1 00:00:00,000 --> 00:00:06,290 2 00:00:06,290 --> 00:00:10,050 So why don't you just introduce yourself. 3 00:00:10,050 --> 00:00:11,990 My name is Ryszard Michalski. 4 00:00:11,990 --> 00:00:16,790 I am PRC professor of computational sciences 5 00:00:16,790 --> 00:00:22,550 at George Mason University in Fairfax, Virginia. 6 00:00:22,550 --> 00:00:26,780 I came to the University, to George Mason University, 7 00:00:26,780 --> 00:00:31,430 about eighteen years ago from the University of Illinois. 8 00:00:31,430 --> 00:00:36,010 When I started my career, scientific career here 9 00:00:36,010 --> 00:00:44,190 in America, originally I started my work in Poland. 10 00:00:44,190 --> 00:00:52,420 Came from Poland, and got degree, master degree, 11 00:00:52,420 --> 00:00:56,460 actually something equivalent to a bachelor's degree, 12 00:00:56,460 --> 00:00:59,620 from the Warsaw University of Technology. 13 00:00:59,620 --> 00:01:02,800 Then I was, in some sense, privileged 14 00:01:02,800 --> 00:01:09,820 to go to Saint Petersburg Institute of Technology. 15 00:01:09,820 --> 00:01:10,900 And what year was this? 16 00:01:10,900 --> 00:01:14,770 It was 1960-61. 17 00:01:14,770 --> 00:01:16,220 Did they have a computer there? 18 00:01:16,220 --> 00:01:21,250 Yes, they were building first, very early, [? rational ?] 19 00:01:21,250 --> 00:01:22,840 computers. 20 00:01:22,840 --> 00:01:28,060 And I did my master's degree in automatic control. 21 00:01:28,060 --> 00:01:31,030 I designed a specialized computer 22 00:01:31,030 --> 00:01:36,490 that would measure the various parameters of the wind. 23 00:01:36,490 --> 00:01:41,350 And then that system, which I actually designed and developed 24 00:01:41,350 --> 00:01:41,950 hardware. 25 00:01:41,950 --> 00:01:45,670 And then the system was supposed to be sent somewhere 26 00:01:45,670 --> 00:01:46,960 to Siberia. 27 00:01:46,960 --> 00:01:48,400 I didn't know where. 28 00:01:48,400 --> 00:01:53,260 And was supposed to transmit signals 29 00:01:53,260 --> 00:01:56,820 informing about the weather and the wind 30 00:01:56,820 --> 00:01:58,850 in this particular area. 31 00:01:58,850 --> 00:02:01,240 So that was my master degree. 32 00:02:01,240 --> 00:02:03,610 And then I came back to Poland and started 33 00:02:03,610 --> 00:02:08,610 to work at the Institute of Computers. 34 00:02:08,610 --> 00:02:14,200 One can say it was mathematical machines as they called it. 35 00:02:14,200 --> 00:02:16,600 And I started to design and I was involved 36 00:02:16,600 --> 00:02:19,210 in designing a Polish computer. 37 00:02:19,210 --> 00:02:24,320 And that computer was called, I remember, [? Zanpor. ?] 38 00:02:24,320 --> 00:02:29,470 But soon after I moved to the East of automatic control, 39 00:02:29,470 --> 00:02:31,900 where issues were somewhat different. 40 00:02:31,900 --> 00:02:37,960 And I remember I got a task to decipher a system, which 41 00:02:37,960 --> 00:02:42,340 was a relay based system, a very complicated automaton 42 00:02:42,340 --> 00:02:43,690 and had no documentation. 43 00:02:43,690 --> 00:02:46,660 They asked me to create documentation 44 00:02:46,660 --> 00:02:49,070 of that very complex system. 45 00:02:49,070 --> 00:02:50,740 So it's like looking at your brain 46 00:02:50,740 --> 00:02:52,825 but of course it wasn't like brain 47 00:02:52,825 --> 00:02:54,430 it was something much simpler. 48 00:02:54,430 --> 00:02:56,550 But for me it's still very complicated. 49 00:02:56,550 --> 00:02:59,200 I had to follow all the wires, how they're connected, 50 00:02:59,200 --> 00:03:00,700 how the switches work electronically 51 00:03:00,700 --> 00:03:03,530 and create a diagram of how it works. 