YT YouTube2Text

YouTube2Text → Transcripts → Richard Feynman Computer Science Lecture - Hardware, Software and Heuristics

Richard Feynman Computer Science Lecture - Hardware, Software and Heuristics — Transcript

by Muon Ray · 13,257 words · 1,803 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:08[Music]
0:47[Music]
1:03okay what I want to tell you about is
1:05computers because they have computers
1:06all over the place and everybody's
1:07talking about him and using him in a
1:10frightened by them or else you find it
1:12very great assistants and so forth and I
1:15wanted to tell you a little bit about
1:17how they work what's inside him what the
1:19idea is what the mechanism is so that
1:23you're gonna get a better idea from that
1:25of what you can expect them to do and
1:27not be able to do and the so on so I
1:31want that's what we're doing we will
1:35expose what a computer really is
1:37actually the first aspect is you see
1:41there are many different kinds of
1:42computers there are computers that you
1:44can put the data in by talking to it
1:46even there are computers you have to
1:48type the data in there are computers
1:50that get that data from the machinery of
1:52an automobile or something and computer
1:55how much they should shift the gears so
1:57they have a automatic transmission but I
1:59mean controlled by an automatic
2:01computers that are you feeling how fast
2:03you're going and what the hill height
2:04strength is that called the input part
2:10of the computer that is the informations
2:12coming from different kinds of sources
2:13their keyboard or telephone sensors
2:17inside the automobile that measure
2:19temperature or acceleration or something
2:22like that and there's another aspect
2:24call output for example the thing that
2:26goes out might be the control of the
2:27gasoline in the automobile it might be
2:30to make pictures on a TV screen that you
2:32can read or look at it might be that it
2:35makes music and make sounds and so forth
2:37now in the plot that's the input and the
2:42output which makes computers look
2:44different is not the part I'm talking
2:47about there is in between the input in
2:50the output the input is converted to
2:52some electrical form then something's
2:55done in the computer and it spits out
2:57another electrical signal which is used
3:00for the output so for instance in the
3:03automobile get thing the various senses
3:05are sending electrical signals to a box
3:07which is the proper computer proper or
3:09the part of the computer that I'm going
3:11to talk about and then in the output of
3:13this box is other electrical
3:16which go to control magnets or something
3:18which turn on and off the gasoline or
3:22else control magnets which pushing it
3:25out a diaphragm very rapidly it makes
3:27sounds
3:28or controls the electrical currents
3:30running in the beam that's going all the
3:32time on the screen and it makes a
3:34picture on the screen and so on
3:35so what we're going to talk about which
3:38is really the computer proper is the
3:40part in middle so it's sort of we don't
3:43care where the stuff comes from we don't
3:45care where it's going on now what is the
3:47stuff in the middle
3:48that's what's manual really manufactured
3:51in the computer of course a real
3:52computer that's sold is advertised and
3:55it's good because it's got a big screen
3:57or it's got a small screen or it's got a
3:59tiny keyboard that your fingers are too
4:01big to push the buttons or it's got a
4:03big keyboard that has so many buttons on
4:04it you can't figure out which one to
4:06push and they have advantages and
4:07disadvantages which is not what we're
4:10talking about would talk about what's on
4:11the inside the central part of the
4:14computer proper which I'm going to call
4:16it the computer problem but the whole
4:18computer of course is the real thing
4:20so but we just talk about the guts in
4:24the middle the thinking part if you like
4:26to imagine that computers think but what
4:29is it I'd like to compare it to a filing
4:34system as a matter of fact I'll
4:36anticipate the end a computer is a
4:41high-class super speed nice streamlined
4:44filing system and I'm going to start out
4:47with an ordinary old-fashioned business
4:49filing system and that tell you ways
4:52that you can change it and improve it
4:53and so forth until it becomes a computer
4:56one of the miseries of life is that
5:01everybody names things a little bit
5:03wrong and so it makes everything a
5:06little harder to understand in the world
5:07than it would be if it were named
5:09differently computer does not primarily
5:12compute in the sense of doing arithmetic
5:15strange although they call them
5:17computers that's not what they primarily
5:19do they primarily are filing systems
5:24people in the computer business say
5:26they're not really computers they're
5:29data
5:30handlers all right that's nice data
5:32handlers would it be a better name
5:33because it gives a better idea of the
5:35idea of a filing system there's a lot of
5:37data on cards and your handler you take
5:39cards out look at them put them back and
5:40so on so that's what the computer does
5:42and I'm going to describe first an
5:45old-fashioned filing system and then
5:47we'll talk about how it's improved for
5:50how it might be improved okay
5:52actually it's improved to the point of
5:54being made automatic and make mechanical
5:56electrical actually and that's our
5:59computer and if old-fashioned filing
6:02system is something that many of you are
6:04perhaps even unfortunately familiar with
6:07it's a whole lot of boxes drawers cards
6:11and things you know which all have stuff
6:15numbers it and fax on them for example
6:18in a business that has lots of salesmen
6:19that maybe that does each salesman has a
6:21card he has his name on it and his
6:23address what his general salary is what
6:26his commission rate is how much he sold
6:28last month and where he lives things
6:31like that let's suppose and it's written
6:33on lots of cards because there's lots of
6:34salesmen in the boxes and then the
6:38problem for a file clerk might be for
6:40example a boss comes in and says would
6:43you figure out oh the commission rate is
6:46given on the cloud like a certain
6:47percentage and how much you sold during
6:49the month but now you have to figure out
6:50how much we have to pay him so what you
6:53do is you take out a card and you look
6:56at how much he sold that month and you
6:58look as commission rate you multiply the
7:00two numbers and you write it on the card
7:02and put it back and do this on card
7:06after card okay what does it mean you're
7:08taking a car then yet looking at some
7:11things and doing something to them that
7:13is looking at some data on the car the
7:16easiest way to think of it to simplify
7:18just a little bit is that we have scrap
7:19paper that we can erase and use again
7:21use again
7:22so we take them we rewrite the numbers
7:24off the card the the amounts before so
7:27while we're doing the multiplication we
7:28can do it with the numbers without
7:30dirtying up the card so we copy the
7:32numbers over and then do the
7:34calculations see the answer and if the
7:36rule is to put the result back on this
7:38man Scott okay so we write it on this
7:40car it may be that we would like to put
7:42this on some other car
7:43maybe cards about expenses that have to
7:46be paid or something you see so you'd
7:48put that somewhere else or perhaps to
7:51make another example that you would like
7:53to add together all well I'll forget
7:54that you want to add together all the
7:56money to know how much you're gonna have
7:57to pay out that's another example but
7:59they're all the same of course that you
8:01take that numbers from the cards and
8:02write them down and recalculate and put
8:05it back in the numbers so you have two
8:06kinds of three kinds of operations
8:08taking cards from places looking at
8:13information on the cloud doing something
8:15to the information on the card writing
8:18things on the card and putting the cards
8:20back okay actually it could be a little
8:24bit simpler you could say you take the
8:25thing out and take the information off
8:27the card immediately put the card back
8:28this save it now you do the calculation
8:31and when you get the answer you pull a
8:33particular card out that what you want
8:35to put the answer on it may be the same
8:37man's car but wherever you want to put
8:39the answer you take it out right at that
8:40put it back okay so we have two
8:42operations picking up a particular card
8:45getting data off of it doing something
8:47to it and then putting numbers onto
8:52cards from our scrap paper on different
8:55cards and that's all if the file clerk
8:57does it to take another example suppose
9:01that we wanted to suppose we have
9:03figured out or the amount of money that
9:04we owe each salesman and we would like
9:06to know how much to draw out of the bank
9:08in total so we need the total amount of
9:10all the sales then we could do that by
9:14having a special card goal of total card
9:16and writing zero on there and this is a
9:20way of kind of I'm talking about the
9:22details of what we have to do it
9:23everybody knows what the do you tell a
9:25file clerk they add the numbers but this
9:27is what you might do or he who's write
9:30zero on this particular thing called
9:32total sales costs in California and do
9:37the fine then you take the first
9:38salesman card you look he's not from
9:39California we want the total amount of
9:41money we have to spend in California
9:43salesmen all right just as an example so
9:46you put that one back take the next card
