►
From YouTube: DevoWorm (2023, Meeting #20): SAM to cell lineages, dev-enviro interactions, triangular planes
Description
Using the Segment Anything Model to visualize 4-D cell lineages. Developmental-environment interactions and the Free Energy Principle, division planes in circular and triangular cells, digital biological twinning and hyperrealistic models of C. elegans (for understanding chemoattractive stimuli). Attendees: Sushmanth Reddy Mereddy, Morgan Hough, Susan Crawford-Young, Amanda Nelson, Bradly Alicea, Jiahang Li, and Richard Gordon.
B
C
D
C
C
C
B
C
C
B
So
it
sounds
good,
so
you're
getting
everything
set
up,
so
this
is
actually
we're
heading
into
the
coding
period.
We
had
the
community
period,
the
last
three
weeks,
so
yeah
and
I,
don't
know
where
I
know
that
hamanchu
has
been
working
on
things,
but
he's
also
been
very
busy.
He
had
exams
and
then
he
also
had
some
other
things
going
on.
So.
C
C
Engineering
department,
this
thing
is
good
and
see
at
all
those
things.
I
mostly
concentrate
on
machine
learning,
other
open
source
works,
I
just
open
a
day
before
exam
and
I
start
reading.
That's
what
happens
with
me
if
I
have
studied
like
from
last
time,
I
could
manage
these
exams,
but
I
didn't
study.
That's
the
main
issue
going
on
even.
E
C
E
Yeah
yeah
I,
don't
know
if
that's
better
yeah
fine
I've
always
had
trouble
with
exams
and
yeah,
basically
like
about
two
days
before
I
start
to
study
my
notes
and
look
things
up
and
then
I
write
the
exam
and
then
I,
maybe
forget
half
of
it.
Anyways.
C
E
I
think
I've
been
a
couple
of
times.
Worse
all
depends
I've,
had
students
do
that
too,
because
no,
this
is
Red,
River
College
and
they
said
they've
been
homeschooled
and
that
they
just
wanted
to
write
the
exams.
I
said
here
they
are
there's
your
midterm.
There's
your
final
good
luck.
D
E
Right
before
the
High
School
courses,
I
was
teaching
they
they
that's
all
they
charged.
E
B
All
right,
so
it
looks
like
geohang,
is
here:
Amanda
is
someone
from
the
other
group
that
I
work
with
and
she's
here,
because
we're
gonna
do
we're
gonna
talk
about
a
paper
that
is
well
we'll.
C
B
All
right,
so
all
right
so
I
have
a
couple
things
today.
I
I
know
that
sushmath
is
busy
with
his
exams,
but
it
looks
like
he's
moving
along
and.
B
C
Yeah,
these
are
some
bugs
coming
up.
We'll
try
to
solve
them.
I
have
done
most
of
the
work.
Just
some
Library
importing
issues
are
there
I
almost
solved
it.
I
will
try
to
update
it
by
this
week,
just
working
on
semi,
segmented,
same
segment
after
that
I
will
convert
them
into
onmx
model
and
I
will
try
to
host
like
a
hugging
face
page
in
hugging
face
spaces
also
like
what
we
did
as
a
development
but
instant
simulation.
C
We
have
models
on
semantic
segmentation
right
right
now,
instant
segmentation
models
will
be
hosted
in
the
hugging
phase
so
that
every
camera
you
can
use
it
so
right
now,
that's
my
work
will
be
and
with
that,
I
will
try
to
add
another
to
the
model,
also
calculating
volume
of
itself.
Actually,
this
is
3D
data
set
right.
We
can
calculate
volume
why
you
told
this
thing:
I
will
try
to
I.
Will
Implement
and
I
will
definitely
upload
into
grid
okay,
so
13
phase
spaces.
C
E
C
Which
I
have
proposing
proposed
proposal?
Actually
the
code
is
everything:
is
there
just
I
need
to
train
it
on
nematodes
data,
I
mean
what
you
call
CL
against
data
and
extract
the
weights
of
those
files
and
again
convert
them
into
hugging
phase
and
uploading
them
in
our
universe.
