:48.455
AND IT'S THAT REALIZATION THAT
THINGS THAT WE MIGHT HAVE KIND
00:41:48.541 --> 00:41:53.927
OF INTUITIVELY ASSUMED TO BE
CREATING CONTEXT FOR CRIME ARE
00:41:54.013 --> 00:41:56.346
NOT IN FACT CORRELATED AT ALL.
00:41:58.350 --> 00:42:00.467
NARRATOR: THE IDEA IS TO
MANAGE THE RISK PRESENTED BY
00:42:00.553 --> 00:42:02.970
CERTAIN PLACES, WITHOUT
ASSUMING THE PEOPLE
00:42:03.055 --> 00:42:05.556
IN THOSE PLACES ARE CRIMINALS.
00:42:05.641 --> 00:42:07.024
CAPT. JAMES SARKOS: IF I'M AN
INDIVIDUAL WHO JUST HAPPENS
00:42:07.143 --> 00:42:09.943
TO LIVE IN AN AREA THAT WE'VE
DEEMED TO BE A HOTSPOT,
00:42:10.029 --> 00:42:11.645
I SHOULDN'T BE STOPPED BY THE
POLICE EVERY TIME I WALK INTO
00:42:11.731 --> 00:42:13.113
THE DOOR OF MY HOUSE.
00:42:13.199 --> 00:42:14.648
AND THAT'S WHAT'S
GREAT ABOUT THIS PROGRAM
00:42:14.734 --> 00:42:15.949
IS WE DON'T DO THAT.
00:42:16.035 --> 00:42:17.818
WE IDENTIFY THE AREA AND
THEN WE TRY TO ADDRESS THE
00:42:17.903 --> 00:42:21.321
GEOGRAPHICAL FEATURES OF
THAT AREA TO IMPROVE CRIME.
00:42:31.500 --> 00:42:34.084
IT MIGHT EVEN BE NOT THAT
SOMETHING PARTICULAR IS THERE,
00:42:34.170 --> 00:42:35.586
BUT MAYBE SOMETHING'S MISSING.
00:42:35.671 --> 00:42:38.555
FOR EXAMPLE, IT MIGHT BE AN
AREA THAT DOES NOT HAVE A
00:42:38.674 --> 00:42:39.973
CLEAR LINE OF SIGHT.
00:42:40.059 --> 00:42:41.175
IT MIGHT BE THAT THAT AREA
DOES NOT HAVE ANY SECURITY
00:42:41.260 --> 00:42:43.594
CAMERAS, SO PEOPLE FEEL
COMFORTABLE COMMITTING A CRIME
00:42:43.679 --> 00:42:46.263
THERE BECAUSE THEY THINK THEY'RE
GONNA GO GET AWAY WITH IT.
00:42:47.183 --> 00:42:48.515
NARRATOR: BUT THE TRULY
PREDICTIVE PART OF THE PROCESS
00:42:48.601 --> 00:42:51.018
IS WHAT HAPPENS AFTER
THE POLICE CLEAN UP
00:42:51.103 --> 00:42:53.386
THE ZONES OF HIGHEST RISK.
00:42:53.489 --> 00:42:56.657
BECAUSE THE MAP NOT
ONLY SHOWS WHERE CRIME IS,
00:42:56.742 --> 00:42:59.159
IT PREDICTS WHERE IT WILL GO.
00:43:00.029 --> 00:43:01.361
CAPT. JAMES SARKOS:
"WHACK-A-MOLE" IS A TERM THAT
00:43:01.447 --> 00:43:03.080
SOMETIMES POLICE OFFICERS FEEL
LIKE THAT'S WHAT WE'RE DOING.
00:43:03.199 --> 00:43:05.032
WE'RE PLAYING WHACK-A-MOLE WHERE
WE HIT A PERSON HERE AND THEN
00:43:05.117 --> 00:43:06.366
CRIME POPS UP OVER THERE,
00:43:06.452 --> 00:43:08.118
AND YOU'RE JUST
CONSTANTLY READJUSTING.
00:43:08.204 --> 00:43:10.070
SO THIS IS TRYING TO GET US
AHEAD OF WHACK-A-MOLE,
00:43:10.172 --> 00:43:11.922
SO THAT WE DON'T,
WE'RE NOT ALWAYS REACTING.
