WEBVTT 00:00:00.000 --> 00:00:01.980 Welcome to the Deep Dive, where we unpack your 00:00:01.980 --> 00:00:04.480 source material and find the insights that matter 00:00:04.480 --> 00:00:07.459 most. Today we're diving deep into the life and 00:00:07.459 --> 00:00:11.400 the monumental legacy of a figure who not only 00:00:11.400 --> 00:00:15.240 defined modern computing, but then in this amazing 00:00:15.240 --> 00:00:18.140 twist of intellectual honesty, spent his later 00:00:18.140 --> 00:00:21.260 years trying to undo the very constraints he 00:00:21.260 --> 00:00:23.960 himself helped create. We're talking about John 00:00:23.960 --> 00:00:26.480 Bacchus. That's exactly right. Our sources paint 00:00:26.480 --> 00:00:28.300 this really compelling picture of a man whose 00:00:28.300 --> 00:00:30.719 career had, you know, three profound and very 00:00:30.719 --> 00:00:34.719 distinct acts. Act one, inventing the first widely 00:00:34.719 --> 00:00:37.700 used high -level language, Fortran. And this 00:00:37.700 --> 00:00:39.640 just instantly brought computing out of these, 00:00:39.659 --> 00:00:41.880 like, specialized labs and into the hands of 00:00:41.880 --> 00:00:43.880 scientists and engineers. Right, for the masses. 00:00:44.020 --> 00:00:46.479 So the scientific masses, yeah. Then act two, 00:00:46.659 --> 00:00:48.979 formalizing how we even describe a language's 00:00:48.979 --> 00:00:50.939 structure through the notation we now know as 00:00:50.939 --> 00:00:53.780 Bacchus -Nor form or BS. that really cemented 00:00:53.780 --> 00:00:55.859 the intellectual foundations of the field. And 00:00:55.859 --> 00:00:59.100 then act three, delivering this fundamental philosophical 00:00:59.100 --> 00:01:02.140 critique of the entire programming paradigm he 00:01:02.140 --> 00:01:04.299 helped establish, that sequential procedural 00:01:04.299 --> 00:01:07.390 von Neumann style. So our mission today is to 00:01:07.390 --> 00:01:09.989 understand not just these three influential contributions, 00:01:10.230 --> 00:01:13.790 but how they all connect. I mean, how the necessities 00:01:13.790 --> 00:01:16.709 of the 1950s led to a revolution. Which then 00:01:16.709 --> 00:01:18.969 demanded a formal structure. And then that structure, 00:01:19.049 --> 00:01:21.109 I guess, revealed its own limitations, which 00:01:21.109 --> 00:01:24.909 led to a philosophical overhaul. Exactly. And 00:01:24.909 --> 00:01:27.209 Bacchus wasn't just foundational. He was heavily 00:01:27.209 --> 00:01:29.920 decorated. We're talking about a recipient of 00:01:29.920 --> 00:01:32.799 the National Medal of Science in 1975 and the 00:01:32.799 --> 00:01:36.040 ultimate prize in computing, the ACM Turing Award 00:01:36.040 --> 00:01:39.799 in 1977. And that Turing Award specifically recognized 00:01:39.799 --> 00:01:42.700 his, quote, profound and lasting contributions 00:01:42.700 --> 00:01:45.579 to the design of practical high level programming 00:01:45.579 --> 00:01:48.400 systems, which is Fortran. Right. And then for 00:01:48.400 --> 00:01:50.920 the formal procedures he created for specifying 00:01:50.920 --> 00:01:53.260 programming languages. That's the BNF part. It's 00:01:53.260 --> 00:01:55.840 this extraordinary recognition of both a practical 00:01:55.840 --> 00:01:58.969 system builder and, you know, a deep. theoretical 00:01:58.969 --> 00:02:01.810 formalist. A builder and a thinker. Yes. But 00:02:01.810 --> 00:02:03.969 what's truly fascinating and what our sources 00:02:03.969 --> 00:02:06.409 really highlight is that before all the accolades, 00:02:06.450 --> 00:02:08.669 before the global scientific impact, his path 00:02:08.669 --> 00:02:11.449 was, well, it was anything but straight. Yeah, 00:02:11.490 --> 00:02:14.030 this isn't the story of some childhood prodigy. 00:02:14.050 --> 00:02:16.150 Not at all. It's the story of delayed genius, 00:02:16.310 --> 00:02:18.870 if anything. Okay, let's unpack this and start 00:02:18.870 --> 00:02:22.330 right at the beginning. The surprisingly circuitous 00:02:22.330 --> 00:02:25.789 route John Bacchus took to even find his way 00:02:25.789 --> 00:02:28.139 into computer science. If you track his early 00:02:28.139 --> 00:02:30.319 academic record, it's honestly just striking 00:02:30.319 --> 00:02:33.080 how much he struggled to find any kind of focus. 00:02:33.560 --> 00:02:37.000 He was born in 1924, grew up in Wilmington, Delaware, 00:02:37.199 --> 00:02:39.199 and he started at the Hill School in Pennsylvania 00:02:39.199 --> 00:02:42.319 where he did not exactly thrive. Right. He later 00:02:42.319 --> 00:02:44.180 said he was just terrible at keeping track of 00:02:44.180 --> 00:02:46.240 things. Which is pretty ironic for the guy who 00:02:46.240 --> 00:02:48.460 would later invent the formal notation for language 00:02:48.460 --> 00:02:50.680 grammar, right? Struggled is definitely putting 00:02:50.680 --> 00:02:53.620 it politely. I mean, the sequence of events right 00:02:53.620 --> 00:02:57.129 after high school, it looks... Less like a path 00:02:57.129 --> 00:02:59.830 to greatness and more like a young man just running 00:02:59.830 --> 00:03:01.689 away from any kind of structure. Oh, completely. 