Latest / The Adults In The AI Room / Mark Anderson on Pattern Discovery Engines - Episode #11

Transcript

Mark Anderson on Pattern Discovery Engines - Episode #11 · The Adults In The AI Room

• 0:36
  SPEAKER_02: So welcome to today's podcast.
• 0:38
  SPEAKER_02: I'm Seth Early.
• 0:39
  SPEAKER_03: And I'm Chris Featherstone.
• 0:41
  SPEAKER_02: And uh Chris, I know uh I know you have uh had a busy week.
• 0:44
  SPEAKER_02: I'm out here in Las Vegas, my least favorite part in the of the world.
• 0:49
  SPEAKER_02: Uh but at least on uh you're closer to your time zone now.
• 0:53
  SPEAKER_03: That's it it's it's refreshing, actually.
• 0:55
  SPEAKER_03: Welcome to the West, where you you know you get to uh inherit all the things from the east as they're up earlier than you, you know.
• 1:04
  SPEAKER_02: Right.
• 1:04
  SPEAKER_02: Well I think the problem is I've been going to bed on West Coast time and getting getting up on East Coast time, so that doesn't work very well for me.
• 1:11
  SPEAKER_03: Well, it it also though it makes you feel really good about yourself when you spend time in the East Coast and then you come to the West, you're like, dude, I'm up early, early bird gets worm, you know, all of the old uh you know, cliches into place.
• 1:22
  SPEAKER_02: I I yeah, I I switch time zones really fast.
• 1:28
  SPEAKER_02: Anyway, so um well let's let's go ahead and get started.
• 1:31
  SPEAKER_02: But before we do, I just want to thank our sponsors, Simpler Media, the Marketing AI Institute and Early Information Science.
• 1:37
  SPEAKER_02: Uh so our guest today is a scientist, an entrepreneur, and a winner of the prestigious Alexandra J.
• 1:45
  SPEAKER_02: Noble Award for Science Discovery in 2021, uh, which is really exciting.
• 1:50
  SPEAKER_02: He's author, among 35 other prestigious authors, of a new paper titled Learning from Learning Machines, a New Generation of AI.
• 1:59
  SPEAKER_02: And he's here today to challenge our fundamental notions about how science is done and what's next for AI.
• 2:05
  SPEAKER_02: Please welcome the CEO of Pattern Computer, Mark Anderson.
• 2:09
  SPEAKER_03: Hey Seth.
• 2:10
  SPEAKER_03: I feel like you know, we should have these little uh audio things that start to do the applause, like yay, you know, and stuff like that.
• 2:16
  SPEAKER_03: But it's it's great to have you on, Mark.
• 2:18
  SPEAKER_03: Really appreciate you spending the time with us.
• 2:21
  SPEAKER_03: I think the one thing that I've that I've been able to, you know, just from our early interactions, is you know, I which I really appreciate is to, I think, and this may be something that we want to get into, but all of the current thinking that I had in preconceived bias and notion, which that's what it is, it really needs to be questioned and thrown against the wall and say, okay, what is are we really doing?
• 2:45
  SPEAKER_03: Because, like as we as we talked and interacted about, you know, just challenging the status quo and the hypothesis that you set and what that looks like.
• 2:53
  SPEAKER_03: And really, it feels like it's the collective work of your life to actually really challenge those ideas and true to try to get to the truest form and sense of how do we actually you know go and get the truth?
• 3:06
  SPEAKER_03: How do we make you know sense of it?
• 3:09
  SPEAKER_03: And then how do we apply it to give us more abilities to to gain more knowledge and insights, right?
• 3:15
  SPEAKER_02: And and Chris, exactly that's what we want to dive into, but I do want to get the all the audience kind of oriented around market.
• 3:22
  SPEAKER_02: There's so much to talk about, but Mark, maybe you could begin and we'll dive into what uh Chris is uh uh suggesting.
• 3:29
  SPEAKER_02: But why don't you tell us a little bit about yourself?
• 3:30
  SPEAKER_02: I know you had a really great 2021.
• 3:33
  SPEAKER_02: Uh we'll talk about that as well.
• 3:34
  SPEAKER_02: Uh, but we also want to know um what led up to it.
• 3:38
  SPEAKER_02: Where did it all start for you?
• 3:39
  SPEAKER_02: And and by the way, I have been listening to the Google Talks and I did uh read uh I'm up to chapter 15, I believe, in the pattern future.
• 3:49
  SPEAKER_02: So I'm just about to uh I'm just about to solve uh climate change.
• 3:54
  SPEAKER_03: Uh which which by the way, I'm I'm sorry for jumping.
• 3:57
  SPEAKER_03: I'm just super excited to have him on, right?
• 3:59
  SPEAKER_03: As you can tell, I got lots of thoughts and questions, ideas.
• 4:03
  SPEAKER_03: Go ahead, Mark.
• 4:04
  SPEAKER_03: We'll let you take the take the stage.
• 4:06
  SPEAKER_01: Well, um, always been a science guy.
• 4:08
  SPEAKER_01: Uh since I was like five, I always knew I was gonna be a science guy.
• 4:11
  SPEAKER_01: I was disappointed when I finally figured out during my PhD program that I wasn't gonna be a science guy, but I think I still am.
• 4:18
  SPEAKER_02: And so I'm a scientist at heart.
• 4:21
  SPEAKER_02: I am as well.
• 4:21
  SPEAKER_02: I have chemistry undergraduate and I still read science publications every week.
• 4:26
  SPEAKER_01: So tell us about it.
• 4:32
  SPEAKER_01: Um as you you referred to, I published three papers, uh, physics uh physical review category papers.
• 4:40
  SPEAKER_01: So so physics theory papers uh this year in about eight weeks.
• 4:44
  SPEAKER_01: So um that that was added to a family of uh three before.
• 4:48
  SPEAKER_01: So I have six in a series called resonance theory.
• 4:52
  SPEAKER_01: Um we can back that if you want to, but uh so I'm still actively working in theoretical physics on my own dime, basically.
• 5:00
  SPEAKER_01: Yeah.
• 5:02
  SPEAKER_02: Um I read I read a lot about your uh resonance theory and and how it course correlated with string theory and uh and really all of the subatomic uh resonances, and it was it was fascinating and it and it actually gave me a little bit clearer understanding of some of the quantum physics, uh, which are always uh beyond me my understanding.
• 5:23
  SPEAKER_02: But uh that was very refreshing.
• 5:25
  SPEAKER_01: I'm happy to hear that.
• 5:26
  SPEAKER_01: I I think that's generally true about resonance, it helps frame things in a new way of seeing physics itself.
• 5:34
  SPEAKER_01: Uh, I'll just say it.
• 5:35
  SPEAKER_01: The the some the summary of all the work in resonance theory is that the uh physical properties of otherwise empty space, space is not empty, uh lead directly to the laws of physics.
• 5:50
  SPEAKER_01: So you start everything, you see everything through space itself and its physical nature instead of seeing all the billiard balls.
• 5:57
  SPEAKER_02: Right, right.
• 5:59
  SPEAKER_02: And and then the other things that were fascinating that that um that I gleaned from your your book uh was the um was was the the the in the flows, the whole idea of everything being a flow.
• 6:11
  SPEAKER_02: And I've always talked about information flows in the enterprise and looking at patterns from those flows.
• 6:17
  SPEAKER_02: In fact, when we talk about information management, we talk about speeding up the information metabolism of the organization.
• 6:23
  SPEAKER_02: So that worked in comp complexity theory, uh in knowledge processes and uh pattern recognition to me really, really resonated.
• 6:32
  SPEAKER_02: I also like the fact that uh you uh you do a lot of meditation and mindfulness.
• 6:37
  SPEAKER_02: That's another thing that I found from the other area.
• 6:39
  SPEAKER_02: So we're really you're really uh you're resonating with me, is what I have to say there.
• 6:43
  SPEAKER_02: But let me let me thank you for the pitch.
• 6:47
  SPEAKER_01: There it is.
• 6:48
  SPEAKER_01: Yeah, and uh uh actually not flows.
• 6:54
  SPEAKER_01: Let me Seth, let me add in to what you said.
• 6:58
  SPEAKER_01: Everyone believes in flow, I think.
• 7:00
  SPEAKER_01: The great discovery there wasn't flows, which um Leonardo himself actually had discovered, it it was flow and interaction as a pairing of complementary fundamental actions.
• 7:13
  SPEAKER_01: And that discovery is probably the most important discovery I've ever made in my entire life.
• 7:17
  SPEAKER_01: And that's when we were talking about earlier that Curtis Wong said you have now achieved what Leonardo aspired to.