That's
the
only
part
left
with
that.
We
can
extract
the
volume,
mean
volume,
some
area
of
the
itself,
and
we
can
view
it
in
a
3D
structure.
I
was
trying
to
use
plotly
and
implement
it.
Hopefully
I
put
I
will
complete
this
work.
C
That's
what
my
update
is
and
now
Brandy
I
will
say
this
GitHub
link
with
you.
If
you
want
check
or
something
like
that,
just
give
it
a
look.
My
updates
will
be
here.
I
will
start
writing
blog,
also
I
I
thought
of
start
writing
it,
but
due
to
other
works,
I
couldn't
write
the
block,
but
by
this
week
I
will
write
the
block.
C
Every
update
will
find
next
few
weeks
plan
in
what
we,
what
I
will
do
because
I
thought
of
making
only
development
model
in
my
proposal,
but
after
talking
with
mayo,
he
told
to
told
me
to
integrate
Sam
segment.
Anything
normal,
so
I
will
rearrange
my
schedule
and
I
will
share
it
with
you
in
the
blog
itself.
You
can
read
and
keep
an
eye
on
it,
and
this
is
the
GitHub
link
actually
where
I
was
keeping
all
my
code
and
keeping
here
in
my
guitar
sorry
here
by
our
Unity.
Please
check
it:
okay,
yeah!
B
Enough,
okay,
that
sounds
great
yeah,
so
thank
you.
Yes
actually
is
what
I
was
gonna.
Actually
mention
is
that
we
have
I
want
to
be
able
to
have,
like
you
know,
push
us
to
some
sort
of
GitHub
wherever
we
know
where
things
are
so
it
looks
like
you
have
that
taken
care
of
and,
and
it
may
end
up,
forking
it
to
give
a
worm.
You
know
just
to
have
it
there
so.
C
C
B
C
B
B
Yeah
no
problem,
yeah
and
I,
don't
know
Manchu.
If
he's
going
to
have
you
know,
I
might
create
a
repo
for
him
on
Diva
worm
or
we
might
just
work
on
his
own
repo,
but
we'll
set
that
up
after
the
meeting.
B
C
C
This
one,
this
is
a
CSV
file
menu.
Oh
yeah,
this
kind
of
3D
thing
this
is
not.
This
is
just
some
kind
of
molecule,
but
imagine
some
Clans
spreading
through
this
3D
file.
One
can
rotate
and
see
it
where
it
is
located.
When
you
keep
some
position,
what
will
be
the
volume
of
that
cell
and
a
video
I
was
thinking
to
build
this
kind
of
thing,
a
3D,
visualization
kind
of
thing,
so
everyone
could
see
in
a
3D
model,
rotate
it
scene
upside
down
and
get
the
centroids
of
each
cell
manually
rather
than
expecting
yeah.
D
If
you
could
rotate,
that
continuously,
it
would
appear,
can
be
three-dimensional.
C
C
C
E
D
D
C
C
C
B
Yeah,
oh
yeah,
so
I
did
post
a
bunch
of
materials
in
the
slack
on
Cell,
lineage
and
developmental
lineage
trees,
and
things
like
that
so
and
that
I
think
so
shmob.
You
probably
found
that
useful.
C
B
B
Soon,
as
possible,
yeah
so
I
tried
to
get
a
nice
set
of
references
in
there
and
we've
we've
done
a
lot
of
work
with
that
in
the
past,
so
I
I
last
week's
meeting
at
the
end,
I
kind
of
summarized
it
and
then
I
have
those
links
in
the
slack.
So
everyone
should
be
able
to
you
know,
get
a
handle.
I
know
that
hamanchi
will
probably
find
that
useful
as
well.
So
it's
in
there
we
we
have
that
that
those
resources
available.
B
B
A
four-dimensional
visualizations
to
some
extent,
but
it's
not
been
on
like
freshly
segmented
data.
It's
you
know
we're
working
from
things
like
the
Epic
data
set,
and
things
like
that.
So
you
know
like
basically,
people
have
tried
to
do
these
kind
of
visualizations,
where
you
know
they
like
kind
of
track
cells.