00:03:01.930 --> 00:03:04.169 He enters the University of Virginia to study 00:03:04.169 --> 00:03:07.090 chemistry, and that lasts less than a year. Why? 00:03:07.719 --> 00:03:09.979 What happened? The records show he was expelled 00:03:09.979 --> 00:03:12.379 for poor attendance. So we're talking about a 00:03:12.379 --> 00:03:14.199 mind that would later design these systems for 00:03:14.199 --> 00:03:16.699 ultimate efficiency. And at first, he couldn't 00:03:16.699 --> 00:03:19.599 grasp the simple efficiency of, you know, showing 00:03:19.599 --> 00:03:22.759 up to class. And this was the mid -1940s, so 00:03:22.759 --> 00:03:25.560 World War II is in full swing. That expulsion 00:03:25.560 --> 00:03:28.800 leads directly to conscription. It does. And 00:03:28.800 --> 00:03:30.960 for a lot of intellectually restless young men 00:03:30.960 --> 00:03:33.960 at the time, that proved to be a surprising and 00:03:33.960 --> 00:03:36.860 I think necessary reset button. But it wasn't 00:03:36.860 --> 00:03:39.500 an immediate jump into high tech. No. He starts 00:03:39.500 --> 00:03:41.860 as a corporal in the U .S. Army, commanding an 00:03:41.860 --> 00:03:44.419 anti -aircraft battery down at Fort Stewart, 00:03:44.560 --> 00:03:47.979 Georgia. It's a very gritty, mechanical, command 00:03:47.979 --> 00:03:50.500 and control kind of role. But it's during his 00:03:50.500 --> 00:03:52.460 time in the service that his actual aptitude 00:03:52.460 --> 00:03:54.860 starts to surface, right? Absolutely. He scores 00:03:54.860 --> 00:03:57.219 remarkably high on one of those military aptitudes. 00:03:57.229 --> 00:03:59.389 tests, you know, the ones designed to sort personnel. 00:03:59.669 --> 00:04:02.469 And that score convinces the army to completely 00:04:02.469 --> 00:04:04.870 shift his path. They saw this raw, high -level 00:04:04.870 --> 00:04:07.710 processing capability and sent him to study engineering 00:04:07.710 --> 00:04:10.169 at the University of Pittsburgh. So the military 00:04:10.169 --> 00:04:13.189 basically recognized this latent talent and gave 00:04:13.189 --> 00:04:15.030 him the structure he couldn't impose on himself. 00:04:15.310 --> 00:04:17.949 But even after that, he's still not in computer 00:04:17.949 --> 00:04:20.779 science. The search continues. It does. From 00:04:20.779 --> 00:04:23.680 Pittsburgh, he moves again. He transfers into 00:04:23.680 --> 00:04:25.800 a pre -medical program at Haverford College. 00:04:26.199 --> 00:04:28.399 He's still searching for that field that would 00:04:28.399 --> 00:04:31.800 engage his unique blend of practical problem 00:04:31.800 --> 00:04:34.779 -solving and abstract thought. And this is where 00:04:34.779 --> 00:04:37.139 his biography takes a really unusual turn. The 00:04:37.139 --> 00:04:39.459 health crisis. A major personal health crisis. 00:04:39.720 --> 00:04:41.220 And the way he responded to it, I mean, it's 00:04:41.220 --> 00:04:43.660 such a strong indicator of the engineer he was 00:04:43.660 --> 00:04:46.399 about to become. It really is. During a hospital 00:04:46.399 --> 00:04:48.720 internship, he's diagnosed with a cranial bone 00:04:48.720 --> 00:04:52.079 tumor. It was serious, life -threatening, but 00:04:52.079 --> 00:04:54.360 it was successfully removed, and the procedure 00:04:54.360 --> 00:04:56.720 resulted in a metal plate being installed in 00:04:56.720 --> 00:04:59.459 his head. After recovering, he tries medical 00:04:59.459 --> 00:05:01.819 school at Flower and Fifth Avenue Medical School, 00:05:01.899 --> 00:05:04.839 but, you know, much like his attempt at chemistry, 00:05:05.100 --> 00:05:07.060 he just found it uninteresting. He dropped out 00:05:07.060 --> 00:05:09.779 after only nine months. So his mind just couldn't 00:05:09.779 --> 00:05:11.879 tolerate the rote memorization, it sounds like. 00:05:12.000 --> 00:05:14.279 I think so. especially after the intellectual 00:05:14.279 --> 00:05:17.180 challenges of war and recovery. So, yeah, he's 00:05:17.180 --> 00:05:19.220 failed out of one college, dropped out of two 00:05:19.220 --> 00:05:21.879 subsequent programs, and he's had major brain 00:05:21.879 --> 00:05:25.000 surgery. That's a really disruptive period. And 00:05:25.000 --> 00:05:27.199 then comes the second operation. This is the 00:05:27.199 --> 00:05:29.459 anecdote that I think just perfectly demonstrates 00:05:29.459 --> 00:05:32.420 his mindset. It is perfect. He needed the original 00:05:32.420 --> 00:05:34.899 plate replaced. But instead of just accepting 00:05:34.899 --> 00:05:38.189 the standard surgical solution, What does he 00:05:38.189 --> 00:05:40.490 do? He designs the replacement himself. He designed 00:05:40.490 --> 00:05:42.730 it himself. He ultimately had a metal plate of 00:05:42.730 --> 00:05:45.290 his own design installed in his head. It's the 00:05:45.290 --> 00:05:48.629 perfect metaphor. He was innately a system designer, 00:05:48.829 --> 00:05:51.470 solving a problem of necessity with custom engineering, 00:05:51.689 --> 00:05:54.110 even when the system was his own body. Exactly. 