0:00:07.360,0:00:11.760
Machine learning is only as good as the[br]training data you put into it.

0:00:11.800,0:00:15.820
So, it's super important to use high quality data, and lots of it.

0:00:16.760,0:00:21.960
But if data is important, it's worth asking where does training data come from?

0:00:22.280,0:00:26.260
Often, computers are collecting training data from people like you and me,

0:00:26.260,0:00:27.860
without any effort on our part.

0:00:28.440,0:00:31.480
A video streaming service might keep track of what you watch, then it can recognize patterns

0:00:31.660,0:00:36.000
in that data to recommend what you might want to watch next.

0:00:37.420,0:00:43.200
Other times, you're directly asked to help, like when a website asks you to spot street signs and photos,

0:00:43.780,0:00:49.280
You're providing training data to help a[br]machine learn to see, and maybe even one day drive.

0:00:52.320,0:00:56.440
Medical researchers can use[br]medical images as training data to teach

0:00:56.520,0:00:59.900
computers how to recognize and diagnose diseases.

0:01:00.300,0:01:05.560
Machine Learning needs hundreds and thousands of images, and training direction from a doctor

0:01:05.640,0:01:09.920
who knows what to look for, before it can correctly identify disease.

0:01:10.520,0:01:15.540
Even with thousands of examples, there can be problems with the computer's predictions.

0:01:15.880,0:01:20.660
If X-ray data is only collected from men, then the computer's predictions may only work for men.

0:01:21.880,0:01:26.300
It may not recognize diseases when[br]asked to diagnose the X-ray of a woman.

0:01:26.620,0:01:30.820
This blind spot in the training data[br]creates something called bias.

0:01:31.260,0:01:36.420
Biased data favors some things, and de-prioritizes or excludes others.

0:01:36.780,0:01:41.800
Depending on how training data is collected, who is doing the collecting, and how the data is fed,

0:01:41.800,0:01:45.340
there is a chance that[br]human bias is included in the data.

0:01:45.880,0:01:50.700
By learning from bias data, the computer may make biased predictions,

0:01:50.780,0:01:54.320
whether the people training the computer[br]are aware of it or not.

0:01:54.760,0:01:58.400
When you are looking at training data, ask yourself two questions:

0:01:58.640,0:02:01.600
Is this enough data to accurately train a computer?

0:02:02.320,0:02:06.860
And, does this data represent all possible scenarios and users without bias?

0:02:07.460,0:02:11.040
This is where you, as the human training, play a crucial role.

0:02:11.160,0:02:14.500
It's up to you to give your machine unbiased data.

0:02:14.500,0:02:18.160
That means collecting tons of examples, from lots of sources.

0:02:19.300,0:02:22.580
Remember, when you pick and choose data for machine learning,

0:02:22.580,0:02:26.660
you're actually programming the algorithm, using training data instead of code.

0:02:27.100,0:02:29.780
The data IS the code.

0:02:30.180,0:02:34.680
The better the data you provide, the better the computer will learn.