They
had
one
person
who
was
tracking
cells.
B
You
know
they
have
a
three-dimensional
space
which
is
the
the
embryo,
and
then
time
would
be
like
these
tracks
that
would
come
out
and
because
the
data
is
comes
out
that
you
know
the
cell
centroid
is
moving
around
a
bit
and
sometimes
it
migrates
and
then
it
divides.
So
you
have
like
two
things
that
come
out
of
a
single
centroid,
and
so
you
have
to
track
all
that
and
so
the
different
ways
you
can
do
that.
B
But
you
know
so
yeah,
don't
I
I'll,
maybe
I'll
try
to
get
those
things
together
and
and
showcase
them
and
see
what
they
look
like,
but
yeah.
So
yeah,
good
luck
with
that!
I,
don't
know,
I,
don't
know
what
your
ideas
are.
For
that
I
mean
we
haven't
really
gotten
to
canonical
sort
of
approach
where
this
is.
This
is
what
we
want
to
do
so:
okay,
yeah
great.
Well,
thank
you
for
the
updates.
B
Your
social
looks
like
Morgan's
here,
hello,
Morgan
and
then
such
mods
GitHub
repository
for
gsoc,
okay,
so
yeah.
They
said
that
Amanda's
here
today
and
is
our
first
meeting
so
welcome
again
Amanda.
One
of
the
reasons
she's
here
is
because
I'm
going
to
cover
a
paper
that.
A
B
New
one
by
Michael
Levin,
but
it's
a
regulative
development.
So
let's
see
this
is
it
here
so
this
as
you
can
see
my
screen?
B
Okay.
So
this
is
the
new
paper
biosystems
paper.
Regulative
development
is
a
model
for
the
origin
of
life
and
artificial
life
studies.
So
I
found
this.
You
know
we
found
this
in
the
coming
out
in
the
journals
and
it
looks
early
and
so
Chris
Fields
actually
does
he's
I
think
he's
a
retired
academic.
He
did
a
lot
of
stuff
with
information,
Theory
and
physics
and
quantum
physics
he's
currently
teaching
a
course
through
the
active
inference
Institute
on
information
and
life,
or
something
like
that.
B
I
can't
remember
what
the
name
of
it
is,
but
it's
really
interesting
stuff
how
to
think
about
information.
So
it's
like
a
lot
of
quantitative
approaches
and
information
and
of
course
you
know
Michael
Levin,
and
so
what
they
do
in
this
paper
is
they're
using
this
idea
of
regulative
development
as
a
model
for
the
origin
of
life.
So
this
is
really
kind
of
interesting
to
a
number
of
things.
We're
doing
in
the
group
here
as
I
figured
I'd
cover
it.
B
B
So
you
know
without
getting
into
the
math,
which
is
pretty
in-depth.
It's
basically,
this
energy
minimization
Paradigm
people
have
done
this
in
different
ways
in
complexity.
Theory
people
have
done
it
with
like
dissipative
systems
and
all
sorts
of
other
approaches.
But
this
is
where
you
know
this
has
been
Unified
as
an
approach
to
some
of
these
problems,
and
so
that's
what
they're?
That's?
B
Where
they're
coming
from
on
this
so
they're
using
the
free
energy
principle,
they
show
how
generic
thermodynamic
requirements
on
bi-directional
information
exchange
between
a
system
and
its
environment
can
generate
complexity.
So
this
is
where
you
have
this
interaction
between
the
system
and
its
environment.
There's
this
bi-directional
information
exchange,
so
information
from
the
environment
to
the
to
the
developing
system,
and
then
the
system
to
the
environment,
and
so
that's
how
you
generate
complexity.
B
This
leads
to
the
emergence
of
hierarchical,
computational
architectures
in
systems
that
operate
sufficiently
far
from
thermal
equilibrium.
So
one
of
the
things
about
the
free
energy
principle
is
it
allows
you
not
only
to
think
about
energy
minimization,
but
it
allows
you
to
think
about
things
that
are
far
from
equilibrium
and
so
I.
