How does artificial intelligence learn? - Briana Brownell

709,016 views ใƒป 2021-03-11

TED-Ed


ืื ื ืœื—ืฅ ืคืขืžื™ื™ื ืขืœ ื”ื›ืชื•ื‘ื™ื•ืช ื‘ืื ื’ืœื™ืช ืœืžื˜ื” ื›ื“ื™ ืœื”ืคืขื™ืœ ืืช ื”ืกืจื˜ื•ืŸ.

ืชืจื’ื•ื: Ido Dekkers ืขืจื™ื›ื”: Naama Lieberman
00:09
Today, artificial intelligence helps doctors diagnose patients,
0
9829
5167
ื”ื™ื•ื, ื‘ื™ื ื” ืžืœืื›ื•ืชื™ืช ืขื•ื–ืจืช ืœืจื•ืคืื™ื ืœืื‘ื—ืŸ ืžื˜ื•ืคืœื™ื,
00:14
pilots fly commercial aircraft, and city planners predict traffic.
1
14996
5042
ืœื˜ื™ื™ืกื™ื ืœื”ื˜ื™ืก ืžื˜ื•ืกื™ื ืžืกื—ืจื™ื™ื, ื•ืœืžืชื›ื ื ื™ ืขืจื™ื ืœื—ื–ื•ืช ืชื ื•ืขื”.
00:20
But no matter what these AIs are doing, the computer scientists who designed them
2
20038
4250
ืื‘ืœ ืœื ืžืฉื ื” ืžื” ื”ื‘โ€œืž ื”ืืœื” ืขื•ืฉื•ืช, ืžื“ืขื ื™ ื”ืžื—ืฉื‘ ืฉืชื›ื ื ื• ืื•ืชืŸ
00:24
likely donโ€™t know exactly how theyโ€™re doing it.
3
24288
2750
ื›ื ืจืื” ืœื ื‘ื“ื™ื•ืง ื™ื•ื“ืขื™ื ืื™ืš ื”ืŸ ืขื•ืฉื•ืช ืืช ื–ื”.
00:27
This is because artificial intelligence is often self-taught,
4
27038
3875
ื–ื” ื‘ื’ืœืœ ืฉื‘ื™ื ื” ืžืœืื›ื•ืชื™ืช ื”ืจื‘ื” ืคืขืžื™ื ืžืœืžื“ืช ืืช ืขืฆืžื”,
00:30
working off a simple set of instructions
5
30913
2374
ื›ืฉื”ื™ื ืขื•ื‘ื“ืช ืžืกื˜ ืคืฉื•ื˜ ืฉืœ ื”ื•ืจืื•ืช
00:33
to create a unique array of rules and strategies.
6
33287
3584
ื›ื“ื™ ืœื™ืฆื•ืจ ืžืขืจืš ื™ื™ื—ื•ื“ื™ ืฉืœ ื—ื•ืงื™ื ื•ืืกื˜ืจื˜ื’ื™ื•ืช.
00:36
So how exactly does a machine learn?
7
36871
2625
ืื– ืื™ืš ื‘ื“ื™ื•ืง ืžื›ื•ื ื” ืœื•ืžื“ืช?
00:39
There are many different ways to build self-teaching programs.
8
39496
3000
ื™ืฉ ื”ืจื‘ื” ื“ืจื›ื™ื ืฉื•ื ื•ืช ืœื‘ื ื•ืช ืชื•ื›ื ื•ืช ืฉืžืœืžื“ื•ืช ืืช ืขืฆืžืŸ.
00:42
But they all rely on the three basic types of machine learning:
9
42496
3916
ืื‘ืœ ื›ื•ืœืŸ ืžืกืชืžื›ื•ืช ืขืœ ืฉืœื•ืฉื” ืกื•ื’ื™ื ื‘ืกื™ืกื™ื™ื ืฉืœ ืœืžื™ื“ืช ืžื›ื•ื ื”:
00:46
unsupervised learning, supervised learning, and reinforcement learning.