00:43:12.041 --> 00:43:13.290
WE WANNA BE AHEAD OF THAT.
00:43:13.375 --> 00:43:15.459
WE DON'T WANNA REACT,
WE WANNA BE RESPONDING BEFORE
00:43:15.544 --> 00:43:17.878
WE HAD TO REACT.
00:43:25.855 --> 00:43:27.387
PROF. JOEL CAPLAN: WE'VE
APPLIED THIS TO CRIME,
00:43:27.473 --> 00:43:31.424
ROBBERY, SHOOTINGS, BURGLARY,
DOMESTIC VIOLENCE,
00:43:31.527 --> 00:43:36.029
MOTOR VEHICLE THEFT, AND
A NUMBER OF OTHER CRIME TYPES.
00:43:42.905 --> 00:43:45.038
BUT WE'VE ALSO MADE RISK
TERRAIN MODELING FREE.
00:43:45.124 --> 00:43:46.540
RESEARCHERS
HAVE DOWNLOADED IT AND
00:43:46.625 --> 00:43:50.127
UTILIZED THE FREE RESOURCES
TO STUDY MARITIME SHIPPING,
00:43:50.246 --> 00:43:52.445
LOOKING AT PIRACY
ON THE EARTH'S OCEANS.
00:43:52.548 --> 00:43:57.668
TO LOOK AT CHILD ABUSE, AND
CHILD DROWNINGS AND SUICIDE.
00:43:57.753 --> 00:44:01.171
IT'S BEEN USED IN
ENVIRONMENTAL SCIENCE,
00:44:01.257 --> 00:44:03.757
POLLUTION CONTROL,
URBAN PLANNING.
00:44:03.843 --> 00:44:05.192
THE LIST GOES ON.
00:44:05.294 --> 00:44:10.013
AND IT'S BASICALLY A TOOL THAT
COULD APPLY TO ANY OUTCOME,
00:44:10.099 --> 00:44:13.934
AS LONG AS THE OUTCOME HAS
A SPATIAL CHARACTERISTIC.
00:44:23.329 --> 00:44:24.912
NARRATOR: THE
RISK TERRAIN MODEL
00:44:24.997 --> 00:44:27.865
ONLY LOOKS AT
PLACES, NOT PEOPLE.
00:44:27.950 --> 00:44:31.118
THERE IS A REASON
FOR THIS BEYOND BIAS.
00:44:31.220 --> 00:44:34.288
OUR NEW ELECTRONICALLY-ENHANCED
FORTUNE TELLERS RISK BEING
00:44:34.373 --> 00:44:37.490
OVERWHELMED BY ALL
THE DATA POURING IN.
00:44:37.593 --> 00:44:39.092
MARY BETH GERHARDT: THERE IS
AN ENORMOUS AMOUNT OF DATA
00:44:39.178 --> 00:44:40.177
THAT WE'RE DEALING WITH.
00:44:40.296 --> 00:44:42.296
AND IT CAN GO ON AND ON
BECAUSE THE MODELS
00:44:42.381 --> 00:44:44.932
THAT WE'RE LOOKING AT,
THEY'RE TAKING INTO
00:44:45.017 --> 00:44:46.300
ACCOUNT OBSERVATIONAL DATA,
00:44:46.385 --> 00:44:48.435
THEY'RE TAKING INTO
ACCOUNT SATELLITE DATA.
00:44:48.520 --> 00:44:50.103
IT'S REALLY ENDLESS.
00:44:50.189 --> 00:44:52.139
TOM SCHENK: IF YOU ALWAYS JUST
TRY TO THROW IN EVERYTHING,
00:44:52.224 --> 00:44:53.106
YOU'RE GONNA GET
A LOT OF NOISE.
00:44:53.192 --> 00:44:55.225
THINGS THAT DON'T MAKE SENSE.
00:44:58.314 --> 00:44:59.563
NATE SILVER: WE CAN
GET OVERWHELMED
00:44:59.648 --> 00:45:02.149
WITH DATA PRETTY EASILY.
