Eric Drexler | MSEP: What, Why, and How?

good to be here uh talking to a

foresight group it's I guess my first

face to face

talk to a group in three years having

spent a lot of time in covid lockdown in

England

I did use my my status as a co-founder

of the foresight Institute to negotiate

for a little bit more time which I will

use in part to give some some context

some perspective on what drew me out of

uh the AI area which is where I've been

in the last seven years

uh asking where does AI go as it goes in

the directions it seems to be going

broadly speaking

often the future of humanity Institute

at Oxford and in the orbit of deepmind

so what dragged me back was realizing

that after all of this time people who

are concerned about the long-term future

have not taken account of what physics

tells us about what can be built using

atomically precise Machinery that can

among other things be used to build

atomically precise machinery and a wide

range of High Performance Products

I see the the effect of altruism

movement uh thinking about the future

future centuries and assuming that we

will still be making for example

spacecraft out of sheet metal

implicitly assuming that not explicitly

so why has there been a failure to

update on this possibility my conclusion

was that you can stand around talking

about it and uh putting out texts on

Applied Physics all that you want but

without a clear and Vivid picture of

future prospects they will not be

understood without a community of people

who are visibly working towards toward

those those purposes they won't be

understood

so accelerating progress toward Advanced

molecular machinery building on the work

being done by people in this room

making it clear what some of the paths

forward are and where they lead I think

can make a tremendous difference to the

human future and I guess I could

incidentally say should greatly increase

funding for work in molecular machines

so uh

how to make these things visible and

clear

well we need simulations we need design

environments we need to be able to go

beyond building you know designing and

building simple machines to building

complex systems

so the we need a new name if anyone has

a better name than msap for something

that serves as a molecular systems

engineering platform that would be great

but a number of points here I've already

said something about concrete vision and

goals and why those are important uh I'd

like to talk about design by refinement

top down design as a way of thinking

about complex systems and how to how to

develop them

the different worlds of DNA protein and

dense covalent structures as as media

for design

and then

since the topic is uh you know a a

specific

plan a specific approach a specific

project for building a design platform

uh talk about the virtues of open

architecture and large what I'm calling

large surface area and making that work

well and finally an invitation to

contribute where the main question will

be

I barely touched this whoop

okay this this is mysterious

there was a switch on the side of a of a

computer at MIT which had it was

attached to a wire that led nowhere and

had a switch that the one position it

said magic and in the other position it

said more magic it wasn't connected to

anything but if you turn the switch to

computer failed

um okay so yes that's dangerous

I think we're okay

thank you oh

think about the invitation to contribute

the what we're most interested in is

what people need to to do their work

better and what what chunks of

functionality should be built that we

don't have and what chunks of

functionality should be brought into the

same workspace so we can better Design

Systems

okay

now that is not advancing

this is not an a

presentation mode at the moment

there we go thank you

yeah so I wanted to incorporate by

reference a whole lot of earlier talks

which have touched on very crucial

themes and then at 3 pm cut off because

I actually had to finish the

presentation turned it in but

so it's been known for a while that

dense covalent machines structures where

you do not have uh uh conformational

degrees of freedom

can be used to build machines that are

like those that we see in the

macroscopic world in a car you'll find a

transmission that has a planetary gear

and a universal joint and differential

gears that enable the parts to move

together and work as a system

well uh you can do systems with the

kinematics and function at

extraordinarily high frequencies like

gigahertz range and extremely high power

densities uh you know many megahertz

megawatts per cubic centimeter because

of Elementary scaling laws make

planetary gears

uh universal joints

differential gears

but all of these were designed

on workstations that by present

standards have no computational capacity

at all

at Xerox Park in 1992.

