Idea Machines
Focusing on Research with Adam Marblestone [Idea Machines #33]
A conversation with Adam Marblestone about his new p...
more
Oct 26 2020 1h 6m
Chapter 1 59 sec
Chapter 2 59 sec
just a housekeeping note. We talk about F borrows a lot, and that's just the abbreviation for focus, research organizationsChapter 3 59 sec
like thatChapter 4 59 sec
everything, but, but I think, I think I ran up against some structural issues too, in, in dealing with, the brain. So, so basically one thing we want to do, is to map is make a map of the brainChapter 5 59 sec
thickness or lessChapter 6 59 sec
or gray scale, you know, is there a membrane present here or not? and then you have to stitch together imagesChapter 7 59 sec
going to be, just a, a single investigator science, Well, so it depends on how you think about it. One, one, one way to think about it is if you're just talking about scaling up, quote, unquote, just talking about scaling up the existing, technologies, which in itself entails a lot of challengesChapter 8 59 sec
they're ultimately there. Although I, you know, I was, I was a grad student, did a lot of pipetting also, but, But ultimately they're grad student. So are there in order to distinguish themselves as, as scientists and publish their own papers and, and really generate a unique academic sort of brand really for their workChapter 9 59 sec
aspectsChapter 10 59 sec
interesting to say the least. but yes, not only my incentives. but also everybody else's incentives are to, to maybe spend say 60% of their time doing some academically novel things for their thesis and only spend 40% of their time on, on building the connectome systemChapter 11 59 sec
is a government yeah. Led, you should, if you do requires a lot of systems engineering, there's probably a lot of work that is not academic interestingChapter 12 59 sec
once that flywheel sort of start spinning, then you have you have it. And so, and so that, that is a nonacademic research project and also the physics and astronomy communities, I think have more of a track record and pipeline overallChapter 13 59 sec
that sort of medium science, as opposed to small science, which is like a, you know, academic or one or a few labs working together, Or big science, which is like the human genome project was $3 billionChapter 14 59 sec
cost of genome sequencing through, through new technologies was brought down, basically by a million fold or so is, is, is, how George Church likes to say it, inventing new technologies, bringing them to a level of, of readiness where they can then be, be used catalyticallyChapter 15 59 sec
innovation, the, the, the permanence. So, so the, the flip side of the permanence is that, I guess, how are you going to convince people to do this, this, like this temporary thing, whereChapter 16 59 sec
opportunity in the sense that I think this is a very good match for a number of things that the government really would care about. and the government has, has, has the money, and resources to do this, but philanthropic is also one we should considerChapter 17 59 sec
if this is the only way to do it, and if it's incredibly important. So it really is a, it's a medium scale moonshots. you would have to be extremely passionate about it. That being said, there are reasons I think in approximate sense why one might want to do it both in terms of initiating one and in terms of sort of B being part of themChapter 18 59 sec
so there are a number of groups of people that I've seen in sort of, if you want critical mass labs or environments where they're working together, actually, despite perhaps the. Incentive to, to, differentiate where they're working, does a group of three or four togetherChapter 19 59 sec
talk about this as, as nonprofit organizations. these are the kinds of projects where, you would be getting a relatively small team together to basically create a new industryChapter 20 59 sec
so, so have you thought about how to sort of make sure that that lives on. Well, this is a tricky thing as we've discussed, in a number of settingsChapter 21 59 sec
host the data and provide it to the worldChapter 22 59 sec
and that's a problem that I think, we sometimes see with, externalized research models, like DARPA ARPA models, that try to. achieve more coordination and, and, and goal driven among otherwise, somewhat uncoordinated entities like contractors and, and universities that, that are working on programs, but then they, they, they, they achieve that coordination by then, managing the process and, with an fro, I think it will be closer toChapter 23 59 sec
figuring out how to avoid, Micromanagement seems like it's going to be really tricky because it's sort of like once you get to that amount of money, I like, have you, have you thought about, like how, like, if you could do some kind of like actually, well, I'll, I'll give her the, the, the, the, the, the thing that the cruxy thing is like this, I think there's a huge amount of trust that needs to happen in itChapter 24 59 sec
between accountability and, Sort of like doing the things that need to get done. I agree with that and Efros, we're going to navigate that. YeahChapter 25 59 sec
personalities, interests, and sort of conflicting politics, ending up. Fragmenting that resource into a million pieces. So, so I think this is a problem that you see a lot with billion dollar scale projects, major international and national initiativesChapter 26 59 sec
fro need to have a somewhat objective perspectiveChapter 27 59 sec
project. And so I think you need something like a program manager driven process to initiate the fro and figure out is there appropriate leadership and goals and our livable as reasonable, Yeah, that seems the way, at least the way that it's presented in the paper, it, it feels a little bit chicken and egg in thatChapter 28 59 sec
fundingChapter 29 59 sec
really strong interest, people committing, to really be involved is difficult, because it is a big change to people's normalChapter 30 59 sec
people. But, but there is a bit of a chicken and egg problem. In the sense that it's, it's not so much as here's an fro, would you please fund to me it's we need to go and figure out where there might be Faros to be had, and then who is interested in those problems as well to, to fund and support those thingsChapter 31 59 sec
sort of like refine the idea? and then sort of like, like you will drive that loop hopefully to, Getting a, an fro funded that's right. And there's, there's further chicken and egg to itChapter 32 59 sec
then start prototyping a fewChapter 33 59 sec
you know, there's a lot of venture capitalists there. They're pretty aggressive about funding and one could make an argument that, if it's, if it, it really is going to be inevitable and it really is going to start a new industryChapter 34 59 sec
Things that is driving technology right now. So, so I'm in no way saying that fro is, are somehow superior to two startups, in any generalized wayChapter 35 59 sec
