1
00:00:06,070 --> 00:00:07,120
Hi, my name's John.

2
00:00:07,510 --> 00:00:10,140
I lead the search and machine
learning teams at Google.

3
00:00:12,130 --> 00:00:14,230
I think it's amazingly inspiring

4
00:00:14,230 --> 00:00:16,214
that people all over the world

5
00:00:16,215 --> 00:00:19,160
turn to search engines to
ask trivial questions

6
00:00:19,160 --> 00:00:20,930
and incredibly important questions.

7
00:00:20,930 --> 00:00:23,450
So it's a huge
responsibility to give them

8
00:00:23,450 --> 00:00:24,864
the best answers that we can.

9
00:00:26,710 --> 00:00:30,610
Hi, my name's Akshaya and 
I work on the Bing search team.

10
00:00:30,910 --> 00:00:33,190
There are many times where
we will start looking

11
00:00:33,190 --> 00:00:35,800
into artificial intelligence
and machine learning,

12
00:00:35,830 --> 00:00:39,010
but we have to address how are
the users going to use this,

13
00:00:39,140 --> 00:00:42,390
because at the end of the day,
we want to make an impact to society.

14
00:00:43,780 --> 00:00:45,400
Let's ask a simple question.

15
00:00:45,820 --> 00:00:48,070
How long does it take to travel to Mars?

16
00:00:49,330 --> 00:00:50,950
Where did these results come from

17
00:00:51,370 --> 00:00:54,100
and why was this listed
before the other one?

18
00:00:55,700 --> 00:00:58,150
Okay, let's dive in and
see how the search engine

19
00:00:58,150 --> 00:00:59,860
turned your request into a result.

20
00:01:00,690 --> 00:01:03,360
The first thing you need to
know is when you do a search,

21
00:01:03,430 --> 00:01:06,480
the search engine isn't actually
going out to the World Wide Web

22
00:01:06,480 --> 00:01:08,010
to run your search in real time.

23
00:01:08,140 --> 00:01:10,610
And that's because there's
over a billion websites

24
00:01:10,610 --> 00:01:14,140
on the internet and hundreds more are
being created every single minute.

25
00:01:14,140 --> 00:01:16,210
So if the search engine
had to look through

26
00:01:16,240 --> 00:01:18,690
every single site to
find the one you wanted,

27
00:01:18,690 --> 00:01:20,120
it would just take forever.

28
00:01:20,500 --> 00:01:21,940
So to make your search faster,

29
00:01:21,970 --> 00:01:24,940
search engines are constantly
scanning the web in advance

30
00:01:25,420 --> 00:01:28,560
to record the information that might
help with your search later.

31
00:01:28,930 --> 00:01:31,270
That way, when you search
about travel to Mars,

32
00:01:31,630 --> 00:01:33,700
the search engine
already has what it needs

33
00:01:33,700 --> 00:01:35,728
to give you an answer in real time.

34
00:01:36,250 --> 00:01:37,540
Here's how it works.

35
00:01:37,900 --> 00:01:42,010
The internet is a web of pages
connected to each other by hyperlinks.

36
00:01:42,400 --> 00:01:44,680
Search engines are
constantly running a program

37
00:01:44,680 --> 00:01:47,380
called a Spider that cross
through these web pages

38
00:01:47,380 --> 00:01:49,040
to collect information about them.

39
00:01:49,780 --> 00:01:51,550
Each time it finds a hyperlink,

40
00:01:52,090 --> 00:01:55,000
it follows it until it
has visited every page

41
00:01:55,030 --> 00:01:57,240
it can find on the entire
internet.

42
00:01:57,335 --> 00:01:59,170
For each page the spider visits,

43
00:01:59,200 --> 00:02:02,320
it records any information
it might need for a search

44
00:02:02,500 --> 00:02:05,650
by adding it to a special
database called a search index.

45
00:02:07,166 --> 00:02:09,530
Now, let's go back to
that search from earlier

46
00:02:09,590 --> 00:02:11,990
and see if we can figure
out how the search engine

47
00:02:11,990 --> 00:02:13,333
came up with the results.

48
00:02:13,640 --> 00:02:16,460
When you ask how long does
it take to travel to Mars,

49
00:02:16,640 --> 00:02:18,860
the search engine looks
in each of those words

50
00:02:18,920 --> 00:02:21,410
in the search index to
immediately get a list

51
00:02:21,410 --> 00:02:24,500
of all the pages on the
internet containing those words.

52
00:02:24,890 --> 00:02:26,870
But just looking for these search terms

53
00:02:26,870 --> 00:02:28,760
could return millions of pages,

54
00:02:28,760 --> 00:02:31,110
so the search engine needs
to be able to determine

55
00:02:31,110 --> 00:02:33,120
the best matches to show you first.