00:45:05.154 --> 00:45:06.653
BRENNA BERMAN: I DON'T THINK
THERE'S SUCH THING AS TOO MUCH
00:45:06.739 --> 00:45:08.872
DATA, BUT WHEN YOU'RE THINKING
ABOUT HOW MUCH DATA YOU REALLY
00:45:08.991 --> 00:45:12.359
WANNA BE RESPONSIBLE FOR, YOU
NEED TO THINK CAREFULLY ABOUT
00:45:12.461 --> 00:45:14.995
WHAT YOU'RE GOING TO USE THE
DATA FOR, WHAT VALUE IT CAN
00:45:15.080 --> 00:45:18.248
BRING TO THE CLIENTS OR THE
RESIDENTS THAT YOU SERVE,
00:45:18.334 --> 00:45:19.917
AND THE LENGTHS THAT YOU'RE
GONNA HAVE TO GO TO TO
00:45:20.002 --> 00:45:22.536
PROPERLY MANAGE AND
PROTECT THAT DATA.
00:45:25.257 --> 00:45:27.808
NARRATOR: IN OUR EFFORT TO
MAKE THINGS EASIER, FASTER,
00:45:27.893 --> 00:45:31.094
AND MORE EFFICIENT, WE HAVE
GIVEN BIRTH TO A POWERFUL
00:45:31.180 --> 00:45:34.381
TECHNOLOGY WE MAY
NOT FULLY UNDERSTAND.
00:45:34.483 --> 00:45:35.432
NATE SILVER: WHEN
PEOPLE START TO SAY,
00:45:35.517 --> 00:45:38.885
"WELL, THE COMPUTER
HAS A MIND OF ITS OWN,"
00:45:38.988 --> 00:45:40.687
THAT'S WHEN
I GET WORRIED, RIGHT?
00:45:40.773 --> 00:45:43.824
BUT IF YOU GENERALLY DON'T
UNDERSTAND WHY A COMPUTER
00:45:43.909 --> 00:45:48.612
PROGRAM DOES WHAT IT DOES,
THEN THAT'S VERY DANGEROUS.
00:45:53.919 --> 00:45:56.119
ERIC SIEGEL: "WITH GREAT POWER
COMES GREAT RESPONSIBILITY."
00:45:56.205 --> 00:45:59.089
THAT'S FROM SPIDERMAN.
00:46:01.043 --> 00:46:02.876
THIS STUFF IS POWERFUL.
00:46:02.962 --> 00:46:05.595
IT'S NOT GOING TO BE
UNIVERSALLY INDICTED.
00:46:05.714 --> 00:46:08.181
YOU CAN'T JUST TURN IT OFF.
00:46:10.719 --> 00:46:12.219
IT'S TOO VALUABLE.
00:46:12.304 --> 00:46:15.689
IT'S HELPING TOO MANY PEOPLE,
AS WELL AS ORGANIZATIONS WITH
00:46:15.774 --> 00:46:17.090
TOO MUCH INTENSITY.
00:46:17.192 --> 00:46:18.608
IT'S NOT GOING AWAY.
00:46:18.727 --> 00:46:20.560
IT'S LIKE A KNIFE, YOU KNOW,
YOU CAN USE A KNIFE IN A GOOD
00:46:20.646 --> 00:46:24.564
WAY, YOU CAN USE A KNIFE IN
A BAD WAY, NOBODY'S GOING TO
00:46:24.650 --> 00:46:27.117
OUTLAW KNIVES IN GENERAL.
00:46:29.538 --> 00:46:31.655
NARRATOR: WE BUILT PREDICTIVE
MACHINES TO REDUCE THE
00:46:31.740 --> 00:46:35.275
UNCERTAINTY OF LIFE.
00:46:35.377 --> 00:46:37.160
BUT THE COMPLETE
PICTURE OF THE FUTURE
00:46:37.246 --> 00:46:41.331
MAY ALWAYS BE UNKNOWABLE.
00:46:42.334 --> 00:46:46.303
EVEN FOR PROPHETS AND
SEERS, HUMAN OR MACHINE,
00:46:46.422 --> 00:46:48.472
LIFE IS FULL OF SURPRISES.