9:48is it from California put it back next
9:50card in California yes then the next
9:52instruction is to look at how much money
9:54you are with this man and write that by
9:57- what's on the total card you take the
9:59number that's on the total card and the
10:00number that's on this man's card add
10:02them together and write that back on to
10:05the total card right you're keeping the
10:07total running see so the next time when
10:09you take another card that has a number
10:11on it
10:12just take the new totally you just
10:13cooked up from before there was no
10:15trouble last time you had zero but this
10:17time it's getting harder ya headed ohon
10:19and put the Sun back on the total card
10:21so you see you got a kind of a steady
10:23thing and when you're all done going
10:24through all the file the total has been
10:26changed and changed and changed and
10:28changed in the last thing you read is
10:30the real total of all the numbers okay
10:31that's what the file court might do in
10:34that situation and now what we're going
10:39to try to improve the filing system to
10:43improve it you see we have to have a
10:45fairly intelligent file click a file
10:47click that can multiply and add and so
10:50on read cards get instructions from the
10:53boss and remember what they are
10:56naturally and know how to do all these
10:58things has to know how to where to go to
11:02run up and down and pick the cards out
11:04of the boxes and I has to read the place
11:06to find the cards and know where they
11:07are at so it's a lot of stuff well
11:12another one comes along we got a new
11:14file clerk I got a new file clerk for
11:16you say to me five times faster than the
11:20old one oh well we'll hire but you
11:24discover suddenly that he doesn't know
11:25how to do multiplication well how in the
11:28hell is he going to handle their finding
11:32the Commission's by multiplication this
11:34guy isn't any good but he's five times
11:37faster so it's very tempting if we can
11:39only get him to be able to multiply so
11:42what do you do he is sending to school
11:43too much waste of time better idea to
11:47teach him had him you tell him how to
11:48multiply as follows I've taught he's
11:50very fast at picking up cards and
11:52putting him down so he takes the cards
11:55and he looks at say two numbers and you
11:58tell him to look at the last end of the
11:59number one's an eight and one's a seven
12:01right then that instruction looked at
12:04card number 78 in a multiplication file
12:07when card number 78 no multiplication
12:10says
12:1056 the multiplication table which we
12:15worked so hard to learn in school is
12:16easy to put on a set of cards on each
12:20card which has which is labeled by a
12:22number let's say the card number 78 or
12:247/8 the card label I don't have to
12:27convert it to a number of card called
12:29multiply seven eight answers sixteen
12:32there's written on a card so all he has
12:35to do is pick up the right card he's
12:36very good at that and he's very fast see
12:38so he sees the seven eight take the
12:39cockpit sink okay and so it goes he
12:41doesn't have to know how to multiply you
12:43tell them a series of instructions what
12:45to do exactly you give him in the
12:47instructions as to which cards to look
12:49for at what time so when you say
12:51multiply you don't even tell them that
12:52your tongue look in this box now take
12:54that card out and write down write down
12:56what that one says it's so on so you see
12:58that we can describe multiplication as a
13:01definite procedure a precise procedure
13:05of handling numbers in a kind of a dumb
13:07way and things that would have to be
13:11memorized are never a problem
13:14memorization is no problem that's what a
13:16filing system is for that's what cards
13:18are for we write everything on the cards
13:20every damn thing is suppose you've
13:21learned where all this effort is written
13:24on a card so with cards we can make this
13:27fast filing clerk actually able to
13:30multiply of course it's slower to have
13:34to look up each time in the
13:35multiplication table because he has no
13:37memory were worth a damn four numbers of
13:39this coin so he can't even remember that
13:41his use of seven times eight fifty five
13:43times but it still has to look it up
13:44again so he's slower in the sense of the
13:48number of operations to multiply than
13:51the original file clerk was but he's
13:53really faster because altogether he
13:56handles the card so much faster that in
13:58spite of the fact that if there's a lot
14:00of multiplication problems he isn't five
14:03times faster than other Clerk but he
14:05might be two times faster because he
14:06slowed up in the process of
14:08multiplication so our filing clerk does
14:12not know how to know how to multiply
14:14well we'll get a faster one
14:17I got another one that's ten times
14:19faster than last night
14:21he can't add no problem right I mean we
14:25have a adding table too right no problem
14:30okay he's easy he's better he can't add
14:32he can multiply can't handle numbers at
14:35all but he can compare it read and check
14:37that what it says on the cards is the
14:40same he doesn't even know that these are
14:43numbers he sees this shape and compares
14:46it to this shape and takes the cards out
14:48according to the instructions okay now
14:51where does he get his instructions
14:53they're in a series of cards that says
14:56do this then do this then do this you
14:58see so we have a sequence of card with
15:01instructions on them he doesn't even
15:03have to remember the instructions he's
15:06rather stupid the first filing cloture
15:10could say go get the card that
15:13instruction meant go into room five
15:17opened the drawer push your fingers
15:20through the card lift and so forth okay
15:22so if you want to get the card to read
15:25it and then carry it back and write it
15:27down you could have a list of
15:29instructions for this idiot instead of
15:31saying get the card you say look in
15:33these set of instructions please first
15:35instruction walk over to the file second
15:37rope pull out the drawer but if at home
15:39so does it done and so forth so all he
15:41has to do is keep track of which
15:43instruction comes after which he's
15:46really quite stupid but man is he fast
15:49dumber but faster okay the key of it all
15:53is dumber but faster right and so what
15:57we're having is the filing systems with
15:59ever-increasing speed of the file clerk
16:03but ever stupider file clerks okay and
16:07they're getting better assistant Chris
16:09brevis file everything now the last
16:12fellow that we've got in this not the
16:15last one but the willit next to the last
16:17one is unable to remember which
16:21instructions on
16:23well what's the next instruction oh but
16:27he can add one he knows how to change
16:29something with one he has a little
16:30counter but he pushes a button and it
16:32goes up one okay so all he has to
16:36remember is the following push the
16:41button on theirs a little register that
16:43he can read okay that's a push that we
16:46call it the program counter or
16:48instruction count all right push the
16:50button that changes the instruction
16:52count by one that's the only button he
16:53knows how to push that okay that's the
16:55first thing spike read the number there
16:58on the program counter and take that
17:01card out that instruction number out
17:03okay and that says go get this paper or
17:06transfer this so whatever it is you see
17:10because we have already put all the
17:12instructions in a certain order and then
17:14put the card back you're finished do the
17:16intern take out read the instruction and
17:18through the instruction and put the card
17:20back okay then the next cycle he has
17:23push the button on the program counter
17:25read the program counter it's the next
17:28number you know anybody couldn't
17:30remember so he has it there it's the
17:31next card takes the next corner and it
17:34tells him what to do next and so on okay
17:36so he doesn't even have to remember
17:38where he's at
17:40as long as he's got this one little
17:41gadget inside which I've beginning to
17:44think you could see could be done
17:45mechanically that is we can make a
17:48device that all you push a button and
17:49number changes by one right
17:52in fact so that that almost got it but
17:58we find out very much fast at ten times
18:00faster than this marvel who was so
18:03stupid but so fantastically fast which
18:06ten times faster than this fellow but he
18:09can't read letters he doesn't know what
18:12they mean doesn't know what the numbers
18:14mean okay so how is he gonna what can he
18:17do if he can't read anything he can't do
18:19anything I grant you that I grant you
18:21that but he can distinguish two signs
18:25one from the other which we'll call high
18:27and low or black and white or one and
18:31zero or mark and Norma whatever all
18:34right just he only can read
18:36not any kind of letter but just one
18:37little spot it's yes let's say we always
18:40make it red and blue you see is it red
18:42or is it blue so we make a lot of spots
18:44for him red blue spots it's all he can
18:46read what I need to eat how can we make
18:47him do anything how can he know what an
18:50a is suppose we got letters we want to
18:52write the letters for names or something
18:54like that so we would like to instruct
18:57him in the letters for the names to see
19:00so instead of saying we tell him which
19:03Oh first better to give a better example
19:05by this time because they have to use so
19:07many cards in so many files the files
