My Life Has Been Lucky! with Shun-Ichi Amari

5:18 Shun-Ichi Amari: Okay, thank you for coming everybody.

5:22 It's my great honor and pleasure to be here in San Diego,

5:29 and I will deliver a short lecture looking back on my life.

5:40 How I studied and how [inaudible].

5:47 I never dreamed of getting the Kyoto Prize,

5:52 it was a big surprise, the package on my door.

5:55 My life has been lucky.

5:59 It's very lucky.

6:03 I loved mathematics in my elementary and high school days.

6:12 One of the things I remember,

6:15 is the high school teacher took us to the University of Kyoto mathematics,

6:25 mathematics department.

6:26 And they had on the board some interesting problems.

6:33 And if we can solve one of them, then they gave one piece of a pencil.

6:41 And I tried many solutions.

6:45 But one thing I couldn't solve is this problem.

6:54 So X power equal X power Y equals Y power X.

7:02 X is two Y equal four is the answer.

7:08 And the question [inaudible] all the rational numbers [inaudible].

7:17 I tried.

7:20 I couldn't so I go back home.

7:24 After dinner I also think of [inaudible] maybe [inaudible].

7:32 I got idea and next day I [inaudible] University of Tokyo.

7:39 So I solved it.

7:41 One pencil.

7:44 However, when I prepared this slide.

7:50 I tried to solve this problem.

7:53 I couldn't.

7:55 So, maybe I'm not better in my high

8:00 school [inaudible] and I entered the University of Tokyo.

8:09 I struggled to become some researcher in science or engineering.

8:18 However, there is a lot of peace movement

8:22 or movement against government or against America imperialism.

8:29 And I was involved in that movement.

8:34 What does [inaudible].

8:35 After one and a half years they

8:40 decide which course you choose the higher education.

8:47 And say mathematics department, physics department,

8:53 electrical engineering department, those require very high score.

9:00 But my skill was low, unfortunately.

9:06 So I thought at that time, there was some explanation that some researcher some

9:16 professors created a new course named mathematical Engineer.

9:23 Okay?

9:24 The friends of those professors, They worked in aeronautics in the world wide

9:35 war second then Japan was defeated that, you know,

9:45 and they prohibit research on aeronautics.

9:50 It might be dangerous.

9:53 And so those professors changed the name into applied mathematics.

10:01 But after the [inaudible], it end the prohibition,

10:08 they come back created new aeronautics department,

10:14 [inaudible] and joyfully, I went back to those new department.

10:21 But some strange professor, so no more aeronautics.

10:27 Okay, it's too narrow.

10:29 We should apply much wider and apply modern mathematics.

10:36 The main engineering or some other problems.

10:41 So they created mathematic engineering course, very small course.

10:47 Professor are only four, but the students are five.

10:52 Okay.

10:53 And so, and at that time, they are yeah,

11:00 they themselves are just struggling to found a new mathematical engineer.

11:09 Okay?

11:09 So they did very hard work.

11:12 They worked hard.

11:14 So this is maybe the beginning of my very lucky life, okay?

11:22 And they say, Professor, you are free.

11:28 You can do anything you want, okay?

11:31 For mathematical, science and engineering.

11:34 But you should open your eyes very wide.

11:38 And to think about any phenomena or problems to [inaudible], if you like it,

11:47 you should do it deeply, but never on the inside,

11:54 your eye should be much larger.

12:00 So, I mentioned the mathematical engineering.

12:06 Then mathematical engineering.

12:10 It sounded not so strange.

12:13 But at that time, it was very

12:17 strange and mathematical science still, not a popular.

12:21 Mathematics, very popular.

12:24 In Japan, pure mathematics is a top in the world.

12:31 They say application of mathematics.

12:35 That nonsense.

12:37 Okay?

12:37 Those two dropped out, [inaudible] application of mathematics.

12:43 Mathematician should be purer and purer.

12:50 But, as I mentioned, it is beginning of my luck.

13:01 So my graduate study is say very different subject.

13:08 In master course, I try to analyze electrical networks.

