Jack Park with Peter Kaminski, 2023-12-06
Author: Peter Kaminski Issue: 2024-01-03
Jack Park with Peter Kaminski, 2023-12-06
Aviation
Thanks for speaking with me, Jack. You have a lot of interesting history. I remember one day you were talking about stopping by Scaled Composites...
Well, I hung out with Burt Rutan for a brief period of time.
Peter Garrison wrote for Flying Magazine, but he also built a home-built airplane, kind of like mine. And he actually flew his to Japan and flew it to Europe, in fact, which took a lot of gas, but he did it.
And I hopped in his airplane and we went out and I met Burt Rutan, and I told Rutan that I was building windmills. And he says, “Oh, you gotta see what I’m doing.” And he was building a windmill too.
He was using this new airfoil that had been invented at some research lab in Glasgow. It turned out that he needed that airfoil for the front canard wing on his VariEze airplane because most airfoils didn’t generate enough lift, and this was a high lift airfoil, so he put it on. In fact, when he first put it on, it made so much lift, it almost threw the airplane over backwards. So he had to cut about a foot off of each end of the canard. So I was around it. He never let me fly the thing. I wanted to, but he never let me.
I didn’t realize you had a home-built.
I built an airplane called a Thorp T-18, sort of right out of high school. I was in the Experimental Aircraft Association and there was this article by John Thorp on the new kind of airplane. And I showed it to my father and he says, “You know, that’s the guy that designed the Little Dipper and the Big Dipper at Lockheed.” My father was a model builder at Lockheed. So he went over and got the models he had built of Little Dipper and Big Dipper, and we drove up to John Thorp’s house and showed him the models. And he got all excited. So I bought the plans and came home and started building it. And it took three years to build. And then a couple of years later, I crashed it in Shasta Lake. It’s always something.
What was the T-18 like?
It was all metal. It’s a low wing, two seater, taildragger. In other words, two wheels in the front, one in the back. And it used a surplus four-cylinder Lycoming engine that had been used in Vietnam to power generators. And what you had to do was saw the generator mount off the front of the engine block. And then you basically overhaul it and put it on your airplane with a propeller and away you go.
Sweet. I’m a little jealous, but...
Well, you know, that’s almost 60 years ago. And so I look back and I kind of wish I could do it again, but now wait a minute, not at my age. I mean, there’s guys my age doing it all the time, but look, I kind of want to be a part of the team that helps save the planet and flying my airplane isn’t gonna do that, so...
Totally fair.
Yeah, you have to make choices. And I think that’s my definition of being an adult. When you finally learn how to make choices, you know what I mean?
Yeah, making responsible choices for the times.
Back in the day, how did you end up flying an airplane?
In 1952, I was around eight years old, one Saturday morning, my father says, “Come on out in the front lawn.” And we went out on the front lawn. Now we lived a couple of blocks from Burbank Airport, it was called Lockheed Air Terminal at the time, it wasn’t Burbank Airport yet.
And all of a sudden, this gigantic flying wing, a B-49 just comes sailing over the house. And I said, “That’s it, I’m gonna be a jet pilot.” And so it was from then on that I was game to do that. And in fact, all the way up until draft age, I was pretty sure I was going to fly in the Air Force or Navy and fly F-4s or something or other.
But then also at age eight, I wrecked one of my eyes. And so it turned out I was 4F and didn’t get to. So I said, “Screw it, I’ll build and fly my own.” And that’s how it kind of got away.
Software, XML, Topic Maps
It’s a fun story. So later, you decided to save the Earth instead of play around?
I’m not gonna save the Earth. It would be stupid to think that I can save the Earth. Instead, it’s going to take a team of people, kind of a viral infection that will grow. And I wanna be a part of that viral infection. I have my ideas, you have yours, and Jerry has his. And that’s what I like about OGM–the possibilities seem fairly exponential within the tribe.
Now, again, it’s always like herding cats. There is no meow mix that’s gonna make us all work together in some super fashion, but we stumble around.
How did you get into computers?
When I got my first big Silicon Valley job, the guy looked at my resume and he says, “My God, you’re a hacker.” And so he put me in his skunk works. Not on a line doing the coding. I was in the “let’s build new sh*t” team, So, life for me has been kind of like this fairy tale.