• 7:22
  SPEAKER_01: So the idea that there were these two complementary things behind chaos or complex complexity mathematics, even the people at Santa Fe Institute, who I've talked to individually who seem to agree, but I've never been down there, um, you know, they lack a fundamental theory.
• 7:37
  SPEAKER_01: Here it is.
• 7:39
  SPEAKER_01: So literally, if you ask Murray Cantor, who's a pretty well-known IBM fellow, when we had him at our fire conference, his comment was, Mark, this applies to everything.
• 7:48
  SPEAKER_01: I said, Everything is that everything in the universe.
• 7:51
  SPEAKER_01: This is a truly fascinating, fun, deep, uh rather wonderful idea that once you get your arms around everything looks now through that lens of not just of space being not empty, but now we can also do flow and interaction as the two complementary fundamental actions that drive everything in your life.
• 8:12
  SPEAKER_03: Will you will you double click on that con on that uh comment that Curtis made, you know, because um it's one it's it it's not it shouldn't be you know taken lightly about what that meant and the work that Curtis did and how that applies and really what Leonardo you know Da Vinci was doing, but anyway, will you double click on that just for a sec?
• 8:33
  SPEAKER_03: Because I think it's fascinating the readers would or the listeners would love it.
• 8:37
  SPEAKER_01: I I will.
• 8:38
  SPEAKER_01: So it took me 30 years to do this, not not of intense effort.
• 8:42
  SPEAKER_01: I just kept it took a long time.
• 8:44
  SPEAKER_01: And um when I finally got it figured out, because I run the future in review conference, I had the luxury of being able to ask.
• 8:51
  SPEAKER_01: I was already asking people to come with flow in their mind.
• 8:54
  SPEAKER_01: So I went back to our top day speakers and said, Can you do flow and interaction?
• 8:58
  SPEAKER_01: And every one of them said, Yeah, that's a perfect fit.
• 9:01
  SPEAKER_01: So Bill Janeway, a lot of very famous people came in.
• 9:04
  SPEAKER_01: Uh, Murray Cantor was there, and so we did a whole day of whether you're doing politics or economics or the the shape of leaves or or coasts or complexity mathematics and the structure of the eye and so on.
• 9:16
  SPEAKER_01: And it all, you know, flow and interaction.
• 9:19
  SPEAKER_01: And then um one of the people there was the founder of Dent Conference, and he asked, Could I come and do this at Dent?
• 9:26
  SPEAKER_01: I said, Sure, be happy to.
• 9:27
  SPEAKER_01: How much time do I have?
• 9:28
  SPEAKER_01: He said, eight minutes.
• 9:30
  SPEAKER_01: So my EA and I trained to do an eight-minute talk, include so the whole theory plus eight different fields.
• 9:38
  SPEAKER_01: And then um we went and gave a talk, and Curtis was there at Dent.
• 9:42
  SPEAKER_01: And at the speaker's um wine event later, I saw Curtis, and I know him, and I said, What do you think of that of that piece we did on flow and interaction?
• 9:51
  SPEAKER_01: And he said, Mark, you have achieved what Leonardo aspired to.
• 9:56
  SPEAKER_01: And I thought, you know, he's not just blowing smoke.
• 9:59
  SPEAKER_01: Curtis is a pretty serious guy.
• 10:00
  SPEAKER_01: And I went back and I reread his uh Curtis is Bill Gates' curator for what's called the Codex Lester.
• 10:07
  SPEAKER_01: It's a Leonardo article, and and I went back and reread the Codex Lester.
• 10:12
  SPEAKER_01: And here's what here's what Leonardo was doing.
• 10:15
  SPEAKER_01: He was standing on bridges looking at water and seeing the flow of a stream and finding a fundamental truth in that.
• 10:26
  SPEAKER_01: And and he would look, you see, like a woman's hair.
• 10:28
  SPEAKER_01: You say that's the same thing you're seeing there as you're seeing in that stream.
• 10:33
  SPEAKER_01: So fractal he had a well, yeah, that's we could go there too.
• 10:40
  SPEAKER_01: But um, but what he was missing was, and that's why Curtis said it the way he said it Leonardo didn't get interactions.
• 10:47
  SPEAKER_01: He got the flow part right, but he hadn't had the chance to get the other part.
• 10:52
  SPEAKER_03: That's interesting.
• 10:53
  SPEAKER_03: Yeah, yeah.
• 10:54
  SPEAKER_03: That's super interesting.
• 10:55
  SPEAKER_03: Finding the yeah, just looking at the flow pattern, right?
• 10:58
  SPEAKER_03: And finding that patterns on top of it.
• 11:02
  SPEAKER_03: Yeah, yeah, yeah.
• 11:03
  SPEAKER_03: That's interesting.
• 11:05
  SPEAKER_01: I go far, I go so far as to say that the ultimate practice someday for me will be what you just said to do pattern discovery against flow interactions.
• 11:14
  SPEAKER_01: That's it.
• 11:16
  SPEAKER_03: That's where it comes together.
• 11:17
  SPEAKER_03: I I at some point I do, I you know, I this is it's so fascinating because obviously this does and will um at some level apply to artificial intelligence, machine learning, and how we get business derivatives out of that, right?
• 11:31
  SPEAKER_03: To really make insightful decisions.
• 11:32
  SPEAKER_03: But absolutely before we get to all that, right?
• 11:35
  SPEAKER_03: Um, I think, you know, Seth, you were gonna jump into some areas around, you know, some of the backgrounds that that Mark actually you know works on and stuff.
• 11:43
  SPEAKER_02: Sure.
• 11:43
  SPEAKER_02: Well, I know that uh I know from your book that uh from an early age you were really at uh building, as you say, your biocomputer by being exposed to uh classical music and scientific experiments and uh mechanical uh problems and you know, and that really set your your your mind up uh from a very early age to be in that mode of uh of identifying patterns and taking very diverse concepts and finding relationships between them.
• 12:14
  SPEAKER_02: And of course, you went on to you know get multiple degrees and talk about that path of of pattern discovery and then how that kind of evolved in your career.
• 12:24
  SPEAKER_01: Sure.
• 12:26
  SPEAKER_01: So uh I uh managed to get myself fired from a job um where I was um I built the first the world's first whale uh research-based whale museum.
• 12:40
  SPEAKER_01: And um uh then I got myself fired because they wanted to take all the money out of it, and we had a deal not to take all the money out of it, and then uh yeah, I they tried to make me CEO and then they fired me.
• 12:49
  SPEAKER_01: So suddenly I'm on the street, and uh I had I had been really bothered by light, and I went to Stanford, but I was really actually angry with the way that they taught physics there.
• 13:01
  SPEAKER_01: And uh it was juvenile, it was embarrassing, and uh and they were saying things that didn't make any sense in terms of like conservation of energy, and it just stuck in my craw.
• 13:09
  SPEAKER_01: So I thought, well, here's a chance, you know, do something interesting, come back to that.
• 13:14
  SPEAKER_01: And I what I did was um really fun.
• 13:17
  SPEAKER_01: I had I had a um waterfront house and I had an eight-foot table, like a dining room table, and 150 feet of butcher paper and all these colored pens, and I started looking at all the force laws of physics uh with an eye toward coming back to the lightning again.
• 13:31
  SPEAKER_01: And what I was trying to do was find the mathematical patterns in physics.
• 13:37
  SPEAKER_01: Instead of studying physics like at Stanford, which was a complete, I'll say it again, I didn't like it, but you know, so try something orthogonal to that, right?
• 13:46
  SPEAKER_01: Take take whatever you're good at in pattern recognition and apply that, period.
• 13:50
  SPEAKER_01: Just apply that raw to math in that science and see what you get.
• 13:57
  SPEAKER_01: And so I had all, you know, yellow, red, and pink and green.
• 14:00
  SPEAKER_01: And so I so and and the beautiful thing compared to a computer, for instance, is you've got eight feet by three feet, and if we go like this twice, you got 20 three times as much.
• 14:10
  SPEAKER_01: That's a pretty big area you can look at, right?
• 14:12
  SPEAKER_01: So um, it wasn't that hard to do this, and it worked.
• 14:17
  SPEAKER_01: I'm not gonna go into detail, but it worked.
• 14:18
  SPEAKER_01: So I was able to make a discovery uh right away that no one had made before, and I reduced all these laws to a single mathematical formula, which I proved by putting into a Commodore 64 uh program.
• 14:30
  SPEAKER_01: All right, so um on a color TV, right?
• 14:34
  SPEAKER_01: So that was great, and I realized uh something right then, and I checked it out with a bunch of physics guys to make sure I was on the right track.
• 14:41
  SPEAKER_01: I thought, okay, a Ninkin poop can use this toolkit of pattern recognition to make major discoveries in physics that had been unknown.