13:15 It has some combinatorial structure, and I used modern homology theory to solve

13:26 those problems in terms of algebraic and topology.

13:32 And then I try to differential geometry of material.

13:40 Very different field.

13:42 The material, there's a lot of fine structure.

13:48 They call it dislocation and fracture material happened by motion dislocation.

13:57 And it is very good.

13:59 Described by Riemannian Geometry, it is one tablet my professor created,

14:10 but at the same time in the UK or in Germany,

14:15 similar theory appear and they discussed a lot of things.

14:23 Finally, I have interesting something [inaudible] theory.

14:30 I'm very impressed by reading [inaudible] theory book.

14:37 And I tried to make some theory of single

14:42 transformation or singular transformation in terms of geometry.

14:48 This is my Ph.D.

14:51 study.

14:52 Okay.

14:53 I was very lucky because at that time,

15:00 in Japan, they say, and technology engineering, science, very important.

15:08 Maybe the influence [inaudible] Russia unique then the government

15:17 decided to enlarge those science and engineering department.

15:25 Implies they created a lot of new professor position.

15:32 So there are a lot of [inaudible].

15:37 After the graduation I [inaudible],

15:41 they become the associate professor at the Kyushu University.

15:46 Okay.

15:47 And to my surprise and to my luck.

15:53 And in Kyushu University, I studied what subjects interesting.

16:01 At that time, Perceptron is invented by Rosenblatt, an American psychologist.

16:10 He proposed some learning machine.

16:14 Machine should learn as our brain.

16:18 Okay?

16:19 And then we can create some artificial intelligence things.

16:28 Now there's another school of artificial intelligence.

16:33 But the Rosenblatt, so that certain [inaudible] neural networks.

16:40 And to train that one,

16:43 maybe the universal artificial intelligence machine can be realized.

16:51 But at that time, it's impossible.

16:56 The computer facility computability is so poor.

17:03 They cannot handle the problem that you're interested in.

17:08 At that time, however, the theory of learning machine is very interesting.

17:19 And at that time, only the last day neuron we can learn.

17:28 We can train only last neuron and intermediate neurons,

17:33 if we could train it too, and maybe it much higher bit.

17:40 But at that time, it's impossible.

17:42 Why?

17:43 The reason is they used 01 combinatorial neuron.

17:50 They called McCulloch-Pitts neurons.

17:54 However, if we use a narrow type neuron, now they use a narrow type neuron.

18:01 Then, if we change the connection weights everywhere slightly,

18:08 we can find the effects.

18:11 It gives to the last neurons of the neurons.

18:16 That, in mathematics, it's a differentiation.

18:21 If there are many snaps, it's gradient.

18:26 I proposed gradient [inaudible] of the train,

18:33 certain neural network, it's so simple.

18:37 It's so obvious.

18:39 But at that time, no one said so.

18:42 I tried, and I did computer simulation.

18:46 It worked well.

18:48 To my surprise, after 20 years later, some American professor,

18:55 including Hinton rediscovered this type of learning method.

19:01 But at the same time, I am interested in why our brain works so well.

19:16 It's like neural networks here.

19:19 But it works really well.

19:23 I just studied something physiology, neuroscience, book, very difficult.

19:31 But reading, they describe details of neurons, synapses, and [inaudible].

19:39 But still, how we think the unknown.

19:43 Even now, it's unknown.

19:46 We have mind, we have consciousness.

19:50 We can do mathematics in the brain.

19:53 Brain surely can do that, but it's a very different thing.

20:00 I think maybe we use some mathematical model,

20:08 simplified model of neural networks.

20:12 But analyze deeply those simple model.

20:16 It's simple.

20:18 We can do analyze and just study,

20:22 what is the ability and disability or limitation of this simple model,

20:29 then we can go one step further to understand what their brain works.

20:37 It's mathematical neuroscience.

20:41 Some physiologist, my friend, said, Amari, you are doing the theory in the head.

20:54 We are doing theory experiment on the half using monkey, rats, all those things.

21:04 You are the heavenly, academia.

21:09 Yes.

21:10 I think it's okay.