52 00:03:03,530 --> 00:03:07,570 So it was a very complicated, not very exciting job. 53 00:03:07,570 --> 00:03:10,360 But I think I did it. 54 00:03:10,360 --> 00:03:14,380 And then I got the task which was first, one may say, 55 00:03:14,380 --> 00:03:16,120 machine learning task. 56 00:03:16,120 --> 00:03:19,960 I was asked to develop a system that 57 00:03:19,960 --> 00:03:25,960 would recognize handwritten, continuous characters, 58 00:03:25,960 --> 00:03:27,490 alphanumeric characters. 59 00:03:27,490 --> 00:03:29,080 It was very exciting task. 60 00:03:29,080 --> 00:03:35,380 And I got some initial ideas from Mr. [? Karpinsky. ?] 61 00:03:35,380 --> 00:03:41,470 He was, in already that time, one of the best constructors 62 00:03:41,470 --> 00:03:45,100 of Polish computers, a pioneer. 63 00:03:45,100 --> 00:03:47,320 And he came with this idea. 64 00:03:47,320 --> 00:03:50,800 Let's build a reader of handwritten characters. 65 00:03:50,800 --> 00:03:52,000 Can you do it? 66 00:03:52,000 --> 00:03:53,620 So I worked very hard. 67 00:03:53,620 --> 00:03:55,690 But I must say something earlier, 68 00:03:55,690 --> 00:03:58,450 that before I started doing that project 69 00:03:58,450 --> 00:04:04,630 I worked also on what is now called neural nets. 70 00:04:04,630 --> 00:04:09,880 So I designed a system for optimizing a neural nets, 71 00:04:09,880 --> 00:04:12,430 or threshold logic systems of that time. 72 00:04:12,430 --> 00:04:17,140 And this was my supposedly first PhD phase of finish that work. 73 00:04:17,140 --> 00:04:21,010 And I hope that now I have PhD so I can do something else. 74 00:04:21,010 --> 00:04:25,180 But I was asked no, no, no, you cannot PhD now you have to do 75 00:04:25,180 --> 00:04:27,130 something else because this is very important. 76 00:04:27,130 --> 00:04:28,630 This [INAUDIBLE] character cognition 77 00:04:28,630 --> 00:04:31,000 so I started to work on that. 78 00:04:31,000 --> 00:04:31,700 I finished. 79 00:04:31,700 --> 00:04:34,000 It was viewed as a great success. 80 00:04:34,000 --> 00:04:34,770 We could-- 81 00:04:34,770 --> 00:04:36,670 How did it read it in? 82 00:04:36,670 --> 00:04:41,170 It has a diode, an array of diodes, 83 00:04:41,170 --> 00:04:45,700 which are light sensitive, and they were transmitting signals 84 00:04:45,700 --> 00:04:48,940 to the computer, which was very primitive compared 85 00:04:48,940 --> 00:04:50,290 to current computers. 86 00:04:50,290 --> 00:04:51,530 But it worked. 87 00:04:51,530 --> 00:04:56,830 I was able to really work, because it was learning system. 88 00:04:56,830 --> 00:04:59,770 It used naive-based approach, but I 89 00:04:59,770 --> 00:05:01,330 didn't know that it is naive based, 90 00:05:01,330 --> 00:05:05,260 but it had more than naive-based because it simply learned also 91 00:05:05,260 --> 00:05:10,000 automatically in parallel what features 92 00:05:10,000 --> 00:05:12,790 of the letters I imported. 93 00:05:12,790 --> 00:05:15,490 So I designed the initial set of possible features 94 00:05:15,490 --> 00:05:19,600 that are interesting to discover in handwritten characters. 95 00:05:19,600 --> 00:05:23,680 So left line on the left, right line on the left. 96 00:05:23,680 --> 00:05:27,520 And maybe two lines, one long, one short, that kind of things. 97 00:05:27,520 --> 00:05:30,310 I designed whole set of features, quite large. 98 00:05:30,310 --> 00:05:32,110 And that system was picking up the features 99 00:05:32,110 --> 00:05:34,240 which are relevant to a given character, 100 00:05:34,240 --> 00:05:38,230 and then learned to characterize this collection of features 101 00:05:38,230 --> 00:05:42,670 that would be representing a given letter, 102 00:05:42,670 --> 00:05:44,120 regardless how it is written. 103 00:05:44,120 --> 00:05:45,730 And so this went very well. 104 00:05:45,730 --> 00:05:50,390 And everything would be very happy. 