00:05:54.310 --> 00:05:56.810 He received an honorable medical discharge from 00:05:56.810 --> 00:05:59.990 the Army in 1946. And after that, he moves to 00:05:59.990 --> 00:06:02.069 New York City and initially trains as a radio 00:06:02.069 --> 00:06:04.889 technician. And that's the spark, that hands 00:06:04.889 --> 00:06:08.259 -on technical work. That seems to be it. It illuminated 00:06:08.259 --> 00:06:10.579 the underlying principles, the mathematics of 00:06:10.579 --> 00:06:13.100 systems. And that's what finally captivated him. 00:06:13.220 --> 00:06:15.920 He fell in love with pure math. And through that, 00:06:16.000 --> 00:06:18.339 he found his field. So he enrolls at Columbia. 00:06:18.480 --> 00:06:20.620 He enrolls at Columbia University, earns his 00:06:20.620 --> 00:06:23.699 bachelor's in 49 and his master's in 1950, both 00:06:23.699 --> 00:06:26.600 in mathematics. He joined IBM that same year, 00:06:26.720 --> 00:06:29.720 1950. And that's when his life as a revolutionary 00:06:29.720 --> 00:06:32.819 truly begins. And once he was at IBM, the... 00:06:33.800 --> 00:06:36.519 The raw, demanding nature of early computing 00:06:36.519 --> 00:06:39.519 just forced him to address the biggest practical 00:06:39.519 --> 00:06:42.620 problem facing the entire industry, usability. 00:06:42.800 --> 00:06:46.019 It did. And that brings us to Section 2, all 00:06:46.019 --> 00:06:48.779 about Fortran. And to really appreciate its impact, 00:06:48.899 --> 00:06:51.319 you have to ask, what was programming even like 00:06:51.319 --> 00:06:54.600 when Backus joined IBM in 1950? I think we, you 00:06:54.600 --> 00:06:56.699 know, in the modern age, we just take compilers 00:06:56.699 --> 00:06:58.759 and high -level languages for granted. But the 00:06:58.759 --> 00:07:00.699 difference between what we do and what Backus 00:07:00.699 --> 00:07:03.290 was doing. It's almost impossible to really grasp. 00:07:03.470 --> 00:07:06.029 It was unbelievably difficult. His early work 00:07:06.029 --> 00:07:08.589 involved the Selective Sequence Electronic Calculator, 00:07:08.610 --> 00:07:11.149 the SSCC. It was a marvel of electromechanical 00:07:11.149 --> 00:07:13.970 computing. And his first major project was this 00:07:13.970 --> 00:07:16.829 practical, high -stakes, but just painstakingly 00:07:16.829 --> 00:07:18.910 difficult task. Calculating the moon's position. 00:07:19.209 --> 00:07:21.029 Calculating the positions of the moon, which, 00:07:21.050 --> 00:07:23.389 you know, required continuous, accurate data. 00:07:23.610 --> 00:07:25.810 So the reality of doing that was just manual, 00:07:25.949 --> 00:07:28.800 tedious labor. You were writing code directly 00:07:28.800 --> 00:07:31.180 in machine language or, you know, assembly at 00:07:31.180 --> 00:07:34.259 best. You aren't writing A, F, E, plus C. No, 00:07:34.300 --> 00:07:36.379 not at all. You were writing instructions for 00:07:36.379 --> 00:07:39.079 register management, for memory fetches, for 00:07:39.079 --> 00:07:42.819 carry operations. You had to have this intimate, 00:07:42.860 --> 00:07:45.079 deep knowledge of the hardware architecture. 00:07:45.459 --> 00:07:47.720 Down to the specific memory cell. Right. Down 00:07:47.720 --> 00:07:51.100 to which memory cell held which variable and 00:07:51.100 --> 00:07:54.459 how to sequence every single tiny operation through 00:07:54.459 --> 00:07:57.379 the accumulator. The programmer's job was fundamentally 00:07:57.379 --> 00:07:59.939 mechanical. It was just translating mathematical 00:07:59.939 --> 00:08:03.360 logic into this excruciating sequence of hardware 00:08:03.360 --> 00:08:06.410 specific instruction. It was slow. error -prone 00:08:06.410 --> 00:08:09.129 incredibly error -prone and it required immense 00:08:09.129 --> 00:08:11.350 amounts of time just to translate the math into 00:08:11.350 --> 00:08:13.910 something the machine could understand the computer's 00:08:13.910 --> 00:08:16.290 efficiency was completely bottlenecked by the 00:08:16.290 --> 00:08:18.430 programmer's ability to communicate with it manually 00:08:18.430 --> 00:08:20.829 so the computer could calculate in microseconds 00:08:20.829 --> 00:08:23.089 but it would take a programmer weeks to just 00:08:23.089 --> 00:08:25.209 manually lay out the instructions for a single 00:08:25.209 --> 00:08:28.129 complex problem that's it exactly the translation 00:08:28.129 --> 00:08:31.389 time just defeated the speed of the machine and 00:08:31.389 --> 00:08:34.049 this inherent difficulty was the bottleneck it 00:08:34.049 --> 00:08:36.620 was preventing computer from becoming truly useful 00:08:36.620 --> 00:08:40.019 scientific tools. And Backus saw this. He saw 00:08:40.019 --> 00:08:42.899 it clearly. And his first step towards solving 00:08:42.899 --> 00:08:45.580 it was a language called speed coding. The stepping 00:08:45.580 --> 00:08:49.500 stone to Fortran. In 1953, yeah, he developed 00:08:49.500 --> 00:08:53.169 speed coding for the IBM 701 computer. Our sources 00:08:53.169 --> 00:08:55.309 note this was the first high -level language 00:08:55.309 --> 00:08:58.269 created for an IBM machine. It wasn't FORTRAN 00:08:58.269 --> 00:09:00.610 yet, but it was this critical proof of concept. 