so why have we not seen more of this

kind of thing why did this basically

stall it simple machines and the answer

is no they're no good design tools these

things easy to design yes they're fairly

easy but even with with point and click

atom by atom copy paste maneuver draw

bonds

uh it's fun

and simulations work very well because

they're insensitive to small differences

in energy

so they are very nice in terms of what

the physics tells us about their

performance and uh are going to be

characteristic I think there's every

reason to believe of techno important

Technologies at some point

but they're not understood much less the

systems that can be built with them

so a little closer to the present and

and implementability this is a diagram

of three configurations of a device that

is a the framework for a 3D printer a

positional chemistry system and additive

manufacturing system

and it has a number of Virtues it has a

surface area for certain kind of motor

device it has interfaces and a geometry

that enables a rather stiff structure

low thermal low amplitude of thermal

vibrations for given given modulus of

elasticity of the materials

in 2015

was presented uh somewhat before that at

one doe workshop and then in a follow-up

Workshop 2015 that was organized by Adam

marblestone at the University of

Cambridge

that's what led to the funding that went

to uh to Andrew turberfield and William

she working on the 3D

you know the the two-dimensional

positioning mechanism using DNA origami

and uh and strand based actuation

but do you want to use protein or DNA

what are the consequences of those

stiffness thermal fluctuations

components kinematics Dynamics

it's not easy with available design

tools and I would like to have a little

aside at advertisement for protein

this this is a deadly weapon it's made

of a biopolymer keratin it's protein

and you can tell something about its

mechanical properties from

its prop from its its acoustic

properties

and a key fact if you're interested in

machines that can position things

precisely in the presence of thermal

fluctuations is that this has a modulus

of elasticity

something like a hundred times that of

DNA

that means the amplitude of thermal

fluctuations Square goes at the square

root is is 10 times smaller with it for

an equivalent shape with protein that

means that the area circular error

probable the RMS footprint

is a hundred times smaller

which makes a big difference you can get

another order of magnitude or more by

going to strong covalent solids but

proteins get you a long way

so designed by refinement you need to

have a concept and go somewhere with it

so here's a concept 3D printer uh XYZ

control and some kind of of thing that

the motion of which results in material

somehow being deposited and there are a

lot of ways of doing that

the configuration that I showed there is

an example of a concept for design it's

a mechanical concept it's not

analyzed numerically but the geometry is

the right kind of geometry to provide

good performance to take that further

one would want to have a coarse grained

elastic model of the structure to get

more more detailed uh specification of

of shapes and so on to meet the uh the

the

criteria for adequate stiffness

need a motor

well 2D interface with a set of

azobenzene molecules on it that are

tuned to or can get energy from light

antenna from three different wavelengths

is enough to make three independently

controllable reversible stepper Motors

which can be reliable in the pla in the

presence of photo bleaching and

crosstalk so it's a neat scheme

patentable and fact patented uh and if

you want to look into that it's a really

cool way of making a reliable motor

that's light I actuated the trick there

was that azobenzes used to be activated

in the Violet ultraviolet range and to

switch back to relax thermally if

they're Trans State in a Time on the

order of an hour advances in 2011 2012

got the the wavelength the energy

required down to the the red or in later

infrared range and the reverse switching

time tunable down to nanoseconds which

is what makes these practical for a fast

stepper motor

so General pattern concept think in

terms of shape moving beyond that to

solid bodies where at that point one can

start doing physical modeling elastic

body models a good way to do that in the

context of a uniform system of looking

at for for modeling molecular systems is

with an elastic Network model

which is not quite as doesn't have quite

the Fidelity

of a finite element model doesn't

capture the poisons ratio but it can

capture elastic modulus and the the

considerations that matter most for

design and in a way that's continuous

with uh

shapes that are are can be can be made

to uh well my key point is that it fits

with coarse grain and atomistic modeling

can have one modeling universe that

describes both large continua and actual

atomistic detail

could live in the same in the same

modeling world

so moving beyond simple structures we

start work moving into the world of

computer-aided design

complex systems have a lot of parts the

parts work together they have shapes

they have elastic properties they have

kinematics they have forces torques and

so on

the computer aided design work world has

been working this problem for decades

and they have ways of systematically

representing compositions of parts and

keeping track of things like the uh the

computational models that tell us about

the performance of those parts

and the thing on the left there is

called a model tree and it's important

to be able to do that because if one has

ambitious aspirations eventually you're

building things that are very complex

so comments on DNA protein and dense

covalent structures we have cap Nano

which serves the DNA engineering

community I gather it's the most popular

software at this point though it's a an

area in field and Rapid flux

you'll note here that it was embedded as

a Maya plug-in I have the old

capitalization there

so that just illustrates the fact that

one can have a a package that has a lot

of complex domain specific information

in this case about how to design DNA