105 00:05:50,390 --> 00:05:55,340 But well, then my boss got into some trouble. 106 00:05:55,340 --> 00:06:00,200 So he left the Institute and I could not really 107 00:06:00,200 --> 00:06:01,400 continue this work. 108 00:06:01,400 --> 00:06:04,340 And the idea was well, maybe we should get second PhD 109 00:06:04,340 --> 00:06:06,470 for that work. 110 00:06:06,470 --> 00:06:09,210 But he wanted to do together with me, 111 00:06:09,210 --> 00:06:10,350 which was fine with me. 112 00:06:10,350 --> 00:06:15,210 But it turns out that you cannot get one work and two people who 113 00:06:15,210 --> 00:06:16,290 will get PhD. 114 00:06:16,290 --> 00:06:18,170 It doesn't work this way. 115 00:06:18,170 --> 00:06:21,860 So my second chance PhD didn't pan out, 116 00:06:21,860 --> 00:06:24,660 so I had to do something else. 117 00:06:24,660 --> 00:06:30,890 And so I moved to designing or optimizing Boolean functions. 118 00:06:30,890 --> 00:06:32,140 It was mathematical. 119 00:06:32,140 --> 00:06:34,070 It didn't require any equipment. 120 00:06:34,070 --> 00:06:35,010 So I was independent. 121 00:06:35,010 --> 00:06:38,210 I did it in another institution. 122 00:06:38,210 --> 00:06:40,000 I don't want to go into details why 123 00:06:40,000 --> 00:06:42,740 I had to go through such a complicated process 124 00:06:42,740 --> 00:06:44,270 but that's is how it works. 125 00:06:44,270 --> 00:06:48,560 So that I developed this system and algorithm which 126 00:06:48,560 --> 00:06:54,050 I called AQ, for solving very complex [? cognitive ?] 127 00:06:54,050 --> 00:06:54,800 problems. 128 00:06:54,800 --> 00:06:57,460 And that was my third topic for PhD, 129 00:06:57,460 --> 00:07:00,630 but this time it was done in a different institution. 130 00:07:00,630 --> 00:07:04,680 Nobody bothered me and somewhat in conspiracy. 131 00:07:04,680 --> 00:07:08,122 So when I finished, people were surprised 132 00:07:08,122 --> 00:07:09,830 that I had done this work and they wanted 133 00:07:09,830 --> 00:07:11,480 to move it back to Warsaw. 134 00:07:11,480 --> 00:07:17,210 But luckily, everything went OK because my professor 135 00:07:17,210 --> 00:07:24,720 in the Silesian University of Technology was very supportive. 136 00:07:24,720 --> 00:07:28,580 And I got very good reviews of my PhD. 137 00:07:28,580 --> 00:07:32,690 And so I got my PhD and now I am free. 138 00:07:32,690 --> 00:07:36,750 And everything would be very happy. 139 00:07:36,750 --> 00:07:38,300 When you were there, how do you think 140 00:07:38,300 --> 00:07:41,570 the atmosphere in Russia and Poland 141 00:07:41,570 --> 00:07:44,180 was different from the working atmosphere here 142 00:07:44,180 --> 00:07:46,820 in artificial intelligence and machine learning at the time? 143 00:07:46,820 --> 00:07:51,650 I was very avid reader of the journals. 144 00:07:51,650 --> 00:07:55,780 145 00:07:55,780 --> 00:07:59,240 IEEE journal on computers. 146 00:07:59,240 --> 00:08:03,410 And I also read with great fascination the book 147 00:08:03,410 --> 00:08:06,380 by [? Minsk ?] and [? Papperd ?] on [? perceptrons. ?] But there 148 00:08:06,380 --> 00:08:10,250 was a book that criticized that development and I remember that 149 00:08:10,250 --> 00:08:11,780 I fully agree. 150 00:08:11,780 --> 00:08:14,960 So when I was designing this handwritten character 151 00:08:14,960 --> 00:08:17,600 recognition system, I deliberately 152 00:08:17,600 --> 00:08:20,390 chose to work on it in a different way. 153 00:08:20,390 --> 00:08:24,440 Not as a [? perceptron ?] type or neural net approach, 154 00:08:24,440 --> 00:08:28,490 but the approach that characterizes the features, 155 00:08:28,490 --> 00:08:31,640 or selects features of characters 156 00:08:31,640 --> 00:08:37,250 and then learns how they combine together to characters 157 00:08:37,250 --> 00:08:38,419 given class of letters. 