00:09:00.870 --> 00:09:03.169 What did it do? It introduced basic forms of 00:09:03.169 --> 00:09:05.730 abstraction. It essentially let programmers write 00:09:05.730 --> 00:09:08.029 code that was interpreted by a layer of software 00:09:08.029 --> 00:09:10.809 into the necessary machine instructions. It just 00:09:10.809 --> 00:09:12.649 immediately showed there was a huge need for 00:09:12.649 --> 00:09:15.009 this. It proved abstraction was possible. And 00:09:15.009 --> 00:09:17.970 by 1954, the difficulty of low -level programming 00:09:17.970 --> 00:09:21.389 was becoming a real crisis for IBM, commercially 00:09:21.389 --> 00:09:23.860 and intellectually. This is the inflection point. 00:09:23.940 --> 00:09:26.080 This is it. This is where Backus assembles the 00:09:26.080 --> 00:09:28.779 team to define and develop FORTRAN. Formula translation. 00:09:29.120 --> 00:09:31.460 Formula translation designed specifically for 00:09:31.460 --> 00:09:35.120 the new IBM 704 computer, which was significantly 00:09:35.120 --> 00:09:38.679 more powerful than the 701. And Backus's mandate 00:09:38.679 --> 00:09:41.059 was revolutionary. He had to create a language 00:09:41.059 --> 00:09:44.059 that was easy to write, that looked like algebraic 00:09:44.059 --> 00:09:46.980 notation. And this is the key part. That could 00:09:46.980 --> 00:09:49.840 produce code nearly as efficient as handwritten 00:09:49.840 --> 00:09:51.899 assembly. That was the fundamental challenge. 00:09:51.919 --> 00:09:54.440 If the compiled code was too slow, scientists 00:09:54.440 --> 00:09:56.539 would just ignore it and go back to the old hard 00:09:56.539 --> 00:09:59.279 way. So Fortran wasn't just about making programming 00:09:59.279 --> 00:10:02.480 easier. It was about creating an efficient compiler 00:10:02.480 --> 00:10:04.879 that could manage all the hardware logistics 00:10:04.879 --> 00:10:07.940 better than a human could do it manually. That 00:10:07.940 --> 00:10:10.220 is the structural significance. Fortran became 00:10:10.220 --> 00:10:12.720 the first widely used high -level programming 00:10:12.720 --> 00:10:14.940 language because it solved that efficiency problem. 00:10:15.279 --> 00:10:18.000 Think about something like array indexing. In 00:10:18.000 --> 00:10:21.179 assembly, if you wanted to access AIJ, you had 00:10:21.179 --> 00:10:23.899 to manually calculate the memory offset based 00:10:23.899 --> 00:10:26.639 on the array's starting address, the column size, 00:10:26.960 --> 00:10:30.500 and the indices, I and J. It was tedious. And 00:10:30.500 --> 00:10:32.539 Fortran handled all that for you? All of it, 00:10:32.559 --> 00:10:34.779 automatically. And the compiler did it optimally. 00:10:34.860 --> 00:10:38.450 That's the real revolution. computer accessibility 00:10:38.450 --> 00:10:41.669 by abstracting the machinery away. Programmers 00:10:41.669 --> 00:10:44.149 could write formulas that looked like x, y, plus 00:10:44.149 --> 00:10:47.590 z, i, plus 2 .0, and let the compiler worry about 00:10:47.590 --> 00:10:50.669 the registers, the memory allocation, all of 00:10:50.669 --> 00:10:52.970 it. So suddenly computers were practical for 00:10:52.970 --> 00:10:54.789 the people who actually had the problems to solve. 00:10:55.120 --> 00:10:58.539 physicists, chemists, structural engineers. This 00:10:58.539 --> 00:11:01.679 single innovation unlocked computing for scientific 00:11:01.679 --> 00:11:04.379 and engineering research globally. It shifted 00:11:04.379 --> 00:11:07.179 the entire bottleneck from how do I talk to the 00:11:07.179 --> 00:11:10.559 hardware to what mathematical process do I need 00:11:10.559 --> 00:11:12.720 to automate? And that paradigm shift was recognized 00:11:12.720 --> 00:11:15.379 pretty quickly. The success of Fortran led to 00:11:15.379 --> 00:11:18.340 Bacchus getting the W .W. McDowell Award in 1967. 00:11:18.759 --> 00:11:21.480 Yes, it did. But as soon as high level languages 00:11:21.480 --> 00:11:24.019 were successful, they created a new, deeper, 00:11:24.200 --> 00:11:27.039 more philosophical problem. Definition. Definition. 00:11:27.039 --> 00:11:29.559 Exactly. Once you have a powerful tool like Fortran 00:11:29.559 --> 00:11:31.840 and then dozens of other languages following 00:11:31.840 --> 00:11:33.460 its lead, you need a way to make sure everyone 00:11:33.460 --> 00:11:35.360 agrees on what the rules of that language actually 00:11:35.360 --> 00:11:37.500 are. Which leads us directly into act two of 00:11:37.500 --> 00:11:40.460 his career. This is section three, defining the 00:11:40.460 --> 00:11:42.799 language of languages. the Bacchus Narriform, 00:11:43.100 --> 00:11:46.659 or BNF. So once these languages emerged, the 00:11:46.659 --> 00:11:49.139 technical community needed a precise, universal 00:11:49.139 --> 00:11:52.639 dictionary for the grammar. Before this, descriptions 00:11:52.639 --> 00:11:54.539 were just written in English, which is... you 00:11:54.539 --> 00:11:56.860 know, inherently ambiguous. Disastrously ambiguous. 00:11:57.019 --> 00:11:59.360 And the catalyst for standardization came from 00:11:59.360 --> 00:12:01.759 an international context, mostly around the development 00:12:01.759 --> 00:12:04.279 of the Algeol family of languages. And Backus 00:12:04.279 --> 00:12:06.259 was a central figure in that. A central figure. 