Which company was this?
Well, so this was a B2B ontology company. It was called VerticalNet. It was originally called Isadra, but by the time I actually got hired, VerticalNet had bought them out. And the idea was you take a central portal and you use its ontology to map catalogs of all kinds of vendors. And then when buyers come in, they put up their spec and then the ontology does the mixing and matching and makes the connections. And so they’re just a little friction in the middle of this wiring. And XML was kind of new. And they funded me to be on the XML topic maps committee. And so I really got into XML like I’d never imagined I would.
Sounds like fun.
Well, it was fun. And it allowed me then to make a discovery of an opportunity to do a PhD at a university in the UK remotely. And so that didn’t actually come to pass until I had left VerticalNet. Actually they folded when the dot-bomb hit them. I worked for Adam Cheyer at VerticalNet. After that ended, he moved back to SRI and he brought me there. It was at that point that I said, okay, let’s do this PhD. And so I made flights to England and whatever. So it was fun.
Why the UK? How did that happen?
Well, the Open University, they had a speaker come to Stanford at some conference and I was there and the guy says, we’re doing this and this and this and this and this. So I looked them up on the web and I went in the next day and I said, do you accept remote students? Now I already knew about them because I had written to one of them, Simon Buckingham Shum and asked for permission to use his code in my project. He said, “please do and tell us about it”. And so then that guy told Simon about me and Simon wrote and said, we’d like you to come to visit us, we’ll pay your ticket. And so I’m being flown to the UK to interview for a PhD. And I’m thinking, I guess this is as good as it gets.
That’s an awesome story.
And by the way, Doug Engelbart was in London at the same time. So he took a train, came up to Milton Keynes and Engelbart and I and gave some talks. But these people were using an issue-based information system for climate studies. And my PhD was to use topic mapping and AI to tame the conversations, to improve the conversation. So that you can see the lineage of what I’m doing.
Lifecycles of AI
Yeah. What kind of AI would that have been at the time?
Well, at the time we didn’t have LLMs. So the only thing that you really had was like open NLP and core NLP from Stanford and other sources. SpaCy didn’t even exist in those days. And so it was just a bunch of crude hackery, but yes.
So that kind of natural language processing (NLP) wouldn’t be neural nets, would it?
Well, look, AI was invented at Dartmouth at a conference in around what, 1958, 59, somewhere in there. And, you know, Dartmouth is right on that river that separates New Hampshire from Vermont. And so, you know, White River it’s called and there’s all these people in their kayaks and whatever. And Minsky was there and all of the big names in the old days. Now you have to remember that by that time Minsky did a PhD using neural nets. Now his neural nets were not transistors, they were light bulbs and switches, but that’s what he did. And so then shortly thereafter, he and Papert got together and tried to do XOR (exclusive OR) experiments with his neural nets and they didn’t work. So he and Papert wrote a paper that basically put neural nets into a black hole.
It wasn’t until Hopfield at Caltech started talking about his Hopfield networks that people said, oh, wait a minute. But keep in mind that at the same time, there were these guys Rumelhart and the others that were still playing with them. And they wrote the series on connectionism. So they weren’t dead, but they were not AI. They were nothing to do with AI.
They were connectionism. That’s a different game. And look, we and the AI jockeys hijacked from the philosophers this thing called ontology and gave it a whole new sort of world compared to what philosophers would do with it. But there’s a really cool book called “Computational Philosophy of Science.” And I’ll bet you can find a copy used on Amazon. In fact, when you Google the author, you’ll find that the source code for it, it’s in Common Lisp, is online. And I think it’s the best elucidation of AI that I’ve ever seen.
So for me, AI’s history is rooted in frame-based symbolic reasoning. That’s what it was. Minsky’s paper on frames, which was borrowed from Charles Fillmore’s “Case-Frames.” And Case Frames became “FrameNet” at UC Berkeley. And while they are open source, you gotta pull hen’s teeth to get them to give you the link to download the data, but it’s there. And there’s people extending FrameNet into the biomedical domain. Again, it’s all symbols and it’s all structures. Whereas the connectionist people sort of just blew right past that. And it’s true that, okay, from my perspective, I think the guy that got it best was Douglas Lenat.