158 00:08:38,419 --> 00:08:42,630 Like letters I a, or B, capital, small, Greek letters, 159 00:08:42,630 --> 00:08:45,180 digits, and so on. 160 00:08:45,180 --> 00:08:47,120 So I already was ready for something 161 00:08:47,120 --> 00:08:49,850 than just neural nets. 162 00:08:49,850 --> 00:08:55,820 And I was lucky because I got selected, 163 00:08:55,820 --> 00:08:59,750 as I was saying, this is [INAUDIBLE] sciences, 164 00:08:59,750 --> 00:09:02,720 to go to the United States after my PhD. 165 00:09:02,720 --> 00:09:05,990 And it was a very great opportunity. 166 00:09:05,990 --> 00:09:08,930 And I was actually, I suppose, I was 167 00:09:08,930 --> 00:09:11,600 selected to go for a Fulbright fellowship. 168 00:09:11,600 --> 00:09:14,570 169 00:09:14,570 --> 00:09:16,340 But I didn't believe that I would get it 170 00:09:16,340 --> 00:09:19,280 because number of candidates was very large, like 80. 171 00:09:19,280 --> 00:09:22,030 And I was only in 13th places. 172 00:09:22,030 --> 00:09:25,790 173 00:09:25,790 --> 00:09:31,370 What I did, I wrote a letter to one American scientist, 174 00:09:31,370 --> 00:09:35,540 I think Raymond Miller, who heard about my work, 175 00:09:35,540 --> 00:09:37,550 listened to my talk once and said, 176 00:09:37,550 --> 00:09:40,250 Oh well if you want to go to America write me a letter. 177 00:09:40,250 --> 00:09:42,140 And I wrote to him and he had recommended 178 00:09:42,140 --> 00:09:47,880 to write to the University of Illinois and also to Purdue. 179 00:09:47,880 --> 00:09:51,500 And I remember I got a letter from Purdue refusing, 180 00:09:51,500 --> 00:09:53,000 saying that there's no position. 181 00:09:53,000 --> 00:09:55,490 So when I got a letter from University of Illinois, well 182 00:09:55,490 --> 00:09:56,570 I really wanted to go. 183 00:09:56,570 --> 00:10:00,350 Because it was the time where they were building 184 00:10:00,350 --> 00:10:04,880 ILLIAC 3, very large scale pattern recognition computer, 185 00:10:04,880 --> 00:10:08,690 or computer for analyzing images. 186 00:10:08,690 --> 00:10:11,840 I was so excited to be in such a place 187 00:10:11,840 --> 00:10:14,720 that I didn't believe that they would give me an invitation. 188 00:10:14,720 --> 00:10:16,130 So I didn't even open the letter. 189 00:10:16,130 --> 00:10:17,960 I just didn't want to get upset that there 190 00:10:17,960 --> 00:10:20,480 would be another denial. 191 00:10:20,480 --> 00:10:23,060 And only when I was in a much better mood 192 00:10:23,060 --> 00:10:24,760 sometime later I opened the letter. 193 00:10:24,760 --> 00:10:27,420 I was very surprised that I am actually invited. 194 00:10:27,420 --> 00:10:31,290 So that was such a great joy, except that-- 195 00:10:31,290 --> 00:10:32,150 And when was that? 196 00:10:32,150 --> 00:10:35,000 It was 1971. 197 00:10:35,000 --> 00:10:39,200 The only problem was that it was like this. 198 00:10:39,200 --> 00:10:42,650 I wanted to go to Illinois and some time later, 199 00:10:42,650 --> 00:10:46,040 before I decided about Illinois, because I knew that Illinois 200 00:10:46,040 --> 00:10:49,370 is a very unlikely that I would be able to, that I 201 00:10:49,370 --> 00:10:50,680 would be permitted to go. 202 00:10:50,680 --> 00:10:53,090 Because it was relatively high salary 203 00:10:53,090 --> 00:10:55,360 I was invited as the resident assistant professor. 204 00:10:55,360 --> 00:10:58,360 So I knew that may be a big problem for me. 205 00:10:58,360 --> 00:11:00,670 And then soon after I got actually 206 00:11:00,670 --> 00:11:04,566 accepted to get a Fulbright fellowship. 