10
46412
5000
ืœื™ืžื•ื“ ืœื ืžืคื•ืงื—, ืœื™ืžื•ื“ ืžืคื•ืงื—, ื•ืœื™ืžื•ื“ ืขื ื—ื™ื–ื•ืงื™ื.
00:51
To see these in action,
11
51412
1959
ื›ื“ื™ ืœืจืื•ืช ืื•ืชืŸ ื‘ืคืขื•ืœื”,
00:53
letโ€™s imagine researchers are trying to pull information
12
53371
3250
ื‘ื•ืื• ื ื“ืžื™ื™ืŸ ื—ื•ืงืจื™ื ืฉืžื ืกื™ื ืœืžืฆื•ืช ืžื™ื“ืข
00:56
from a set of medical data containing thousands of patient profiles.
13
56621
4208
ืžืกื˜ ืฉืœ ืžื™ื“ืข ืจืคื•ืื™ ืฉืžื›ื™ืœ ืืœืคื™ ืคืจื•ืคื™ืœื™ื ืฉืœ ืžื˜ื•ืคืœื™ื.
01:01
First up, unsupervised learning.
14
61371
3125
ืจืืฉื™ืช, ืœืžื™ื“ื” ืœื ืžืคื•ืงื—ืช.
01:04
This approach would be ideal for analyzing all the profiles
15
64496
3333
ื”ื’ื™ืฉื” ื”ื–ื• ืชื”ื™ื” ืื™ื“ื™ืืœื™ืช ืœื ื™ืชื•ื— ื›ืœ ื”ืคืจื•ืคื™ืœื™ื
01:07
to find general similarities and useful patterns.
16
67829
3625
ื›ื“ื™ ืœื’ืœื•ืช ื ืงื•ื“ื•ืช ื“ืžื™ื•ืŸ ื›ืœืœื™ื•ืช ื•ืชื‘ื ื™ื•ืช ืฉื™ืžื•ืฉื™ื•ืช.
01:11
Maybe certain patients have similar disease presentations,
17
71454
3417
ืื•ืœื™ ืžื˜ื•ืคืœื™ื ืžืกื•ื™ืžื™ื ืžืฆื™ื’ื™ื ืžื—ืœื•ืช ื“ื•ืžื”,
01:14
or perhaps a treatment produces specific sets of side effects.
18
74871
4042
ืื• ืื•ืœื™ ื˜ื™ืคื•ืœ ืžื™ื™ืฆืจ ืกื˜ ืกืคืฆื™ืคื™ ืฉืœ ืชื•ืคืขื•ืช ืœื•ื•ืื™.
01:18
This broad pattern-seeking approach can be used to identify similarities
19
78913
4125
ื”ื’ื™ืฉื” ื”ืจื—ื‘ื” ืœื—ื™ืคื•ืฉ ืชื‘ื ื™ื•ืช ื™ื›ื•ืœื” ืœืฉืžืฉ ืœื’ืœื•ืช ื ืงื•ื“ื•ืช ื“ืžื™ื•ืŸ
01:23
between patient profiles and find emerging patterns,
20
83038
3375
ื‘ื™ืŸ ืคืจื•ืคื™ืœื™ ืžื˜ื•ืคืœื™ื ื•ืœื’ืœื•ืช ืชื‘ื ื™ื•ืช,
01:26
all without human guidance.
21
86413
2291
ื”ื›ื•ืœ ื‘ืœื™ ื”ื“ืจื›ื” ืฉืœ ืื ืฉื™ื.
01:28
But let's imagine doctors are looking for something more specific.
22
88704
3209
ืื‘ืœ ื‘ื•ืื• ื ื“ืžื™ื™ืŸ ืฉื”ืจื•ืคืื™ื ืžื—ืคืฉื™ื ืžืฉื”ื• ืกืคืฆื™ืคื™ ื™ื•ืชืจ.
