1
00:00:03,240 --> 00:00:07,170
Who wrote all this that I am going
to say in the next few minutes?
2
00:00:07,440 --> 00:00:08,800
Myself? A screenwriter?
3
00:00:09,100 --> 00:00:10,930
Or was it an
artificial intelligence?
4
00:00:11,080 --> 00:00:14,715
Only one thing is certain:
immediately after creating this video,
5
00:00:14,739 --> 00:00:17,590
what we explain here
could become obsolete.
6
00:00:17,830 --> 00:00:21,588
It is not at all easy for humans
to distinguish authorship.
7
00:00:21,613 --> 00:00:25,560
And that gives us a measure of
the progress of artificial intelligence.
8
00:00:25,585 --> 00:00:28,074
ChatGPT and similar tools
9
00:00:28,098 --> 00:00:30,535
have the ability to write
texts in their entirety,
10
00:00:30,559 --> 00:00:33,760
precisely and adapted
to the requested style.
11
00:00:33,820 --> 00:00:37,345
And other tools are capable
of creating images from scratch,
12
00:00:37,369 --> 00:00:39,970
respecting the orders
given that we give him.
13
00:00:39,995 --> 00:00:41,955
AI or artificial intelligence
14
00:00:41,980 --> 00:00:43,030
is not a brand new topic.
15
00:00:43,080 --> 00:00:46,850
Early tests of this rapidly
evolving technology
16
00:00:46,874 --> 00:00:50,300
were done as early as in 1943.
17
00:00:50,320 --> 00:00:54,454
But since 2022 it has
fully entered our lives,
18
00:00:54,479 --> 00:00:57,254
and of course, the
discussions with our friends.
19
00:00:57,279 --> 00:00:59,720
But how does all this work?
20
00:00:59,950 --> 00:01:02,210
And how much will
it change our lives?
21
00:01:10,001 --> 00:01:13,167
Artificial intelligence is
the capacity of machines.
22
00:01:13,191 --> 00:01:16,044
To perform tasks that normally
require human intelligence.
23
00:01:16,068 --> 00:01:21,275
These include learning, decision
making, and pattern recognition.
24
00:01:21,300 --> 00:01:23,960
To achieve this, artificial
intelligence developers
25
00:01:23,985 --> 00:01:26,082
use what is called
a neural network.
26
00:01:26,433 --> 00:01:29,910
A neural network can learn
to perform difficult tasks,
27
00:01:29,950 --> 00:01:33,297
like recognizing images
or translating languages.
28
00:01:33,321 --> 00:01:36,538
A neural network learns
through experience.
29
00:01:36,563 --> 00:01:39,977
This means that when it
is trained with a data set,
30
00:01:40,001 --> 00:01:43,840
that data set can adjust its
connections and parameters
31
00:01:44,020 --> 00:01:46,220
to improve the ability to
perform the desired job.
32
00:01:46,270 --> 00:01:50,211
In other words, you can create
and adapt your own algorithms,
33
00:01:50,235 --> 00:01:52,450
through machine learning.
34
00:01:52,660 --> 00:01:57,138
Basically, a neural network
consists of many "artificial neurons",
35
00:01:57,162 --> 00:01:59,860
and are organized in
interconnected layers.
36
00:01:59,920 --> 00:02:02,763
Each neuron receives
certain information.
37
00:02:02,787 --> 00:02:06,270
And performs a small
mathematical operation to process it.
38
00:02:06,350 --> 00:02:10,196
This information is then transmitted
to the neurons in the next layer,
39
00:02:10,221 --> 00:02:14,100
and so on until
reaching the final answer.
40
00:02:14,380 --> 00:02:18,340
That's right, data must be
provided to train the neural network.
41
00:02:18,390 --> 00:02:20,680
There are different types
of learning processes:
42
00:02:20,988 --> 00:02:24,649
1- Supervised learning:
43
00:02:24,650 --> 00:02:26,837
This type of
training is used when
44
00:02:26,862 --> 00:02:28,830
there is a data set that
has been labeled before,
45
00:02:28,855 --> 00:02:29,712
is to say,
46
00:02:29,737 --> 00:02:33,660
when the correct answer
is known for each input.
47
00:02:34,230 --> 00:02:38,100
The AI learns from
this labeled data,
48
00:02:38,180 --> 00:02:41,520
and then you can make
predictions for new data.
49
00:02:42,190 --> 00:02:44,610
2- Unsupervised learning:
50
00:02:45,190 --> 00:02:46,711
In this type of training,
51
00:02:46,736 --> 00:02:49,550
AI does not have
pre-labeled data.
52
00:02:49,630 --> 00:02:52,220
Instead, they are
presented with a set of data,
53
00:02:52,244 --> 00:02:56,660
and then are asked to find
patterns or structures within that data.
54
00:02:57,100 --> 00:03:00,020
This can be useful
for data analysis or
55
00:03:00,044 --> 00:03:02,060
to segment
customers, for example.
56
00:03:02,414 --> 00:03:05,240
3- Learning through
reinforcement:
57
00:03:05,520 --> 00:03:10,460
This type of training is used
to teach AI to make decisions,
58
00:03:10,590 --> 00:03:12,850
through reward and punishment.
59
00:03:12,875 --> 00:03:14,430
They are presented
with a scenario,
60
00:03:14,510 --> 00:03:16,540
and asked to perform an action.
61
00:03:16,800 --> 00:03:19,941
If the action is correct,
the AI receives a reward,
62
00:03:19,965 --> 00:03:23,110
but if the action is not correct,
it receives a punishment.
63
00:03:23,135 --> 00:03:24,165
Over time,
64
00:03:24,190 --> 00:03:28,860
the AI learns to make the right
decisions to maximize rewards.