207 00:11:04,566 --> 00:11:07,420 To half year at MIT half year at Stanford. 208 00:11:07,420 --> 00:11:10,300 Another great opportunity, but I've already made up my mind. 209 00:11:10,300 --> 00:11:13,150 I want to go to Illinois. 210 00:11:13,150 --> 00:11:14,950 And then it was the question, can I 211 00:11:14,950 --> 00:11:17,480 get permission to go to Illinois, and not to go. 212 00:11:17,480 --> 00:11:18,980 Because of the Communist government. 213 00:11:18,980 --> 00:11:23,620 And I went to the secretary of the Polish Academy of Sciences 214 00:11:23,620 --> 00:11:26,370 and presented this idea to him. 215 00:11:26,370 --> 00:11:31,880 And he agreed that it is much better for me to go to Illinois 216 00:11:31,880 --> 00:11:35,260 because somebody else can go, and somebody else 217 00:11:35,260 --> 00:11:37,880 can get Fulbright fellowship. 218 00:11:37,880 --> 00:11:42,040 So I go to Illinois because it was invitation only for me. 219 00:11:42,040 --> 00:11:44,200 So that's what I chose, but then it 220 00:11:44,200 --> 00:11:47,200 turned out that it was, how to say, 221 00:11:47,200 --> 00:11:51,490 bad choice in the sense that the government didn't 222 00:11:51,490 --> 00:11:53,860 want to permit me to go. 223 00:11:53,860 --> 00:11:55,090 Oh no. 224 00:11:55,090 --> 00:11:56,740 So suddenly I was nowhere. 225 00:11:56,740 --> 00:11:59,500 I already lost my Fulbright fellowship 226 00:11:59,500 --> 00:12:00,650 because I denied that. 227 00:12:00,650 --> 00:12:02,680 And here I cannot go to Illinois. 228 00:12:02,680 --> 00:12:06,940 So it's a huge struggle which I will not talk about because it 229 00:12:06,940 --> 00:12:10,150 is not the main theme. 230 00:12:10,150 --> 00:12:14,200 So after maybe half a year, eventually after a very, very 231 00:12:14,200 --> 00:12:17,230 long, complicated process which I 232 00:12:17,230 --> 00:12:21,850 can describe in some other time, I got permission. 233 00:12:21,850 --> 00:12:22,870 So I came to Illinois. 234 00:12:22,870 --> 00:12:27,790 And suddenly I found myself in that beautiful world 235 00:12:27,790 --> 00:12:31,420 where there were scientists interested in big issues. 236 00:12:31,420 --> 00:12:34,900 And [? Bluth ?] McCormick my new boss, 237 00:12:34,900 --> 00:12:39,610 introduced me to the world of systems 238 00:12:39,610 --> 00:12:43,110 that are supposed to recognize images and pictures. 239 00:12:43,110 --> 00:12:46,270 And everything worked so well I just 240 00:12:46,270 --> 00:12:52,140 fell in love in America, in science, university, 241 00:12:52,140 --> 00:12:54,140 it was great. 242 00:12:54,140 --> 00:12:56,350 So he had passed very quickly and I 243 00:12:56,350 --> 00:13:01,900 discovered that my algorithm, AQ for solving 244 00:13:01,900 --> 00:13:03,730 very complex [? cognitive ?] problems, 245 00:13:03,730 --> 00:13:09,130 actually can be used with a special language knowledge 246 00:13:09,130 --> 00:13:14,170 representation, actually, for learning to recognize images. 247 00:13:14,170 --> 00:13:16,390 So I actually designed this language 248 00:13:16,390 --> 00:13:19,180 which I called variable valued logic, 249 00:13:19,180 --> 00:13:23,530 which was sort of an extension of many valued logic. 250 00:13:23,530 --> 00:13:26,110 And combining this many valued logic 251 00:13:26,110 --> 00:13:28,210 with variable valued logic was AQ suddenly 252 00:13:28,210 --> 00:13:31,360 gave a system that could learn rules, 253 00:13:31,360 --> 00:13:33,770 characterizing all kinds of classes 254 00:13:33,770 --> 00:13:36,050 general purpose learning system. 