00:12:06.399 --> 00:12:08.620 He served on the international committees that 00:12:08.620 --> 00:12:11.700 developed both Algeol 58 and then the highly 00:12:11.700 --> 00:12:14.500 influential Algeol 60. And it's important to 00:12:14.500 --> 00:12:17.269 stress Algeol 60. didn't achieve the commercial 00:12:17.269 --> 00:12:19.830 dominance of Fortran. Not even close. But it 00:12:19.830 --> 00:12:22.250 became the de facto standard for publishing algorithms 00:12:22.250 --> 00:12:24.429 just because of its clean, structured design. 00:12:24.690 --> 00:12:26.950 And that design elegance was a direct result 00:12:26.950 --> 00:12:29.710 of the clarity of its formal definition, which 00:12:29.710 --> 00:12:32.649 was entirely due to Backus's invention. In the 00:12:32.649 --> 00:12:35.929 context of the UNESCO report on Algeol 58, Backus 00:12:35.929 --> 00:12:38.909 invented the notation we now know as BNF. So 00:12:38.909 --> 00:12:41.700 what exactly is BNF? For someone who hasn't done 00:12:41.700 --> 00:12:45.120 a deep dive into compiler theory, it sounds esoteric, 00:12:45.120 --> 00:12:47.860 but it's foundational. It is. BNF is a formal 00:12:47.860 --> 00:12:50.019 notation that can describe the syntax of pretty 00:12:50.019 --> 00:12:52.399 much any context -free programming language. 00:12:52.679 --> 00:12:55.620 To understand how revolutionary it was, think 00:12:55.620 --> 00:12:58.139 about what came before. If I used English to 00:12:58.139 --> 00:13:00.600 define a loop, I might write, a loop starts with 00:13:00.600 --> 00:13:03.759 DO, then a variable, an equal sign, an initial 00:13:03.759 --> 00:13:06.899 value, the word TO, a final value, and so on. 00:13:06.980 --> 00:13:09.399 Right, which sounds... Clear enough. But then 00:13:09.399 --> 00:13:11.220 you get into all the edge cases. What if the 00:13:11.220 --> 00:13:13.539 variable has to be an integer? What if the initial 00:13:13.539 --> 00:13:16.500 value can be a complex expression? Angular just 00:13:16.500 --> 00:13:18.940 gets messy with footnotes and exceptions. It 00:13:18.940 --> 00:13:21.799 gets messy immediately. BNF provides a set of 00:13:21.799 --> 00:13:23.840 concise production rules that just eliminate 00:13:23.840 --> 00:13:27.059 all ambiguity. For instance, a simplified BNF 00:13:27.059 --> 00:13:30.610 rule might look like this. Expression. Term expression 00:13:30.610 --> 00:13:32.909 plus term. Okay, break that down. It's defining 00:13:32.909 --> 00:13:35.090 an expression recursively. It says an expression 00:13:35.090 --> 00:13:37.830 is either a single term or it's another valid 00:13:37.830 --> 00:13:40.029 expression followed by a plus sign and another 00:13:40.029 --> 00:13:42.490 term. It's a mathematical definition of grammar. 00:13:43.049 --> 00:13:45.289 So it takes all the ambiguity of natural language 00:13:45.289 --> 00:13:47.409 and just replaces it with mathematical certainty. 00:13:47.610 --> 00:13:50.129 And that precision is why BNF matters so much 00:13:50.129 --> 00:13:52.950 for compiler development. The ability to formally 00:13:52.950 --> 00:13:55.490 specify the language allowed for the eventual 00:13:55.490 --> 00:13:58.190 automation of building the compiler itself. The 00:13:58.190 --> 00:14:01.389 translator. Yes. And that's why BNF was explicitly 00:14:01.389 --> 00:14:04.649 cited in his Turing Award. It established the 00:14:04.649 --> 00:14:07.340 theoretical bedrock for the field. Instead of 00:14:07.340 --> 00:14:10.000 a compiler writer having to interpret some ambiguous 00:14:10.000 --> 00:14:12.519 English manual, they could just read the mathematical 00:14:12.519 --> 00:14:15.440 grammar in BNF and program their compiler to 00:14:15.440 --> 00:14:17.740 recognize tokens according to those exact rules. 00:14:17.980 --> 00:14:21.649 It was transformative. It introduced rigor. Now, 00:14:21.669 --> 00:14:23.230 there were some other approaches. You know, our 00:14:23.230 --> 00:14:25.049 sources mentioned Lisp and APL had their own 00:14:25.049 --> 00:14:27.990 notations. But by the 1970s, the Bacchus -Narr 00:14:27.990 --> 00:14:30.629 context -free specification had become the industry 00:14:30.629 --> 00:14:33.149 standard. And that standardization was critical 00:14:33.149 --> 00:14:36.269 because it enabled automated tools. Things like 00:14:36.269 --> 00:14:40.129 YAC, yet another compiler. Tools like YAC let 00:14:40.129 --> 00:14:42.789 developers feed a BNF description of a language 00:14:42.789 --> 00:14:45.509 to a program, and that program would automatically 00:14:45.509 --> 00:14:48.470 generate a huge chunk of the necessary parsing 00:14:48.470 --> 00:14:52.360 code for a compiler. This just massively accelerated 00:14:52.360 --> 00:14:54.840 the creation of new languages. So this lets us 00:14:54.840 --> 00:14:57.279 connect the dots perfectly. Fortran solved the 00:14:57.279 --> 00:14:59.779 problem of writing programs efficiently. Making 00:14:59.779 --> 00:15:02.340 computing practical. Right. And then BNF solved 00:15:02.340 --> 00:15:04.340 the problem of defining programs rigorously, 00:15:04.480 --> 00:15:07.600 which made language design systematic. It really 00:15:07.600 --> 00:15:09.720 formalized computer science as a discipline. 