• 14:52
  SPEAKER_01: I got offered a PhD program, you know, to come and I didn't do that, but so it was it was real, and I thought this is very interesting.
• 15:02
  SPEAKER_01: There's something here that I didn't expect.
• 15:05
  SPEAKER_01: It wasn't just we all use pattern recognition every day in our brains because we do.
• 15:09
  SPEAKER_01: Yeah, it was more than that.
• 15:11
  SPEAKER_01: And now I know we call it pattern discoveries, but I didn't have a word for it at the time.
• 15:16
  SPEAKER_01: But the short the short answer there was it turns out if you have a very broad funnel, instead of being more and more specialized the way they teach you at Stanford, unspecialized, look very widely, apply a lot of pattern recognition skills to that, you're going to make major discoveries that even the folks who are the world's experts didn't make.
• 15:38
  SPEAKER_01: Or you'll find something out about a so later I started publishing SMS or strategic news service.
• 15:44
  SPEAKER_01: So that's been going on every week since '95.
• 15:47
  SPEAKER_01: And the whole predicate there was to make predictions, accurate predictions, graded predictions in public and turn it into a science.
• 15:55
  SPEAKER_01: Instead of it being an op-ed piece, it was a science.
• 15:58
  SPEAKER_01: Take it seriously.
• 15:59
  SPEAKER_01: And that whole bullshit thing about you can't predict the future.
• 16:03
  SPEAKER_01: Yes, you can.
• 16:05
  SPEAKER_03: I would do it in 95.
• 16:06
  SPEAKER_03: I know you made some pretty good, some, you know, pretty spot-on predictions, and we can go into that some point.
• 16:12
  SPEAKER_03: I think the the key too is with your company pattern computer and with the way that you think about these kind types of things, there's some questions that naturally I think come up, like, okay, well, then one, what data do you need to actually look at the at the to prove the pattern even works, right?
• 16:34
  SPEAKER_03: It may be out there and you may be thinking about it from the perspective of, like you said, orthogonally looking at the bigger picture.
• 16:40
  SPEAKER_03: Is that a pattern that you get validated or invalidated?
• 16:44
  SPEAKER_03: Sometimes you have to do that with data.
• 16:46
  SPEAKER_03: And then how does that also go, maybe either in defiance of let's say hypothesis-based, you know, science, or maybe it's indirect to it?
• 16:57
  SPEAKER_03: Go through those two two things.
• 16:59
  SPEAKER_03: So the data that you need, and then how that actually you know really uh goes against hypothesis-based.
• 17:05
  SPEAKER_03: Let me flip it if I can.
• 17:07
  SPEAKER_01: Do it, do it, yeah, absolutely.
• 17:08
  SPEAKER_01: So the whole point of what I've what we've just discussed is you do not start with a hypothesis.
• 17:14
  SPEAKER_01: And that's true at pattern computer, and that's true in the science of what I'm talking about.
• 17:18
  SPEAKER_01: So so instead of instead of that, uh, you do the opposite.
• 17:22
  SPEAKER_01: You take the data first, which brings us back to your second question.
• 17:25
  SPEAKER_01: Yeah.
• 17:25
  SPEAKER_01: And and uh each story is different, of course.
• 17:29
  SPEAKER_01: There are very important statistical ways of questioning this.
• 17:33
  SPEAKER_01: I want to add something which is important to I think the audience for this program.
• 17:38
  SPEAKER_01: Today, when someone says AI, 99.5% of the time they mean neural networks.
• 17:43
  SPEAKER_01: We don't.
• 17:45
  SPEAKER_01: And they have some really big walls, really big limitations that um are why we don't.
• 17:50
  SPEAKER_05: Yeah.
• 17:51
  SPEAKER_01: Um, so uh uh including explainable AI.
• 17:54
  SPEAKER_01: So we've done that already.
• 17:55
  SPEAKER_01: Uh we're putting on a program next month about that.
• 17:58
  SPEAKER_01: So uh, and as you mentioned, there's a paper we just wrote with 35 co-authors about that.
• 18:02
  SPEAKER_01: So um you you don't have explainability, uh, you don't get make you don't make pattern discoveries.
• 18:07
  SPEAKER_01: Do not make big, big discoveries in general with the tools that Google uses.
• 18:12
  SPEAKER_01: You just don't.
• 18:13
  SPEAKER_01: So um uh instead of having a hypothesis, you start with the data itself.
• 18:20
  SPEAKER_01: You can find out, Chris, if there's missing data.
• 18:22
  SPEAKER_01: So we did this with an aerospace customer where we found out, oh, there are latent variables.
• 18:27
  SPEAKER_01: We can tell you a little bit about them.
• 18:28
  SPEAKER_01: We tell you where they occurred, when they occurred, you go find them.
• 18:31
  SPEAKER_01: And they did, and we helped them improve their manufacturing quality for tier one aerospace company.
• 18:37
  SPEAKER_01: So, you know, it's you don't always have what you want to start with.
• 18:40
  SPEAKER_01: You have to have what we call a Y value.
• 18:43
  SPEAKER_01: You have to know what you are after.
• 18:46
  SPEAKER_01: And we had one fail, the only fail I've ever seen us have, really big fail.
• 18:50
  SPEAKER_01: We had a data set from a very famous person who's on our A B, I won't mention his name, geneticist, and there was no Y value.
• 18:56
  SPEAKER_01: There was nothing to go after.
• 18:57
  SPEAKER_01: It was a bunch of data about healthy young tech guys.
• 19:00
  SPEAKER_01: It was useless, you know.
• 19:01
  SPEAKER_01: It was just so um uh if in the case uh you're asking about here's a pretty good example.
• 19:08
  SPEAKER_01: We went after Metabric, which was a publicly open source data set on triple negative breast cancer.
• 19:15
  SPEAKER_01: Uh Berkeley Labs held it, been around for 20 years.
• 19:18
  SPEAKER_01: Everybody in the world who cares had been through it.
• 19:22
  SPEAKER_01: On average, the average paper written about that data discussed one gene expression or one gene.
• 19:28
  SPEAKER_01: One.
• 19:29
  SPEAKER_01: Three of the papers we found talked about two or three genes.
• 19:31
  SPEAKER_01: Yeah.
• 19:32
  SPEAKER_01: We did our very first run with no hypothesis, but there was a Y value.
• 19:36
  SPEAKER_01: The Y value was did women die or not die?
• 19:39
  SPEAKER_01: Okay, that's what we want to know.
• 19:41
  SPEAKER_01: So we with that in hand, in the data, we could look at this open source data set, and we immediately came up with what now are two different combo drugs, both of which are in their fifth round of positive testing, good tests, at major places like Berkeley Labs and Charles River, and both of which killed 100% of the cancer cells in the in the category, which is the only no-direct treatment category in the world for breast cancer.
• 20:10
  SPEAKER_01: Yeah.
• 20:11
  SPEAKER_01: So uh we we had found 17 combo drugs that we picked out of the, I'll make up a raw a raw number, 50 to 100 gene expressions that we found that had not been found before.
• 20:24
  SPEAKER_01: Not one.
• 20:26
  SPEAKER_01: So imagine if you had a new tool, and again, I'll say it differently.
• 20:30
  SPEAKER_01: This is what what Murray Cantor calls new mathematics.
• 20:32
  SPEAKER_01: We gave him an NDA and he took a look.
• 20:34
  SPEAKER_01: So it's not just a deep neural network.
• 20:36
  SPEAKER_01: This is literally going back to the 50s, picking up some lost threads, come bringing them forward, building new mathematics to do this one job, right?
• 20:46
  SPEAKER_01: Pattern recognition, pattern discovery.
• 20:48
  SPEAKER_01: And that's true, by the way, for the whole stack and the whole architecture of our system.
• 20:52
  SPEAKER_01: So that's what we do.
• 20:53
  SPEAKER_01: That's that's all we do.
• 20:54
  SPEAKER_01: But we didn't have to know anything, really.
• 20:56
  SPEAKER_01: We just had to have the right system.
• 20:58
  SPEAKER_05: Yeah.
• 20:58
  SPEAKER_01: So then you see, you know, 50 to 100 new correlations of gene expressions.
• 21:04
  SPEAKER_01: You can see how they can relate to each other as well as to the problem, to the death of women, and then you can see the mechanism that they've got, and then you can look at this, say, let's have this pre-approved FDA drug and this pre-approved FDA drug together to go after that unknown mechanism and see if we can kill the cancer.
• 21:24
  SPEAKER_02: That's amazing.
• 21:25
  SPEAKER_02: So the inputs for that, the inputs for that is genomics data.
• 21:30
  SPEAKER_02: Yes.
• 21:31
  SPEAKER_02: What else?
• 21:31
  SPEAKER_02: What else did you have?