origami structures and it can then be

embedded in a range of different

environments for visualization and and

manipulation

I'll be talking about a different

environment for embedding this kind of

functionality in a few moments

other areas as uh well one thing that's

very useful is of course coarse grain

model like ox DNA

making these design loops work well

protein engineering another area where

there isn't a lively Community Rosetta

has developed a design software for for

doing work in this in this domain

I'm told that protein engineering

software and DNA design software

together don't really work together

the the one can make substantial

progress by how by by just having the

shape of a protein molecule embedded in

the DNA world so that you can make sure

that the DNA fits geometrically the

shape of the protein forget about

actually modeling anything in terms of

forces but just the excluded volume

I think that's kind of shocking

surely where we want to go is to combine

our different biomolecular Technologies

to use for example the mechanical

properties and and chemical design

flexibility of proteins and the ability

to make huge uh addressable

structures with DNA origami those need

to be brought together systematically

and well

how else are we going to make really

large aperiodic protein structures well

there are ways but this may be very

attractive way we need to bring these

tools into the same environment make the

tools work together

now as a segue to where I'm going in a

moment here I would note that this

picture is actually not a picture of

something from from the Rosetta design

but from folded which comes out of the

same group

and folded is a citizen science game it

has more than 700 000 I don't know how

many players on a given day but 700 000

registered users

so this is a citizen science Community

is very interested in playing with the

molecular world

so this raises this now that moves to

the next question what about the world

of things we can't make things that

physics tells us we will want to make

things that we will be able to make

along the path that we're following

the things that no one can make no one

is going to pay you to design things

usually that can't be made not an

industry only in the more theoretical

activities in Academia

so how does one make progress

in exploring this world and getting a

sense of where we're going then perhaps

working backward to bridge the gap

between biomolecular machines

positional chemistry working perhaps

with dense non-covalent but but ionic

solids or or metal oxide solids that can

be built in aqueous Solutions and have

modulus that's much higher than protein

much finer grain there are bridging

Technologies

so how could we get a clear picture a

lot of work done a whole lot of whole

lot of time going into this into

designing and modeling systems like this

I think Minecraft

millions and millions of users they're

given this world which doesn't look like

any real world it's a bunch of blocks

with with a very simple simple-minded

kind of physics

they added something called Redstone

which allows one to build machines

robots uh working Rubik's Cubes all

sorts of crazy stuff in Minecraft

now if there's seven if Minecraft has

millions and millions of users and fold

it has 700 000 people who've at least

downloaded and registered it I think we

might be able to get a substantial user

base uh playing around with what I will

guarantee you or are fun things to to

build

uh with blocks where you can point at

them and say look

I made this thing and this is what

physics how physics tells us it will

work see the the community curation has

certified that the simulations are are

appropriately done it's a good it's a

proper member of the repository of

designs it's better than anything that

was there before or it's the first

implementation of a certain kind of

machine and you talk about it on social

media and it goes up on the leaderboard

and you have more friends come in and

they play with it and they say this is

fun and perhaps you build out not a a

Minecraft city

but a molecular systems engineering

platform based at least that was would

be under the under the hood I'm sure the

name of the the that you'd see is

different

it would be something on that platform

that would be a view of intricate future

technologies that would be clear and

concrete and could have videos instead

of being science fiction

I'm very tired of the science fiction

I pretty much ate everything

for a decade or two

Okay so

that's the gamified end of the spectrum

they have computational chemistry at one

end and games at the other and the

computational chemistry underlies the

games

open architecture large surface area

want to build on free permissively

licensed open source software

right thing to build on I concluded and

I tried to talk myself out of this just

for exercise and came back to the the

conclusion which is you want to build on

a game engine

and there is a rising star in the game

engine world it is the high quality open

source game engine called Godot good OS

and waiting for Godot meaning you'll

wait forever for all the features to

appear as a large enthusiastic user base

it has capabilities for rendering and

manipulation and interaction and user

interface building and

I did not touch the machine at all I

touched the cow horn and it jostled the

edge of the computer and my watch and

the watches touching this wire here

this is this is this is more magic flick

the switch and it will stop now Okay so

gado

it's basically C plus plus

plugins and IPC communication with other

things it has a scripting language it

has a native scripting language what

we're going to be doing is replacing

that with python

and that makes it easier to integrate

beautiful Graphics from PI Mall not just

picking up pie mall but uh because of

issues with shaders and various things

but using pie Mall as a major leg up and

getting high quality publication quality

highly informative graphics

so that combination gives one malleable

user interfaces visualization

interaction et cetera et cetera et

cetera and python scripting for

everybody and for interfacing with lots