19:09have gotten larger right there are lots
19:11of filing cabinets in many rooms all
19:14over and we this poor fool has to be
19:17able to find the right cabinet now
19:19previously we labeled all the cabinets
19:21with the names of the sections from from
19:23small stoop to zoology and so forth here
19:27and this is in box 34 and this is all
19:31over it's like that places in Esalen
19:32have numbers on them and so forth I'll
19:34show you how we could do that without
19:35having only with red and blue dots on
19:38the on the rooms for those particular
19:41kind of workshop people who cannot read
19:44numbers all right I will still be able
19:47to do it I'm not sure how we'll do it
19:49let's do it with the esselman rooms it
19:50just makes it more interesting I don't
19:52propose this as a practical device but
19:54perhaps it is and it goes as follows
19:56just till you take the map and you
19:59divide the rooms through the middle and
20:01say these are the red first level red
20:04and blue rooms okay then you take the
20:10red rooms and divide them in half let's
20:12say those are the higher part of the
20:13hill in the lower part and say they are
20:15secondary red or blue and the blue ones
20:18the main blue ones have been divided at
20:21the primary and secondary rooms like
20:23this no I see that I don't have red and
20:26blue so I'll use brown and orange it's
20:28perfectly all right had I said brown and
20:30orange it would be more like they
20:31understand that we couldn't be any two
20:34colors okay so and let's first draw
20:40what in yellow a lot of rooms can't see
20:47the rooms okay I don't have enough thing
20:51so these are the rooms will be all right
20:55let's say we had this many rooms I've
20:59got them all laid out in a straight line
21:00but it doesn't make any difference where
21:02they are on the map right we got eight
21:05rooms I know you got more rooms than
21:07eight rooms but just see the scheme will
21:09work just as well for any number of
21:11rooms so what we're gonna do is we're
21:13gonna mark all these rooms with a black
21:15dot as a first dot I mean a brown one
21:19all right and these will get orange dots
21:23okay
21:27now of the brown dotted rooms we think
21:31of the brown dotted rooms and we look at
21:32the next dot which says the first half
21:35of them get brown on that dot and the
21:39second half of those get orange all
21:43right and then the first half of these
21:45get brown and the second half gets
21:49orange I say but still you're filled
21:53okay now you look at the brown brown
21:55rooms so any pet is a brown brown the
21:58brown orange rooms
21:59the orange brown rooms and the black
22:01orange orange rooms of the brown brown
22:03rooms we label the side ones Brown as
22:06the first each pair gets the front end
22:09brown and the other one gets orange and
22:11lo and behold we got eight rooms and
22:15with three little dots each of which can
22:18only have one of two colors we can
22:20identify the room so the guy who has to
22:24locate himself and you can't read
22:26numbers but can read whether a dot is
22:28brown or orange will in this system be
22:30very easy to find his unique room in
22:33fact it's easy to find the room which is
22:35different of course as you already know
22:37the numbers on these rooms around here
22:39do not locate the rooms right but this
22:42new system really in locate the room the
22:45first Brown tells us to kind of go over
22:47here somewhere
22:48the next round swap more is it orange
22:50take that side
22:53so if the filing cabinet can happen this
22:56can be very large and have labels on him
22:59which I like these brown dots and orange
23:01dots but because instead of being eight
23:03there are say a thousand such rooms the
23:06read turns out you only need ten dots to
23:08do it a million such rooms is only
23:10twenty dots because there's so many
23:12different combinations of brown brown
23:14brown orange brown I'm bombed around by
23:16men for twenty long so many ways in a
23:20million different ways so you have that
23:23way you can label a large number of
23:24different cabinets and have a file clerk
23:27which you can only read that's one of
23:29two alternatives and make sense of it
23:31and so we can do it with rooms and also
23:35with the letters of the alphabet because
23:37we take the letters of the alphabet and
23:38write them along and label them the
23:42Bradford the letters on this side of the
23:44alphabet of the brown letters first
23:46primary brown and the others are orange
23:48and never those you make the brown so
23:49these could have been ABCDEF yo so he
23:52could teach him a different way to write
23:54his alphabet
23:55he's so dopey he could only read these
23:57but we smart enough to convert it so he
24:00can be therefore using names and numbers
24:02but they're written in a different code
24:04it's like the dot-dash system of Morse
24:06only slightly more organized
24:09we call it ones and zeros brown and
24:12orange I'd call again whatever you can
24:14only read two possible things okay so
24:17this the new file clerk can only make
24:20out one of two alternatives in any
24:22particular spot of course he has to
24:24realize is do first the first one and
24:26then he's up to the second when he saw
24:27him first makes one choice and then the
24:29next choice and next choice and
24:30according with with this and then
24:32determines the letter of the alphabet if
24:34he had to talk to us and make an output
24:36he's not going to be able to just put
24:39this ones and zeros we're not going to
24:40be able to read it
24:41so suppose he wants to say something and
24:43he wants to put an out but you ask him
24:45for the name of the the best-selling the
24:48guy who sold the most in California and
24:53he he does it with his dots and I'll
24:54tell you a little more about how you can
24:55do that and comes out that his name
24:58starts with an AE say or let's say
25:01starts with a C whereas a C for him it's
25:05a three dot
25:05Sanna - right but he wants to print see
25:08for you what can you do we wanted to see
25:10it look like a scene so what we have
25:14done for him is we imagine he's going to
25:17write this on some kind of a screen that
25:19we have divided into squares okay like
25:24this I should make more and we tell him
25:31whether to mark the square or not in the
25:34electronic device with a pulse so it
25:37lights up on the screen or not by
25:39whether it's a brown or a orange dot
25:41there so he put a brown dot here well
25:44let me first draw the C or C is going to
25:46look like this is where the orange dots
25:47go okay there's the C all right not a
25:55very good see but to see if you add more
25:57dots and a finer screen it would make a
25:59much better see so so we put all these
26:02Browns and so on and then we put into a
26:04filing cabinet under the labeling for
26:08the third letter of the alphabet this
26:10pattern so when it cuz his when he's not
26:13working internally which is you can use
26:15this other system but wants to
26:17communicate with us we told him how to
26:19do it what we get for a seat by putting
26:22in the filing of that that he's supposed
26:24to put out on the screen so when he's
26:27finished looking at that we look at the
26:28little white dots and black areas we see
26:31a shape of a seat so the hardest thing
26:35he has is communicating with us getting
26:37the information converting at the dot
26:39and getting the dots back to the form
26:41that we've we're used to the same thing
26:44can be done with numbers of course
26:45suppose that he had only deal with
26:47numbers from one to a thousand and
26:49twenty zero to a thousand and 23 it's a
26:52happy case where you can divide ten
26:55times you can divide those in half you
26:57say Oh numbers described by the same
26:59bunch of dots by taking a you know the
27:02top half of the numbers or the bottom
27:05half of the numbers I already did it
27:06with numbers
27:07it was the numbers on the rooms so we
27:09can use these dots and so forth to
27:12represent numbers the same way
27:16and now we can use in a particularly
27:19insane I mean stupid a file clerk
27:23perfect and it's only useful of course
27:24if he's really very fast because
27:26everything requires him to looking and
27:28his damn files all the time because he
27:30knows so little and that's what
27:32ultimately has been developed first of
27:35all in order to get up speed instead of
27:37having to get up and walk over to the
27:39file we put the files not on cars but on
27:43little tape decks huh little tape decks
27:47electrical and we send the signal turn
27:49on the tape they can listen to the
27:51signal that comes out we don't have to
27:53run over there we've got a telephone so
27:54to speak you we tell which one to go off
27:56then it sends back what's on that tape
28:00so you see that we can file things not
28:02on pencil and paper but there the
28:04electrical signals or magnetic different
28:09kinds of techniques they're off of
28:10electrically saving information and it
28:13turns out you can do that smaller and
28:15smaller until the filing cabinet is the
28:17speak which contains millions of items
28:21and what happened what you're doing is
28:23you're signaling to the different filing
28:25cabinets depending upon the ones and
28:27zeros there are switches inside that
28:29tells which one it goes to and the
28:31information comes in and out
28:32electrically that's much faster than
28:34carrying cards around and running around
28:35in sneakers and so that's the way that's
28:38done and by the time we got a file clerk
28:40this stupid we can do it mechanically we