56
00:02:33,340 --> 00:02:36,010
This is where it gets tricky
because the search engine

57
00:02:36,010 --> 00:02:38,040
may need to guess what
you're looking for.

58
00:02:38,930 --> 00:02:41,360
Each search engine
uses its own algorithm

59
00:02:41,360 --> 00:02:44,230
to rank the pages based on
what it thinks you want.

60
00:02:44,930 --> 00:02:47,660
The search engine's ranking
algorithm might check

61
00:02:47,990 --> 00:02:50,360
if your search term shows
up in the page title,

62
00:02:50,900 --> 00:02:53,820
it might check if all of the
words show up next to each other,

63
00:02:54,520 --> 00:02:57,020
or any number of other calculations

64
00:02:57,020 --> 00:02:58,610
that help it better determine

65
00:02:58,670 --> 00:03:01,420
which pages you'll want
to see and which you won't.

66
00:03:02,960 --> 00:03:04,960
Google invented the most 
famous algorithm

67
00:03:04,960 --> 00:03:08,530
for choosing the most relevant results
for a search by taking into account

68
00:03:08,560 --> 00:03:11,230
how many other Web pages
linked to a given page.

69
00:03:11,830 --> 00:03:14,140
The idea is that if
lots of websites think

70
00:03:14,140 --> 00:03:15,660
that a web page is interesting,

71
00:03:15,660 --> 00:03:17,940
then it's probably the one 
you're looking for.

72
00:03:18,190 --> 00:03:20,020
This algorithm is called page rank,

73
00:03:20,590 --> 00:03:22,330
not because it ranks web pages,

74
00:03:22,570 --> 00:03:25,210
but because it was named after
its inventor, Larry Page,

75
00:03:25,480 --> 00:03:27,333
who's one of the founders of Google.

76
00:03:27,940 --> 00:03:30,520
Because a website often makes
money when you visit it,

77
00:03:30,820 --> 00:03:32,950
spammers are constantly
trying to find ways

78
00:03:32,950 --> 00:03:35,741
to game the search algorithm 
so that their pages

79
00:03:35,742 --> 00:03:37,931
are listed higher in the results.

80
00:03:38,260 --> 00:03:40,750
Search engines regularly
update their algorithms

81
00:03:40,750 --> 00:03:44,296
to prevent fake or untrustworthy
sites from reaching the top.

82
00:03:44,680 --> 00:03:47,350
Ultimately, it's up to you 
to keep an eye out

83
00:03:47,500 --> 00:03:49,450
for these pages that are untrustworthy

84
00:03:49,690 --> 00:03:52,990
by looking at the web address and
making sure it's a reliable source.

85
00:03:53,680 --> 00:03:55,390
Search programs are always evolving

86
00:03:55,420 --> 00:03:58,420
to improve the algorithms 
wo they return better results,

87
00:03:58,540 --> 00:04:00,460
faster results than their competitors.

88
00:04:01,000 --> 00:04:03,100
Today's search engines
even use information

89
00:04:03,100 --> 00:04:06,820
that you haven't explicitly provided
to help you narrow down your search.

90
00:04:07,150 --> 00:04:10,120
So, for example,
if you did a search for dog parks,

91
00:04:10,240 --> 00:04:12,190
many search engines
would give you results

92
00:04:12,190 --> 00:04:13,840
for all the dog parks nearby,

93
00:04:14,080 --> 00:04:16,260
even though you didn't
type in your location.

94
00:04:17,800 --> 00:04:20,530
Modern search engines
also understand more

95
00:04:20,530 --> 00:04:22,060
than just the words on a page,

96
00:04:22,300 --> 00:04:24,970
but what they actually mean
in order to find the best one

97
00:04:24,970 --> 00:04:26,750
that matches what you're looking for.

98
00:04:27,130 --> 00:04:29,980
For example, if you search 
for fast pitcher,

99
00:04:30,280 --> 00:04:32,300
it will know you're
looking for an athlete.

100
00:04:32,500 --> 00:04:34,450
But if you search for large pitcher,

101
00:04:34,450 --> 00:04:36,730
it will find you options
for your kitchen.

102
00:04:38,420 --> 00:04:41,910
To understand the words better, 
we use something called machine learning,

103
00:04:41,910 --> 00:04:43,985
a type of artificial intelligence.

104
00:04:43,985 --> 00:04:46,050
It enables search
algorithms to search out

105
00:04:46,090 --> 00:04:48,400
not just individual letters
or words in the page,

106
00:04:48,400 --> 00:04:51,280
but understand the underlying
meaning of the words.

107
00:04:53,690 --> 00:04:55,850
The internet is growing exponentially,

108
00:04:56,210 --> 00:04:59,810
but if the teams that design
search engines do our jobs right,

109
00:05:00,080 --> 00:05:04,090
the information you want should
always be just a few keystrokes away.