of existing packages like for example

lamps and lamps in turn the large-scale

atomic molecular massively parallel

simulator supports uh favorites of the

protein engineering community like Amber

and charm Ox DNA though it's apparently

a very slow implementation but I am told

that it will be easy to use the the

certified high efficiency aux DNA and

interface that instead of going through

lamps on that

and handles rigid body mechanics and

custom force fields and all these can be

made to live in the same world with take

your pick of favorite Quantum modeling

package

live in the same world through an

abstraction used in the atomic

simulation environment which I

eventually decide was too heavyweight in

itself but has this cool abstraction

which is you have a thing it's called

that you call a calculator lamps in a

particular mode would be a calculator

Quantum espresso would be calculator oxy

and eight would be a calculator

and your your your authoritative source

of information on the configuration of

the world is on your side of the fence

and you tell these different calculators

where the particles are that they're

supposed to be thinking about they think

about it a while they come back and tell

you the forces on the particles

you put those together

add some other calculated forces that

you made up like something that

simulates the force of a motor that's

not a real physical motor but represents

a boundary condition from a larger

system

and you do a relaxation step or need to

integrate the equations of motion

forward a step that gives you new

positions you send those out to the

calculator again

very flexible very clean very extensible

so picture here is core platform

Graphics manipulation Etc et cetera Etc

and when I speak of surface area I mean

surface area for contributions things

that can be built in parallel

functionality that other people have

that can be brought in

scripting that people can do because

they did a task once they now know what

the configuration of tools is and what

the sequence of operations is and if you

save the configuration and perhaps

perhaps do the do scripting to automate

what you did by hand you now have a

workbench environment

you have a a setup for doing a

particular task in the molecular systems

engineer engineering you were doing so

plugins modeling and design workbenches

uis tool sets educational and game

content uh if you start working with the

good old Community you find people

saying I know how to make this into a

compelling game

okay we'll do it

uh some things on Wiki content that's

another story

but building up a designed and validated

repository of components and systems is

a major part of the objective here as

well as supporting people who are

actually building things in the lab

and that in turn

supports again activities ranging from a

range of computational chemistry

workflows through games in education and

bracket in between are the molecular

systems engineers and for the long term

molecular systems engineering the

stand-ins who will be the citizen

scientists engineers

so picture platform engine extensions to

make it a molecular systems engineering

platform

toolkit workbenches past focused uis and

tool sets and growing repository of

products

so

picture here is one in which you have

mostly core team contributions in the

lower left and mostly community

community contributions above

and with respect to community

contributions I should emphasize the

idea here isn't here's this big thing it

doesn't do everything it's going to

Stone Soup you you bring you bring

different uh different things and then

it does something else it's gonna be a

bunch of functionality

and added functionality where it's a

distinct chunk of functionality like

this is like embedding embed CAD Nano

and Maya embed CAD Nano in msep what

does that look like it looks like

something that is branded as CAD Nano

with a you know reference to the website

and papers to site

it's not some the idea isn't to have

these little little contributions be

hidden away but to push forward

attention and make people responsible

for their work and fixing bugs

okay status newly funded after very long

incubation a lot of thought about money

flowing just this month

so Apple initial funding from Astera

courtesy of a lot of work by uh by Adam

marblestone part of which was persuading

me that perhaps I wanted to play a

leadership role in something like this

and we have

an excellent project manager Pine Peter

HP you should talk to him he's here he

introduced himself earlier

so he's he's not a graduate student and

who's learning computer so learning to

program because he wants to program

something that's useful in his in his

PhD

in this dissertation he's a professional

software developer who's LED million

dollar scale projects before and has a

network of

software Engineers he's worked with and

so we're going to be building to and

Beyond in the next year about a half a

dozen full-time equivalent software

engineers

that's enough to get something done and

there's enough excitement out there that

uh I think we're going to be getting a

lot more than that done

and again lots of parallel work to be

done with Strong's Community Support

I think means that it will be possible

to expand on that build on that surface

area rather rapidly

so finally an invitation to contribute

we need to understand what people need

to need to do their work

so that we can have a road map that

leads to supporting that

we need help to build what you need

which includes help finding the the

software components and systems that can

be usefully integrated

help recruiting contributors and

Leadership

this project is going to need to build

out leadership as well at some point

we're going to need a full-time uh

product manager as they say in the

software world slash project leader

and of course development moves faster

if we have more money especially with

many parallel tasks we expect to have a

a demo at some point in not too many

months in the future that will give some

indication of of where we're going

so with that I will just say thank you

and invite questions

[Applause]

