Would you like to inspect the original subtitles? These are the user uploaded subtitles that are being translated:
1
00:00:00,540 --> 00:00:00,840
All right.
2
00:00:00,860 --> 00:00:06,480
So as we talk more about A.I. and machine learning and statistical modeling it's important to remember
3
00:00:06,480 --> 00:00:12,720
that these tools can be incredibly powerful when they're used appropriately but incredibly dangerous
4
00:00:12,780 --> 00:00:14,230
when they aren't.
5
00:00:14,250 --> 00:00:19,920
Now the thing about tools like this key influence or visual here is that they're great because they
6
00:00:20,130 --> 00:00:23,520
helped make machine learning accessible to everyday users.
7
00:00:23,520 --> 00:00:29,490
But a little scary because those same users often lack the foundational knowledge to understand what's
8
00:00:29,490 --> 00:00:35,370
happening behind the curtain and how to properly interpret the results or make intelligent decisions
9
00:00:35,400 --> 00:00:36,630
based on them.
10
00:00:36,630 --> 00:00:42,810
So speaking of interpreting results I think this is a good time to take a step back take a pause and
11
00:00:42,810 --> 00:00:47,540
review one of the most important rules in statistics and analytics.
12
00:00:47,610 --> 00:00:51,210
Correlation does not imply causation.
13
00:00:51,240 --> 00:00:55,460
Now I'm sure many of you have heard this before especially if you work in data or analytics.
14
00:00:55,530 --> 00:01:02,860
But let's take two minutes and break this down correlation is one two variables x and y move together
15
00:01:03,400 --> 00:01:04,510
kind of like this.
16
00:01:04,600 --> 00:01:11,830
They move in the same direction causation on the other hand is one variable X causes variable Y.
17
00:01:11,830 --> 00:01:16,970
In other words there's a clear cause and effect relationship here now.
18
00:01:17,020 --> 00:01:23,170
What if I were to show you a scatter plot like this where plotting violent crime rate on the y axis
19
00:01:23,590 --> 00:01:31,870
and some mystery variable on the X and based on this 25 30 observations here we've got a very tight
20
00:01:31,870 --> 00:01:36,900
correlation pretty clear linear relationship between the two variables.
21
00:01:36,970 --> 00:01:39,760
So clearly they move in the same direction.
22
00:01:39,760 --> 00:01:47,050
Clearly they're correlated but you might be tempted to think you know that this x axis variable is the
23
00:01:47,050 --> 00:01:50,790
driver behind violent crimes it's causing violent crimes.
24
00:01:50,950 --> 00:01:56,500
And that if only we could cut back on whatever this variable is we might be able to make our streets
25
00:01:56,500 --> 00:01:57,550
safer.
26
00:01:57,550 --> 00:02:02,830
The problem with that is that we're looking at ice cream cones sold and you may be scratching your head.
27
00:02:02,830 --> 00:02:08,050
You may be a little confused and that's totally understandable because the human brain is biased to
28
00:02:08,050 --> 00:02:13,040
look for cause and effect relationships where in fact they don't exist.
29
00:02:13,060 --> 00:02:18,100
So if you're still wondering kind of what's going on here and how this could possibly be true here's
30
00:02:18,100 --> 00:02:27,640
a hint our y axis violent crime rate could just as easily be drowning deaths or forest fires or even
31
00:02:27,700 --> 00:02:30,040
your dreaded crab attack.
32
00:02:30,040 --> 00:02:33,500
So think about what those things have in common.
33
00:02:33,670 --> 00:02:38,950
And you probably start to realize that this has nothing to do with ice cream at all and everything to
34
00:02:38,950 --> 00:02:40,850
do with temperature.
35
00:02:41,140 --> 00:02:47,110
As temperatures rise you have more people out later gathering in public spaces as a result.
36
00:02:47,110 --> 00:02:48,700
Crime rates increase.
37
00:02:48,700 --> 00:02:51,720
You also have more people going to the beach and swimming in the ocean.
38
00:02:51,760 --> 00:02:55,230
So drowning deaths increase and so on and so forth.
39
00:02:55,270 --> 00:03:02,530
So because ice cream sales are such a close proxy temperature we've created a false narrative that paints
40
00:03:02,530 --> 00:03:07,130
a completely misleading story so key takeaways here.
41
00:03:07,130 --> 00:03:12,170
Ice cream does not turn you into a violent criminal does not make you drown.
42
00:03:12,290 --> 00:03:18,040
It does not start forest fires and it certainly does not encourage crabs to attack you.
43
00:03:18,080 --> 00:03:24,680
Now obviously these are silly examples here but the core principle that concept holds true and it's
44
00:03:24,680 --> 00:03:26,830
a really important one to keep in mind.
45
00:03:26,960 --> 00:03:32,800
I'll leave you with one kind of more real world business case here should be a scenario like this.
46
00:03:32,920 --> 00:03:34,580
You know maybe you run a startup.
47
00:03:34,730 --> 00:03:39,380
You've been live for about four months and you're plotting your weekly marketing spend which you've
48
00:03:39,380 --> 00:03:42,270
been ramping up against your total revenue.
49
00:03:42,290 --> 00:03:48,740
Now if you were to imply causation based on this chart these results here you might think that ramping
50
00:03:48,740 --> 00:03:53,120
up your marketing spend is a surefire way to drive more revenue.
51
00:03:53,120 --> 00:03:57,230
And that may be the case it may be true but it also may not.
52
00:03:57,230 --> 00:04:02,720
So the idea is you've got to think about the other factors that might be at play here maybe over this
53
00:04:02,720 --> 00:04:04,340
three or four month period.
54
00:04:04,340 --> 00:04:11,930
You've also been ramping up a new sales team or maybe your organic traffic has been growing due to referrals
55
00:04:11,990 --> 00:04:14,210
or PR or something like that.
56
00:04:14,210 --> 00:04:21,050
So bottom line here be thoughtful about how you interpret these results and Please use caution before
57
00:04:21,050 --> 00:04:23,810
you make big decisions based on these findings.
6244
Can't find what you're looking for?
Get subtitles in any language from opensubtitles.com, and translate them here.