I mean, he did his PhD in ’76 at Stanford called “Automated Mathematicians.” He was in Cordell Green’s group called “Automated Programming.” And he has an “If-Then” program with 150 “If-Then” rules that discovered prime numbers and Goldbach’s Conjecture and a whole lot of other cool stuff. Now he got slammed and kicked pretty hard because he did it in Lisp. And fundamentally Lisp already knows how to do that. So he went and did a postdoc with Herbert Simon, and they said, “Look, lift it above Lisp.” So Lenat and Russell Greiner wrote a program called RLL, Representation Language Language. You and I would know it as a DSL, Domain Specific Language.
He had this high level language that was written in Lisp, but it totally isolated Eurisko, his new program, from Lisp. And it went on and designed a 3D VLSI, and they even got Carver Mead from Caltech to come up to Stanford and build it and prove it would work. It won a naval warfare game twice, and they kicked him out of the fleet, made him an honorary admiral, and said he couldn’t play anymore. And it did a lot of extraordinarily good sh*t, but think about it. It always died when it ran out of what it knew how to do. It had no heuristics for going outside of its box. And what he decided was what it lacked was common sense. That was when they birthed the program called Cyc.
So, that’s AI to me.
Cyc, Frames
Yeah. So, connectionism and symbolics. Is Cyc one of those, or is it something else?
Cyc is fundamentally symbolic. It’s got 111 different reasoning inference engines and this massive, basically symbolic database. Now, Lenat went through an enormous personal evolution. He went in thinking that he could just take Eurisko and start blowing it up.
And quickly learned, no, that wasn’t going to work. And so what they ended up doing was inventing little objects that were called microtheories.
And so the trick was to collect as many microtheories as you could against this massive knowledge base of facts that are out there. And he was trying to make Cyc understand in quotes, this desktop encyclopedia. That was the birthing of it.
So he had to move away from his simple frames and went into a whole bunch of logic-based systems.
Are frames related to symbolics, or...
Well, look, imagine you’ve got a relational database and in that database is a table and the table has these columns. In a frame, those would be called slots. And then it’s got rows, which are instances of, and so that table defines a frame.
And so each instance has its own identifier. So the table would represent “person” instances.
And so it’s going to have these slots called name, birthdate, and other properties. And so a frame is given an identifier. The symbol is the identifier for the frame for Peter Kaminski or Jack Park. Each frame is, in fact, just like a row in the “person” table. And Lisp was built on what are called property lists. And a property is a slot. And a slot is a column in a table. So the mapping to a relational database is one-to-one.
You’re just passing around some numeric ID. Internally it knows that it’s Peter Kaminski, but the program knows nothing about it. It’s just a symbol.
I wonder if you know the “Thousand Brains” theory by Jeff Hawkins. He’s got frames. He says, essentially, that the way we think is by using geographic frames.
Right. He’s operating in a cognitive science domain and not a computer science domain. And so it stands to reason that there’s a mapping between ontologies. And I totally got it when I was reading the book.
I was reading the book on a flight to Hawaii and rapidly taking notes and so forth. The point is that he talks about how there are maps, spatial maps.
And I’m reading it and I’m saying, yes, I get this. It maps to how I’m thinking.
Perhaps, but then I asked my friend who is a retired professor from UCLA med school and he thinks it’s a bunch of horsesh*t.
He promptly sent me a book on neurophysiology. I’ve got it. I haven’t read it yet, but that’s the neurophysiology ontology which is not Hawkins’s cognitive science. This is a complex field; there be dragons.
So what we have to do, you and I and others, is to find the mappings between the ontologies.
Signal-To-Noise Ratio
Yeah. That sounds like fun. So, OGM is kind of going in the right direction? Or at least possibly going in the right direction.
Well, okay. I’ll say it and if this gets published, it gets published. The signal to noise ratio on OGM is pretty low. But when the signal pops, it’s a doozy.
And that’s why I hang around. We all hang around for that signal, and we learn to hit the delete button on the noise.