21:12 But still now, mathematical neuroscience is going

21:17 on, and it's becoming more and more important.

21:27 Then, in 1975, I visited America.

21:37 Michel Arbib, he's here, invited me to join his laboratory.

21:44 Also, he has an interest in brain, mathematics, and combination on those things.

21:52 I think my English is very poor still now,

21:58 but maybe if I visit America, maybe my English becomes much better.

22:08 My daughter, one of them here, speak English very fluent.

22:16 I'm not.

22:18 But and nine months of the life in America,

22:26 it opened my eyes very widely to the international academia.

22:37 At the university, there are many visitors from all over the world.

22:42 They do speak, and students hear those things,

22:48 becoming they are very internationally educated.

22:57 Unfortunately, yes, in the airplane from Japan to America,

23:07 it's a Korean airline.

23:10 At that time, they request [inaudible] to put

23:14 the earphone and to watch the movie.

23:17 I tried, totally, not understanding.

23:23 At that time, I felt okay in the town flight.

23:29 I can understand very well that movie.

23:33 However, the entire flight, I paid $3.

23:39 But still, it was very different.

23:45 But one thing interesting, I traveled from Massachusetts to California by car,

23:55 taking three months, a long travel, looking very night, sightseeing places.

24:04 Now here is some picture, I was young,

24:11 obviously, and my daughter smallest [inaudible] here.

24:18 Then the 1985, [inaudible],

24:28 they say the neuron-like machine can do something very good,

24:35 internationally, a big write-up on neural network research.

24:43 But before that, it's called winter period

24:50 and those neural networks or artificial digits.

24:57 But in Japan, we don't have any budget in America, Europe.

25:06 The large budget or this subject because it looks useless.

25:15 In Japan, we have almost no budget for research.

25:20 For them, that [inaudible], we have freedom instead to do our research,

25:30 really and we tried to think

25:33 about mathematical neuroscience or machine learning,

25:38 and we collaborated with those neuroscientists.

25:44 They are working on the ground.

25:46 I'm working on the table.

25:48 Then, Japan became something advanced country in theory of neural networks.

26:03 I'm one of those who did many

26:06 of those stochastic gradient descent learning method,

26:13 I did associative module model for neuro

26:18 fluid dynamics and many of these things.

26:22 Then, suddenly, at that time, still, the Japan financial economy is going up,

26:34 but still university, it's difficult to go abroad.

26:40 Then I met Michel Arbib, and after that, he invited me to visit America.

26:53 Then there was a boom in 1985.

26:58 Suddenly, a telephone call came to me.

27:03 At that time, no direct telephone; the university took some place of telephone,

27:12 and they called me Professor Amari, here is the international call from America.

27:19 It's English.

27:21 But I couldn't do it.

27:25 [inaudible] connect.

27:28 It's the Neural [inaudible] Neural Network Conference in America.

27:37 It is in San Diego.

27:40 They say they'd like to invite me as a speaker.

27:49 All the expenses they pay.

27:54 I said, yes.

27:57 This is the beginning, my debut of international community.

28:06 Then Japan got better and better some difficult so quickly.

28:15 There are a lot of one to go abroad.

28:19 Then after that, I did reach, but some time after,

28:30 then I did research on the mathematical

28:39 neuroscience or machine learning about 10 years.

28:44 Then I felt I was an expert on this subject, doing many things.

28:56 It's maybe that that might be big mistake.

29:03 I should search for other interesting field to take.

29:11 I think something at the time,

29:17 the information scientists say homotopic science or [inaudible],

29:28 or those dealing with information, they use algebra,

29:34 the mathematical analysis, mathematical logic, and those things.

29:41 But we couldn't find any geometrical set.

29:47 Geometry is a wider perspective.

29:54 We need the science of information.

30:01 I just worked on the statistics.

30:09 Statistics, they use the not single probability,

30:16 but other candidates to estimate a family of probability distributions,

30:25 say the Gaussian distribution.

30:27 It has something meaningful.

30:30 In Gaussian case, say simple [inaudible] with two dimension.

30:38 What is the two-dimensional Gaussian distribution?