28:45can tell whether two dots are the same
28:46for instance we can ask him when the two
28:48looked the same do this so if there's
28:50two brown ones or two orange ones do
28:52this now it's rather easy to make
28:55mechanical devices which will do things
28:57like that for example instead of brown
29:02and orange dots you could have
29:03electrical voltages which are either up
29:06or down
29:07you know there's current dah no there
29:08isn't so instead of a dot of different
29:11colors we just have a voltage or no
29:13voltage on a particular wire at a given
29:15moment or if you would like but this
29:18isn't the way it's the electrical one is
29:19the way it's done but if you would
29:20prefer you can imagine you made a
29:22hydraulic computer with pipes with
29:24pressure with water going through and if
29:27you look at the pipe and water squirt
29:29that's a black dot and the waters no
29:31water there that's the other kind so
29:33it's about that easy so you'd make the
29:34pipes go squirt squirt not squirt squirt
29:37squirt not squirt that's that this one
29:39is this room right now what we need is a
29:42situation to compare things that only
29:44squirts of two things let's suppose we
29:47were looking for both dots be brown only
29:50then should you make a brown dot in
29:52other words two pipes such that if both
29:54pipes have water in them then only then
29:58should water come out the other one what
30:00we need only a valve a valve operated by
30:03water we have a pipe like this with a
30:07valve which is a thing you don't turn it
30:10on let the water through the waters
30:11trying to go through here it's turned
30:12off by a faucet which only takes a small
30:15turn and has sticking out above it
30:17instead of an ordinary handle a flat
30:20vane okay that we're going to scrape
30:22with water and that's going to push the
30:25vane around and open the valve so here's
30:27the for the things coming in with the
30:29water to operate that so when the water
30:32goes here it turns the water on there
30:33now if I put two such the bowels in
30:36succession in the pipe okay
30:39and I had the water coming in need two
30:42different pipes then it was only work
30:45that is water will only come out of here
30:46if both of these pipes are squirting
30:48because if I got only one pipe squirting
30:51in the other not this valve was off but
30:52this on so water don't get through on a
30:54counter that's in the way
30:55and vice-versa okay so this is what you
30:58call an and you get result here when
31:02both this and that are on then you get
31:04that on you could do an or just as well
31:08when you get a result if either one is
31:10on but that's different kind of a logic
31:14and the way to do that is simple enough
31:16just arrange the pipes so that the water
31:21that's coming in is split into two
31:25possible pipes let me see how I draw
31:27this yes well the brown and the orange
31:29is just an accident of fate of having
31:32put the orange cap on the brown
31:36gadget that some would straighten that
31:38out right now and here we go the pipes
31:42go like this the input is here but we go
31:46with the two valves this way here's the
31:47one with the other valve okay and the
31:50two squares let's say this one squirts
31:52here and this one squirts here you got
31:54to get some geometry as three-dimension
31:56got to get this pipe on top of this pipe
31:57wouldn't be trouble anyway the idea is
32:00that these then come together and you
32:03might ask under what circumstances does
32:05water come out of there well clearly if
32:07either this is open or that's open to
32:09either goes around this way or it goes
32:11around this way or both that's an or you
32:14get choked out only when one or the
32:16other or both is on and so we could say
32:18very simple comparisons of these black
32:21these dots or Bennet ones and zeroes or
32:25fresh it water and no water or
32:28electricity and no electricity
32:29voltage or no voltage electrons or no
32:32electron and so we can make the device
32:35simple enough now that we don't need a
32:38human being anymore
32:39to do the file work and the whole system
32:43is now a computer it's all done
32:45electrically and it has this kind of
32:48switches in it valves are electrical
32:51valves are called transistors and how
32:53they work in detail of how the electrons
32:56come in here affect the electrons going
32:57there I won't describe but they're just
33:00granting that a transistors are like is
33:02the analog of a vowel but for electric
33:04current then you're going to understand
33:06that large numbers of transistors lots
33:08of these vowels making comparisons
33:10switching where the current goes to find
33:12the right filing place in the Box where
33:15all the stuff is filed and so forth
33:16would constitute a computer and that all
33:19computer does is what a filing cabinet
33:21system can do it can do everything that
33:26an ordinary most things that an ordinary
33:28file clerk system can do it can multiply
33:32numbers because it's got we store the
33:34numbers we store the combinations they
33:36can add it can do all these things even
33:37though it for hardly do anything
33:41and you might ask is there some kind of
33:43designer computers that can do more than
33:45other kinds of computers and the answer
33:47is no it turns out once you begin to
33:49understand this that if you try to add
33:51more complicated things you can always
33:52put them back in filing systems and use
33:56the same system to do a more elaborate
33:57thing so all the computers can do the
34:00same things but maybe they take longer
34:02or so it may be a matter of efficiency
34:04but they can all do the same things
34:06they're all Universal inside if they're
34:09big enough their limitations are how big
34:10is your filing system maybe there's not
34:13enough room for all the cards you want
34:14to put in and things like that but
34:16that's the only the only differences
34:18between the between the different
34:20machines and any rate that is supposed
34:23to be an expose of what the heck's
34:25inside the machine now let's ask what
34:28the miss such a machine could do first
34:32of all the same machine can be used for
34:35more than one problem by just having
34:37different instructions list some
34:39instruction what to do do this then that
34:40to do that to deep deep deep deep and
34:42another problem you do something
34:43entirely different this might have been
34:46a business with salesmen and so on and
34:48now you want to do oral company reports
34:52and listings of what they discover when
34:54they drill all right so all new numbers
34:57or all new instructions of how to handle
34:58a numbers everything what do you do by
35:02another filing system it's more
35:04expensive to do that than to do the
35:06following
35:08throw away all the cards from the
35:10business company
35:11wait a minute suppose you want to come
35:12back in all right take everything that's
35:14on the cards of the business company and
35:16make photocopies and store them in the
35:18basement now erase all the cards so if
35:20you want to come back we'll come back
35:21then you go down the basement from the
35:23time year last year when you were doing
35:25the oil company problem and pick up from
35:29the photocopies all this stuff that was
35:31supposed to be on the cards for oil
35:33business and then just fill out fill up
35:36all the cards with the oil companies job
35:39with the instructions for oil companies
35:41and our dumb file clerk is now an oil
35:43company file system okay because he's
35:47got new instructions and
35:50these things that in the basement the
35:52analogy that I had wanna make is those
35:54things are dead not operating and I just
35:57stored they're not easy to get at it's
36:00hard to get down to the basement it
36:01takes a long time to pill pull them up
36:03and set them in all those cards but you
36:06do that just once in a while when you
36:08want to change your problem when you've
36:10got them all in then you can rapidly
36:11take them and put them in compare them
36:13and do this and do this let the filing
36:14cop click is supposed to do so we have
36:18cassette tapes or hard disks of floppy
36:22disks or various and sundry different
36:25ways of taking a large numbers of
36:27instructions out and sticking them on
36:30the disk and leaving them there putting
36:32them in your back pocket throwing them
36:33around in the boxes while the computer
36:35is doing something else then if you want
36:37to change a computer to a word processor
36:38instead of a a game or whatever it is
36:41then you just take out one this put the
36:44other disk kid puts new stuff into the
36:47computer all new instructions done file
36:50clerk working like a son-of-a-bitch and
36:53a big memory outside and that's why
36:56computers could do so many things
36:58apparently the same computer it doesn't
37:01do so many things it always does the
37:02same thing it follows instructions and
37:04the instructions are very done and it is
37:07low-level ok that's the end of the
37:14expose but they would like to make some
37:16remarks and some things about because
37:19people like to think of things more than
37:21just how does it work they would like to
37:23know what's the future why can't we
37:26improve this looks what we've got now we
37:29have this enormous memory with lots of
37:31data all around which is sort of sitting
37:33there and this one little section where
37:35something's happening which I call the
37:37file click before which is the systems
37:39which go and get something pull it in
37:41here and do something over you better