Noise is okay because if you don’t have a large network, you’re not going to get enough noise to stir up the brilliance. Noise is a very important part of cognition. Without it, you’re not going anywhere.
Take the Boltzmann machine. Simulated annealing. Or another term is hill climbing.
As Stuart Kaufman says, if it’s a rugged fitness domain, a rugged mountain, you take a leap off of some little cliff and you’re liable to land out in the boonies. But the whole point of the Boltzmann machine is to bump you off of the local maximum and try to start climbing again to find the higher peak. Hill climbing. We’re all doing that, it’s the game.
Communication
So how does the viral team, Team Virus, save the world? Like how do we get out of where we are, or how do we move forward – or do we need to get out of where we are?
I’ll give you Park’s theory. Was it Margaret Mead that said that conversations were the only way anything gets done?
You know, I’m not a reductionist at heart. I’m kind of trying to be a whole systems person. But if you try to reduce the world’s problems to anything that are core, conversation lies at the bottom of it. I mean, when we were living in caves, we sat around fires and we mumbled and jumbled and whatever, it’s conversation. That’s human nature to find a way to communicate. In fact, if you think about it, communication starts at the ion level. Atoms bonding. From bonding, you get organs and from organs, you get organisms and from organisms, you get societies and from societies, whatever. You can see this progression and it’s all about communication.
Starting with covalent and ionic bonding. So give me a break folks, communication is core. And so if you think about that, then you kind of scan the horizon. You know, I have to admit it was Jeff Conklin and Engelbart and the PhD research that put me onto this ‘cause I didn’t see it coming. You start by taming town halls. And the magic that goes on when you make a dialogue map of what’s going on in a town hall and people look at it and say, oh, wait a minute. Geez, Joe, I didn’t realize that’s what you were saying ‘cause I wasn’t listening, be honest now.
And it’s, you know, Otto Scharmer gets it too. This whole thing of Theory U, it’s this, you’ve got to calm down and find out who you are in the middle of this.
And so it’s in that space that I’m operating. And that space, and that’s where folks like us would want to put energy to make communication better.
Somehow we all need to get along.
You have to spend a lot of time with your neighbors, because first of all, everything is based on trust. The minute you broach a wicked problem, that person better be the equivalent of in love with you before you go there because otherwise it becomes explosive. So you really have to, you gotta spend a lot of time with them.
Trust
Trust is at the bottom line. So I try to build ecosystems, environments where trust will grow. And you can see the reason that we went with the SenseCraft project.
In January of 2010, I defended a thesis proposal. And it was well along the way towards doing the dissertation and getting a PhD when a Carnegie Mellon professor stood up in a face-to-face meeting in Palo Alto one morning, 8:30 in the morning. And he stood up and he says, “How could we have civil conversations about politics online?”
Everybody kind of knew I was in that space so they said, “Talk to Park.” And so he and I went out to have coffee. And it was on that day we formed TopicQuests.
My answer was “World of Warcraft meets Global Sensemaking”. That’s because I had read all this literature in my PhD research that went into the social life in guilds and the social production and the trust metrics and all the rest of it. Then I met the co-author of a book on using avatars in business. And so I knew that the mechanism was there.
So from mid-2010 on, we were building prototypes, hacking Java servlets and experimenting. But I could never get anything that my team liked. And then along came Marc-Antoine [Parent] and I said, “Screw it. Marc-Antoine, you get to be the chief architect.” And he said, “Oh, okay.”
Marc-Antoine took it in new directions but they’re still the same genre. But his work is brilliant. His work just goes so far beyond anything I would have thought of. You know, pure signal.
A Timeline of the Planet
Look, I did a, my second invited talk in Tokyo was 2017. And when they asked me to do it, they said, “Oh, by the way, be sure and put an emphasis on citizen science.” And the talk was a keynote to a conference where they were building a Japanese open source Watson. Believe it or not, very high tech.
And so I decided, what’s the point of having this Watson if you don’t put it to use? So I said, “What if we built a timeline of the planet, citizen science?” This is, it’s this gigantic timeline online that people enroll in guilds to contribute to. And they go on these quests to discover everything from the fall of the dinosaurs to the great religions and all of the rest.