• 21:34
  SPEAKER_01: See good seek data on RNA uh for um not a very large number of women, uh around a thousand, I think.
• 21:41
  SPEAKER_02: So that's I mean there'd be something about gene expression or these are gene expressions.
• 21:48
  SPEAKER_01: RNA I'm sorry, RNA seq data is gene expressions.
• 21:51
  SPEAKER_02: Yeah.
• 21:52
  SPEAKER_02: Okay.
• 21:53
  SPEAKER_02: And then uh what about uh what about the um the results or the clinical uh or observation?
• 22:00
  SPEAKER_02: Stuff in terms of how these who died and who didn't.
• 22:05
  SPEAKER_02: Okay.
• 22:06
  SPEAKER_02: Yeah.
• 22:08
  SPEAKER_02: And and and let me ask this question.
• 22:11
  SPEAKER_02: If uh how is the why, and this might be semantics, or maybe there's a uh deeper thinking here that I'm not seeing, how is the why different than the hypothesis?
• 22:21
  SPEAKER_02: Isn't that premised on there is no hypothesis in a why question?
• 22:27
  SPEAKER_01: Look, let me put it this way.
• 22:28
  SPEAKER_01: In every company, I'm gonna make a little assumption here, until we showed up.
• 22:33
  SPEAKER_01: Every company probably that wanted to go after this from Merck on down had a hypothesis.
• 22:39
  SPEAKER_01: What if it's this gene?
• 22:40
  SPEAKER_01: What if it's this enzyme?
• 22:42
  SPEAKER_01: What if it's this thing?
• 22:43
  SPEAKER_01: You know, what if it's what lifestyle problem drinking too much, or and whatever it was, it's one or two hypotheses together.
• 22:50
  SPEAKER_01: We didn't do any of that.
• 22:51
  SPEAKER_01: None, zero.
• 22:53
  SPEAKER_01: So we did not have a hypothesis.
• 22:55
  SPEAKER_01: The y value is not the hypothesis, it's what you're after, it's the solution you're after without knowing what hypothesis to make.
• 23:04
  SPEAKER_01: So the way we got hypotheses was after we ran the data.
• 23:08
  SPEAKER_01: Then it gave you more questions.
• 23:10
  SPEAKER_01: We call the machine our discovery, and validate it's a hypothesis generator.
• 23:16
  SPEAKER_01: And it doesn't come from our bias or our teacher at Stanford or something, it comes from the data itself.
• 23:22
  SPEAKER_03: Yeah, it's almost like you're like you're saying, you know, listen, our y value is the outcome is yes, people die or not die.
• 23:30
  SPEAKER_03: Yes.
• 23:30
  SPEAKER_03: Now let's look backwards to the data that we do have, and let's look at all the genetic markers and the patterns that are there just to come up with is are those the are those the causal things that are or correlative or correlative.
• 23:43
  SPEAKER_03: Yeah, exactly.
• 23:44
  SPEAKER_03: I think that's I I at some point too.
• 23:46
  SPEAKER_03: I want to bring that the whole discussion of the flow and interaction back to this because how do that does that help you formulate the mathematics, you know, for the engine to actually go through and find those patterns?
• 23:57
  SPEAKER_03: Because there's there's data that you like you said, there's it it'll tell you where you're missing data also, and then and then think about it from the perspective of, oh, well shit, maybe we need to go get some of that information and bring it in, yeah, right?
• 24:11
  SPEAKER_03: Right.
• 24:11
  SPEAKER_03: So anyway, I don't know if that if that if I'm if I'm hopefully you know pairing those together because the the so I'm still you know spinning out on the flow and interaction type of of uh physics there and how that actually works with well, I want to frame this a little bit together, you know what I mean?
• 24:27
  SPEAKER_01: I want to frame this a little bit better, Chris, because we're we're not done with the program you and I have just laid out here in this hour.
• 24:33
  SPEAKER_01: So uh that's you know that's 10 years of work, probably.
• 24:35
  SPEAKER_01: Yeah, yeah um the the work that we did on the mathematics for the pattern computer system is not based on flow and interaction, it is based on pattern recognition.
• 24:46
  SPEAKER_03: Okay.
• 24:46
  SPEAKER_01: So pattern discovery and so um and a bunch of other things.
• 24:51
  SPEAKER_01: So uh we built engines more than one uh over the last six years that are the best in the world that we know of at that.
• 25:00
  SPEAKER_01: Yeah, and our our North Star was be the be the best purpose built first and best purpose-built system for finding patterns, period.
• 25:11
  SPEAKER_01: Period.
• 25:12
  SPEAKER_01: Now, how you you apply that to flow and interaction or to something else is in the future.
• 25:18
  SPEAKER_03: Yeah, okay.
• 25:19
  SPEAKER_03: Yeah, I I didn't want to conflate the two.
• 25:21
  SPEAKER_03: That's what I was asking, right?
• 25:22
  SPEAKER_03: You want to make sure that that I'm clear and that everybody else is clear also.
• 25:25
  SPEAKER_03: I definitely want to do that work, but we haven't done it.
• 25:28
  SPEAKER_03: Well, we'll have to have part three, four, six, seven, twelve, whatever it is.
• 25:33
  SPEAKER_03: So yeah.
• 25:35
  SPEAKER_02: Well, I do I do want to take a moment just to remind people that we're uh speaking with with Mark Anderson, who is CEO of Patterns Computer, and also like to uh uh once again thank our sponsors, Marketing AI Institute, Simpler Media, and Early Information Science.
• 25:49
  SPEAKER_02: So with that, let's let's carry on.
• 25:51
  SPEAKER_02: And um, Mark, so you know, one of the things that I thought was really interesting about your observation, and and I've actually made this kind of assertion too in the past, is that uh the patterns are there.
• 26:03
  SPEAKER_02: It's not a matter of opinion and it's not a matter of something you make up, it's like it's it's what's in nature, it's like it's it's but it is the nature of things.
• 26:12
  SPEAKER_02: The patterns are there and it's a matter of identifying them.
• 26:15
  SPEAKER_02: Do you want to expand on that a little bit?
• 26:17
  SPEAKER_02: Because it really does say that this is the way uh the world works, yes, uh, based on what you observe and what we can understand, and then finding the patterns within those uh uh whatever that realm is, whatever that domain is.
• 26:32
  SPEAKER_02: Do you want to talk more about that?
• 26:34
  SPEAKER_01: Yeah, sure.
• 26:35
  SPEAKER_01: And we made up a little elevator speech last week, Seth.
• 26:38
  SPEAKER_01: Uh see if I can get it right.
• 26:40
  SPEAKER_01: Um, who knew that the world was made of patterns?
• 26:44
  SPEAKER_01: We all did.
• 26:46
  SPEAKER_01: Who knew that you needed a pattern computer to understand them?
• 26:49
  SPEAKER_01: We did.
• 26:51
  SPEAKER_01: And the you know, the stunning, the stunning thing when you go back and look at our story, we built we built pattern computer originally at at our fire conference on stage.
• 27:00
  SPEAKER_01: But you know, it's like the the idea that no one had done it yet, that was 2015.
• 27:06
  SPEAKER_01: That's a that's an amazing fail.
• 27:09
  SPEAKER_01: Yeah, an amazing opportunity for us because with all the work that had gone into computer design and chip design and mathematics and programming, uh Lisp, I mean, make a long list of all the technologies that we had come up with by 2015.
• 27:23
  SPEAKER_01: And it hadn't apparently occurred to someone to build a pattern computer.
• 27:29
  SPEAKER_01: When in fact, nature, as you just pointed out, I mean, all the problems around us, whether it's airplane manufacturing or whether it's how does a leaf work, are all based on patterns, you know, and our brain, our brains work in pattern recognition, I believe, because our brains evolved in a world made of nature.
• 27:48
  SPEAKER_01: And and so to be effective, we had to be good at that, right?
• 27:52
  SPEAKER_02: Right, right.
• 27:53
  SPEAKER_02: We're pattern, we're pattern making machine or identifying machines, right?
• 27:57
  SPEAKER_02: When we start to learn as a child, we're looking for patterns and we're looking, and that's what's imprinting upon us, and that's what's causing building the circuitry to identify more patterns.
• 28:08
  SPEAKER_01: It's kind of surprising that no one had thought to build this until we built it.
• 28:13
  SPEAKER_02: Um tell me how the the fundamental when you say what we did was different.
• 28:20
  SPEAKER_02: What is fundamentally different?
• 28:22
  SPEAKER_01: Well, every part of every part of the system is proprietary except for the chips.
• 28:29
  SPEAKER_01: Even the Linux got stripped down to very thin layer, and we re we rebuilt all of the the drivers for Linux uh for those chips.
• 28:36
  SPEAKER_01: And the management layer above that uh is ours alone.