You
know
that's
another
theme
again
from
complexity
Theory,
where
you
have
these
edge
of
chaos
behaviors
or
these
far
from
equilibrium
behaviors.
If
you.
C
B
Of
a
system
like
a
a
typical
entropic
system,
a
closed
system,
that's
usually
an
equilibrium
system,
so
you
usually
can
find
the
equilibrium
solution
for
that
system,
and
you
know
it's
always
pretty
much
going
to
describe
it
now.
An
open
system
like
life,
which
doesn't
you
know,
has
an
open
system.
There's
energy
coming
in
from
the
environment.
B
There's
you
know,
there's
entropy,
but
there's
also
a
lot
of
energy
coming
in
a
lot
of
those
systems
operate
in
what
they
call
a
far
from
equilibrium
regime,
which
means
that
you
have
these
multiple,
stable
States
and
you
have
things
that
move
from
state
to
state
and
sometimes
they
collapse
into
disorder,
and
so
this
is,
of
course,
what
we
see
in
life
all
the
time
we
see
this
in
cells.
B
We
see
this
another,
you
know
organisms,
and
so
that's
why
we're
interested
in
this
type
of
approach,
not
just
because
of
energy
minimization,
but
because
of
the
open
systems.
Aspect
where
you
have
this
energy
fluctuation
in
this
setting,
the
environment
of
any
system
increases
its
ability
to
predict
system
Behavior.
B
So
this
is
the
enroll.
The
environment
is
to
do
this
engineering
or
I
guess
a
better
way
to
put
this
is
that
it
sort
of
supervises
the
the
development
or
evolution
of
the
system.
So
you
have
this
engineering.
Maybe
the
environment
is
setting.
You
know
a
pathway
forward
for
the
system
where
it's
you
know,
filtering
information
or
something
like
that.
D
We've
had
a
problem
with
this:
okay:
where
is
it
how
it's
it's
a
it's:
an
old
idea,
the
environment,
influence
of
development
yeah?
And
yes,
if
you
cook
an
egg,
it
will
hard
boil:
okay,
okay,
however,
our
Excellence
grow
very
nicely,
even
in
distilled
water,
in
a
constant
temperature.
So
you're
you're
wondering
what
on
Earth
information
could
the
environment
be
provided?
Yeah.
D
Yeah,
this
is
just
a
how
should
I
say
a
formulation
about
myth.
B
Yeah
yeah,
so
it
really
depends
on
the
sort
of
the
context.
You
know
there
are
going
to
be
some
things
that
are,
you
know,
determine
you
know
deterministic
and
some
things
that
are
going
to
be
stochastic.
So
that's
I
think
that's
what
they're
getting
at
with
this.
But
you
know
it
depends
on,
of
course,
if
it's
Behavior
or
if
it's
like
embryogenesis
or
what
yeah.
D
But
if
it's,
if
it's
embryogenesis,
we
know
embryos
need
a
range
of
temperature
right.
We
need
a
certain
amount
of
water
and
I'm,
not
sure
anything
else.
Yeah
or
the
information
from
the
environment
may
be
minimal
or
not
at
all
right.
It's
just
what
what
I
call
permissive
it's
possible
for
things
to
go
forward,
but
because
you
could
have
an
environment
where
that
is
impossible.
Yeah.
B
D
You
freezer
cook
the
embryo,
for
example,
or
dry.
It
out
happens
so
the
whole
concept
I
think
it
means
because
I
discuss
this
problem.
D
Yeah,
that's
true
because
some
components
are
sensitive
to
gravity,
but
again
it's
the
environment
is
permissive,
but
I
don't
see
that
it
provides
any
information.
D
D
E
It's
not
I.
D
E
D
D
D
B
So
it's
sort
of
a
passive
versus
active
component,
so
passivity
would
be
like
just
allowing
for
certain
things
to
happen
over
others,
whereas
acting
would
be
just
saying
that
this
is
the
way
it's
going
to
be
in
certain
things,
or
only
certain
things
are
possible.