They’re accumulating this literature and each dot on the timeline is a topic in the topic map. You go to the topic map and you can see where this is going, but the kids are doing this kind of research.
So in my talk preparation, I found out that Bill Gates was watching a video on something called deep history. And he takes a million dollars of his own and gives it to UC Berkeley. And they built that timeline game, except it’s built in C# and, you know, it has to use SQL Server and not really open source services.
But if I had the time, I would transliterate it out to Java and make it real. But the point is other people already saw this. But it’s a real concept and it was actually tested in a couple of classrooms. So the question is, where did it go? Why didn’t they do any more with it?
That I don’t understand. But there is a drive to get people actively engaged in our planet, its history and its future. And it’s what we’re building this technology to help do.
And OGM can be a part of it. Not all of OGM. But obviously there are the little islands of greatness in OGM that can contribute.
And that’s the whole theory.
Aviation, Back Around
Thank you, Jack.
You never did ask me how I crashed my airplane.
I didn’t want to think about it! Okay, how did you crash your airplane?
Well, my roommate and I, his parents were up in Portland and we were going to fly up to Portland. And we’re flying along from Los Angeles towards Portland up the central valley. And we’re in one of those high pressure days, which means we’re in a headwind, massive headwind. The airplane will indicate 150 miles an hour, but our speed over the ground was nothing like that. So when we got up to Shasta Lake, I said, screw it. And I dropped down on the lake. That way I got out of the winds and I’m blasting along. And there’s a cable that suspends a marina and I hit the damn cable.
The cable tried to saw the wing off but it didn’t. The cable broke before it finished. But the point is that I realized something was wrong. I didn’t know what. I did the right thing, as I just killed the engine. Started looking for a place to land. And the only place to land was in the water. That’s what I did.
How far from shore?
Actually, that was the problem. It was getting perilously close to shore. I thought there was this cove and I was gonna turn up this cove and it wasn’t a cove at all. And so I just ditched it probably 50 feet from shore, but the banks are vertical in that area. And the airplane descended onto a shelf, a local minimum at 80 feet of water.
Wow. And then what?
Well, I came up a couple of days later and hired a diver to go down and find it. And he found it and tied a rope on it. And then we had a big power boat come along and we pulled it backwards over to a boat launching ramp and a Jeep was up the ramp. So he grabbed the plane and pulled it up the ramp and I had a trailer and took the wings off and put it on the trailer. But the spar... so, the wing goes out horizontally and then about four feet out, it has the dihedral.
It’s not dihedral from the fuselage, it’s dihedral halfway out. And the cable had gone in right at the joint where the two wing panels meet, and there was a tiny amount of spar material left. Had I put the power on and tried to fly out of it, that could have broken.
If it had broken, we wouldn’t be having this conversation.
Afterwards, Adam Cheyer used to tell new hires to come and ask me about my “submarine.”
(laughing) So it never got back together?
I actually started rebuilding it and then it was about that time that I met this nice lady and sort of didn’t. I ended up selling all the components that I had started with.
I mean, the airplane itself, the fuselage was thrown away. It was kind of twisted. The wings were gone, but there were parts from it, the horizontal tail and other stuff that was perfectly fine. So I was in the process of assembling a new airplane.
But I just sort of started going another direction. The thing is, the lady I met wanted to live on a farm. And so I got interested in wind power. Now a windmill is nothing but a glider with a bearing...
Coming around full circle!
And I’d been studying aeronautical engineering. That was my undergraduate work. So I wrote a book about wind power and the wind power people would say, well, how do you know where to site it?
So then I started building microprocessor-based weather stations and it started accumulating gobs of data. And the farmers came to me and said, “Hey, I see you’ve got a tower there. Could we predict fruit frost?” And water needs, and more. You can see where this is going.
And then I read a magazine article in Byte Magazine in 1982 about artificial intelligence. And I thought, oh sh*t, this is the answer. And so you can see the Markov chain here, the process.
Not something that I would have predicted from one end to the other. I guess that’s the way Markov chains work...
It really is a pretty nonlinear Markov.
Related:
- Peter Kaminski (author)
- 2024 (year)
- Topics: AI and Technology, Events and Gatherings, Tools and Platforms