• 28:39
  SPEAKER_01: And and so, and everything above that is ours alone.
• 28:42
  SPEAKER_01: So every part of that is proprietary.
• 28:44
  SPEAKER_01: We were in stealth mode for three years uh before we did our first coming out party in in San Francisco, called it Splash One, and and we started showing results, but uh we were we didn't even have the same name.
• 28:55
  SPEAKER_01: We're called Coventry Computer.
• 28:58
  SPEAKER_01: So we were really careful to be quiet about it and worked our butts off, uh mostly mathematics and in engineering.
• 29:05
  SPEAKER_01: Uh very, very smart veterans out of Amazon and Microsoft and other places uh in math and in architecture.
• 29:14
  SPEAKER_01: And so we built the whole architecture from scratch.
• 29:19
  SPEAKER_01: And are you go ahead, sorry, I mean to cut you off.
• 29:22
  SPEAKER_01: I had a hint that this was going to work, not just from the work I'd done uh in my other career, but um, I wrote a paper about three years before that called The Most Important Chip Not Yet Invented, and I called it the pattern recognition processor for the same reason.
• 29:39
  SPEAKER_01: And so I wrote this paper, it was a concept design, and then I got a call from a guy who became a good friend of mine, um, Darmendra Moda, who was secretly the project lead on an eight-year DARPA IBM project called True North.
• 29:56
  SPEAKER_01: And True North turned out to be the very first pattern recognition processor, but I hadn't heard of it, right?
• 30:03
  SPEAKER_01: So he invited me over to help launch the thing at Almaden, and I did that.
• 30:06
  SPEAKER_01: And so when I saw the amount of effort worldwide that had gone into that chip, I realized, okay, this isn't just an idea, this is an idea whose time has come.
• 30:20
  SPEAKER_01: And if someone types of problems, if somebody like if someone like IBM and DARPA had said that we're gonna put whatever it is, 200 million into this, then it was like, okay, who's gonna build a computer?
• 30:32
  SPEAKER_05: That was the obvious thing.
• 30:36
  SPEAKER_02: So so the you've you've addressed uh this the some life sciences uh issues, you've you've you're looking at at uh treatment of breast cancer.
• 30:46
  SPEAKER_02: What other classes of problems are you currently focusing on or directing the organization towards?
• 30:54
  SPEAKER_02: So you know clearly you got this great uh valuation from a business model and projections, and and I and did you say it was a two billion dollar valuation?
• 31:03
  SPEAKER_02: 1.2 to make 1.2, sorry, sorry, 1.2.
• 31:07
  SPEAKER_02: I exaggerated uh it'll be it'll be two soon enough.
• 31:10
  SPEAKER_02: So that's amazing.
• 31:11
  SPEAKER_02: So what are the what is what's the trajectory?
• 31:14
  SPEAKER_02: Where are you going with this and where do you see the biggest opportunities for this approach to solve these types of intractable problems?
• 31:24
  SPEAKER_01: Well, first of all, it's agnostic, so it's you know you get to pick what field you want to be in.
• 31:29
  SPEAKER_01: It works equally well in agriculture, I think, as it does in bio or as it does in markets or as it does in uh airplane design.
• 31:39
  SPEAKER_01: So um we we have to be careful what to pick because we're still a small team.
• 31:44
  SPEAKER_01: Um we picked bio for a number of reasons.
• 31:46
  SPEAKER_01: A, it's the hardest.
• 31:48
  SPEAKER_01: Yeah, B, it's the most competitive, C, it had the most venture money going into it.
• 31:52
  SPEAKER_01: Uh D, we we love saving lives.
• 31:55
  SPEAKER_01: I mean, that's just a team unifier.
• 31:57
  SPEAKER_01: You want to have a great, happy team, work on that, right?
• 32:00
  SPEAKER_01: Yeah, so um, and another thing, EFNG, when you succeed, you don't have to beg for approval.
• 32:08
  SPEAKER_01: I like that.
• 32:10
  SPEAKER_01: Because you know what I don't like is like peer review and committees.
• 32:14
  SPEAKER_01: So you know if you if you kill the cancer cells, you killed the cancer cells.
• 32:19
  SPEAKER_01: That's pretty important.
• 32:21
  SPEAKER_01: So, as an example, last year, 2021, as part of the magic year for us, we decided one good way to showcase how good the system is would be to take the top five cancer killers and and move the needle on all of them.
• 32:36
  SPEAKER_01: Now imagine you're a startup in computing and you did that, right?
• 32:40
  SPEAKER_01: That that proof there, I think that's our proof case right there.
• 32:43
  SPEAKER_01: And and by saying the top five killers, what you're really saying, I think, is those are the most competitive arenas by definition.
• 32:52
  SPEAKER_01: Yeah.
• 32:52
  SPEAKER_01: So so we did it.
• 32:54
  SPEAKER_01: And um, pretty excited about that.
• 32:56
  SPEAKER_01: We call it Starbrite project.
• 32:58
  SPEAKER_01: And it's not just breast cancer, it's also lung cancer and ovarian cancer and prostate cancer and uh on and on.
• 33:05
  SPEAKER_02: So far along have you gotten with with those efforts?
• 33:09
  SPEAKER_02: It sounds like you're pretty farthest along with breast cancer.
• 33:11
  SPEAKER_01: How the other efforts uh breast cancer is the furthest along where we actually are are in our fifth round of testing uh treatments.
• 33:18
  SPEAKER_01: We're about to begin testing treatments for ovarian.
• 33:21
  SPEAKER_01: Uh, we have each one has its own story, of course.
• 33:24
  SPEAKER_01: We are working with the Fred Hutch Cancer Research Center on uh colorectal cancer, where we have been working for about two years now on the largest data set in the world of that type.
• 33:34
  SPEAKER_01: And what we've done is we have been able to advance our own mathematics while we were doing this, particularly in clustering, which is quite interesting.
• 33:42
  SPEAKER_01: And so we took, I don't call it AI, I call it AM, advanced mathematics.
• 33:46
  SPEAKER_01: And we were able to um to look at what they had done, which is quite good.
• 33:52
  SPEAKER_01: And then we we are able to beat their general numbers.
• 33:54
  SPEAKER_01: This is a gigantic data set.
• 33:57
  SPEAKER_01: So you're looking at 155,000 patients worldwide, seven different consortia doing the collection, uh, 39 million features per person.
• 34:07
  SPEAKER_01: Wow.
• 34:08
  SPEAKER_01: So the run, the run, you know, do the general run with all the possible correlatives in that run.
• 34:15
  SPEAKER_05: Yeah.
• 34:15
  SPEAKER_01: Yeah, yeah.
• 34:16
  SPEAKER_01: Okay, so you get it.
• 34:17
  SPEAKER_01: And um, we were able to go from, I think they were around the in the low 30s, or in the low 60s, we got we beat them by about five points in general.
• 34:26
  SPEAKER_01: Um, and that's a lot of lives saved if you get it right.
• 34:29
  SPEAKER_01: But then we got up to 70, 80, and 100% by doing the clustering.
• 34:34
  SPEAKER_01: So we're just about to go into a special report with them and explain what we found.
• 34:37
  SPEAKER_01: And so it's kind of a pre-announcement, but um this is a big deal for us.
• 34:41
  SPEAKER_01: So when you get the knowledge based on gene expressions of that type, what that all what that frees you then to do, as we were just discussing with breast cancer, is that frees you to do diagnostics, it frees you to do treatments, and ultimately, I hope prevention.
• 35:00
  SPEAKER_02: And and you're referring to the patient uh data as what?
• 35:03
  SPEAKER_02: Is that considered clinical or observational data?
• 35:07
  SPEAKER_02: What what are you?
• 35:08
  SPEAKER_01: The data in that set was is at with 39 million features per patient, was very wide and included everything from environmental related problems to or or you know data to um full genome, uh gene expression.
• 35:22
  SPEAKER_02: Okay, and it's all and and and and physical, you know, uh all of the variables and geographic and biomarkers and and then environmental stuff.
• 35:33
  SPEAKER_01: So you really you look it's clinical and observational and on environmental data, it sounds like a lot of yeah, and we're not done yet, but we already have done what I just mentioned to you with primarily with the genomic data.
• 35:49
  SPEAKER_01: Right, right.
• 35:50
  SPEAKER_02: And then what other problems?
• 35:51
  SPEAKER_02: So if you were to say listen if someone was listening to this and they worked for a large, you know, uh conglomerate or global manufacturer, one of those an industrial firm, what kinds of things would you uh be able to solve for them, or what kinds of problems would would be appropriate for a company like yours?
• 36:09
  SPEAKER_01: Right.