It's
kind
of
subtle
until
you
see
like
well,
I
mean
it's
subtle
until
you
actually
set
up
the
experiment,
I
guess
and.
D
Then
the
you
know,
the
notion
gets
into
a
nonsense
state
where
you
say
everything
that
happens
inside
an
embryo.
There's
a
local
environment.
D
B
Yeah
yeah:
let's
see
what
okay
so
then
we'll
see
if
they
they
Define.
Any
of
that
in
the
you.
D
B
Have
a
feeling
that
maybe
they
don't
but
let's
see
when
seen
in
this
light,
regulative
development
becomes
an
environmentally
driven
process
in
which
the
parts
are
assembled
to
produce
a
system
with
predictable
Behavior
yeah.
B
D
So
the
environment
is
gone.
Okay,.
B
Well,
no
human
Engineers
but
I
think
what
they're
doing
they
have
this
whole
Paradigm
that
they
don't
have
in
this
paper,
which
is
like
they're
trying
to
engineer
living
systems
and
developmental
systems.
Yeah.
D
B
So,
let's
see
they
talk
about
multi-scale,
competency,
architectures,
and
so
this
goes
to
the
the
stuff
that
they
talk
about.
They
talk
about
like
how
a
lot
of
systems
have
this
these
levels
of
Competency,
which
they
have
to
achieve
in
order
to
have
a
certain
level
of
complexity,
and
so
you
have
like
you
know
simple
systems
and
complex
systems
and.
B
That
there
are
these
different
competencies
that
they
have
to
have
so
in
an
MCA
components
at
each
scale,
are
competent
to
perform
the
functions
appropriate
to
that
scale
with
that
explicit
top-down
instructions.
So
this
is
where,
like
basically
the
more
complex
you
get
if
you're
a
top-down,
explicit
top-down
instructions,
you
need
things
emerge
bottom
up,
and
you
know
so
there's
this.
So
they
give
an
example
of
human
cells.
They
don't
have
to
be
told
how
to
divide
by
the
brain
or
any
larger,
larger
scale
system.
B
D
Any
yeah
just
because
there's
no
brain
an
early
embryogenesis
goes
after
all,
Michael
Michael
Levin
has
been
talking
about,
for
example,
electric
Fields
yeah,
which
are
Global
in
nature
and
might
provide
some
of
this
top-down
instruction.
How
do
we
call
that
environment?
Because
it's
generated
by
the
embryo.
B
B
So
yeah,
so
this
is
just
kind
of
good.
This
whole
section
kind
of
goes
through
like
these
top
down
and
bottom-up
mechanisms.
So
the
way
they
think
about
top
down
is
top
down
instructions
versus
bottom-up
sort
of
self-organization.
So
this
is
something
again
that
you
know
there's
a
lot
of
work
in
the
complexity.
B
Literature
on
you
know,
bottom-up
self-organization,
so
cells
will
interact
through
maybe
simple
rules
and
they'll
form
some
pattern,
and
then
the
top-down
instructions
basically
tell
that
pattern,
but
again
that
that
that
language
is
a
little
bit
fuzzy
in
terms
of
what
the
top-down
instructions
are
doing.
D
B
D
So
I
mean
for
the
individual
cell
I
suppose
it
is,
but
you
know
it's
about
it's
constructed
by
the
embryo
and
then
has
consequences
in
the
top
dog
action:
okay,
but
the
magic.
We
call
this
micro
environment
I.
Think
it's
misleading
yeah.
B
Yeah,
so
that
that's
kind
of
the
approach
they're
using
you're
talking
about
these
competencies,
they're
talking
about
self-organization
and
these
top-down
signals
versus
bottom
up
sort
of
self-organization
and
things
like
that.
So
they
kind
of
go
through
this
the
introduction
here
and
then
they
start
talking
about
the
free
energy
principle,
which
is
this
least
action
principle
or
energy,
minimization
principle
and
there's
a
lot
of
jargon
associated
with
the
free
energy
principle.
So
it's
it's
kind
of
one.
B
It's
kind
of
hard
to
like
go
into
it
in
like
five
minutes,
so
I
probably
won't
do
that.
