TED2003
Marvin Minsky: Health and the human mind
Filmed:
Readability: 3.5
606,909 views
Listen closely -- Marvin Minsky's arch, eclectic, charmingly offhand talk on health, overpopulation and the human mind is packed with subtlety: wit, wisdom and just an ounce of wily, is-he-joking? advice.
Marvin Minsky - AI pioneer
Marvin Minsky is one of the great pioneers of artificial intelligence -- and using computing metaphors to understand the human mind. His contributions to mathematics, robotics and computational linguistics are legendary and far-reaching. Full bio
Marvin Minsky is one of the great pioneers of artificial intelligence -- and using computing metaphors to understand the human mind. His contributions to mathematics, robotics and computational linguistics are legendary and far-reaching. Full bio
Double-click the English transcript below to play the video.
00:18
If you ask people about what part of psychology do they think is hard,
0
0
6000
00:24
and you say, "Well, what about thinking and emotions?"
1
6000
3000
00:27
Most people will say, "Emotions are terribly hard.
2
9000
3000
00:30
They're incredibly complex. They can't -- I have no idea of how they work.
3
12000
6000
00:36
But thinking is really very straightforward:
4
18000
2000
00:38
it's just sort of some kind of logical reasoning, or something.
5
20000
4000
00:42
But that's not the hard part."
6
24000
3000
00:45
So here's a list of problems that come up.
7
27000
2000
00:47
One nice problem is, what do we do about health?
8
29000
3000
00:50
The other day, I was reading something, and the person said
9
32000
4000
00:54
probably the largest single cause of disease is handshaking in the West.
10
36000
6000
01:00
And there was a little study about people who don't handshake,
11
42000
4000
01:04
and comparing them with ones who do handshake.
12
46000
3000
01:07
And I haven't the foggiest idea of where you find the ones that don't handshake,
13
49000
5000
01:12
because they must be hiding.
14
54000
3000
01:15
And the people who avoid that
15
57000
4000
01:19
have 30 percent less infectious disease or something.
16
61000
4000
01:23
Or maybe it was 31 and a quarter percent.
17
65000
3000
01:26
So if you really want to solve the problem of epidemics and so forth,
18
68000
4000
01:30
let's start with that. And since I got that idea,
19
72000
4000
01:34
I've had to shake hundreds of hands.
20
76000
4000
01:38
And I think the only way to avoid it
21
80000
5000
01:43
is to have some horrible visible disease,
22
85000
2000
01:45
and then you don't have to explain.
23
87000
3000
01:48
Education: how do we improve education?
24
90000
4000
01:52
Well, the single best way is to get them to understand
25
94000
4000
01:56
that what they're being told is a whole lot of nonsense.
26
98000
3000
01:59
And then, of course, you have to do something
27
101000
2000
02:01
about how to moderate that, so that anybody can -- so they'll listen to you.
28
103000
5000
02:06
Pollution, energy shortage, environmental diversity, poverty.
29
108000
4000
02:10
How do we make stable societies? Longevity.
30
112000
4000
02:14
Okay, there're lots of problems to worry about.
31
116000
3000
02:17
Anyway, the question I think people should talk about --
32
119000
2000
02:19
and it's absolutely taboo -- is, how many people should there be?
33
121000
5000
02:24
And I think it should be about 100 million or maybe 500 million.
34
126000
7000
02:31
And then notice that a great many of these problems disappear.
35
133000
5000
02:36
If you had 100 million people
36
138000
2000
02:38
properly spread out, then if there's some garbage,
37
140000
6000
02:44
you throw it away, preferably where you can't see it, and it will rot.
38
146000
7000
02:51
Or you throw it into the ocean and some fish will benefit from it.
39
153000
5000
02:56
The problem is, how many people should there be?
40
158000
2000
02:58
And it's a sort of choice we have to make.
41
160000
3000
03:01
Most people are about 60 inches high or more,
42
163000
3000
03:04
and there's these cube laws. So if you make them this big,
43
166000
4000
03:08
by using nanotechnology, I suppose --
44
170000
3000
03:11
(Laughter)
45
173000
1000
03:12
-- then you could have a thousand times as many.
46
174000
2000
03:14
That would solve the problem, but I don't see anybody
47
176000
2000
03:16
doing any research on making people smaller.