• 36:10
  SPEAKER_01: So we'd like to say give us your biggest problem, which is a bad way to start any kind of relationship.
• 36:14
  SPEAKER_01: But um as an example, uh we're helping a mining company.
• 36:20
  SPEAKER_01: Um they want to double their throughput, uh their uh their iron miners.
• 36:25
  SPEAKER_01: And so we told them no coal mining, iron is fine.
• 36:28
  SPEAKER_01: And uh we do care about that.
• 36:31
  SPEAKER_01: Um and they gave us some information about um their process, and we were able to very quickly find things about the machines that they use that were wrong, wrong meaning not performing well, costing them a lot of money.
• 36:46
  SPEAKER_01: And there were things that they I don't think even expected.
• 36:49
  SPEAKER_01: So uh with the data they were giving us, we could do things that were a little bit surprising in terms of very, very specific information about chemicals and types of ore and sizes of crushers and all kinds of things that mean a lot of money to them, but um weren't obviously available to them through their current techniques.
• 37:11
  SPEAKER_01: Uh in the aerospace example, um, there was a company that was making a very important um uh type of equipment.
• 37:20
  SPEAKER_01: Uh you don't want it to fail when you're flying.
• 37:22
  SPEAKER_01: And um, they were all the things were passing the tests component by component, but the assembly was failing increasingly.
• 37:30
  SPEAKER_01: Interesting.
• 37:31
  SPEAKER_01: And so we were able to find out enough about that to help them turn that around.
• 37:35
  SPEAKER_02: Um tell me tell me if there are there any downsides or weaknesses or flaws in the approach.
• 37:42
  SPEAKER_02: What is the approach not appropriate for?
• 37:45
  SPEAKER_01: Uh I don't know of any flaws.
• 37:46
  SPEAKER_01: I think you you would probably say there are things that it's better at or not so good at, probably.
• 37:52
  SPEAKER_01: And um so we, as an example, we didn't start out being good at computer vision.
• 37:57
  SPEAKER_01: Uh, we haven't even tried, we have experts on NLP, but we aren't doing NLP right now.
• 38:02
  SPEAKER_01: Uh but we got involved in this computer vision thing at the request of a defense firm.
• 38:07
  SPEAKER_01: And so we learned more and more about that.
• 38:10
  SPEAKER_01: We had already done something that worked out well, and so here's what we did.
• 38:14
  SPEAKER_01: We were able to apply normal techniques, including deep learning, to computer vision and do what anybody would do.
• 38:21
  SPEAKER_01: Then we were able to improve it because we have a really good math team.
• 38:24
  SPEAKER_01: So we were able to make it proprietary.
• 38:26
  SPEAKER_01: That's still you're still in a country where that's kind of what a normal AI company would do.
• 38:30
  SPEAKER_05: Yeah.
• 38:31
  SPEAKER_01: Then we did something uh beyond that.
• 38:33
  SPEAKER_01: We took one of our discovery engines and applied that to the output.
• 38:37
  SPEAKER_01: And then we had something really special because we had two different types of engines working together.
• 38:42
  SPEAKER_01: A hybrid.
• 38:44
  SPEAKER_01: Yeah, the second of which was a world-class world beater in very high dimensional reduction in very complex data sets and without losing the um the original information of the connections.
• 38:57
  SPEAKER_01: Yeah, right.
• 38:58
  SPEAKER_01: And so we call that Leonard uh island, that's our code name for it.
• 39:01
  SPEAKER_01: It's our pattern discovery, our first pattern discovery engine.
• 39:04
  SPEAKER_01: And you know, at that point, you can rank what you've got.
• 39:09
  SPEAKER_01: So uh in one thing we did, we had to take a universe of 10 to the 40th correlations in a microbiome, drop it down to 39 proteins and then nine proteins in about 10 days.
• 39:21
  SPEAKER_01: Wow.
• 39:21
  SPEAKER_01: Larry Smarr brought us that problem, who is the co-inventor of personalized medicine, and he did, you know, he's got his own supercomputer, he couldn't do it.
• 39:28
  SPEAKER_01: So, so this system is extremely good at that.
• 39:31
  SPEAKER_01: We could we could go into computer vision then.
• 39:33
  SPEAKER_01: We we took a problem where um it turns out at that time, at least about four years ago, that um you couldn't use x-rays of children on pneumonia.
• 39:43
  SPEAKER_01: They had to do biopsies, and that was painful and and not good.
• 39:48
  SPEAKER_01: And so um we found that we could use x-rays with our system to a identify that the child had pneumonia, b, where it was localized in the lung, which was hard to do.
• 40:00
  SPEAKER_01: C, more importantly, was it bacterial or viral, all from x-rays?
• 40:04
  SPEAKER_01: And so if you're in Africa and you're, you know, how many miles away from a lab, that time delay can be life or death.
• 40:12
  SPEAKER_01: And and so this was a meaningful thing, which we still haven't pursued enough, but it worked.
• 40:18
  SPEAKER_01: Um, we know it works.
• 40:19
  SPEAKER_01: So that kind of thing was our weakest link, I think.
• 40:24
  SPEAKER_01: And but we could improve it over time and make it better.
• 40:27
  SPEAKER_03: I hate the fact that most um, you know, most of really great technology that gets out in the in the wild is used for nefarious activities first before business figures out how to use it.
• 40:37
  SPEAKER_03: So, how do you keep it, how do you keep it locked up so that you can't put this in the hands of those that want to use it to, you know, the bad guys.
• 40:44
  SPEAKER_03: Yeah, exactly, to do predictions on, you know, like I know you'd done some some good predictions on oil prices and things like that, but the the what where they use that for is to, you know, is like the old boiler rooms where they hedge, you know, their own to you know, either take money from the market people, whatever, and actually cause problems, right?
• 41:04
  SPEAKER_03: So how do you how do you keep it locked up and and used for good?
• 41:08
  SPEAKER_03: We lock it up.
• 41:09
  SPEAKER_03: So you're not gonna publish this as a open source or anything like that and just keep it.
• 41:14
  SPEAKER_01: Murray Cantor, uh, who I mentioned earlier, when he when he came to our our uh we he came to an annual math retreat, which we have, we just had our latest in Berkeley, and he he was so excited, he said, this has to be a paper that was at 10 o'clock in the morning.
• 41:27
  SPEAKER_01: And we said, No, Murray, it's not a paper, you know.
• 41:29
  SPEAKER_01: And by the end of the day, he said, This is a Springer book, you got to do a Springer book.
• 41:33
  SPEAKER_01: So Murray, we're a company.
• 41:35
  SPEAKER_01: But the the real reason isn't just to honor our shareholders, it's it's that what you said, Chris.
• 41:41
  SPEAKER_01: So I'll give you an example.
• 41:42
  SPEAKER_01: I won't name the bank.
• 41:43
  SPEAKER_01: A bank came to us uh after a private meeting with the doe in uh in Chicago, and they wanted us to help them out.
• 41:50
  SPEAKER_01: And um they were asking about flows of this and flows of that, and we we declined.
• 41:57
  SPEAKER_01: We worked uh for about uh four weeks on what they where we're going, and then we said we're not interested in this project.
• 42:03
  SPEAKER_01: Yeah, good.
• 42:04
  SPEAKER_01: So we have the ability to control our destiny, yeah.
• 42:07
  SPEAKER_01: And uh we've done we've done enough good things we don't have to go to the dark side.
• 42:11
  SPEAKER_01: Um, I'll give you an example.
• 42:13
  SPEAKER_01: Um, COVID, right?
• 42:14
  SPEAKER_01: So um Berkeley Labs said a couple years to us ago that they thought our math was a good fit for what's called hyperspectral uh work, um, broadband light.
• 42:26
  SPEAKER_01: Yeah, and uh when COVID appeared, they said again, you know, take a look.
• 42:30
  SPEAKER_01: We did we now have the best COVID detection system in the world for high throughput.
• 42:34
  SPEAKER_01: We're monitoring 400 competitors, so three seconds per test, yeah, about two bucks almost free.
• 42:42
  SPEAKER_01: Um and and no reagents.
• 42:44
  SPEAKER_01: What is the no reagents?
• 42:46
  SPEAKER_01: Yeah, what is what is the mechanism?
• 42:48
  SPEAKER_03: That's using a center for COVID tests, but yeah, to to basically work through all the patterns.
• 42:54
  SPEAKER_03: It's hyper structural.
• 42:55
  SPEAKER_03: Yeah, so we can find patterns.
• 42:59
  SPEAKER_02: What are the markers you're looking at for the COVID test?
• 43:02
  SPEAKER_02: Is it it's it's not a uh it's not an antigen test?
• 43:05
  SPEAKER_02: Is it no what is the input?
• 43:07
  SPEAKER_02: Yeah.