30:44 What its structure is?

30:48 We need some principle.

30:52 The idea if we are given a number of the observation from one distribution,

31:05 we have a graph form of that distribution.

31:11 Its histogram right this way.

31:16 We need a very smooth two-dimensional Gaussian distribution.

31:22 We need to project this observation

31:26 in the two-dimensional manifold of probability distribution.

31:34 Then, what is the prediction?

31:37 Prediction is just a straight run.

31:44 Such that it crosses at a right angle.

31:50 But what is a right angle?

31:53 We need something metric to define the right angle.

32:00 We need a fine connection to be what is a straight run.

32:09 Of course, it's a curved space to view the learn of differential geometry.

32:19 I tried those things very hard and I was very happy.

32:27 I met Sir David Cox in Japan and presented him my idea,

32:37 my design of the differential geometry or statistics.

32:44 Cox said almost nothing.

32:49 After returning to London, he wrote me,

32:56 your idea might be opening a new way to statistics.

33:06 We need to study, check those ideas internationally.

33:14 He organized a workshop on differential geometry or statistics in London.

33:23 I was invited.

33:26 I've met many famous statistics people.

33:31 One of them is C.R.

33:35 Rao.

33:36 He died two years ago, at the age of 103.

33:43 Cox died also, maybe nearly 100.

33:49 [inaudible] from America, and Efron from Stanford came.

33:57 So, Cox recommended me to many international conferences to be

34:06 a speaker so that the idea of neural and [inaudible] spreads widely.

34:15 This was very lucky.

34:18 At that time, I met Komarov in Russia.

34:28 Of course, he died now.

34:31 But those things very lucky.

34:36 I called my theory information geometry.

34:41 It was geometry of statistics.

34:46 Maybe statistics geometry might be better,

34:54 but I think, no, it's not limited to statistics.

35:00 It should be open to more wider subjects and now

35:07 information geometry is applied to many areas of physics,

35:14 mathematics, and so on.

35:16 It becomes wider and wider.

35:19 Of course, we use information geometry to analyze an intelligence

35:26 while the studying the ability of those deep neural networks.

35:35 But also, it's happy that and okay, and spring up people and visit me.

35:45 There's a Japan chapter of spring.

35:49 When they visit me, we are thinking about to publish a new journal,

35:58 Information Geometry.

36:00 How do you think about that?

36:02 I said, oh great idea.

36:07 However, it's not profitable.

36:09 There is already that.

36:11 At that time spilling out people, no no no we are looking for the 10,

36:22 20 years later, it might harm big audience, big leaders.

36:29 We are just setting such science.

36:33 It develops 20 years later.

36:36 We think Information Geometry might be such one.

36:42 I'm so glad.

36:45 It's now 80 years after, and Springer just united to his nature,

36:54 and they're such more profitable journals.

37:02 They say, there's a problem in continuing Information Geometry.

37:09 There are not so many leaders, they contributors.

37:14 That's the problem.

37:16 They are in fact doing very well.

37:25 I mentioned second neuro boom happened,

37:31 but still, unfortunately, the neural network technology.

37:37 We cannot use it.

37:39 The other time computer facility running, not so good.

37:46 Still, not so good.

37:48 It's much better than the other time, but still not good.

37:55 Now the thought was of course AI boom, neuro network boom is coming, okay?

38:07 I was astonished at seeing those new AI technology, say ChatGPT.

38:16 They do very well.

38:19 Gen AI and special.

38:22 I cannot use that word, but to see very helpful to doing something and writing

38:33 a letter of congratulations which I think I'll ask to write, okay?

38:40 Now, I draw the rough manuscript, asks ChatGPT to read make it much better.

38:54 They do very simple thinking.

38:58 I studied English for the 60 years or more, still very well.

39:05 But AI much, much better.

39:10 It's very surprising.

39:13 Much better.

39:16 Yes, at that time, retirement of University of Tokyo or the 60

39:26 then I should retire and just to search for some another job

39:33 and there the RIKEN that is Physical and Chemical Science Institute in Japan

39:42 and Masao Ito for the neuroscientist found that it's neuroscience inside RIKEN.