48
178000
3000
03:19
Now, it's nice to reduce the population, but a lot of people want to have children.
49
181000
5000
03:24
And there's one solution that's probably only a few years off.
50
186000
3000
03:27
You know you have 46 chromosomes. If you're lucky, you've got 23
51
189000
5000
03:32
from each parent. Sometimes you get an extra one or drop one out,
52
194000
6000
03:38
but -- so you can skip the grandparent and great-grandparent stage
53
200000
4000
03:42
and go right to the great-great-grandparent. And you have 46 people
54
204000
5000
03:47
and you give them a scanner, or whatever you need,
55
209000
3000
03:50
and they look at their chromosomes and each of them says
56
212000
4000
03:54
which one he likes best, or she -- no reason to have just two sexes
57
216000
5000
03:59
any more, even. So each child has 46 parents,
58
221000
5000
04:04
and I suppose you could let each group of 46 parents have 15 children.
59
226000
6000
04:10
Wouldn't that be enough? And then the children
60
232000
2000
04:12
would get plenty of support, and nurturing, and mentoring,
61
234000
4000
04:16
and the world population would decline very rapidly
62
238000
2000
04:18
and everybody would be totally happy.
63
240000
3000
04:21
Timesharing is a little further off in the future.
64
243000
3000
04:24
And there's this great novel that Arthur Clarke wrote twice,
65
246000
3000
04:27
called "Against the Fall of Night" and "The City and the Stars."
66
249000
4000
04:31
They're both wonderful and largely the same,
67
253000
3000
04:34
except that computers happened in between.
68
256000
2000
04:36
And Arthur was looking at this old book, and he said, "Well, that was wrong.
69
258000
5000
04:41
The future must have some computers."
70
263000
2000
04:43
So in the second version of it, there are 100 billion
71
265000
5000
04:48
or 1,000 billion people on Earth, but they're all stored on hard disks or floppies,
72
270000
8000
04:56
or whatever they have in the future.
73
278000
2000
04:58
And you let a few million of them out at a time.
74
280000
4000
05:02
A person comes out, they live for a thousand years
75
284000
4000
05:06
doing whatever they do, and then, when it's time to go back
76
288000
6000
05:12
for a billion years -- or a million, I forget, the numbers don't matter --
77
294000
4000
05:16
but there really aren't very many people on Earth at a time.
78
298000
4000
05:20
And you get to think about yourself and your memories,
79
302000
2000
05:22
and before you go back into suspension, you edit your memories
80
304000
5000
05:27
and you change your personality and so forth.
81
309000
3000
05:30
The plot of the book is that there's not enough diversity,
82
312000
6000
05:36
so that the people who designed the city
83
318000
3000
05:39
make sure that every now and then an entirely new person is created.
84
321000
4000
05:43
And in the novel, a particular one named Alvin is created. And he says,
85
325000
6000
05:49
maybe this isn't the best way, and wrecks the whole system.
86
331000
4000
05:53
I don't think the solutions that I proposed
87
335000
2000
05:55
are good enough or smart enough.
88
337000
3000
05:58
I think the big problem is that we're not smart enough
89
340000
4000
06:02
to understand which of the problems we're facing are good enough.
90
344000
4000
06:06
Therefore, we have to build super intelligent machines like HAL.
91
348000
4000
06:10
As you remember, at some point in the book for "2001,"
92
352000
5000
06:15
HAL realizes that the universe is too big, and grand, and profound
93
357000
5000
06:20
for those really stupid astronauts. If you contrast HAL's behavior
94
362000
4000
06:24
with the triviality of the people on the spaceship,
95
366000
4000
06:28
you can see what's written between the lines.
96
370000
3000
06:31
Well, what are we going to do about that? We could get smarter.
97
373000
3000
06:34
I think that we're pretty smart, as compared to chimpanzees,
98
376000
5000
06:39
but we're not smart enough to deal with the colossal problems that we face,
99
381000
6000
06:45
either in abstract mathematics
100
387000
2000
06:47
or in figuring out economies, or balancing the world around.
101
389000
5000
06:52
So one thing we can do is live longer.
102
394000
3000
06:55
And nobody knows how hard that is,
103
397000
2000
06:57
but we'll probably find out in a few years.