yeah thank you for sharing your vision

uh how do you see the integration with

experiment both in terms of obtaining

data from experiment as well as a

feeding bacterialization of some of

those ideas

yeah workflows that go through the

experimental world and back in again on

this side of the of the the interface

with experiment you have have problems

of

keeping data aligned uh the experimental

data aligned with the simulation data

aligned with the model that generated it

and not getting confused about that

that's a database and consistency

management kind of problem that that

burden could be taken on inside this

this framework and in fact the mechanism

should be there to do that because it's

needed for a wide range of purposes

beyond that it's a question of very

specific workflows you know how are you

how are you using your experimental data

to update your models how are using your

models to to to guide your

experimentation and that is something

where the the aim is to provide tools

that make it easy to describe and

Implement those workflows

but we need you to tell us how that

works and perhaps uh you to uh to direct

the implementation perhaps with some

professional software engineering

support to make that go well and be a

good solid usable product

um I guess my question is really late

but um so isn't the bottleneck uh the

actual synthesis of these machines more

than the modeling the long-term uh

strong covalent solid machines or oh I

mean the current generation atomic scale

machines uh protein engineering and

structural DNA oh I mean like so like

for example like DNA is very easy to

assemble because of the productivity and

and protein is made by cells yes

but so how do you make how do you

actually synthesis synthesize this very

complex uh do you mean dense covalent

solids the ones that have the gears and

so on yeah for example the answer is

that today you don't we're in the

process of building tools that will

enable us to build tools that will

enable us to build tools that will move

into that domain part of the conceptual

difficulty that people have had is that

these machines are

almost childishly simple they look like

macroscopic machines you know surely

this is just some silly idea real

molecules don't work like it physics

tells us this is what you want to build

if you could physics tells us that you

can build this kind of thing if you have

tools that are made of comparable you

know materials of comparable stiffness

and and fine granularity and you work

with extremely reactive molecular

fragments in construction if you want to

grow a something like Silicon or silicon

carbide or diamond or or graphene those

those kinds of structures the growth

species tend to be small and highly

reactive how do you make that work well

chemists when they want to direct direct

reactivity tend to use uh you know the

inherent differential reactivity of

molecules or they use protective groups

you can think of a machine doing

positional synthesis and sort of the

ultimate protecting group The reactive

molecule only touches the work days

where it's supposed to if you have

adequate adequate control of thermal

thermal vibration amplitudes you can

Target one location rather than another

with angstrom Precision if you have that

then you can use highly reactive

molecular fragments that can form

multiple covalent bonds in in one

deposition operation which means you can

build up these very rigid polycyclic

structures one does not want to try to

do that today or tomorrow some people

have looked at this and said oh we're

going to go off and try to do it in high

ultra high vacuum with STM and because

because Eric said it was a good idea and

I've always said no I've always said

that's a bad idea

look at my my first paper on the subject

it was 1981. it was at the root of the

citation tree for protein engineering

until people got tired of citing papers

that old

because I said that's the way forward

and now we're getting there

I'm trying to be the devil's advocate

here obviously

um

so so a lot of us can build

wonderful structures

but nature doesn't use them for example

nature doesn't use gears with few

exceptions nature doesn't use CIS trans

isomerization for motion with few

