AI Rewind: A Year of Amazing Machine Learning Papers

This video shows a curated list of the latest breakthroughs in AI by release date with a clear video explanation, link to a more in-depth article, and code. The videos include great commentary on the future of AI by many important people in the field such as Fei-Fei Li, Luis Lamb, Gary Marcus, and more.

Enjoy the video!

Chapters:

0:00 Hey! Tap the Thumbs Up button and Subscribe. You'll learn a lot of cool stuff in 2021, I promise.

0:28 2020, A year in review

9:06 Where do you want AI to go?

References:

• Complete article: https://github.com/louisfb01/Best_AI_paper_2020
• GitHub repo: https://github.com/louisfb01/Best_AI_paper_2020

Follow me for more AI content:

• Instagram: https://www.instagram.com/whats_ai/
• LinkedIn: https://www.linkedin.com/in/whats-ai/
• Twitter: https://twitter.com/Whats_AI
• Facebook: https://www.facebook.com/whats.artifi...
• Medium: https://medium.com/@whats_ai
• Youtube: https://www.youtube.com/channel/UCUzGQrN-lyyc0BWTYoJM_Sg/join

Video Transcript

00:00

where do you want ai to go what what

00:02

would make you happy

00:03

what is the objective here the objective

00:05

function for those of us who are

00:06

building ai

00:08

um or trying to inform it from other

00:10

fields

00:11

what would count in success where do you

00:12

want to take it to

00:17

[Music]

00:50

[Music]

00:57

[Music]

01:18

[Music]

01:43

[Music]

01:54

[Music]

02:39

[Music]

03:24

[Music]

03:30

[Music]

03:51

[Music]

04:16

[Music]

04:29

[Music]

04:50

but it's a trick no no it's much more

04:52

than that uh whosoever be he were they

04:54

shall have the power

04:57

whatever man it's a trick please be my

05:00

guest

05:02

come on really yeah oh this is gonna be

05:05

beautiful clint you've had a tough week

05:07

we won't hold it against you if you

05:08

can't get it up you may have seen this

05:10

before

05:10

right yeah all right

05:39

[Music]

06:24

[Music]

06:30

[Music]

06:51

[Music]

07:16

[Music]

07:27

[Music]

08:13

[Music]

08:17

researchers and artists around the world

08:19

have been using style game 2 to do all

08:20

kinds of interesting work

08:22

at nvidia we've been pushing the

08:23

frontier of using gans to synthesize

08:25

extremely realistic images

08:28

the catch is that you need to train them

08:30

on extremely large quantities of data

08:32

so to tackle this challenge we've

08:33

invented a new approach for stylegan 2

08:35

that we call adaptive discriminator

08:37

augmentation or ada

08:38

this lets us reduce the number of

08:40

training images by a factor of 10 a

08:42

factor of 20 or more while still getting

08:44

great results

08:56

[Music]