• 43:08
  SPEAKER_01: Oh, you're asking that same question, Seth.
• 43:10
  SPEAKER_01: There is no hypothesis.
• 43:12
  SPEAKER_01: There is no hypothesis.
• 43:14
  SPEAKER_01: Oh, I just understand there is a Y value.
• 43:17
  SPEAKER_01: And the Y value is COVID, right?
• 43:20
  SPEAKER_01: We check everything we did against the gold standard, which is PCR, yeah.
• 43:28
  SPEAKER_05: Yeah, yeah.
• 43:29
  SPEAKER_01: Awesome.
• 43:30
  SPEAKER_02: What is secret sources?
• 43:32
  SPEAKER_02: What are the inputs on on that COVID uh determination?
• 43:36
  SPEAKER_01: So, yeah, two drops of saliva, three seconds, uh 98.5% balanced accuracy.
• 43:42
  SPEAKER_01: That's that's a better right there.
• 43:43
  SPEAKER_01: And we can push it to 100 if we're willing to take a little hit on the on the backside, it goes down about 94 percent.
• 43:50
  SPEAKER_01: Um, so we can we can capture everybody who's sick going on their way into a cruise liner or a baseball game.
• 43:55
  SPEAKER_01: That's awesome.
• 43:56
  SPEAKER_01: And and so you have a sensor on it on a
• 44:00
  SPEAKER_02: Essentially, that is processing some signal from the saliva and then uh correlating that with with the illness.
• 44:07
  SPEAKER_01: So I appreciate your curiosity, uh Seth.
• 44:12
  SPEAKER_01: You're jumping into their uh IP.
• 44:14
  SPEAKER_03: That's enough then.
• 44:15
  SPEAKER_03: That's about them.
• 44:17
  SPEAKER_03: I love that.
• 44:18
  SPEAKER_03: That's a let me let me ask you this.
• 44:20
  SPEAKER_03: Let me ask you this so we don't have to have your attorneys jumping in all over to lock this this wonderful podcast down to just the four of us listening.
• 44:29
  SPEAKER_03: That's it, right?
• 44:30
  SPEAKER_03: Um how how is that well you you you mentioned uh processing power, right?
• 44:37
  SPEAKER_03: And you know, as we get better, right?
• 44:39
  SPEAKER_03: We had another gentleman, we're talking about Moore's Law, it's always been there, right, with with shrinking nanometers on this and that.
• 44:45
  SPEAKER_03: How is that gonna help?
• 44:46
  SPEAKER_03: And what is it that you need, right, for you to continue the work?
• 44:50
  SPEAKER_03: Because we're gonna get hopefully more, you know, more powerful processing, but at the same time, do you guys just need to to get to that chip level and design your own, or are the you know the off-the-shelf ones that are there powerful enough to get where you need?
• 45:04
  SPEAKER_01: Well, we were hoping to use True North, and then the Air Force uh research lab took it over, so you can't buy it.
• 45:09
  SPEAKER_05: Yeah.
• 45:10
  SPEAKER_01: Um, the idea of Moore's Law is uh not important to us.
• 45:16
  SPEAKER_01: Okay.
• 45:16
  SPEAKER_03: That's why that's what I figured that you needed more specialized processing.
• 45:20
  SPEAKER_03: Right.
• 45:21
  SPEAKER_01: So this idea of a pattern recognition processor is very interesting to us.
• 45:25
  SPEAKER_01: Yes.
• 45:25
  SPEAKER_01: And there are people who are going that direction now, even even Google with you know TensorFlow and so on.
• 45:30
  SPEAKER_01: But I mean, there are going to be more and more chips that will be, you know, DeepMind will probably build one.
• 45:35
  SPEAKER_01: There'll be more and more chips that are at least trying to go to AI.
• 45:38
  SPEAKER_01: Unfortunately for us, they're mostly trying to go to deep neural networks.
• 45:41
  SPEAKER_01: That doesn't help us at all.
• 45:42
  SPEAKER_03: I don't care about that.
• 45:43
  SPEAKER_03: Well, that's what we're that's what we're doing at Amazon with inferentia, right?
• 45:46
  SPEAKER_03: With the inferentia chips.
• 45:48
  SPEAKER_03: They're normal.
• 45:48
  SPEAKER_01: That's the normal track.
• 45:49
  SPEAKER_01: Yeah, but there are people like IBM that you know that that were trying to go toward the pattern recognition processor, and there'll be more and more of those.
• 45:57
  SPEAKER_01: So that's that's we're not going to make chips ever.
• 45:59
  SPEAKER_01: I don't want to make chips, but but um we have done a lot of other things in software and in mathematics to avoid some of the pitfalls that others have encountered.
• 46:10
  SPEAKER_01: We've learned from them.
• 46:12
  SPEAKER_01: I'll give you one example.
• 46:14
  SPEAKER_01: When you try, well, let's put it this way when Larry Smarr was trying to do his work on the microbiome on a supercomputer, you can blow up the supercomputer memory stack because everything grows at n squared using the normal neural networks.
• 46:29
  SPEAKER_01: Yeah, and so we found a system that is only linear like that.
• 46:34
  SPEAKER_01: So we don't blow up the stack that fast.
• 46:37
  SPEAKER_01: Yeah, and that allows us to do a lot of things, go faster, do more, put better.
• 46:42
  SPEAKER_01: Yeah.
• 46:43
  SPEAKER_01: Um, we also found a way to run parallel systems that um I think, and I think we know this, and I think that Intel knows this.
• 46:52
  SPEAKER_01: We when the when the um Phi first came out, we called up Intel because we had some deep context there, and we said, you know, we got your new chip, and here's the deal, and we're running some tests on it, and we could run it faster and with higher capacity of all their um paths of all the the parallel pathways than they could.
• 47:12
  SPEAKER_01: That's awesome.
• 47:12
  SPEAKER_01: So um we had a really good proprietary toolkit that does that, you know.
• 47:18
  SPEAKER_01: So at every step, we've tried to design this system to be more efficient and more focused on this one task.
• 47:28
  SPEAKER_03: Yeah, that's that's the you know, the the neural nets are trying to take everything at the same time, right?
• 47:34
  SPEAKER_03: And you know, designed for the optimal use case of uh you know, of applying whatever to what you can have as a managed service as opposed to some.
• 47:41
  SPEAKER_01: I can see them as being incremental, Chris.
• 47:43
  SPEAKER_01: Yeah, you know, I think of the problem of Google advertising, and I say to myself, I want to sell more shoes to girls between the ages of 12 and 18.
• 47:51
  SPEAKER_05: Yeah.
• 47:51
  SPEAKER_01: And you know, and a three percent improvement is worth a lot of money.
• 47:55
  SPEAKER_01: But you know what that's not gonna do is like tell you what is a girl, yeah, or why.
• 48:00
  SPEAKER_01: Yeah, you know, yeah, you know, yeah, so uh it's just not very interesting.
• 48:05
  SPEAKER_01: That's awesome.
• 48:06
  SPEAKER_05: Yeah.
• 48:07
  SPEAKER_01: Well, I have another quick oh go ahead, Seth.
• 48:10
  SPEAKER_02: No, I was just gonna ask um when you start looking at these problems, uh, obviously you're looking at data and you're you're looking for patterns in data.
• 48:20
  SPEAKER_02: And a lot of organizations uh do not have their data house in order.
• 48:24
  SPEAKER_02: So did you want to talk a little bit about you know what needs to be in place and do you know, do you need things like you know ontologies or knowledge architectures or reference uh data uh to to to you know uh to have a good foundation for what you're doing?
• 48:42
  SPEAKER_02: Talk a little bit about that, if you would.
• 48:44
  SPEAKER_01: Sure.
• 48:45
  SPEAKER_01: And I'll start by saying I'm not the guy to talk about this, but we have guys.
• 48:49
  SPEAKER_01: We have guys.
• 48:49
  SPEAKER_01: No worries.
• 48:50
  SPEAKER_01: So um worries.
• 48:51
  SPEAKER_01: Uh I I was smart enough, I was just smart enough to recognize the difficulty of what you're describing as being the most dangerous thing in our path.
• 49:00
  SPEAKER_01: Because I had complete faith in our own technology, but I also had rather complete faith in how screwed up the data world is in general and how like you know how how quickly you can get screwed up just by having garbage in, garbage out.
• 49:15
  SPEAKER_01: And and we've had people give us like sheets of paper on PDFs, it's like you know, here's our data.
• 49:20
  SPEAKER_01: Modern companies, modern companies.
• 49:23
  SPEAKER_01: So um uh to put it quickly, it's no secret to anyone in this business that you need to have good data and it has to be clean.
• 49:32
  SPEAKER_01: Now, having said all that, um we can do if that's true, we can do a lot.