39:55 Then Ito invited me, Amari son RIKEN is very good.

40:01 Why are you not joining us?

40:05 I joined, it's a paradise.

40:09 Lot of research funds, and working much more science.

40:15 We don't need any big funds.

40:18 But we have a lot of the thing and I do research free.

40:26 Also artificial intelligence or special neuroscience

40:32 by using information in the geometry.

40:37 At the age of 80, I retire.

40:46 I think, okay, it's very free time and science

40:51 of the thing I continue as my hobby.

40:57 Now, I like to play Go game.

41:01 It's like chess, okay?

41:07 Then the COVID got here.

41:13 After retirement and I visit the RIKEN every day, but just a COVID came,

41:21 they say, don't come to RIKEN, said once a week is okay.

41:28 I'm still reaching RIKEN once a week to play Go game.

41:43 Now I'm astonished at the power of artificial intelligence and to think

41:51 about what is the difference from human and artificial intelligence.

42:00 Present AI is so wonderful and progress is so rapid.

42:07 Maybe next year, much better, next one, next much better.

42:14 I believe our brain is much more powerful.

42:21 Something wonderful.

42:25 Then, in this period as scientists what our society, our culture is to go.

42:39 There are lots of optimism and pessimisms.

42:45 Some would say, AI [inaudible], think paradise, we can live in paradise.

42:54 All productive works done by AI.

43:01 We don't do anything.

43:03 But we have sufficient basic income.

43:07 Not small income.

43:11 Sufficient in price every day.

43:15 Good wines.

43:17 Good beers and very delicious foods.

43:21 It's enough.

43:23 Do you think this is happy society?

43:29 I don't think so.

43:31 It's not paradise, but it domestication of humans by AI.

43:42 You see the pig or chicken or those animals.

43:50 They can eat freely.

43:52 Are they happy?

43:54 I don't know.

43:55 They don't know too.

43:57 But for human, such a life is not happy.

44:05 Human likes to work because our brain and human are the result of evolution.

44:20 Our evolution, human society.

44:28 Then, those by evolution, we reach to our mind,

44:37 consciousness, and logic, mathematics, science, and so on.

44:48 To work is by the [inaudible].

44:55 We need to work.

44:57 We need to communicate with our other humans.

45:02 It's society.

45:05 The feeling we do work and do something accomplishment,

45:12 feeling of accomplishment.

45:15 That are very joyful.

45:18 We did.

45:19 We did this, okay?

45:26 We need to work, even if there are no pay,

45:32 but the food and drinks from the other side free because we never stop working.

45:42 Work is sometimes painful.

45:48 But after accomplishing it, it's changed into very joyful structure.

45:55 Pain and joy just integrate it and we want to work by something to create,

46:05 to generate or just from curiosity.

46:12 Maybe, what is the work?

46:14 We need to create works.

46:18 That okay?

46:20 Amateur work is okay.

46:23 Amateur science very good.

46:26 Now they say the science compete to each other.

46:31 This discovery, I'm [inaudible] than you one day.

46:37 Then let's keep that it's progress secret.

46:43 That ridiculous.

46:45 We really discuss it.

46:48 There could be many simultaneous discoveries inwards overall.

46:55 Which is earlier, that no problem.

46:59 We enjoy together all our findings are very close, not great.

47:07 Then, that work is just to play.

47:13 Work and play united.

47:16 Also the painful work and the joyful work is united.

47:24 That is our society should go.

47:31 However, have to reach that paradise.

47:38 There are a lot crisis say military use of AI is now very fashionable.

47:52 It might go to the worldwide war.

47:58 Also, the AI is promoted by those big money companies.

48:11 They want to earn money.

48:13 They do proceed AI only for money.

48:18 That are very dangerous.

48:21 From that, disparity of people might increase, increase, increase.

48:33 Politically, there are a lot of populism.

48:39 Also, AI might help with the rights of populism

48:45 and we're thinking about just our society civilization,

48:53 the freedom, and equality and the dignity of humans,

49:03 those things we should keep and we should create not domestication,

49:12 but some new civilization to a our future society.