01:32
These physicians want to create an algorithm
23
92371
2208
ื”ืจื•ืคืื™ื ื”ืืœื” ืจื•ืฆื™ื ืœื™ืฆื•ืจ ืืœื’ื•ืจื™ืชื
01:34
for diagnosing a particular condition.
24
94579
2709
ื›ื“ื™ ืœืื‘ื—ืŸ ืžืฆื‘ ืžืกื•ื™ื.
01:37
They begin by collecting two sets of dataโ€”
25
97288
2583
ื”ื ืžืชื—ื™ืœื™ื ืขืœ ื™ื“ื™ ืื™ืกื•ืฃ ืฉื ื™ ืกื˜ื™ื ืฉืœ ืžื™ื“ืข -
01:39
medical images and test results from both healthy patients
26
99871
3375
ื“ื™ืžื•ื™ื™ื ืจืคื•ืื™ื™ื ื•ืชื•ืฆืื•ืช ื‘ื“ื™ืงื•ืช ืฉืœ ืžื˜ื•ืคืœื™ื ื‘ืจื™ืื™ื
01:43
and those diagnosed with the condition.
27
103246
2417
ื•ืฉืœ ืืœื• ืฉืžืื•ื‘ื—ื ื™ื ืขื ื”ืžื—ืœื”.
01:45
Then, they input this data into a program
28
105663
2583
ืื–, ื”ื ืžื›ื ื™ืกื™ื ืืช ื”ื ืชื•ื ื™ื ืœืชื•ื›ื ื”
01:48
designed to identify features shared by the sick patients
29
108246
3375
ืฉืžื™ื•ืขื“ืช ืœื–ื”ื•ืช ืชื›ื•ื ื•ืช ืžืฉื•ืชืคื•ืช ืœื—ื•ืœื™ื
01:51
but not the healthy patients.
30
111621
2125
ืื‘ืœ ืœื ืืฆืœ ื”ื‘ืจื™ืื™ื.
01:53
Based on how frequently it sees certain features,
31
113746
3083
ื‘ื”ืชื‘ืกืก ืขืœ ื”ืชื›ื™ืคื•ืช ืฉื‘ื” ื”ื™ื ืžื–ื”ื” ืชื›ื•ื ื•ืช ืžืกื•ื™ืžื•ืช,
01:56
the program will assign values to those featuresโ€™ diagnostic significance,
32
116829
4042
ื”ืชื•ื›ื ื” ืชืฉื™ื™ืš ืขืจื›ื™ื ืœืžืฉืžืขื•ืช ื”ืื‘ื—ื•ื ื™ืช ืฉืœ ืชื›ื•ื ื•ืช ื”ืืœื•,
02:00
generating an algorithm for diagnosing future patients.
33
120871
3708
ื•ืชื™ื™ืฆืจ ืืœื’ื•ืจื™ืชื ืœืื‘ื—ื•ืŸ ื—ื•ืœื™ื ื‘ืขืชื™ื“.
02:04
However, unlike unsupervised learning,
34
124579
3167
ืขื ื–ืืช, ื‘ื ื™ื’ื•ื“ ืœืœื™ืžื•ื“ ืœื ืžืคื•ืงื—,
02:07
doctors and computer scientists have an active role in what happens next.
35
127746
4625
ืœืจื•ืคืื™ื ื•ืœืžื“ืขื ื™ ืžื—ืฉื‘ ื™ืฉ ืชืคืงื™ื“ ืคืขื™ืœ ื‘ืžื” ืฉืงื•ืจื” ื”ืœืื”.
02:12
Doctors will make the final diagnosis
36
132371
2083
ืจื•ืคืื™ื ื™ืขืฉื• ืื‘ื—ื•ืŸ ืกื•ืคื™
02:14
and check the accuracy of the algorithmโ€™s prediction.
37
134454
2917
ื•ื™ื‘ื“ืงื• ืืช ื“ื™ื•ืง ื”ืชื—ื–ื™ื•ืช ืฉืœ ื”ืืœื’ื•ืจื™ืชื.