• 49:37
  SPEAKER_01: So we've built really beautiful ingestion engines that can take that clean data, um, even if it's got missing boxes or stuff that can fill it in.
• 49:46
  SPEAKER_01: If we understand from an SME how to fill it in, we can do what's needs what needs to be done to make it better.
• 49:53
  SPEAKER_01: Um, and we can do it very, very fast.
• 49:55
  SPEAKER_01: Uh, we can manipulate it wonderfully.
• 49:58
  SPEAKER_01: Uh, we can move this around, move that around, run parts of it instead of all of it.
• 50:01
  SPEAKER_01: We can compress it, we can run it when it's compressed already and not decompress it.
• 50:05
  SPEAKER_01: A lot of things we can do.
• 50:06
  SPEAKER_01: We can do it in the in the cloud or do it at our own data center.
• 50:09
  SPEAKER_01: So um, we're very, very good at that, but it's got to be decent data.
• 50:13
  SPEAKER_02: The price of admission is still good data.
• 50:16
  SPEAKER_02: And you know, despite a lot what a lot of people think, you know, AI is not gonna fix your bad data, right?
• 50:21
  SPEAKER_02: It can it can help in certain circumstances, but yes, it's still the price of admission.
• 50:26
  SPEAKER_01: Well, well, what we have found, and this I think is important for your audience again, if they're used to what we said a minute ago, if they're used to using neural networks or deep neural networks, and they've gotten so far with whatever they're trying to do, they'll have their own set of frustrations and humble moments about that in whatever problems they're trying to solve, whether in mining or biology or cancer care or whatever it is.
• 50:50
  SPEAKER_01: And so the the intuitive sense they may have of the limits there don't have to be.
• 50:58
  SPEAKER_01: And interesting, yeah.
• 51:00
  SPEAKER_01: We published a thing that's on our website called Three Things.
• 51:03
  SPEAKER_01: If you go to our website and look in a press release, it's three things.
• 51:05
  SPEAKER_01: And what we did pretty quickly, about two weeks, we had one of our guys, really smart guy, brilliant guy, um, take new existing science papers and improve on.
• 51:18
  SPEAKER_01: And in general, you know, these guys, scientists, were using AI to do some kind of work.
• 51:22
  SPEAKER_05: Yeah.
• 51:23
  SPEAKER_01: So in one case, it was the Wisconsin breast cancer data set, and that was all well known.
• 51:27
  SPEAKER_01: Um, so what he did was he did three of those and he beat them in each in each case.
• 51:32
  SPEAKER_01: And it sometimes it took two hours, you know, it was pretty quick.
• 51:35
  SPEAKER_01: The data ingestion takes longer than the runs.
• 51:38
  SPEAKER_01: So um the point being that there's more there, and the same thing was true for breast cancer, as I mentioned.
• 51:44
  SPEAKER_01: So there's more available there than you thought.
• 51:48
  SPEAKER_01: So even if the data set is fine, you know, you think you've hit it all, you haven't.
• 51:53
  SPEAKER_01: Right.
• 51:55
  SPEAKER_05: A neural network won't find it all.
• 51:58
  SPEAKER_03: Yeah, is it some some areas too then, you know, Mark, where you're like, you know what?
• 52:04
  SPEAKER_03: The neural net's fine for that use case.
• 52:07
  SPEAKER_03: Absolutely fine.
• 52:07
  SPEAKER_03: Like, like you brought up NLP and ASR and doing you know, speech rec and all that kind of stuff, where it's like, listen, you know, you're just you're wanting to do call center data, you know, or call center phone calls to find sentiment analysis on how people are doing.
• 52:20
  SPEAKER_03: That's pretty vanilla.
• 52:22
  SPEAKER_01: And I haven't tried that one, but I'm I'm just gonna agree with you.
• 52:25
  SPEAKER_03: Yeah, yeah.
• 52:26
  SPEAKER_03: I mean, and and I think that's one of the key things is you know, is the use cases that are that that's super good for, and that this is applied for as well.
• 52:36
  SPEAKER_03: You you know, it's not like you know, one over the other, it's an and like listen, there's gonna be areas where you gotta figure it out, right?
• 52:44
  SPEAKER_01: I'm not saying don't use them, I'm just saying that there's a particular use.
• 52:47
  SPEAKER_01: Yeah, yeah, right.
• 52:48
  SPEAKER_02: And a hybrid combination for you.
• 52:52
  SPEAKER_01: Yeah, sorry.
• 52:53
  SPEAKER_01: Well, one way we we express this is what we're good at, which is we say the closer you are to a complex data set, very high dimensional data set, the better we'll do compared to a normal whatever.
• 53:06
  SPEAKER_01: Yeah, yeah.
• 53:08
  SPEAKER_01: Okay.
• 53:08
  SPEAKER_01: So um, you know, if all you're trying to do is sell gym shoes, you probably don't need us.
• 53:14
  SPEAKER_03: I mean, listen, unless you're trying to predict if you know rubber prices are gonna go up and maybe that's yeah, we can do that actually.
• 53:21
  SPEAKER_03: Yeah, exactly.
• 53:22
  SPEAKER_03: So yeah, of yeah, you know, the number Indonesia, yeah, yeah.
• 53:27
  SPEAKER_03: The number of squeaks on a floor from a rubber shoe is gonna you know predict the butterfly effect of rubber prices.
• 53:32
  SPEAKER_03: Yeah.
• 53:33
  SPEAKER_01: We've done property values and and why did cars sell in 1980 and all kinds of fun, you know, uh demo projects.
• 53:41
  SPEAKER_01: Yeah.
• 53:42
  SPEAKER_01: That's awesome.
• 53:43
  SPEAKER_02: Well, I love the approach, love the thinking.
• 53:45
  SPEAKER_02: And unfortunately, we are at the top of the hour and we've come run out of time.
• 53:50
  SPEAKER_02: It's really been a tremendous pleasure speaking with you today, Mark and uh and uh and Chris.
• 53:56
  SPEAKER_02: And before I close, I just want to remind everyone of our sponsors, Simpler Media, Early Information Science, and the Marketing AI Institute.
• 54:03
  SPEAKER_02: And um, there should be show notes uh uh that will have some uh URLs for uh about Mark and about uh his company.
• 54:12
  SPEAKER_02: And uh again, thank you, Mark.
• 54:14
  SPEAKER_02: Uh tremendous, and thank you, Chris.
• 54:16
  SPEAKER_03: Yeah, thank you.
• 54:17
  SPEAKER_03: Mark, um, where can people, you know, like we'll have the the show notes, but um, you know, maybe tell people real quickly about your book and then what you're doing next so people know.
• 54:27
  SPEAKER_01: So uh the the book the book book is the pattern future, it's on Amazon.
• 54:32
  SPEAKER_01: Awesome, easy to find.
• 54:34
  SPEAKER_01: Um, and uh the other book you'd have to get from from Strategic News Service.
• 54:39
  SPEAKER_01: There we go.
• 54:39
  SPEAKER_01: Thank you.
• 54:40
  SPEAKER_02: And uh not not that they're gonna see this, but I held it up for people.
• 54:45
  SPEAKER_02: Yeah, for people that imagine imagine the book.
• 54:48
  SPEAKER_05: Yeah, it's right there.
• 54:49
  SPEAKER_05: Yeah, we go.
• 54:50
  SPEAKER_05: It looks like that.
• 54:52
  SPEAKER_05: Yeah, and then um uh idea looks like this.
• 54:55
  SPEAKER_05: It looks more like this.
• 54:57
  SPEAKER_01: So uh if they go to uh patterncomputer.com, they'll see the papers we talked about today.
• 55:05
  SPEAKER_01: If they go to um Stratnews.com, they will see uh and and futureinreview.com, they'll see the the other subjects we talked about.
• 55:16
  SPEAKER_01: And there are a lot of different things inside those places that they can they can read.
• 55:20
  SPEAKER_02: Well we need you back, and we'll love to we'd love to have you back, Mark.
• 55:24
  SPEAKER_02: This is a lot of fun, and I know we just scratched the surface in so many of these areas, but this has been this has been great.
• 55:30
  SPEAKER_02: Thank you.
• 55:30
  SPEAKER_02: Thank you.
• 55:32
  SPEAKER_02: Thanks for sharing everything together.
• 55:35
  SPEAKER_02: Yeah, okay.
• 55:36
  SPEAKER_02: Bye everyone.
• 55:37
  SPEAKER_02: We'll see you next time.
• 55:43
  SPEAKER_00: Thanks for listening to the Early AI podcast with your host Jeff Early at Chris Beaterson.
• 55:50
  SPEAKER_00: We hope you enjoyed the program and took away some ideas that you can start implementing your organization right now.