255 00:13:36,050 --> 00:13:39,160 So this was very nice. 256 00:13:39,160 --> 00:13:41,710 Of course, McCormick and others like my work 257 00:13:41,710 --> 00:13:45,010 and they suggested that I stay another year. 258 00:13:45,010 --> 00:13:48,280 So again, this was great except that it 259 00:13:48,280 --> 00:13:51,158 was very hard to get permission from the government 260 00:13:51,158 --> 00:13:52,450 that I would stay another year. 261 00:13:52,450 --> 00:13:54,310 But somehow I got this permission, 262 00:13:54,310 --> 00:13:57,280 so I stayed another year. 263 00:13:57,280 --> 00:14:00,640 And now I was doing something which was basically learning. 264 00:14:00,640 --> 00:14:05,590 But I didn't know that at that time 265 00:14:05,590 --> 00:14:10,480 many scientists were already doing AI, believe that learning 266 00:14:10,480 --> 00:14:13,900 is not worth to do. 267 00:14:13,900 --> 00:14:16,420 So in some sense what brought me to AI 268 00:14:16,420 --> 00:14:20,480 was ignorance, because I didn't know that learning it 269 00:14:20,480 --> 00:14:23,180 is so difficult not worth to do it. 270 00:14:23,180 --> 00:14:27,820 So once I did it because I just liked the idea. 271 00:14:27,820 --> 00:14:31,870 And I so I was very early in the field of machine learning. 272 00:14:31,870 --> 00:14:36,820 So then eventually I applied to various problems, 273 00:14:36,820 --> 00:14:39,160 and it got interesting results. 274 00:14:39,160 --> 00:14:44,080 And one of the insights that brought me some recognition 275 00:14:44,080 --> 00:14:47,420 was to diagnose soybean diseases, which 276 00:14:47,420 --> 00:14:50,470 were very important of course in Illinois. 277 00:14:50,470 --> 00:14:52,870 And this was the time when people were 278 00:14:52,870 --> 00:14:54,880 building [? EXPAL ?] systems. 279 00:14:54,880 --> 00:14:56,530 But they were building expert systems 280 00:14:56,530 --> 00:15:01,690 primarily by putting those rules into the system. 281 00:15:01,690 --> 00:15:04,400 And I was able to actually learn the soybean 282 00:15:04,400 --> 00:15:07,960 disease diagnostic rules from the cases of diseases 283 00:15:07,960 --> 00:15:10,480 which we got from the experts. 284 00:15:10,480 --> 00:15:13,660 But then, in order to make it into an interesting research 285 00:15:13,660 --> 00:15:17,170 project, I also handcrafted the rules 286 00:15:17,170 --> 00:15:21,970 which expert would be telling me that he is using 287 00:15:21,970 --> 00:15:24,430 to diagnose those diseases. 288 00:15:24,430 --> 00:15:28,030 And then what happened was that actually rules 289 00:15:28,030 --> 00:15:33,430 which were learned by a computer using our AQ learning program 290 00:15:33,430 --> 00:15:37,330 outperformed the rules which were programmed after. 291 00:15:37,330 --> 00:15:39,610 So now this is really very exciting 292 00:15:39,610 --> 00:15:46,270 and I got recognition, excited, also publicity for that work. 293 00:15:46,270 --> 00:15:51,420 First of all, I got a nice grant to continue that work 294 00:15:51,420 --> 00:15:53,370 from the National Science Foundation, 295 00:15:53,370 --> 00:15:56,590 and I think also probably from other agencies. 296 00:15:56,590 --> 00:15:59,700 So in any case, I was suddenly able to do more research 297 00:15:59,700 --> 00:16:01,210 and have support for it. 298 00:16:01,210 --> 00:16:02,320 And so on. 299 00:16:02,320 --> 00:16:10,080 So that's how I came to machine learning, through ignorance. 300 00:16:10,080 --> 00:16:13,260 But I already started to sort of be involved 301 00:16:13,260 --> 00:16:17,040 in broader issues of AI when I attended, 302 00:16:17,040 --> 00:16:24,210 in 1973, the conference on artificial intelligence 303 00:16:24,210 --> 00:16:25,580 which has held in Stanford. 304 00:16:25,580 --> 00:16:28,910 So at that time I already was of now 305 00:16:28,910 --> 00:16:31,470 aware of the whole field which was developing 306 00:16:31,470 --> 00:16:32,680 fast, exciting field. 