02:17
Then computer scientists can use the updated datasets
38
137871
2917
ืื– ืžื“ืขื ื™ ืžื—ืฉื‘ ื™ื›ื•ืœื™ื ืœื”ืฉืชืžืฉ ื‘ืžืื’ืจ ื”ืžื™ื“ืข ื”ืžืขื•ื“ื›ืŸ
02:20
to adjust the programโ€™s parameters and improve its accuracy.
39
140788
3500
ื›ื“ื™ ืœื›ื•ื•ื ืŸ ืืช ื”ืคืจืžื˜ืจื™ื ืฉืœ ื”ืืœื’ื•ืจื™ืชื ื•ืœืฉืคืจ ืืช ื”ื“ื™ื•ืง ืฉืœื•.
02:24
This hands-on approach is called supervised learning.
40
144663
3166
ื”ื’ื™ืฉื” ื”ืžืขื•ืจื‘ืช ื”ื–ื• ื ืงืจืืช ืœืžื™ื“ื” ืžืคื•ืงื—ืช.
02:27
Now, letโ€™s say these doctors want to design another algorithm
41
147829
3167
ืขื›ืฉื™ื• ื‘ื•ืื• ื ื’ื™ื“ ืฉื”ืจื•ืคืื™ื ื”ืืœื” ืจื•ืฆื™ื ืœืขืฆื‘ ืืœื’ื•ืจื™ืชื ืื—ืจ
02:30
to recommend treatment plans.
42
150996
1750
ื›ื“ื™ ืœื”ืžืœื™ืฅ ืขืœ ืชื•ื›ื ื™ืช ื˜ื™ืคื•ืœ.
02:32
Since these plans will be implemented in stages,
43
152746
2833
ืžืื—ืจ ืฉื”ืชื•ื›ื ื™ื•ืช ื”ืืœื• ื™ื™ื•ืฉืžื• ื‘ืฉืœื‘ื™ื,
02:35
and they may change depending on each individual's response to treatments,
44
155579
3959
ื•ื”ืŸ ืื•ืœื™ ื™ืฉืชื ื• ืœืคื™ ื”ืชื’ื•ื‘ื” ืฉืœ ื›ืœ ืžื˜ื•ืคืœ ืœื˜ื™ืคื•ืœ,
02:39
the doctors decide to use reinforcement learning.
45
159538
2833
ื”ืจื•ืคืื™ื ืžื—ืœื™ื˜ื™ื ืœื”ืฉืชืžืฉ ื‘ืœืžื™ื“ื” ืžื—ื•ื–ืงืช.
02:42
This program uses an iterative approach to gather feedback
46
162746
3167
ื”ืชื•ื›ื ื” ื”ื–ื• ืžืฉืชืžืฉืช ื‘ืชื”ืœื™ืš ืžื—ื–ื•ืจื™ ื›ื“ื™ ืœืืกื•ืฃ ืžืฉื•ื‘
02:45
about which medications, dosages and treatments are most effective.
47
165913
4583
ืขืœ ืื™ื–ื” ืชืจื•ืคื•ืช, ืžื™ื ื•ื ื™ื ื•ื˜ื™ืคื•ืœื™ื ื”ื›ื™ ืืคืงื˜ื™ื‘ื™ื™ื.
02:50
Then, it compares that data against each patientโ€™s profile
48
170496
3000
ืื–, ื”ื™ื ืžืฉื•ื•ื” ืืช ื”ืžื™ื“ืข ืžื•ืœ ืคืจื•ืคื™ืœ ืฉืœ ื›ืœ ืžื˜ื•ืคืœ
02:53
to create their unique, optimal treatment plan.
49
173496
2833
ื›ื“ื™ ืœื™ืฆื•ืจ ืชื•ื›ื ื™ืช ื˜ื™ืคื•ืœ ื™ื™ื—ื•ื“ื™ืช ื•ืื•ืคื˜ื™ืžืœื™ืช.