need to develop a tool or a platform in order to explore a whole very wide space of potential applications. maybe you have something like a new method of microscopy or something, or a new way to measure proteins in the cell or things like that, that, you know, you could target it to a very particular, if you want product market fit application, where you would be able to make the most money on that and get the most traction, the soonestChapter 36 59 sec
sometimes you have things that really are meant to be. either generalized platforms or public goods, public data, or knowledge to underlie an entire field. And if you work to try to take the path, the shortest path to the Roadster, you would end up not producing that platformChapter 37 59 sec
radiosChapter 38 59 sec
is, this is a. Sort of like a regime that I'm really interested in and a, just on the transistor example, I've, I've looked at itChapter 39 59 sec
end up. Having the world we have today because I fell labs. It probably goes down this path where it's not part of the core product. and so they just sort of like do some vaguely interesting things with it, but are never incentivized to like, you know, invent, like the, the planner processing method or anythingChapter 40 59 sec
an open license. So this is actually a very long way of asking the question of, if F borrows are going to have a huge impact, it seems like they should default toChapter 41 59 sec
Yeah. Well, there's, there's a lot, a lot there, I think, to loop back to you. So, so I think, right, so, so this idea that we've talked a bit about as sort of default openness, so, so things that can be open for maximum impact should be openChapter 42 59 sec
a system that can be used to, to, or underlie the discovery of a whole new sets of classes of drugs and so on. But you're not so much focused on the drugs themselves. Now, that being said, right. if I invest in an SRO, and I've enabled this thing, rightChapter 43 59 sec
than, than the full capture of, of, of somethingChapter 44 59 sec
10 startups. so if I'm the investor, I might like to be involved in all 10 of the new industry, right. And the way to do that would be to create a platform with which I can explore, but then I have a longer time horizonChapter 45 59 sec
similar with open AI, you know, in a way, for example, converting, you know, into, into a, for profit or at least a big arm of it being, being the for-profit, and keeping all the peopleChapter 46 59 sec
know it until you see the data. There's also potentially applications for something like drug screening, where you could put a bunch of different, Kind of some CRISPR molecules or drug perturbations on, on a, on a brain and then look at what each one does to their, the synopsis or, and look at that in a, in a brain region specific way and sort of have ultra high, but connect to them based drug screeningChapter 47 59 sec
with, say like philanthropic or government funding, then going to fund a thing that proceeds to make a couple of people very wealthyChapter 48 59 sec
be a kind of, A startup that just, just enhances one, one person. It should be something that really contributes very broadly to economic growth and understanding of the universe and all that. But it's almost inevitableChapter 49 59 sec
clear. Cut. it's. We we've talked before about that, like fro is, could like scale up a process or build a proof of concept of, of a technologyChapter 50 59 sec
was some instance within which, a financially controlling entity to sort of arbitrarily bumped drug prices way high, right. A particular drug. and then w was, you know, was regarded as an evil person then, and maybe that's rightChapter 51 59 sec
would massively drop the costChapter 52 59 sec
this moonshot catalog. how do you tell the difference between people who have these really big ideas who are like hardcore legit? but like maybe a little bit crazyChapter 53 59 sec
question of prioritizing, which are the most important, but there's a huge number of. Process innovations or system building innovations that are needed across many, many fieldsChapter 54 59 sec
decades old. So there's sort of an obvious opportunity to sort of make like a CAD software for fusion, for example, you know, that the, the, it doesn't, it's not actually crazy. It's actually just really basic stuff. In some cases, I think they're ones where we'll need more roadmapping and more bringing people together to really workshop the idea, to really have people that are more expert than me say, critique each other and see what'sChapter 55 59 sec
need a roadmapping approach, but it's maybe not quite ready to, to just immediately do an froChapter 56 59 sec
connect them of brain, but how do you actually do activity map of entire brain?Chapter 57 59 sec
district. and, and I imagine that that would be counterproductive towards the goals of. so do you, do you have any sense of like how to, how to get around thatChapter 58 59 sec
matterChapter 59 59 sec
that you need to sort of plunk down, like have the confidence to plunk down $50 million. so, so we need sort of a, a, what I would see as a sustainable, way of. Getting to the point of fro type projectsChapter 60 59 sec
problems with research that we talk about are just about, The sort of conformity and specialization of really idea based exploratory, like completely uncertain researchChapter 61 59 sec
even though I'm so excited about borrows and how much they can unlock, because I think that this is one of two or three categories that has been, you know, under emphasized by current systems or has systems currently have struggled with itChapter 62 59 sec
get the PhD and all those different Alexey goosey has this nice blog post is oriented toward biomedical, but saying basically that in order to get through the system, you need to do 10 or 15 things simultaneouslyChapter 63 59 sec
DARPA stuff. We probably need more darker agencies for other problems. Even though I've, I've sort of said that I think Rose can solve some problems that DARPA DARPA will struggle withChapter 64 59 sec
like, ah, there's some research let's give some money to the research and then magical things will happen actually saying like, okay, like, like how does this work? Like what, and then what can we do for these, these specific situation? Yes. I think as you've identifiedChapter 65 59 sec
How it's not a it's it's a system. Kevin has felt set, said it said it well. And so in some ways it's been designed, but really our scientific systems are something that has evolved into large degree. No one has designed it. It's not. Something that's designed to be optimal is it's a, it's a emergent property of many different people's incentivesChapter 66 59 sec
think DARPA, by the way is a similar thingChapter 67 47 sec
of, of, of what they want of, you know, here here's, here's what I would do, very specifically, but I'm also interested in talking to people that, See problems with the current systems and want to do something and want to learn about, other highly specific fro ideas that others might have, and how to enable those