49:19 Of course, education is very important.

49:23 How to retrieve the AI.

49:28 AI's dangerous because and it's very useful.

49:33 We ask AI to do this work.

49:37 They did.

49:38 That implies our thinking is decreasing.

49:45 Joy of thinking or opinion think it's very important.

49:52 Sometimes we think about deeply, one day, one month, even 10 years.

50:00 Some time the very good idea came into our mind.

50:08 That's a very joyful thing.

50:14 I hope that you should create new society, new civilization.

50:24 Many of you, I'm not satisfied with the current

50:31 situation or civilization or a political situation.

50:37 But there are a lot of difficulty in conquering and create new civilization.

50:46 It's up to you.

50:49 It's no more me.

50:51 I'm just looking, and maybe I live only a few years.

50:59 But it's you who create a new [inaudible] civilization.

51:07 Yes, paradise of human beings.

51:12 Thank you very much.

51:21 Thank you.

51:32 Rose Yu: Thank you very much.

51:34 Amari Shun-Ichi for the very inspiring talk.

51:38 This is a time for question and answering.

51:42 We have students here.

51:44 Anybody have questions for Amari Shun-Ichi please line up behind the mic.

51:52 Speaker 1: Which open problem in information geometry or neural computation.

52:00 What do you consider the most important?

52:04 Shun-Ichi Amari: Yeah.

52:06 I think Meta AI or Large Language Model

52:12 develop so quickly and they believe bigger is better.

52:20 Those skill of neural networks becoming

52:24 larger and larger and basically, it worked.

52:32 But there should be some limits.

52:36 Now, I feel we don't have enough theory

52:42 to explain why the current neural networks or transformers,

52:49 all those things work so weird.

52:52 It's poverty of theory now.

52:57 We say big AI, they are created.

53:02 They need big grant.

53:04 But without certain large money, we can create theories.

53:11 Those thing is open to not America or China, but even Japan.

53:18 Japan is now poor.

53:21 Not poor as six years ago.

53:26 It is still a risk.

53:28 But we can do that.

53:31 So it's a mission of surgical passing to understand why it works so good.

53:41 They say something emerging.

53:46 But what emerging?

53:48 We don't know.

53:50 So maybe we can think about new theory,

53:56 which makes it possible to construct a smaller scale, very wide actions.

54:05 Speaker 1: [FOREIGN] Rose Yu: Let's take a question from my left hand side.

54:16 Speaker 2: Hello.

54:17 My name is Isabella.

54:19 I'm from Tijuana, Mexico.

54:21 My question is, with the evolution of AI, how is the new generation?

54:27 Can we still live freely as a society to continue to develop the use of AI

54:33 to do good and not create danger and live in this paradise that you talk about?

54:40 Shun-Ichi Amari: I do not understand.

54:43 Talk slowly.

54:44 Rose Yu: So would you mind the last part of the question be slow.

54:56 Speaker 2: I'm going to read.

54:58 Rose Yu: She wanted to know how we would live

55:01 with AI in modern days as they are becoming so powerful.

55:06 Then really, I guess, as a newer generation,

55:09 to really live in this paradise that you were describing before.

55:14 What is the optimal relationship that we should develop with AI?

55:19 Shun-Ichi Amari: That's a very difficult question.

55:23 Nobody knows the answer.

55:26 But we should find the way.

55:30 Now so those war are going on.

55:37 Also the disparity of people increasing.

55:42 They have to overcome this bad situation.

55:46 We don't know.

55:48 We know also and democracy is something in Christ.

55:55 Under the democracy, there are Satan, Hitler or those people might keep it.

56:04 So we simply think about to be from the very side.

56:11 Of course, from AI side and smarter

56:17 still small scale AI prevailing for other people.

56:26 Also, the decision between our neighbors.

56:32 It's really important under democracy to keep democracy.

56:38 Those things are all up to you.

56:41 Not me.

56:43 Yes.

56:44 Good.

56:45 Speaker 2: Thank you.