104
399000
3000
07:00
You see, there's two forks in the road. We know that people live
105
402000
3000
07:03
twice as long as chimpanzees almost,
106
405000
4000
07:07
and nobody lives more than 120 years,
107
409000
4000
07:11
for reasons that aren't very well understood.
108
413000
3000
07:14
But lots of people now live to 90 or 100,
109
416000
3000
07:17
unless they shake hands too much or something like that.
110
419000
4000
07:21
And so maybe if we lived 200 years, we could accumulate enough skills
111
423000
5000
07:26
and knowledge to solve some problems.
112
428000
5000
07:31
So that's one way of going about it.
113
433000
2000
07:33
And as I said, we don't know how hard that is. It might be --
114
435000
3000
07:36
after all, most other mammals live half as long as the chimpanzee,
115
438000
6000
07:42
so we're sort of three and a half or four times, have four times
116
444000
3000
07:45
the longevity of most mammals. And in the case of the primates,
117
447000
6000
07:51
we have almost the same genes. We only differ from chimpanzees,
118
453000
4000
07:55
in the present state of knowledge, which is absolute hogwash,
119
457000
6000
08:01
maybe by just a few hundred genes.
120
463000
2000
08:03
What I think is that the gene counters don't know what they're doing yet.
121
465000
3000
08:06
And whatever you do, don't read anything about genetics
122
468000
3000
08:09
that's published within your lifetime, or something.
123
471000
3000
08:12
(Laughter)
124
474000
3000
08:15
The stuff has a very short half-life, same with brain science.
125
477000
4000
08:19
And so it might be that if we just fix four or five genes,
126
481000
6000
08:25
we can live 200 years.
127
487000
2000
08:27
Or it might be that it's just 30 or 40,
128
489000
3000
08:30
and I doubt that it's several hundred.
129
492000
2000
08:32
So this is something that people will be discussing
130
494000
4000
08:36
and lots of ethicists -- you know, an ethicist is somebody
131
498000
3000
08:39
who sees something wrong with whatever you have in mind.
132
501000
3000
08:42
(Laughter)
133
504000
3000
08:45
And it's very hard to find an ethicist who considers any change
134
507000
4000
08:49
worth making, because he says, what about the consequences?
135
511000
4000
08:53
And, of course, we're not responsible for the consequences
136
515000
3000
08:56
of what we're doing now, are we? Like all this complaint about clones.
137
518000
6000
09:02
And yet two random people will mate and have this child,
138
524000
3000
09:05
and both of them have some pretty rotten genes,
139
527000
4000
09:09
and the child is likely to come out to be average.
140
531000
4000
09:13
Which, by chimpanzee standards, is very good indeed.
141
535000
6000
09:19
If we do have longevity, then we'll have to face the population growth
142
541000
3000
09:22
problem anyway. Because if people live 200 or 1,000 years,
143
544000
4000
09:26
then we can't let them have a child more than about once every 200 or 1,000 years.
144
548000
6000
09:32
And so there won't be any workforce.
145
554000
3000
09:35
And one of the things Laurie Garrett pointed out, and others have,
146
557000
4000
09:39
is that a society that doesn't have people
147
561000
5000
09:44
of working age is in real trouble. And things are going to get worse,
148
566000
3000
09:47
because there's nobody to educate the children or to feed the old.
149
569000
6000
09:53
And when I'm talking about a long lifetime, of course,
150
575000
2000
09:55
I don't want somebody who's 200 years old to be like our image
151
577000
6000
10:01
of what a 200-year-old is -- which is dead, actually.
152
583000
4000
10:05
You know, there's about 400 different parts of the brain
153
587000
2000
10:07
which seem to have different functions.
154
589000
2000
10:09
Nobody knows how most of them work in detail,
155
591000
3000
10:12
but we do know that there're lots of different things in there.
156
594000
4000
10:16
And they don't always work together. I like Freud's theory
157
598000
2000
10:18
that most of them are cancelling each other out.
158
600000
4000
10:22
And so if you think of yourself as a sort of city
159
604000
4000
10:26
with a hundred resources, then, when you're afraid, for example,
160
608000
6000
10:32
you may discard your long-range goals, but you may think deeply
161
614000
4000
10:36
and focus on exactly how to achieve that particular goal.