02:56
As the treatments progress and the program receives more feedback,
50
176329
3500
ื›ืฉื”ื˜ื™ืคื•ืœ ืžืชืงื“ื ื•ื”ืชื•ื›ื ื™ืช ืžืงื‘ืœืช ืขื•ื“ ืžืฉื•ื‘,
02:59
it can constantly update the plan for each patient.
51
179829
3292
ื”ื™ื ื™ื›ื•ืœื” ืœืขื“ื›ืŸ ืืช ื”ืชื•ื›ื ื™ืช ื›ืœ ื”ื–ืžืŸ ืœื›ืœ ื—ื•ืœื”.
03:03
None of these three techniques are inherently smarter than any other.
52
183121
3375
ืืฃ ืื—ืช ืžืฉืœื•ืฉ ื”ืฉื™ื˜ื•ืช ื”ืืœื• ืœื ื—ื›ืžื” ื™ื•ืชืจ ื‘ื‘ืกื™ืกื” ืžื”ืื—ืจื•ืช.
03:06
While some require more or less human intervention,
53
186496
2667
ื‘ืขื•ื“ ืฉื›ืžื” ืžื”ืŸ ื“ื•ืจืฉื•ืช ื™ื•ืชืจ ืื• ืคื—ื•ืช ื”ืชืขืจื‘ื•ืช ืื ื•ืฉื™ืช,
03:09
they all have their own strengths and weaknesses
54
189163
2416
ืœื›ื•ืœืŸ ื™ืฉ ื”ื—ื•ื–ืงื•ืช ื•ื”ื—ื•ืœืฉื•ืช
03:11
which makes them best suited for certain tasks.
55
191579
2250
ืฉื”ื•ืคื›ื•ืช ืื•ืชืŸ ืœืžืชืื™ืžื•ืช ื‘ื™ื•ืชืจ ืœืžืงืจื™ื ืžืกื•ื™ืžื™ื.
03:14
However, by using them together,
56
194329
2375
ืขื ื–ืืช, ืขืœ ื™ื“ื™ ืฉื™ืžื•ืฉ ื‘ื›ื•ืœืŸ ื™ื—ื“,
03:16
researchers can build complex AI systems,
57
196704
2875
ื—ื•ืงืจื™ื ื™ื›ื•ืœื™ื ืœื‘ื ื•ืช ืžืขืจื›ื•ืช ื‘โ€œืž ืžื•ืจื›ื‘ื•ืช,
03:19
where individual programs can supervise and teach each other.
58
199579
3292
ืฉื ืชื•ื›ื ื•ืช ื‘ื•ื“ื“ื•ืช ื™ื›ื•ืœื•ืช ืœืคืงื— ื•ืœืœืžื“ ืื—ืช ืืช ื”ืฉื ื™ื™ื”.
03:22
For example, when our unsupervised learning program
59
202871
2958
ืœื“ื•ื’ืžื”, ื›ืฉืชื•ื›ื ื•ืช ื”ืœืžื™ื“ื” ื”ืœื ืžืคื•ืงื—ื•ืช ืฉืœื ื•
03:25
finds groups of patients that are similar,
60
205829
2334
ืžื•ืฆืื•ืช ืงื‘ื•ืฆื” ืฉืœ ืžื˜ื•ืคืœื™ื ื“ื•ืžื™ื,
03:28
it could send that data to a connected supervised learning program.
61
208163
3375
ื”ื™ื ื™ื›ื•ืœื” ืœืฉืœื•ื— ืืช ื”ืžื™ื“ืข ื”ื–ื” ืœืชื•ื›ื ืช ืœืžื™ื“ื” ืžืคื•ืงื—ืช ืฉืžื—ื•ื‘ืจืช ืืœื™ื”.
03:31
That program could then incorporate this information into its predictions.