56:46 Speaker 3: Hi, my name is Erin Alape.

56:49 Throughout your presentation, you mentioned lots of changing situations

56:55 like the way you experience your education,

56:59 different cultures, different languages, and even the change in our critical

57:05 thinking skills that AI has inevitably brought.

57:09 So I wanted to ask, what could you share to us,

57:13 young folks about overcoming change in such a technological era?

57:21 Shun-Ichi Amari: So, we cannot stop development of technology.

57:26 It's so powerful and it's so convenient.

57:31 The modern AI so helpful.

57:36 We should use it.

57:38 But AI, it's not the AI use ourselves to find the answer.

57:49 The important thing we should use, thinking, considering, very carefully.

58:02 Don't rely on the answer of AI, there might be lots of hallucination.

58:09 So, I'm very afraid of the people, the child, can you use AI very conveniently.

58:25 So we should not lose our thinking ability,

58:33 joy of thinking, and all those thinking deeply, say one year.

58:43 Maybe suddenly a new idea come up with my mind.

58:50 That those new idea, AI cannot do.

58:55 So we are very careful not losing our working,

59:02 thinking, and communicating to each other.

59:07 Not communicate with AI, but communicate with the human.

59:14 That's a very important factor, I think.

59:17 Rose Yu: Thank you.

59:19 I can take a question from my right hand side.

59:21 Speaker 4: [FOREIGN] My question is,

59:34 you talked before about your work in geometrical information,

59:40 and I wanted to ask is this different information for conscious,

59:44 and if it is, how can we describe it mathematically, consciousness?

59:51 Shun-Ichi Amari: What is the last part of the question?

59:54 Information?

59:55 Rose Yu: Information of our consciousness.

59:58 Shun-Ichi Amari: Consciousness.

59:59 Rose Yu: Yes.

59:59 Speaker 4: Can it be described by mathematics?

1:00:02 Shun-Ichi Amari: Yes.

1:00:04 Consciousness is a big problem.

1:00:07 It emerged in our mind and brain by evolution.

1:00:14 I think the consciousness,

1:00:18 it's such a simple thing that I know what I want to do by myself.

1:00:27 That's important to communicate to each other.

1:00:32 When we work together,

1:00:35 we need to transmit some idea I'm intending to do to the other people.

1:00:45 So that being of the consciousness.

1:00:56 Recently, there are some theories of consciousness

1:01:01 in the say information and integrated theory or those things.

1:01:10 But still, insufficient.

1:01:13 Consciousness is part of our mind that evolved through long history.

1:01:22 So, simple thing or machine to have consciousness.

1:01:31 That is if machine or AI knows what they are doing now,

1:01:41 currently, maybe after following prompt, they are doing some work.

1:01:49 But if AI understand the true meaning

1:01:53 of prompt put and task which they do themselves,

1:02:01 they do [inaudible] some reason.

1:02:04 Currently the stochastic reading is logical.

1:02:10 But they should do more logical part, and now the logical part is going

1:02:18 on, but I have to combine nicely those stochastic

1:02:24 thinking and logical thinking and knowing when the intention

1:02:31 those thing is necessary for the future math, and it's a being of consciousness.

1:02:40 But consciousness is much more delicate things.

1:02:46 I believe if there is a machine, AI who has personality,

1:02:53 it won`t seem to sense of justice and so on.

1:03:07 It's very dangerous.

1:03:09 One can easily create machine like Trump.

1:03:14 Everything is there.

1:03:17 If we got money, that's okay.

1:03:20 I am great.

1:03:21 Or Hitler or Putin or there are many such people.

1:03:30 Now, if such AI emerge, it's so dangerous.

1:03:36 This is a problem of the personality or consciousness and the machine.

1:03:43 Speaker 4: Hi, [FOREIGN].

1:03:47 Shun-Ichi Amari: Hi.

1:03:47 Rose Yu: You have a question?

1:03:51 Speaker 5: Hello, Dr.

1:03:53 Amari.

1:03:53 It's a pleasure to meet you.

1:03:55 My name is Camila Ela.

1:03:57 I really admire your career and your work.