65
00:03:28,885 --> 00:03:29,975
An example,
66
00:03:30,000 --> 00:03:33,550
if we want a computer to
learn to recognize faces,
67
00:03:33,680 --> 00:03:37,220
and give us information about
what appears in each of them
68
00:03:37,244 --> 00:03:39,610
we will provide it with
many labeled images.
69
00:03:39,800 --> 00:03:42,937
The computer observes
patterns in the data,
70
00:03:42,962 --> 00:03:46,830
and creates a model to
identify faces in new images.
71
00:03:47,130 --> 00:03:50,059
And why is all
this so significant?
72
00:03:50,083 --> 00:03:54,147
Mainly because it is a technology that
can be used in many areas of our lives.
73
00:03:54,172 --> 00:03:58,459
In healthcare, artificial intelligence
systems diagnose diseases
74
00:03:58,483 --> 00:04:01,940
and can help doctors
predict treatment outcomes.
75
00:04:02,080 --> 00:04:06,193
It is also capable of automating
most of the work done by computer,
76
00:04:06,218 --> 00:04:07,700
with increasingly
better results.
77
00:04:07,790 --> 00:04:10,646
Design work,
writing, forecasts...
78
00:04:10,670 --> 00:04:15,242
It is developing rapidly and has
increasingly impressive capabilities.
79
00:04:15,266 --> 00:04:17,690
But be careful, there
are also critical voices.
80
00:04:17,715 --> 00:04:18,667
And thank goodness:
81
00:04:18,691 --> 00:04:22,052
because we have already been warned
about the dangers of this technology.
82
00:04:22,076 --> 00:04:25,867
On the one hand, as
artificial intelligence advances,
83
00:04:25,891 --> 00:04:27,850
new ethical dilemmas arise.
84
00:04:27,875 --> 00:04:28,775
For example,
85
00:04:28,800 --> 00:04:32,710
Who is responsible if an autonomous
vehicle causes an accident?
86
00:04:32,730 --> 00:04:35,698
How can we ensure that
artificial intelligence systems
87
00:04:35,722 --> 00:04:38,170
do not discriminate against
certain groups of people?
88
00:04:38,236 --> 00:04:40,998
Is it ethical to make decisions
that can harm humans
89
00:04:41,022 --> 00:04:43,497
creating artificial intelligences
that will take control?
90
00:04:43,535 --> 00:04:45,050
Who owns the technology?
91
00:04:45,160 --> 00:04:47,586
Who audits or
controls the decisions?
92
00:04:47,610 --> 00:04:50,296
These are some of
the ethical dilemmas
93
00:04:50,320 --> 00:04:53,445
that need to be addressed
as technology advances.
94
00:04:53,470 --> 00:04:54,505
And what about languages?
95
00:04:54,530 --> 00:04:57,587
How does this look from
minority language communities?
96
00:04:57,626 --> 00:04:58,955
There is no round
answer about the
97
00:04:58,980 --> 00:05:01,710
minority languages
and artificial intelligence:
98
00:05:01,980 --> 00:05:04,346
For example, automatic
translation systems
99
00:05:04,370 --> 00:05:08,472
can facilitate communication between
speakers of different languages,
100
00:05:08,498 --> 00:05:11,312
and that could
have a positive effect
101
00:05:11,336 --> 00:05:13,510
in the care and promotion
of minority languages.
102
00:05:13,560 --> 00:05:15,648
It is possible, thanks
to these technologies,
103
00:05:15,673 --> 00:05:18,265
to be able to receive any
content in our language,
104
00:05:18,289 --> 00:05:21,470
without the need for anyone's
translation or dubbing work.
105
00:05:21,630 --> 00:05:23,190
But all that
glitters is not gold.
106
00:05:23,430 --> 00:05:26,156
Artificial intelligence
systems are biased and
107
00:05:26,180 --> 00:05:30,960
There is also the risk of
reproducing linguistic discrimination.
108
00:05:30,985 --> 00:05:34,853
It is likely that there is not enough
information on minority languages,
109
00:05:34,878 --> 00:05:37,928
and therefore they make more
mistakes than in other languages,
110
00:05:37,952 --> 00:05:39,829
or have lack of details,
111
00:05:39,853 --> 00:05:43,618
and can't distinguish details
about smaller linguistic communities.
112
00:05:43,642 --> 00:05:46,280
There are around 7000
languages in the world.
113
00:05:46,430 --> 00:05:48,097
How many of them will have
114
00:05:48,121 --> 00:05:50,880
the opportunity to use the
capabilities of artificial intelligence?
115
00:05:51,070 --> 00:05:53,110
For example, the models used
116
00:05:53,134 --> 00:05:55,844
are intended for
data-heavy languages
117
00:05:55,868 --> 00:05:57,654
those linguistic communities
118
00:05:57,678 --> 00:05:59,816
that tend to have more
technological resources.
119
00:05:59,840 --> 00:06:01,870
That's why it's important,
120
00:06:01,890 --> 00:06:04,451
in addition to having its
own data available and free,
121
00:06:04,475 --> 00:06:07,615
to investigate models
that require fewer resources
122
00:06:07,639 --> 00:06:11,350
or collaborate with other linguistic
communities with similar characteristics.
123
00:06:11,400 --> 00:06:13,210
Scared, delighted?
124
00:06:13,250 --> 00:06:14,699
What do you think?
125
00:06:14,723 --> 00:06:17,850
Everything goes very fast
and it is not easy to keep up.
126
00:06:17,874 --> 00:06:20,750
But don't forget,
information is power.
127
00:06:20,930 --> 00:06:24,900
We will try from here to do our
bit so that everyone can access it.
128
00:06:25,220 --> 00:06:27,926
Understanding
gives us sovereignty.