exceptions and yet nature does things

that we would kill to do so so the

question is

what are we trying to what do we want to

do and maybe that's what needs to be

clarified and then go back and see what

are the things

that we need to achieve to get to that

goal

yeah backward chaining from goals is

very much the way that one wants to

proceed abstractly speaking uh clarify

the goal find what the steps are

necessary to get there and the better

your design environment is the more one

can have a

complex Target have confidence that it

would work if you made the parts and

back out from there and start working on

parts that fit the large scheme instead

of parts then sort of hoping they might

be useful for something

though it's a good exercise to do that

initially

uh but in terms of of non-biological

models which is a grand example of using

that that methodology

why do our brains uh conduct electrical

signals in little tubes full of salt

water

we don't do that in our computers we

have nanoscale Electronics which is you

know finer finer scale than synapses and

the signal Transit transmission speeds

are a healthy fraction at the speed of

light instead of being measured in you

know like single or double digit meters

per second

why are our bones made of a mineral

material that is deposited from aqueous

solution rather than being say the

strength of carbon fiber composite

because we have evolutionary constraints

and those apply all the way through

cellular architecture molecular machine

architecture protein

is an amazing material

but in terms of its engineering

properties it is crap it's low modulus

low stability low strength well actually

spider silk is kind of competitive with

some very good engineering materials

but why why do we have that well it's

made by ribosomes why ribosomes well at

some point back before the the last

Universal common ancestor the life

stabilized on three three three base per

codon system that worked with the the 20

economical amino acids and that's what

we do and the cell is full of salt water

because we came from the ocean

if you're not conveying to that you end

up with you ask what does physics tell

us and you find that you can make

materials and systems that are by many

many engineering uh uh criteria orders

of magnitude better

sure you want to make something that's

that's better than what you can find uh

in nature but the point Still Remains

what is that thing that you want

um uh tell me what application and I

might be able to say more details give

you something in more detail well you

you suggest an application where where

we can we can get there and beat nature

oh I I have already did sorry yeah

another another time yes

I'd really like to talk about things

that might be within reach of this

community that are stretch goals and uh

could uh make have an enormous uh

medical impact uh the the the word to

tag for that is mechanism I think

exosome but with something complicated

inside that can do some uh rudimentary

computation and sensing

so Eric to follow up and comment from my

colleague here critical path analysis

what we need is not motivation in terms

of the final product I mean even the

crude things that we've got already are

plenty of motivation

what we need is is to to work out how to

build the machines to make the machines

to make the machines

I mean the tools you're developing will

do that too

um but what I'm suggesting is that that

should be the focus if you want to speed

things up if you want to get the

community and you know the world able to

explore this technology it's not what

we're aiming at that we need to

understand it's how to make it

um I think that both are very important

for reasons that are in commensurate

it's very hard to compare the value of a

greater Clarity a vision of long-term

future it's very hard to compare the

compare the value of that with more

rapid progress and contributions to

aiding the experimental design and test

cycle

I think both are important and I don't

think they I think they're they're

synergistic rather than interfering

this project is going to have surface

area and if we have enough funding we

can build it build out in both

directions at the same time and share

some of the the work and Tooling in the

process