1:04:00 I would like to ask you,

1:04:02 do you think artificial intelligence will ever surpass human intelligence?

1:04:09 Rose Yu: Whether artificial intelligence

1:04:13 will eventually surpass human intelligence?

1:04:15 Shun-Ichi Amari: A very different question?

1:04:20 Human intelligent, say some scientist knew to Einstein.

1:04:29 Then our user did and also the current AI, they do interpret a prompt.

1:04:43 You're thinking about many examples.

1:04:46 They think this new situation should be set and sun.

1:04:53 However, human do not satisfy with those.

1:05:00 They got given problem they would answer.

1:05:05 It's okay, but it's not such.

1:05:11 From ancient to date, they know the movement of stars and they have some

1:05:20 law how stars going on and even all the time, they can do such interpolation.

1:05:30 However, they don't ask why.

1:05:37 But anyway, the Kepler find that this planet go

1:05:45 around the sun in something in like down more so.

1:05:54 It explained every situation very well.

1:06:01 But Newton did not satisfy with good answer.

1:06:07 Why?

1:06:08 Why is an answer about us while other planet go around.

1:06:16 And he invented something, the law of gravity,

1:06:22 the concept of force or concept of mass,

1:06:27 acceleration, something very new things and combine

1:06:32 those thing and create a new unified theory.

1:06:38 The current AI couldn't do that.

1:06:41 In future, is it possible we need some

1:06:47 motivation or curiosity to ask further and further.

1:06:52 And it's very difficult shall be the current AI to have that curiosity.

1:07:00 I cannot say anything about in future.

1:07:04 It might be possible.

1:07:07 But it takes some years, I guess.

1:07:14 You got it by evolution of many, many, many years.

1:07:21 Rose Yu: Great.

1:07:22 Thank you, Amari.

1:07:23 What's the question from my left hand side?

1:07:26 Speaker 6: Good afternoon.

1:07:27 My name is Al Renoso.

1:07:28 I'm from Lazaro Cardenas High School in Tijuana, California.

1:07:31 I wanted to ask you what advice would you give to young students who want

1:07:35 to contribute to future of AI and what

1:07:38 skills or fields should we focus on studying?

1:07:41 Rose Yu: So, he wanted to know, as for young people,

1:07:45 do you have advice about what skills

1:07:48 and subjects of learning that they should focus.

1:07:52 Shun-Ichi Amari: It's a very difficult question.

1:07:55 I say, don't stop using AI.

1:07:59 It's very convenient, too.

1:08:02 But remember you use the AI.

1:08:08 You are the master.

1:08:11 And you think as things don't AI know the answer which computer gives.

1:08:19 You should think about is it right or no?

1:08:24 So I'll advise you not take easygoing way.

1:08:33 Whatever the AI is useful, convenient.

1:08:38 Don't rely on that.

1:08:41 At the same time, you should consider, think about the whole story Rose Yu:

1:08:49 of the world and the problem you are facing.

1:08:53 We'll take one last question from this side.

1:08:56 Speaker 7: Hello.

1:08:57 My name is [inaudible], and I'm from Southwest High School.

1:09:00 I wanted to know your opinion about AI wasting natural resources for us.

1:09:06 Rose Yu: So she is pointing the fact that says modern

1:09:12 AI models use a lot of energy to train these models.

1:09:17 What is your opinion about that?

1:09:20 Shun-Ichi Amari: Now, the AI is very energy consuming.

1:09:25 It cannot continue further.

1:09:30 But maybe, I don't know 10 years or 20 years or something,

1:09:36 we invent new technology.

1:09:39 We need not so great energy.

1:09:45 A brain works they say only 20 watt power.

1:09:50 That is like the small tube.

1:09:53 So energy problem now the big obstacle making the AI and [inaudible].

1:10:04 But I think not near future, but in some future,

1:10:13 we can invent new technology, how to conquer that problem.

1:10:20 Rose Yu: Sounds good.

1:10:21 Let's hand again to your student last question

1:10:24 and especially Amari Sensei for the wonderful QA.

1:10:29 [APPLAUSE]