37:43put it there and it takes something else
37:44and put it here and put it there put it
37:46there and what a great so this poor fire
37:48clerk is working perpetually with a
37:49tremendous rate of speed going
37:52practically crazy and he does in fact I
37:54just estimated something like 10 million
37:56times faster than an ordinary file click
37:58that's 10 times faster seven times if
38:02you know what I mean
38:02ten tons ten tons but it spam fast and
38:09that's the electrical things of that
38:11fast and but he's working very hard but
38:17the rest of the system which is the
38:19memory and you've heard they made now
38:20memories with 256 K bytes K just means a
38:23thousand and twenty four so they mean
38:25256 times 1,024 eight dot were symbols
38:31in one tiny chip okay it's a lot and you
38:34put several chips and you got a terrific
38:36computer except for the guy the file
38:39click which is called the central
38:40processor who has to look at this this
38:44program can't do in check and change it
38:45and look at the next instruction and
38:47going going alright and he's working
38:49very hard so an obvious way to improve
38:51the thing is to get another file clerk
38:54why not have an army of file clerks
38:57turns out at first that it looked
38:59extremely difficult to do that because
39:01there's all kinds of technical difficult
39:04problems suppose that he one guy has
39:06taken a number out and he's getting
39:08ready to change it he's gonna do
39:10something and then change the number the
39:11man just sold something he wants to
39:12change his he and then put it back and
39:15while he's doing that the other file
39:17clerk is taking the card he doesn't know
39:19whether it's before or after the first
39:21file clerk has changed the number and
39:22it's very important to know that because
39:24he's going to use that number there so
39:26that's to be a certain amount of
39:27organization among the file clerk and
39:29don't forget that kind of dumb so the
39:31organization must also be fairly done
39:33and has turned out for many years it was
39:36a belief that it was difficult or almost
39:40impossible to easily arrange a system
39:42with many file clerks they were always
39:44turned out to be easier to make one file
39:46clerk worked very hard that the made
39:48several working because of all the what
39:50you might call bugs and difficulties
39:52that come from them all using the same
39:53information and changing it on each
39:56other and so on and organizing that the
40:00part of laying out the instructions is
40:02called a programming and a programmer
40:06works out how to put what instructions
40:08to write
40:10the machinery the wires memory in the
40:14boxes where to put the cards but it's
40:15not an analogy and the little processor
40:17in the middle with little switches it's
40:19called the hardware so it always turned
40:24out that with a little more clever
40:26software which means cleverer programs
40:29this single file clerk could do what the
40:33others would do because the others were
40:34confounded by the problem of
40:35interference but recently it's been
40:39shown that it's a misunderstanding and
40:43with clever new ways of analyzing the
40:46thing we have now figured out how to
40:49make things run with numbers of file
40:51clerks there's two different general
40:56system there's one system which has
40:58something you know a reasonable number
40:59like 64 128 file clerks all working on
41:03it saying similar they work on almost
41:05separate parts of the memory that's the
41:07way it's organized and only to once in a
41:09while exchange information so they're
41:11not doing anything to each other's cards
41:14they're each working in separate
41:15sections and there's also a machine
41:18called a connection machine which has
41:2064,000 processors in it which is a sort
41:24of a leap into the blue wild blue yonder
41:26at a time when they were trying to do
41:27with two processes and if the 64
41:30thousand process that could all be made
41:31to work at the same time it'll be 64
41:34thousand times faster than it was before
41:36just like that because they're all
41:38working together the only problem is to
41:41organize the problem and the program so
41:44that you can do the job in a way that
41:46does use 64,000 guys at the same time it
41:50turns out that's not as hard as it
41:51looked at first and that's a thing for
41:53the future I'm talking about the future
41:55they're building such a machine and
41:56they're learning how to program it that
41:59is how to arrange the 64,000 processes
42:01which by the way are each in fact
42:04somewhat stupider than the program is
42:07that I was talking about before because
42:09they can't even read the code for their
42:11instructions but I mean they know as
42:14instructions read what's in that file
42:17place that means go over switch this
42:20switch that take you go over get the
42:21file open the drawer closes
42:23for they have they can't even read that
42:26so that they have to be instructed on
42:28the goal and so there's another machine
42:31on the outside that instructs some more
42:32what to do but it has to tell them all
42:34because see they all have to do the same
42:36thing at the same time because the
42:39instructions go to everybody do this and
42:41they all do it but if they all do it
42:44you're going to be in a great difficulty
42:45if they all do exactly the same thing
42:47and the only opportunity they have the
42:48only thing you two can do that's
42:49different is that you can instruct or
42:52they can keep a little memory thing that
42:54says don't pay attention to this
42:55instruction in other words you can get
42:57some to do something and the others to
42:59do nothing then next time you wanted
43:01them to do something if you wanted one
43:03to do one thing and want to do another
43:04you have to do it in two steps
43:05do this you stay put through this you
43:09stay bored and so on okay and so on so
43:13it's rather interesting because in
43:15trying to manage the sixty-four thousand
43:17they all have to be much simpler and all
43:19the time it gets simpler it gets sort of
43:20slower but the question is whether we
43:22gain from the sixty-four thousand what
43:24we lose from the slowness and it depends
43:26we learn by experience that's the future
43:29finally I would like to just say
43:31somewhere it's about what the computers
43:34can do and you'll see they can do only
43:37what you would be expected to be able to
43:39do with a filing system and a reasonable
43:41file clerk I'll explain an example of a
43:44unreasonably good file clerk later
43:46that's better than one that we can make
43:47with machinery ok so far we seem to be
43:50able to imitate any file clerk real far
43:53click with machinery but I'll mention a
43:55few cases that we don't know how to do
43:57that what we need to do of course is to
44:02get a definite procedure to convert what
44:06you want the clerk to do into an
44:07absolutely definite procedure we every
44:10step is precisely defined in this stupid
44:12way exactly and if you can do that then
44:17you can get the system to work now will
44:20the system think if you could define
44:24exactly a definite procedure tell
44:26somebody exactly how to think which by
44:28the way is what the members of my
44:30workshop have expected me to do for them
44:33to tell him exactly step-by-step what to
44:36do to think if I could do that I could
44:39do it for a computer and I wouldn't have
44:41you know they could get the computer the
44:42thing but nobody knows what we do or how
44:45to define exempt which correspond to
44:48something abstract like thinking but
44:51nevertheless we could make the machines
44:53which play chess which was supposed to
44:54be something that takes thinking and
44:56what exactly do they do what definite
44:59procedure can you make with a big file
45:02system very fast operating gadget by
45:05which the thing could be used to play
45:07chess
45:07never mind the input and output the
45:10information of where their pieces on the
45:11chessboard are converted into numbers or
45:13black dots and so forth ones and zeroes
45:17they're called inside the machine and
45:20then it proceeds so never mind a
45:21translation the idea is in a given
45:24position which is a whole lot of
45:25information that's where it is
45:27you want to know which way to move so
45:29you say well there are many ways I could
45:30move you look at the rules for how to
45:32move the night goes this way to bring
45:34you this way so on and you just list all
45:36the different ways it could move and you
45:40see all the new gap possible positions
45:42then you say in each one of those which
45:44way could the other guy move they're
45:46getting pretty big isn't it
45:47but there are big machines and they've
45:49got lots of storage space so there's how
45:51does it what's the next room and if he
45:53does that what can I do and you just
45:55check every one of the positions after
45:57let's say four or five or six half moves
46:00these are called happen with one side
46:02moves to the other side move and see
46:04which of the final positions
46:06you're a better off and better off might
46:09be some little rule like you haven't
46:10been in check I mean haven't had your
46:12King lost that means you're not dead and
46:14you've got more pieces than the other
46:16guy okay so you look down six steps down
46:20to see which move is the best to make
46:23where you got the best possible outcome
46:25and you make that move then your
46:28opponent makes a move then you start
46:30again this time going down six again of
46:32course you're going deeper now because