00:15:09.980 --> 00:15:12.320 He had basically given the world its first great 00:15:12.320 --> 00:15:15.460 high -level language and then the mechanism to 00:15:15.460 --> 00:15:17.360 define every language that would come after it. 00:15:17.440 --> 00:15:20.159 And if he had retired right then... In 1970, 00:15:20.440 --> 00:15:22.539 he'd be remembered as one of the greats. One 00:15:22.539 --> 00:15:24.620 of the all -time greats. But he wasn't finished. 00:15:25.200 --> 00:15:28.779 In the late 1970s, he enters Act III, a phase 00:15:28.779 --> 00:15:31.000 that involves a deep, self -imposed critique 00:15:31.000 --> 00:15:33.419 of everything he had helped build. And this is 00:15:33.419 --> 00:15:36.259 where his career becomes truly fascinating. In 00:15:36.259 --> 00:15:38.259 the late 70s, after achieving all this fame for 00:15:38.259 --> 00:15:40.779 creating the dominant programming paradigm, this 00:15:40.779 --> 00:15:44.200 sequential procedural style, he pivots to critique 00:15:44.200 --> 00:15:46.980 it fundamentally. It shows a mind that just wasn't 00:15:46.980 --> 00:15:49.940 interested in resting on past achievements. He 00:15:49.940 --> 00:15:52.539 was constantly striving for a more elegant, more 00:15:52.539 --> 00:15:55.419 expressive form of computation. And the public 00:15:55.419 --> 00:15:59.000 declaration of this pivot was his 1977 Turing 00:15:59.000 --> 00:16:01.580 Award lecture. The lecture that shook the industry. 00:16:01.940 --> 00:16:04.879 Can programming be liberated from the von Neumann 00:16:04.879 --> 00:16:08.100 style? That title is the whole thesis. He was 00:16:08.100 --> 00:16:10.360 railing not against Fortran specifically, but 00:16:10.360 --> 00:16:12.679 against the foundational hardware architecture 00:16:12.679 --> 00:16:16.190 that Fortran was forced to obey. The von Neumann 00:16:16.190 --> 00:16:18.470 architecture. Okay, we need to define the von 00:16:18.470 --> 00:16:21.370 Neumann style clearly because it basically defines 00:16:21.370 --> 00:16:23.710 almost every computing device we use today. It 00:16:23.710 --> 00:16:26.210 does. The von Neumann architecture dictates that 00:16:26.210 --> 00:16:28.009 the computer operates through a single channel 00:16:28.009 --> 00:16:30.230 connecting the CPU where the processing happens 00:16:30.230 --> 00:16:33.210 and memory where data and instructions are stored. 00:16:33.370 --> 00:16:35.450 And the CPU just executes instructions one after 00:16:35.450 --> 00:16:37.690 another, sequentially. Sequentially. Constantly 00:16:37.690 --> 00:16:40.409 fetching data from memory, processing it, storing 00:16:40.409 --> 00:16:43.509 the results back, one word at a time. Backus 00:16:43.509 --> 00:16:46.710 called this the von Neumann. Because the CPU 00:16:46.710 --> 00:16:49.950 spends most of its time just waiting for data 00:16:49.950 --> 00:16:52.309 to be shuffled back and forth across this narrow 00:16:52.309 --> 00:16:55.659 channel. Yes. And Backus argued that our programming 00:16:55.659 --> 00:16:58.220 languages evolved to just mirror this mechanical 00:16:58.220 --> 00:17:01.519 process. Procedural languages like Fortran, C, 00:17:01.700 --> 00:17:04.420 Pascal, they were all designed around the fundamental 00:17:04.420 --> 00:17:07.400 concept of the assignment statement. The assignment 00:17:07.400 --> 00:17:09.799 statement being, you know, X plus one. Exactly. 00:17:09.839 --> 00:17:11.980 And that statement is nothing more than saying, 00:17:12.019 --> 00:17:15.500 go to this specific memory location, X, fetch 00:17:15.500 --> 00:17:18.519 the value, do an operation like plus one, and 00:17:18.519 --> 00:17:20.799 then store the new result back in that same location. 00:17:21.589 --> 00:17:24.069 So he's saying that by making Fortran successful, 00:17:24.650 --> 00:17:27.650 they had basically institutionalized a constraint 00:17:27.650 --> 00:17:29.950 that was imposed by the hardware. That's the 00:17:29.950 --> 00:17:32.349 core of his critique. He argued that this style 00:17:32.349 --> 00:17:34.789 was fundamentally limited because it forces the 00:17:34.789 --> 00:17:37.890 programmer to focus on mechanical, low -level 00:17:37.890 --> 00:17:40.190 details of data movement, what he called word 00:17:40.190 --> 00:17:42.509 -at -a -time programming, rather than the high 00:17:42.509 --> 00:17:44.269 -level mathematical structure of the problem 00:17:44.269 --> 00:17:46.230 itself. So he believed programmers were being 00:17:46.230 --> 00:17:48.369 forced to think like the machine's memory fetcher, 00:17:48.470 --> 00:17:51.529 not like a pure mathematician. Yes, a loss of 00:17:51.529 --> 00:17:54.630 expressive power and elegance. He saw the entire 00:17:54.630 --> 00:17:57.690 industry as being trapped in an imperative morass 00:17:57.690 --> 00:17:59.970 where programs were just a sequence of state 00:17:59.970 --> 00:18:02.789 changes rather than clean mathematical transformations. 