0:03
is about automating the ineffable.
0:06
It is about unlocking a second mode of communication
0:09
I think that's incredibly powerful.
0:10
I'm Cassie Kozyrkov.
0:12
I serve as chief decision scientist.
0:13
I'm with Google Cloud.
0:16
What I'm really passionate about is turning information
0:19
I work on decision intelligence.
0:21
It's the applied side of machine learning.
0:23
In a nutshell, machine learning is
0:27
about explaining your wishes to computers in a different way
0:28
from traditional programming.
0:30
So think about it like this.
0:31
Traditional programming-- to get a computer
0:34
to do something for you, you have
0:35
to give explicit instructions to the computer.
0:40
Machine learning is asking for what you want with examples
0:41
instead of instructions.
0:44
The biggest challenge that I see marketers face with machine
0:45
learning is how to get started.
0:48
There's the research side of machine learning.
0:50
And then there's the applied side.
0:53
And the problem area for a lot of businesses
0:56
is that they don't know which one they want to be in.
0:59
The research side is all about building machine
1:01
learning algorithms for other people to use.
1:04
That's a lot like building kitchen appliances,
1:06
whereas applied machine learning is
1:09
all about cooking and cooking at scale.
1:11
And that is a completely different art.
1:13
Just because you know how to build the microwave,
1:15
doesn't make you a great chef and vice versa.
1:18
And so what they end up doing-- and this
1:20
is the big mistake of why businesses and marketers will
1:22
fail-- is they think they need to build their own microwave
1:26
They don't realize that to apply machine learning
1:28
to a specific business problem, you
1:30
don't need to reinvent the wheel and build that algorithm
1:33
There are warehouses upon warehouses of these algorithms
1:35
created by research teams.
1:38
And once it's created, everyone else can take it off the shelf
1:41
and start applying it to innovating in their models.
1:44
The key to success here is rigorous testing,
1:48
making sure that what you asked for has been delivered.
1:52
Keep testing the model on new data to see if it's working.
1:54
And bear in mind, there's only so much
1:56
that the hottest algorithm can help you
1:59
with if your data isn't very good.
2:02
Put your effort into making sure that your data,
2:05
your ingredients are relevant to the task.
2:07
And if you have the most beautiful ingredients,
2:09
you can get away with a very simple kitchen.