09:06

as a scientist i want to

09:10

push the scientific knowledge and

09:12

principles of

09:13

ai further and further and the first

09:15

thing is start with

09:17

some really fundamental laws and

09:21

principles of

09:22

ai i keep saying that i still feel our

09:24

ai

09:25

era is pre-newtonian physics that we are

09:28

still studying phenomenology

09:30

and engineering but there is going to be

09:34

a moment

09:34

or a set of moments where we're starting

09:37

to understand

09:38

the principles of intelligence and that

09:41

is

09:42

the scientists in me we're in the head

09:45

of a citizen i guess

09:47

and directing stanford's human center ai

09:51

institute

09:52

i want this to be a technology that can

09:55

an idealistic way really better human

09:58

conditions

09:58

right it's so profound it's so

10:02

horizontal

10:03

it's so um it has so much human and

10:07

societal impact

10:10

it can be very very bad and can be very

10:12

very good

10:13

so there i would like to see a

10:16

framework of this technology being

10:19

developed and deployed

10:21

in the most benevolent way

10:25

what comes to mind very often so i hang

10:27

out because of the family i'm born into

10:30

with people who work at state

10:32

departments who deal with international

10:34

laws military laws etc

10:36

and there's a there's a there's a big

10:38

concern here with militarization of ai

10:40

right

10:41

danny just mentioned human decision

10:43

making humans in the loop why people are

10:45

trying to negotiate

10:46

treaties limiting should we limit should

10:49

they always be a human

10:50

in the loop when we when we build

10:52

dangerous machines

10:54

you know kill a robot as well as is

10:56

happening as we all know

10:58

in the defense world and and as will

11:01

happen so what about the growing threats

11:03

of militarization of ai which is ongoing

11:06

and what should we

11:07

uh should we do anything about it or

11:10

pessimistic can we do anything about it

11:12

or is that

11:15

genie out of the box i want the costs

11:18

and the benefits of ai to be

11:21

evenly distributed across society both

11:24

in the united states

11:26

and globally and i want the public to

11:29

trust that that's what's being done

11:31

and i think that's impossible without

11:34

changes to law

11:36

just impossible um i like one of the

11:38

sayings of michael raving

11:40

from harvard and hebrew university said

11:42

that it's great that

11:44

uh computer science has not been around

11:46

for 2000 years

11:48

and we are at a stage where very very

11:50

important results

11:51

occur in front of our eyes i also like a

11:54

saying by

11:56

alan turing in his mind paper where he

11:58

said we can only see a short distance

12:01

ahead

12:02

but we can see plenty there that needs

12:04

to be done

12:05

and as fei-fei said as scientists i want

12:08

ai to advance but since ai is having an

12:11

impact

12:12

as big as physics had in the 20th

12:15

century

12:15

engineering heading and 20th century all

12:18

the ethical issues

12:19

all the biases and all the social

12:21

implications that margaret and others

12:23

and here in francesca

12:25

have been studying they are key as

12:28

responsibility of our

12:29

ourselves as scientists and in terms of

12:32

uh

12:32

technical technically speaking what we

12:35

are trying to do

12:36

is to conciliate traditions of ai so

12:39

that we cannot see

12:41

and as you said gary before you've been

12:43

saying since the

12:44

late 90s we have to converge we have to

12:47

look for convergency

12:49

you cited the only colonial cited other

12:51

prominent scientists today

12:53

we need a way of seeing that several

12:56

techniques can contribute to this

12:57

endeavor of making ai

12:59

fairer ai less bias and ai

13:03

to make something very positive for us

13:06

as humanity we need uh as scientists to

13:09

see

13:09

our fields in a very uh human humanistic

13:13

way

13:14

so that not only the technical stuff

13:16

advances but we also have to

13:18

be guided by serious and by effective

13:20

ethical principles

13:22

laws and norms as ryan said it's hard to

13:25

do to do that at the moment we are at

13:27

the beginning of uh

13:29

an ai cambrian explosion as several

13:32

people mention here but we need to be

13:34

very aware

13:35

of the social ethical and global

13:38

implications

13:39

that ai has these days we have to be

13:41

concerned about the north south

13:44

divide about the different cultures in

13:46

order to regulate it properly

13:49

we cannot see it only from a single

13:52

cultural perspective so

13:53

that's what i want to see ai researchers

13:56

doing to be they have to be concerned

13:58

about the technical results

14:00

the outstanding results ai has been

14:02

showing but also

14:03

we have to be to care about other people

14:06

about

14:06

other people's other countries and over

14:09

and over all

14:10

for the global health of the planet

14:12

thank you very much

14:13

and merry christmas happy new year to

14:16

everyone thank you gary and vincent for

14:18

the

14:18

brilliant debate you brought today and

14:20

for the scientists

14:22

i am reminded of the old african proverb

14:24

that i'm sure you all know which is it

14:25

takes a village to raise a child

14:27

clearly it will take a village as we've

14:29

seen today

14:30

to raise an ai that is ethical robust

14:32

and trustworthy