62
211829
3500
ื”ืชื•ื›ื ื” ื™ื›ื•ืœื” ืœืฉืœื‘ ืืช ื”ืžื™ื“ืข ื‘ืชื—ื–ื™ื•ืช ืฉืœื”.
03:35
Or perhaps dozens of reinforcement learning programs
63
215871
2792
ืื• ืื•ืœื™ ืขืฉืจื•ืช ืชื•ื›ื ื•ืช ืœืžื™ื“ื” ืžื—ื–ืงืช
03:38
might simulate potential patient outcomes
64
218663
2291
ื™ื•ื›ืœื• ืœื“ืžื•ืช ืชื•ืฆืื•ืช ืคื•ื˜ื ืฆื™ืืœื™ื•ืช ืฉืœ ืžื˜ื•ืคืœื™ื
03:40
to collect feedback about different treatment plans.
65
220954
2750
ื›ื“ื™ ืœืืกื•ืฃ ืžืฉื•ื‘ ื‘ื ื•ื’ืข ืœืชื•ื›ื ื™ื•ืช ื˜ื™ืคื•ืœ ืฉื•ื ื•ืช.
03:43
There are numerous ways to create these machine-learning systems,
66
223704
3125
ื™ืฉ ืื™ืŸ ืกื•ืฃ ื“ืจื›ื™ื ืœื™ืฆื•ืจ ืืช ืžืขืจื›ื•ืช ืœืžื™ื“ืช ื”ืžื›ื•ื ื” ื”ืืœื•,
03:46
and perhaps the most promising models
67
226829
1834
ื•ืื•ืœื™ ื”ืžื•ื“ืœื™ื ื”ื›ื™ ืžื‘ื˜ื™ื—ื™ื
03:48
are those that mimic the relationship between neurons in the brain.
68
228663
3416
ื”ื ืืœื• ืฉืžื—ืงื™ื ืืช ื”ื™ื—ืกื™ื ื‘ื™ืŸ ื ื•ื™ืจื•ื ื™ื ื‘ืžื•ื—.
03:52
These artificial neural networks can use millions of connections
69
232079
3292
ืจืฉืชื•ืช ืขืฆื‘ื™ื•ืช ืžืœืื›ื•ืชื™ื•ืช ื™ื›ื•ืœื•ืช ืœื”ืฉืชืžืฉ ื‘ืžื™ืœื™ื•ื ื™ ื—ื™ื‘ื•ืจื™ื
03:55
to tackle difficult tasks like image recognition, speech recognition,
70
235371
4417
ื›ื“ื™ ืœื”ืชืžื•ื“ื“ ืขื ืžืฉื™ืžื•ืช ืžื•ืจื›ื‘ื•ืช ื›ืžื• ื–ื™ื”ื•ื™ ืชืžื•ื ื•ืช, ื–ื™ื”ื•ื™ ื“ื™ื‘ื•ืจ,
03:59
and even language translation.
71
239788
2041
ื•ืืคื™ืœื• ืชืจื’ื•ื ืฉืคื•ืช.
04:01
However, the more self-directed these models become,
72
241829
3292
ืขื ื–ืืช, ื›ื›ืœ ืฉื”ืžื•ื“ืœื™ื ื”ืืœื” ื”ื•ืคื›ื™ื ืœื”ื™ื•ืช ืžื•ื›ื•ื•ื ื™ื ืขืฆืžื™ืช,
04:05
the harder it is for computer scientists
73
245121
2125
ืงืฉื” ื™ื•ืชืจ ืœืžื“ืขื ื™ ืžื—ืฉื‘
04:07
to determine how these self-taught algorithms arrive at their solution.
74
247246
3833
ืœืงื‘ื•ืข ืื™ืš ืืœื’ื•ืจื™ืชืžื™ ื”ืœืžื™ื“ื” ื”ืขืฆืžื™ืช ื”ืืœื” ืžื’ื™ืขื™ื ืœืคืชืจื•ืŸ.
04:11
Researchers are already looking at ways to make machine learning more transparent.