46:34the game is preceded a little bit
46:35further and that's all there is to it
46:37what the machines do is just a
46:39tremendous amount of trying different
46:41positions human being doesn't the human
46:44being tries about 35
46:45the sponsor he can tell when he's in a
46:47position he tries about 35 looks at
46:50different moves before he makes a
46:52decision whether the machine makes
46:54something like 35 million positions to
46:57check that doesn't make the Machine
46:59better than the other guy we don't
47:00understand how we do it we cannot give a
47:02direct procedure for this the man does
47:05it by recognizing a pattern oh this
47:08night can can this night can fork the
47:11the king and queen of somethings if it
47:13goes here so I have to watch out about
47:14that square he sees patterns at least it
47:17describes it that way we really don't
47:18know how it work and this business of
47:21the describing patterns and so forth is
47:23something that we haven't been able to
47:25put into a precise form when he says it
47:27looks oh I just see that it's a good
47:28idea to try this and I tried it we don't
47:30know how he does that so we can't make
47:32it play like a human plays but we can
47:35make it play better than almost all in
47:37human machines computing machines have
47:42been used for instance for designing
47:44machinery or designing something in
47:46particular designing a computing machine
47:48for instance or something like that
47:50in a computing machine yet that the
47:52different elements the memory and so on
47:54and different locations with wires in
47:56such a way that the wide the wire
47:58lengths are shortest so that the time is
48:00quickest and the paths that are being
48:02used most on closest to where you want
48:04to use them and things like that and so
48:07you try to figure where should you lay
48:08them out on this chip of silicon and
48:11there has been quite a problem of design
48:14we have engineers that worked very hard
48:15at designing these things and working
48:17these things out in all we do in the
48:20computer now is to try it - try large
48:24numbers of possible arrangements we just
48:29give it an instruction put it here put
48:30this here then this this this this and
48:32how much why do you need now put here
48:34here here here here over there and then
48:37pick the one out which has the biggest
48:38more ones the biggest of the lowest the
48:44best quality whatever it is you do to
48:46measure how good this particular design
48:48is like the amount of time
48:51that it takes to do a calculation or how
48:54much silicon you use or something like
48:55that so we can select from a very large
48:58numbers of alternatives the best
49:00alternative
49:01what narrowly a person doing that
49:03doesn't waste his time with obviously
49:05useless alternatives but we do not know
49:08how to put - well that idea of the
49:11obviously useless if we try to make to
49:14shop the rule that it's useless we miss
49:16sometimes a good opportunity that would
49:18have been a good thing so that that's as
49:20far as we can go with that you've
49:22probably heard that machines have been
49:24used to the dock to be almost like
49:28doctors expert systems which can
49:30diagnose disease now how can a damn file
49:32clerk why it's easy isn't it how would
49:35you diagnose disease you write down all
49:37the symptoms and then you'd like to have
49:40a book that you could look up that says
49:41here's all the symptoms and this is what
49:43you go and in a complicated cases you
49:46list all the symptoms and in there it
49:48says it's either cholera morbus of the
49:50Borba sore its back there a see Miele
49:53whatever it is you see however you can
49:55tell the difference by looking with
49:57ultraviolet like up the nostril like but
50:00what nobody I don't bother to look with
50:02ultraviolet light off at the nostril so
50:04you don't know the answer to that so the
50:06thing comes back and says look with
50:07ultraviolet on the nostril and tell me
50:09whether it gets through or not say yes
50:11okay it's back to see me so this filing
50:15system contains all this stuff plus the
50:17incomplete decisions that require more
50:19information and once it's been arranged
50:22so that when it comes to a place it's
50:23incomplete it simply sends back and says
50:25make the following test and tell me the
50:27result their thing doesn't work as well
50:30as a real doctor but anyway you get some
50:34idea of how machines can be doctors how
50:37machines can be chess players and our
50:39machines can be designers of equipment
50:42or layouts for businesses or whatever
50:44routing for trucks and so on but just
50:47trying what kinds of rulings in seeing
50:49which is the best that's all there is to
50:51it
50:52it's a filing system and anything that
50:55you can put in a filing system and think
50:57of how a big filing system could do it
51:00especially things that you just have to
51:02remember a lot of stuff or try out a lot
51:04of possibilities
51:05those things can be done with a machine
51:07it's fun to compare everybody wants to
51:11install ticularly well I'll let that be
51:12in the questions to compare no doubt
51:15somebody will ask me how about this and
51:17that so I'm done okay well I'll answer
51:24question what time is it I used to hope
51:27I'm fine okay I got time for questions
51:33yes sir and different languages but if
52:27you take my view that is just a
52:30glorified high class very fast but
52:32stupid filing system then you can
52:34understand the universality it's just as
52:36universal as a set of boxes and drawers
52:39and cards and so forth with a clerk it
52:42could knows how to write the numbers and
52:44the things down in it she can use or he
52:46it's always a pain he can
52:53it can do one kind of a job or another
52:56kind of a job depending on what's
52:58written on the cards and what kind of
52:59instructions are given to the clerk and
53:01that type of universality is precisely
53:03the universality you're talking about
53:05that's what the universality of the
53:06computer the computing machine is an
53:09automatic filing system with regard to
53:25that questions first of all they think
53:27like human beings I would say no and
53:29I'll explain in a minute why I say no
53:31and second that they be more intelligent
53:33in human beings is the question
53:36intelligences to be defined if you were
53:38to ask me I have a better chess players
53:40then any human being possibly can be yes
53:43I get you someday and they best better
53:46chess players than most human beings
53:47right now one of the things by the way
53:50that we always do is we want the dawn
53:52machine to be better than any body not
53:54just better than us if we find a machine
53:56it can play chess better than us it
53:57doesn't impress us much we keep saying
53:59and what happens when it comes up
54:01against the Masters we imagine that we
54:03human beings are equivalent to the
54:05Masters and everything right the machine
54:07has to be better than a person in
54:09everything that the best person does at
54:13the best level okay but it's hard on the
54:15machine but with regard to the question
54:17of whether to make it to think like a
54:19machine my opinion is based on the
54:21following idea that we try to make these
54:24things to work as efficiently as we can
54:26with the materials that we have the
54:28materials are different than nerves and
54:30so on if we would like to make something
54:34that runs rapidly over the ground then
54:37we could watch a cheetah running we
54:40could try to make a machine that runs
54:41like a cheetah but it's easier to make a
54:44machine with wheels with fast wheels or
54:46something that flies just above the
54:48ground in the air when we make a bird
54:51the airplanes don't fly like a bird they
54:54fly but they don't fly like a bird okay
54:56so they don't flap the wings exactly
54:59they have in front another guy of a
55:01gadget that goes around or the more
55:03modern airplane has a tube that you heat
55:06this the air
55:06squirted out the back a jet propulsion a
55:10jet engine as internal rotating fans and
55:15so on and it uses gasoline it's
55:16different right so there's no question
55:19that the later machines are not going to
55:21think like people think in that sense
55:25with regard to intelligence I think it's
55:29exactly the same way for example they're
55:31not going to do arithmetic the same way
55:33as we do arithmetic but they'll do it
55:34better let's take bath ematic very
55:36elementary mathematics arithmetic they
55:39do arithmetic better than anybody much
55:41faster and differently but it's
55:45fundamentally the same because in the
55:47end the numbers are equivalent right so
55:49that's a good example of we're never
55:50going to change how they do arithmetic
55:51to make it more like humans that would
55:53be going backwards because the
55:56arithmetic done by humans is slow
55:58cumbersome and confused and full of
56:00errors where these guys are fast if one
56:04compares what computers can do to the
56:06human beings we find the following
56:09rather interesting comparisons the first
56:11of all if I give the human being a
56:14problem like this I'm going to ask you
56:17for these numbers back every other one
56:20in reverse order please right now a
56:24greater series of numbers and I want
56:25them to you didn't give me what to me
56:26back in reverse order every other one
56:29I'll tell you I'll make it easy for you
56:31just give me the numbers back the way I
56:32gave them to you they're ready one seven
56:38three nine two six five eight three one