162
618000
4000
10:40
You throw everything else away. You become a monomaniac --
163
622000
3000
10:43
all you care about is not stepping out on that platform.
164
625000
4000
10:47
And when you're hungry, food becomes more attractive, and so forth.
165
629000
4000
10:51
So I see emotions as highly evolved subsets of your capability.
166
633000
6000
10:57
Emotion is not something added to thought. An emotional state
167
639000
4000
11:01
is what you get when you remove 100 or 200
168
643000
4000
11:05
of your normally available resources.
169
647000
3000
11:08
So thinking of emotions as the opposite of -- as something
170
650000
3000
11:11
less than thinking is immensely productive. And I hope,
171
653000
4000
11:15
in the next few years, to show that this will lead to smart machines.
172
657000
4000
11:19
And I guess I better skip all the rest of this, which are some details
173
661000
3000
11:22
on how we might make those smart machines and --
174
664000
5000
11:27
(Laughter)
175
669000
5000
11:32
-- and the main idea is in fact that the core of a really smart machine
176
674000
5000
11:37
is one that recognizes that a certain kind of problem is facing you.
177
679000
5000
11:42
This is a problem of such and such a type,
178
684000
3000
11:45
and therefore there's a certain way or ways of thinking
179
687000
5000
11:50
that are good for that problem.
180
692000
2000
11:52
So I think the future, main problem of psychology is to classify
181
694000
4000
11:56
types of predicaments, types of situations, types of obstacles
182
698000
4000
12:00
and also to classify available and possible ways to think and pair them up.
183
702000
6000
12:06
So you see, it's almost like a Pavlovian --
184
708000
3000
12:09
we lost the first hundred years of psychology
185
711000
2000
12:11
by really trivial theories, where you say,
186
713000
3000
12:14
how do people learn how to react to a situation? What I'm saying is,
187
716000
6000
12:20
after we go through a lot of levels, including designing
188
722000
5000
12:25
a huge, messy system with thousands of ports,
189
727000
3000
12:28
we'll end up again with the central problem of psychology.
190
730000
4000
12:32
Saying, not what are the situations,
191
734000
3000
12:35
but what are the kinds of problems
192
737000
2000
12:37
and what are the kinds of strategies, how do you learn them,
193
739000
3000
12:40
how do you connect them up, how does a really creative person
194
742000
3000
12:43
invent a new way of thinking out of the available resources and so forth.
195
745000
5000
12:48
So, I think in the next 20 years,
196
750000
2000
12:50
if we can get rid of all of the traditional approaches to artificial intelligence,
197
752000
5000
12:55
like neural nets and genetic algorithms
198
757000
2000
12:57
and rule-based systems, and just turn our sights a little bit higher to say,
199
759000
6000
13:03
can we make a system that can use all those things
200
765000
2000
13:05
for the right kind of problem? Some problems are good for neural nets;
201
767000
4000
13:09
we know that others, neural nets are hopeless on them.
202
771000
3000
13:12
Genetic algorithms are great for certain things;
203
774000
3000
13:15
I suspect I know what they're bad at, and I won't tell you.
204
777000
4000
13:19
(Laughter)
205
781000
1000
13:20
Thank you.
206
782000
2000
13:22
(Applause)
207
784000
6000
ABOUT THE SPEAKER
Marvin Minsky - AI pioneerMarvin Minsky is one of the great pioneers of artificial intelligence -- and using computing metaphors to understand the human mind. His contributions to mathematics, robotics and computational linguistics are legendary and far-reaching.
Why you should listen
Marvin Minsky is the superstar-elder of artificial intelligence, one of the most productive and important cognitive scientists of the century, and the leading proponent of the Society of Mind theory. Articulated in his 1985 book of the same name, Minsky's theory says intelligence is not born of any single mechanism, but from the interaction of many independent agents. The book's sequel,The Emotion Machine (2006), says similar activity also accounts for feelings, goals, emotions and conscious thoughts.
Minsky also pioneered advances in mathematics, computational linguistics, optics, robotics and telepresence. He built SNARC, the first neural network simulator, some of the first visual scanners, and the first LOGO "turtle." From his headquarters at MIT's Media Lab and the AI Lab (which he helped found), he continues to work on, as he says, "imparting to machines the human capacity for commonsense reasoning."
Marvin Minsky | Speaker | TED.com