00:18:03.309 --> 00:18:05.829 And this lecture is sometimes seen as his apology 00:18:05.829 --> 00:18:08.660 for creating Fortran. Is that accurate? It was 00:18:08.660 --> 00:18:10.960 misinterpreted. It wasn't an apology. It was 00:18:10.960 --> 00:18:13.720 an intellectual evolution. The problem was that 00:18:13.720 --> 00:18:16.380 his proposed solution, function level programming 00:18:16.380 --> 00:18:20.480 or FLP, was often mistakenly confused with existing 00:18:20.480 --> 00:18:23.599 functional languages like Lisp. Which used lambda 00:18:23.599 --> 00:18:26.539 calculus. Right. And Backus was aiming for something 00:18:26.539 --> 00:18:29.319 different, something algebraically cleaner. His 00:18:29.319 --> 00:18:31.660 antidote was to move toward a style that focused 00:18:31.660 --> 00:18:34.200 entirely on combining functions to transform 00:18:34.200 --> 00:18:37.220 data rather than assigning values to variables. 00:18:37.680 --> 00:18:39.559 And to explore this idea, he designed a language 00:18:39.559 --> 00:18:42.880 called FP. FP for function -level programming. 00:18:43.220 --> 00:18:45.759 It was his direct attempt to embody these ideas, 00:18:45.880 --> 00:18:48.079 and it was heavily, heavily influenced by Kenneth 00:18:48.079 --> 00:18:50.380 Iverson's APL. A programming language. Yeah. 00:18:50.420 --> 00:18:52.359 Which was already pretty radical. Very radical. 00:18:52.440 --> 00:18:55.180 It was focused on array processing and used this 00:18:55.180 --> 00:18:58.480 non -standard character set to denote these powerful, 00:18:58.579 --> 00:19:01.200 primitive operations that worked on entire arrays 00:19:01.200 --> 00:19:04.980 at once. FP inherited that spirit. So how did 00:19:04.980 --> 00:19:07.670 it work? No variables? No loops? No. It operated 00:19:07.670 --> 00:19:10.269 using a small set of combinators, which are functions 00:19:10.269 --> 00:19:12.450 that take other functions as arguments and combine 00:19:12.450 --> 00:19:15.109 them to create more completes functions. The 00:19:15.109 --> 00:19:17.809 key is the programmer never directly interacts 00:19:17.809 --> 00:19:20.009 with variables or assignment statements. So if 00:19:20.009 --> 00:19:22.089 you wanted to sum a list of numbers. You wouldn't 00:19:22.089 --> 00:19:24.369 write a loop and assign the running total to 00:19:24.369 --> 00:19:26.650 a variable. You compose a function that maps 00:19:26.650 --> 00:19:29.150 an addition operator across the entire data structure 00:19:29.150 --> 00:19:32.450 at once. It was an algebra of programs. That 00:19:32.450 --> 00:19:35.430 sounds incredibly rigorous, but also a huge departure 00:19:35.430 --> 00:19:38.279 from the mainstream. How was its adoption? Adoption 00:19:38.279 --> 00:19:41.140 was niche. An FP interpreter was distributed 00:19:41.140 --> 00:19:44.299 with the 4 .2 BSD Unix operating system, which 00:19:44.299 --> 00:19:46.140 is interesting, but implementations were few. 00:19:46.240 --> 00:19:48.160 It was used mostly for educational purposes. 00:19:48.539 --> 00:19:50.960 The conceptual leap was just too big. Too big 00:19:50.960 --> 00:19:53.740 for an industry already so heavily invested in 00:19:53.740 --> 00:19:56.940 procedural C and Fortran. But Backus didn't give 00:19:56.940 --> 00:19:59.799 up. He spent the latter part of his career trying 00:19:59.799 --> 00:20:02.359 to refine this concept, creating a successor 00:20:02.359 --> 00:20:05.769 language called FL. FL. For function level. Right. 00:20:05.930 --> 00:20:08.309 What was the fate of this second attempt? FL 00:20:08.309 --> 00:20:11.009 was a long -term intensive internal research 00:20:11.009 --> 00:20:13.910 project at IBM. Backus was focused on making 00:20:13.910 --> 00:20:16.490 it even more rigorous so that you could mathematically 00:20:16.490 --> 00:20:19.789 manipulate programs the way a mathematician manipulates 00:20:19.789 --> 00:20:22.809 equations. But unlike Fortran, which IBM pushed 00:20:22.809 --> 00:20:26.069 globally, FFL just stayed internal. Unfortunately, 00:20:26.109 --> 00:20:28.539 yes. Development stopped when the research project 00:20:28.539 --> 00:20:30.599 finished, and the source code for the compiler 00:20:30.599 --> 00:20:33.140 was never made public. Only a few documenting 00:20:33.140 --> 00:20:35.359 papers are left. Why do you think it failed to 00:20:35.359 --> 00:20:37.640 gain traction when the original was such a colossal 00:20:37.640 --> 00:20:39.559 success? It was a collision of philosophy and 00:20:39.559 --> 00:20:42.599 timing. By the 1980s, when FL was being developed, 00:20:42.740 --> 00:20:44.359 functional programming was gaining traction, 00:20:44.599 --> 00:20:46.920 but it was centered around lambda calculus and 00:20:46.920 --> 00:20:50.319 static typing languages like ML or what would 00:20:50.319 --> 00:20:53.380 become Haskell. And FL was different. FL, following 00:20:53.380 --> 00:20:56.579 that APL lineage, was philosophically... distinct. 00:20:56.700 --> 00:20:59.279 It focused on the concatenation of primitive 00:20:59.279 --> 00:21:02.559 operations using these powerful array centric 00:21:02.559 --> 00:21:05.960 operators. It felt alien to the academic functional 00:21:05.960 --> 00:21:08.619 movement. So his vision was kind of caught between 00:21:08.619 --> 00:21:10.720 two worlds. Exactly. It was caught between the 00:21:10.720 --> 00:21:13.759 highly practical procedural world he was critiquing 00:21:13.759 --> 00:21:16.279 and the mathematically rigorous lambda calculus 00:21:16.279 --> 00:21:18.240 world, which he didn't fully embrace either. 