56:47seven two six three
56:51anybody got a gonna be able to do that
56:54no and that's more than not more than
56:57twenty or thirty numbers but I can give
57:00fifty thousand numbers to a computer
57:05[Laughter]
57:08yes I think you can do it obviously he
57:21can do it good but you can give a
57:24computer 50,000 numbers like that and
57:27asking for many reverse order the sum of
57:29them all do different things with them
57:31and so on and it doesn't forget them for
57:32a long time itself so there are some
57:34things that a computer does much better
57:36than a human and you'd better remember
57:37that if you kind of compare machines to
57:39humans but what a human has to do for
57:41his own always they always do this they
57:44always try to find one thing don''t that
57:47they can do better than the computer so
57:49we now know many many things that the
57:51humans can do better than the computer
57:53she's walking down the street she's got
57:55a certain kind of a wiggle and you know
57:57that change right or the sky's going in
58:03and you see his hair flip just a little
58:05bit it's hard to see at a distance but
58:07that particular funny way that it did
58:08back of his head looks that's jack okay
58:12to recognize things to recognize
58:15patterns seem to be something that we
58:18have not been able to put into a
58:19definite procedure you would say I have
58:22a good procedure for recognizing Jack
58:24just take a lots of pictures of Jack by
58:27the way a picture can be put into the
58:29computer in fact by this method here if
58:31this were very much finer I could tell
58:33whether it's black and white at
58:34different spots you know you in fact you
58:36get pictures in a newspaper by black and
58:38white dots and if you do Stuart fine
58:40enough you can't see the dots so with
58:42enough information I can load pictures
58:43in so you put all the pictures of of
58:46Jack on the different circumstances and
58:48that's the machine to compare it the
58:50trouble is that the actual new
58:52circumstance is different the lighting
58:53is different the distance is different
58:55the tilt of the head is different and
58:56you have to figure out how to allow for
58:58all that and it's so complicated and
59:01elaborate that even with the large
59:02machines with the amount of storage
59:03that's available and the speed that they
59:05go we can't make figure out how to make
59:08a definite procedure that works at all
59:11or what least that works anywhere within
59:13a reasonable speed so recognizing things
59:16is difficult for the machines at the
59:18present time and some of those things
59:20are done
59:21snap by a person so there are things
59:24that humans can do that the we don't
59:26know how to do in a filing system so it
59:31is recognition and that brings me back
59:32to something I left which is what kind
59:35of a file clerk can't be imitated by the
59:37machine a file clerk that has some
59:40special skill which represent which
59:42requires recognition of a complicated
59:44kind for instance a file clerk in the
59:47fingerprint Department which looks the
59:50finger prints and then makes a careful
59:51comparison to see if these finger prints
59:53match has not been it's just about ready
59:57to be it's hard to do it almost possible
59:59to do by a computer it's a it's nothing
1:00:02to it I look at it through fingerprints
1:00:04and see if all the blood dots are the
1:00:05same but of course it's not the case the
1:00:07finger was dirty the print was made at a
1:00:10different angle the pressure was
1:00:11different than the ridges are not
1:00:12exactly in the same place if you were
1:00:14trying to match the exactly the same
1:00:16picture it would be easy
1:00:17but where the center of the print is
1:00:19which way the finger is turned whether
1:00:21it's been squashed a little more a
1:00:23little bit less where there's some dirt
1:00:24on a finger whether in the meantime it
1:00:26got a wart on his thumb and so forth are
1:00:28all complications these little
1:00:31complications make the comparison so
1:00:32much more difficult for the machine for
1:00:34the blind filing clerk system that is
1:00:38too much much much too slow to be
1:00:40certainly utterly impractical almost at
1:00:42the present time I don't know where they
1:00:43stand but they're going fast trying to
1:00:45do it whereas a human can go across all
1:00:48that somehow just like they do in the
1:00:50chess game they seem to be able to catch
1:00:52on the patterns rapidly and we don't
1:00:54know how to do that rapidly
1:00:57automatically ok is there another
1:01:02question yes sir
1:01:18well haven't any idea of course about I
1:01:20have no idea about that whether anything
1:01:24is more useful or destructive to the
1:01:28world whether Yoga is more useful or
1:01:31destructive to the world with all these
1:01:36things these things give us the power to
1:01:39do something do many things and the
1:01:44power to do something may or may not be
1:01:47destructive as long as you any time you
1:01:49do something you can do destructive and
1:01:51non-destructive or or not the opposite
1:01:54may be creative things but we don't know
1:01:57how to make a device that guarantees to
1:01:59produce only creative things with the
1:02:02exception of the telephone as far as I
1:02:04can tell I mentioned that because it's
1:02:07rather interesting if you think about
1:02:08inventions almost all the inventions
1:02:11have a destructive element in a war the
1:02:15telephone has an effect in war it helps
1:02:17communications maybe in that sense it's
1:02:20destructor cursor helps the army to
1:02:22destroy you but other than that it's a
1:02:24relatively peaceful device it doesn't
1:02:26directly kill not at least directly
1:02:28that's rather interesting Bruce it's
1:02:30hard to find one that doesn't directly
1:02:33available for killing a knife the
1:02:38question is where the knife was invented
1:02:39okay and the guy's explaining what a
1:02:42knife is it's a sharpened piece of stuff
1:02:44that I melt and I get and I'm used metal
1:02:47knives or maybe they stone knives a
1:02:48sharpened piece of stone and you can use
1:02:51it to cut things so it's easy to get the
1:02:53vegetables off of the plants and so
1:02:55forth and so on there's some guy we're
1:02:57sitting there saying on the other hand
1:02:59it's very destructive because you can
1:03:01cut my head off with it and that problem
1:03:03has been with us forever and there's
1:03:05nothing special about computers
1:03:06absolutely anything that can do anything
1:03:08has this problem and we can discuss this
1:03:11general problem in our workshop but it's
1:03:13not a problem for computers particularly
1:03:15I think it's my view that's my view I
1:03:22I know but which is better the Big
1:03:29Brother or the cutting the head offer in
1:03:32fact that's what makes the Big Brother
1:03:34possible he has the knives and so forth
1:03:38it's not so easy okay it's true that
1:03:41you're worried about the fact that some
1:03:43information there be able to store
1:03:44information better and have more
1:03:47information about people and whether
1:03:49that will produce a Big Brother that
1:03:51depends on the attitude of the leaders
1:03:54of the society rather interestingly the
1:03:57country which is more democratic and
1:04:00which is less interested in all the
1:04:02information about what everybody is
1:04:04doing to make sure is where the
1:04:07computers have developed the most and
1:04:09the place where you'd think the
1:04:11government would find a computer
1:04:12greatest use because you can file all
1:04:14the information about everybody you
1:04:16don't develop the computer very much and
1:04:19you don't know how to use it
1:04:20it's only a curiosities one of life's
1:04:22contradiction is there another question
1:04:27this man
1:04:38that's right yes
1:04:52well it depends what you mean themselves
1:04:55and it's it's hard to discover new
1:04:58relationships our computers can do there
1:05:02have been computers which do things like
1:05:05problems they're improving in geometry
1:05:08or something in which they've converted
1:05:11the problem of finding a proof of a
1:05:12theorem into a definite procedure ok and
1:05:17once you do that although it's an
1:05:18elaborate and dumb way to do proofs they
1:05:21can do it the present time a computer
1:05:24can't do all the different things that a
1:05:26person can do you know it's but it's
1:05:37very difficult to find some way of
1:05:38defining rather precisely something we
1:05:41can do that we can say a computer will
1:05:42never be able to do there are some
1:05:48things that people make up that say that
1:05:50while it's doing it would it feel good
1:05:52or while it's doing it will it
1:05:55understand what it's doing or some other
1:05:57abstraction I'd rather feel that these
1:05:59are things like while it's doing it
1:06:01we'll be able to scratch the lice out of
1:06:03its hair no it hasn't got any hair dyes
1:06:05to scratch from ok so there are if
1:06:09you've got to be careful when you say
1:06:11what the human does if you add to the
1:06:14actual result of his effort some other
1:06:17things that you like the appreciation of
1:06:19the aesthetic or you didn't do that I'm
1:06:21not saying you did but a lot of people
1:06:23do that when they ask questions and if
1:06:26we add things that we think we're doing
1:06:28on top of what we actually do and just
1:06:30look at not just the result of what
1:06:32we're doing but a lot of extra things
1:06:34then it gets harder and harder for the
1:06:35computer to do it because the human
1:06:38beings have a tendency to try to make