75
251079
4459
ื—ื•ืงืจื™ื ื›ื‘ืจ ืžื—ืคืฉื™ื ื“ืจื›ื™ื ืœื”ืคื•ืš ืœืžื™ื“ืช ืžื›ื•ื ื” ืœืฉืงื•ืคื” ื™ื•ืชืจ.
04:15
But as AI becomes more involved in our everyday lives,
76
255538
2916
ืื‘ืœ ื›ืฉื‘โ€œืž ื”ื•ืคื›ืช ืœืžืขื•ืจื‘ืช ื™ื•ืชืจ ื‘ื—ื™ื™ื ื• ื”ื™ื•ื-ื™ื•ืžื™ื™ื,
04:18
these enigmatic decisions have increasingly large impacts
77
258454
2792
ืœื”ื—ืœื˜ื•ืช ื”ืื ื™ื’ืžื˜ื™ื•ืช ื”ืืœื• ื™ืฉ ื”ืฉืคืขื” ื’ื“ืœื”
04:21
on our work, health, and safety.
78
261246
2875
ืขืœ ื”ืขื‘ื•ื“ื”, ื”ื‘ืจื™ืื•ืช ื•ื”ื‘ื˜ื™ื—ื•ืช ืฉืœื ื•.
04:24
So as machines continue learning to investigate, negotiate and communicate,
79
264121
4958
ืื– ื›ืฉืžื›ื•ื ื•ืช ืžืžืฉื™ื›ื•ืช ืœืœืžื•ื“ ืื™ืš ืœื—ืงื•ืจ, ืœืขื‘ื“ ื•ืœืชืงืฉืจ,
04:29
we must also consider how to teach them to teach each other to operate ethically.
80
269079
5209
ืื ื—ื ื• ื—ื™ื™ื‘ื™ื ื’ื ืœื—ืฉื•ื‘ ืื™ืš ืœืœืžื“ ืื•ืชืŸ ืœืœืžื“ ืื—ืช ืืช ื”ืฉื ื™ื” ืœืคืขื•ืœ ื‘ืฆื•ืจื” ืืชื™ืช.
ืขืœ ืืชืจ ื–ื”

ืืชืจ ื–ื” ื™ืฆื™ื’ ื‘ืคื ื™ื›ื ืกืจื˜ื•ื ื™ YouTube ื”ืžื•ืขื™ืœื™ื ืœืœื™ืžื•ื“ ืื ื’ืœื™ืช. ืชื•ื›ืœื• ืœืจืื•ืช ืฉื™ืขื•ืจื™ ืื ื’ืœื™ืช ื”ืžื•ืขื‘ืจื™ื ืขืœ ื™ื“ื™ ืžื•ืจื™ื ืžื”ืฉื•ืจื” ื”ืจืืฉื•ื ื” ืžืจื—ื‘ื™ ื”ืขื•ืœื. ืœื—ืฅ ืคืขืžื™ื™ื ืขืœ ื”ื›ืชื•ื‘ื™ื•ืช ื‘ืื ื’ืœื™ืช ื”ืžื•ืฆื’ื•ืช ื‘ื›ืœ ื“ืฃ ื•ื™ื“ืื• ื›ื“ื™ ืœื”ืคืขื™ืœ ืืช ื”ืกืจื˜ื•ืŸ ืžืฉื. ื”ื›ืชื•ื‘ื™ื•ืช ื’ื•ืœืœื•ืช ื‘ืกื ื›ืจื•ืŸ ืขื ื”ืคืขืœืช ื”ื•ื•ื™ื“ืื•. ืื ื™ืฉ ืœืš ื”ืขืจื•ืช ืื• ื‘ืงืฉื•ืช, ืื ื ืฆื•ืจ ืื™ืชื ื• ืงืฉืจ ื‘ืืžืฆืขื•ืช ื˜ื•ืคืก ื™ืฆื™ืจืช ืงืฉืจ ื–ื”.

https://forms.gle/WvT1wiN1qDtmnspy7