00:21:18.420 --> 00:21:21.789 But the ideas didn't die. Not at all. The legacy 00:21:21.789 --> 00:21:24.630 of FL is seen clearly in later versions of the 00:21:24.630 --> 00:21:27.410 J programming language, which is Iberson's successor 00:21:27.410 --> 00:21:31.430 to APL. J is a modern, powerful array processing 00:21:31.430 --> 00:21:33.390 language that carries many of the structural 00:21:33.390 --> 00:21:36.410 insights and the mathematical elegance that Backus 00:21:36.410 --> 00:21:39.490 was searching for in FP and FL. So his critique, 00:21:39.650 --> 00:21:42.029 if not his specific language, was enduringly 00:21:42.029 --> 00:21:44.789 relevant. It was. His core philosophical drive 00:21:44.789 --> 00:21:47.329 continued to influence that entire lineage of 00:21:47.329 --> 00:21:49.750 array processing languages. It is just so rare 00:21:49.750 --> 00:21:51.950 to find a figure in any field who so fundamentally 00:21:51.950 --> 00:21:54.829 shaped its practical foundation, then formalized 00:21:54.829 --> 00:21:57.250 its theoretical structure, and then stood back 00:21:57.250 --> 00:21:59.710 to deliver it. a devastating philosophical critique 00:21:59.710 --> 00:22:02.690 of the entire system. It really is. He created 00:22:02.690 --> 00:22:04.829 the digital freeway, established the rules for 00:22:04.829 --> 00:22:06.849 building all the future roads, and then said, 00:22:06.930 --> 00:22:09.230 you know what? This whole infrastructure is fundamentally 00:22:09.230 --> 00:22:12.309 limiting our speed. It speaks to his relentless 00:22:12.309 --> 00:22:15.170 intellectual energy. He was never satisfied with 00:22:15.170 --> 00:22:18.190 good enough. So his career spanned the absolute 00:22:18.190 --> 00:22:20.869 necessity of early computing, from the manual 00:22:20.869 --> 00:22:23.849 toil of calculating lunar positions on the SSCC 00:22:23.849 --> 00:22:27.200 to the abstraction of speed coding. The revolution 00:22:27.200 --> 00:22:29.539 of practical science with Fortran. The formal 00:22:29.539 --> 00:22:31.579 definition of all language structure through 00:22:31.579 --> 00:22:34.019 BNF. And finally, the profound philosophical 00:22:34.019 --> 00:22:37.059 critique of that whole world with FP and FL. 00:22:37.259 --> 00:22:39.680 And the recognition he received truly reflects 00:22:39.680 --> 00:22:42.440 that triple legacy. It does. IBM Fellow in 63, 00:22:42.859 --> 00:22:45.960 McDowell Award in 67, the National Medal of Science 00:22:45.960 --> 00:22:49.539 in 75, followed by the Turing Award in 77. Later 00:22:49.539 --> 00:22:52.460 honors included the Draper Prize in 93, and he 00:22:52.460 --> 00:22:54.400 was named a Computer History Museum Fellow in 00:22:54.400 --> 00:22:57.319 1997. He passed away in 2007 in Ashland, Oregon, 00:22:57.519 --> 00:22:59.920 leaving behind this monumental, complex, and 00:22:59.920 --> 00:23:02.509 ultimately challenging legacy. And as a final, 00:23:02.589 --> 00:23:06.230 slightly more whimsical honor, the asteroid 6830 00:23:06.230 --> 00:23:08.589 John Bacchus was named for him. A true giant. 00:23:09.039 --> 00:23:10.859 But for the learner, for you listening to this, 00:23:11.000 --> 00:23:13.240 I think the most important part of Backus' work 00:23:13.240 --> 00:23:16.200 is the lasting question he posed. The idea that 00:23:16.200 --> 00:23:18.940 programming is constrained not by our intelligence, 00:23:19.039 --> 00:23:21.140 but by the physical architecture of the computer. 00:23:21.359 --> 00:23:24.359 Right. If Backus, the creator of Fortran, ultimately 00:23:24.359 --> 00:23:27.140 argued that the underlying sequential von Neumann 00:23:27.140 --> 00:23:30.400 -style architecture limits how powerful and expressive 00:23:30.400 --> 00:23:34.000 our programs can be, this raises a really persistent 00:23:34.000 --> 00:23:36.900 and important question. And that is? Nearly 50 00:23:36.900 --> 00:23:39.579 years after his critique, Are the programming 00:23:39.579 --> 00:23:42.240 languages you use, Python, Java, JavaScript, 00:23:42.559 --> 00:23:44.960 are they still fundamentally constrained by that 00:23:44.960 --> 00:23:47.759 sequential, word -at -a -time, memory -shuffling 00:23:47.759 --> 00:23:50.559 design of the processors they run on? Or have 00:23:50.559 --> 00:23:52.940 model kerygimes, like object -oriented design 00:23:52.940 --> 00:23:55.839 or massively parallel processing, truly found 00:23:55.839 --> 00:23:58.259 ways to liberate us from that mechanical sequence? 00:23:58.599 --> 00:24:00.339 It's something to consider the next time you 00:24:00.339 --> 00:24:02.740 design a system or write a line of code. Are 00:24:02.740 --> 00:24:04.980 you instructing the machine on the logical steps 00:24:04.980 --> 00:24:07.140 of your problem, or are you just efficiently 00:24:07.140 --> 00:24:09.839 managing the flow? of data across the von Neumann 00:24:09.839 --> 00:24:10.460 bottleneck.