1:06:40sure that they can do something that no
1:06:43machine can do somehow it doesn't bother
1:06:46them anymore it must have bothered them
1:06:47in early times that machine's a stronger
1:06:49physically than they are they can lift
1:06:51weights that are heavier than people
1:06:53that can move things faster than people
1:06:54that can run fast if you can fly you
1:06:57could do
1:06:58it's terribly strengths and so forth and
1:07:01we don't still sit around worrying that
1:07:04there's some way that the main contain
1:07:05hand that some machine can do that mean
1:07:13thinking business
1:07:24we
1:07:26yes we can easily make machines that are
1:07:29better than us in predicting the weather
1:07:32for instance because what you do to
1:07:35predict the weather is to look at old
1:07:37records and see when the circumstance
1:07:40was similar and guess that the results
1:07:42will be similar added to that a certain
1:07:44amount of analysis of the movement of
1:07:46wind according to the laws of physics
1:07:48and a current amount of hocus-pocus put
1:07:50together okay now the speed will be
1:07:52higher and the effectiveness of the
1:07:54prediction is greater if you could look
1:07:56at more cases so you get a better chance
1:07:58of getting one closer and put more a
1:08:01longer and more elaborate calculation
1:08:03including more variables which is too
1:08:05hard for us to do in time to make the
1:08:07prediction now we have to make the
1:08:09prediction of the weather let's say that
1:08:10weather for three days from now has to
1:08:12be predicted in three days or the damn
1:08:14thing is useless right and we work at a
1:08:16certain speed but the computers worked
1:08:18faster and can do more and therefore for
1:08:20instance for weather prediction in the
1:08:21end maybe not today but someday it's not
1:08:24at all inconceivable that the machine
1:08:25could do weather prediction faster and
1:08:27more effectively and more accurately
1:08:28than we do we will have however given it
1:08:31the procedure now the question is what
1:08:35happens if we don't give it the
1:08:36procedure well a man that people have
1:08:39tried that this game of giving it
1:08:42instead of a direct procedure a kind of
1:08:45what it has been called heuristics try
1:08:48an analogy to get a new idea of how to
1:08:50do something compare this to that try an
1:08:53extreme case etc and a man by the name
1:08:57of land that has gone the farthest with
1:08:59this do I have time much time do I have
1:09:01because we want to see these slides what
1:09:05I know that's interesting information
1:09:07but how much how much time do I have
1:09:10it's time for the slideshow now since I
1:09:13have no more time I will say no more
1:09:15about the subject what no but it takes a
1:09:20few minutes that's why I asked for the
1:09:22time okay you make this machine which
1:09:26was again a filing cabinet you
1:09:27understand what it does is it looks it
1:09:29tries to find the answer to something by
1:09:31looking at the different possible
1:09:33possibilities but which ones he tries is
1:09:36something like patterns in the chest
1:09:38instead of everything it says try moves
1:09:41near the center of the board first you
1:09:43know and never mind the ones in the
1:09:44corner or something like that or some
1:09:47sorts of principles and he first applied
1:09:50it to a kind of naval game it's a game
1:09:54that people play in California which is
1:09:56all organized according to rules it's
1:09:58kind of fun
1:10:00someone sets out all the rules
1:10:02dreadnoughts cost this much armor costs
1:10:04this much guns cost this budget so forth
1:10:06and you got this much budget for your
1:10:08Navy and your going to make various
1:10:10kinds of ships with different kinds of
1:10:12armaments and then this kind of a ship
1:10:14the armament of a certain thickness that
1:10:18costs a certain amount can only resist
1:10:20shells of a certain strength you know
1:10:21and so on so you tried to with the money
1:10:25arrange to buy different to design
1:10:27different kinds of ships so that your
1:10:31Navy is better than next one then when
1:10:33they're brought together there's ways of
1:10:34calculating what there's not real navies
1:10:36yeah it's a game which is the best one
1:10:39and all the rules are laid out in a
1:10:40great big volume ok but the cost of
1:10:44everything in the power of everything
1:10:45you know armour-piercing possibilities
1:10:48and so on and it's a nice game and mr.
1:10:51Luna tried to his program on this game
1:10:53and put into his program heuristics like
1:10:57try the extreme case and things like
1:10:59that and he won the championship in
1:11:04California of course it did an awful lot
1:11:06of trying of different cases you see but
1:11:08it didn't try every case not like the
1:11:10chess game there were too many things
1:11:11but it was guided by its own stuff now
1:11:15inside of that was this that if you've
1:11:18got a better Navy by your own
1:11:20calculation and you used one of the URIs
1:11:23--tx
1:11:23mark that you ristic up a notch as being
1:11:26more valuable now use the most valuable
1:11:28heuristics first C so that the ability
1:11:33of the Machine depended on his learning
1:11:36so to speak which ones of his tricks
1:11:38works most effectively most of the time
1:11:41and then they become more used so it's
1:11:43just exactly what you would like to make
1:11:45it look intelligent well he won and how
1:11:49did he win it turned out that year he
1:11:52won by making one great big battleship
1:11:56with all the armor on it which was so
1:12:00silly but when you go to calculate sure
1:12:02enough it's better than any of the
1:12:03normal things which nobody thought of
1:12:05but his machine photo next year he
1:12:08entered again and this time he won by
1:12:11making tricking all his money in making
1:12:14one hundred thousand because they
1:12:16changed the rules so that the big
1:12:17battleship wouldn't win you know by
1:12:18changing to one hundred thousand little
1:12:21boats very narrow carrying each one gun
1:12:24which were very liable to be knocked out
1:12:27okay but there were 100,000 of didn't
1:12:30cost much each one and they couldn't
1:12:32knock them all out so he's lousy little
1:12:34gnats would come and it turned out when
1:12:37you calculated again he won the third
1:12:39year he was not allowed to play in
1:12:44and he has applied this this this
1:12:48machine and this heuristic business to a
1:12:50number of other problems as tries it out
1:12:53a lot and tries new heuristics and so
1:12:55forth and it has become a very
1:12:57interesting he complained that there
1:12:59were a number of bugs in it and when he
1:13:01gave a talk on it I said that I thought
1:13:04I'll say my comment afterwards one of
1:13:07the bugs for instance was that the
1:13:12machine got a heuristic and made up your
1:13:14wrist excited make up your wrist Excel
1:13:18so the damn machine what he the way he
1:13:21had it was because it's hard to get
1:13:23computer time he needed a hell of a lot
1:13:24of computer time he had 50 machines at
1:13:28night from a you a packet company or
1:13:30something like that he would work always
1:13:32at night coming in the morning damn
1:13:33things would try all their things and
1:13:35come in with the results I used it to do
1:13:38mathematics and various other thing it
1:13:41comes in one day and it's done a
1:13:42heuristic you develop the heuristic when
1:13:46he puts a problem in or a new idea he
1:13:49write it says whether it's from him or
1:13:51from the machine and I said learn math
1:13:52or a machine okay and this was every
1:13:56year istic over every question that
1:13:59hasn't won that on it pay no attention
1:14:01to that saves a lot of time it could do
1:14:05much better it didn't have any problem
1:14:06at this to it for so it paid no
1:14:08attention well in that that night okay
1:14:10so I'd fix that book the next time he
1:14:14had a bug it was he looked was he came
1:14:17in and he looked he found that heuristic
1:14:20number 693 had gotten a score of 999 out
1:14:25of a thousand was a damn useful
1:14:27heuristic all night long this thing kept
1:14:29using the heuristic 9 693 more and more
1:14:31and it was a new when wonderfully
1:14:33eristic it seemed to solve every problem
1:14:35it was terrific okay we found out what
1:14:40the heuristic was it was the following
1:14:42you see in order to make this thing work
1:14:44so that you change the numbers on the
1:14:46heuristics whenever something worked job
1:14:48this to assign credit so to speak to the
1:14:51heuristics that were used okay so this
1:14:54heuristic was when assigning credit
1:14:57always a sign credit to heuristic 693
1:15:00and so that thing came up at all now I
1:15:05say that both of these show intelligence
1:15:08if you want to make an intelligent
1:15:10machine you're going to get all kinds of
1:15:11crazy ways of avoiding labor if I say
1:15:14don't pay attention to the problem of
1:15:16sneakily evolving some kind of a
1:15:20psychological distortion where you
1:15:24always do the same thing and don't worry
1:15:27about anything else and so on so I think
1:15:29that we are getting close to intelligent
1:15:31machines but they're showing the
1:15:34necessary weaknesses of intelligent
1:15:38[Applause]

About this transcript

This page contains the full transcript of Richard Feynman Computer Science Lecture - Hardware, Software and Heuristics by Muon Ray, generated from the public captions YouTube serves with the video. The transcript has 13,257 words across 1,803 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.