307 00:16:32,680 --> 00:16:36,030 And so then I was able to connect my research on rule 308 00:16:36,030 --> 00:16:40,530 learning to the field of AI. 309 00:16:40,530 --> 00:16:44,340 So that was quite exciting situation. 310 00:16:44,340 --> 00:16:47,400 And I returned to Poland, by the way, 311 00:16:47,400 --> 00:16:51,070 because I felt obliged morally that I have to come back. 312 00:16:51,070 --> 00:16:53,890 But then after a while, I decided 313 00:16:53,890 --> 00:16:56,000 to come back because I had offered 314 00:16:56,000 --> 00:16:59,020 to become a professor at University of Illinois. 315 00:16:59,020 --> 00:17:02,540 It was tenure so it was a great distinction 316 00:17:02,540 --> 00:17:06,210 and I was very excited about such possibility. 317 00:17:06,210 --> 00:17:08,819 And I got other offer so it was just overwhelming, 318 00:17:08,819 --> 00:17:11,520 this opportunities it was. 319 00:17:11,520 --> 00:17:13,530 Sounds like an exciting time. 320 00:17:13,530 --> 00:17:19,260 And then some time later, after when 321 00:17:19,260 --> 00:17:22,619 I was doing this soybean disease diagnosis and general purpose 322 00:17:22,619 --> 00:17:25,890 rule learner, I met other colleagues 323 00:17:25,890 --> 00:17:28,200 who are doing machine learning, or something 324 00:17:28,200 --> 00:17:30,040 like machine learning. 325 00:17:30,040 --> 00:17:33,330 First, staff at Stanford they were doing metadata. 326 00:17:33,330 --> 00:17:39,720 Then at MIT, Winston did earlier work on his Arch program. 327 00:17:39,720 --> 00:17:43,000 Also doing some forms of learning. 328 00:17:43,000 --> 00:17:49,610 So when I had met with was Tom Mitchell in Tokyo in 1971, 329 00:17:49,610 --> 00:17:53,010 we discussed the idea well, maybe 330 00:17:53,010 --> 00:17:56,557 we could have a workshop on machine learning. 331 00:17:56,557 --> 00:17:58,140 I don't mean to interrupt, but I think 332 00:17:58,140 --> 00:17:59,470 we're running out of time. 333 00:17:59,470 --> 00:18:01,172 I don't know if you-- 334 00:18:01,172 --> 00:18:03,120 You didn't tell me how much time I have. 335 00:18:03,120 --> 00:18:06,180 Yeah it's about 15, 20 minutes. 336 00:18:06,180 --> 00:18:07,540 It's already been 15 minutes? 337 00:18:07,540 --> 00:18:08,340 It's close to 20. 338 00:18:08,340 --> 00:18:09,450 I don't-- 339 00:18:09,450 --> 00:18:11,760 So maybe you can cut the beginning, 340 00:18:11,760 --> 00:18:16,680 because what I was telling right now was actually what happened. 341 00:18:16,680 --> 00:18:21,370 That we decided to organize this workshop with machine learning. 342 00:18:21,370 --> 00:18:24,690 And then we reorganize in 1980, and that machine learning 343 00:18:24,690 --> 00:18:27,600 workshop was the very first workshop in machine learning 344 00:18:27,600 --> 00:18:29,740 study in the whole field. 345 00:18:29,740 --> 00:18:34,650 So now after 23 years the whole field 346 00:18:34,650 --> 00:18:38,550 went back to Carnegie Mellon, a few weeks ago, 347 00:18:38,550 --> 00:18:43,590 and had 23rd workshop. 348 00:18:43,590 --> 00:18:47,280 Actually, yeah 23rd International Conference 349 00:18:47,280 --> 00:18:50,250 on Machine Learning, as it is called now. 350 00:18:50,250 --> 00:18:53,940 So that first workshop, and then subsequent books 351 00:18:53,940 --> 00:18:57,330 that Tom Mitchell and Jaime Carbonell and I produced, 352 00:18:57,330 --> 00:19:02,610 Volume I, Volume II, Machine Learning Apprentice Approach 353 00:19:02,610 --> 00:19:06,120 was the beginning of the whole field, which grew 354 00:19:06,120 --> 00:19:08,190 into this very large field now. 355 00:19:08,190 --> 00:19:09,930 Well thank you so much. 356 00:19:09,930 --> 00:19:13,700 What an interesting journey.