Do
they.
D
B
B
Right
so
in
particular,
we
will
see
that
treating
AB
initio
self-organization
is
an
analog
of
regulative
development,
so
AB
initio
means
from
first
principles
so
you're
getting
your
sort
of
self-organization
from
first
principles,
you're
kind
of
making
that
determination
of
something
being
self-organized-
and
you
know
when
you're
looking
at
a
system
and
then
you
they
use
that
that
as
an
analog
of
regulative
development,
and
so
this
challenge
is
too
deeply
entrenched
ideas.
First,
it
questions
the
near
Universal
assumption
that
any
AB
initio
Model
must
result
in
a
self-replicating
system.
B
So
this
is
where
you
need
to
have
a
self-replicating
system,
maybe
at
the
beginning
of
life,
so
regulative
development,
of
course.
B
For
those
who
don't
know
and
there's
a
couple,
people
in
here
haven't
heard-
maybe
heard
that
term
before
that's
the
kind
of
development
where
we
have
cells
that
have
signals
that
sort
of
regulate
with
state
with
their
fate
is
going
to
be
so
in
C
elegans,
the
Fate
is
deterministic,
a
cell
will
divide,
but
it
will
become
a
certain
type
of
cell
and
you
don't
really
need
a
lot
of
self-organization
to
form
say
like
C,
elegans
adult
you
just
need
to
have
the
instructions
unfold,
and
you
know
you
need
to
have
a
decent
set
of
surroundings
in
the
embryo.
B
For
that
to
happen.
In
other
words,
you
can't
have
any
ablations
or
mutations
happen
in
regulative
development.
You
often
get
a
bunch
of
precursor
cells
that
you
know
divide
with
their
clonal
so,
depending
on
the
signal
that
they're
given
they
can
take
on
a
certain
fate,
so
you
can
in
regular
regulative
development.
You
can
see
weird
sort
of
changes
in
development.
You
can
induce
say,
for
example,
interesting
changes
in
development
if
you
give
it
the
proper
signals,
so
the
signals
need
to
be
there
in
order
to
work.
B
It
would
have
development
to
happen,
so
that's
how
they
bring
that
up,
but
anyways.
They
kind
of
talk
about
some
of
these
things
about
how
these
kind
of
models
must
result
in
a
self-replicating
system.
Evolutionary
models
are
narrow
and
onwards
have
strongly
coupled
variation
with
inheritance,
either
through
meiotic
or
mitotic
cell
division.
B
B
We
will
suggest
that
models
in
which
the
needed
parts
are
generated
by
the
environmental
processes
that
are
almost
weakly
coupled
to
the
systems
of
Interest
are
also
worth
consideration
and
then
variation
generated
by
weekly
coupled
processes
is
consistent
with
Evolution
at
the
global
scale,
but
does
not
depend
on
natural
selection
at
any
single
local
scale
thus
provides
a
non-darwinian
source
of
order,
which
I
guess
means
that
they're
sort
of
I'm
never
really
sure
what
what
they're
getting
at
there.
B
But
such
week,
coupling
is
exemplified
by
situations
involving
self-organizing
systems
that
include
engineered
and
manufactured
components.
We
suggest
that
weak
coupling
models
may
be
realistic
in
other
settings
as
well.
So.
B
That
there's
this
you
know
you
can
use
a
developmental
system
in
general.
It
conforms
to
these
sort
of
darwinian
principles
of
reproduction
and
self-replication,
but
in
engineered
systems
which
they're
also
interested
in
you,
don't
necessarily
need
that
you
can
have
order
from
some
other
source
and
I.
Think
that's
what
they're
getting
at,
but
they're,
not
like
clear
about
that.
So.
B
B
I,
don't
know
if
they
I,
don't
think
they
talk
about
RNA
world
I.
Think
it's
yeah.
It's
not
they're,
not
pointing
to
specific
things
in
the
origin
of
life.
It's
just
kind
of
like
yeah.
D
Well,
that's
why
I
said
this
paper
looks
like
formalizing
mythology,
yeah.
A
B
So
yeah
there's
a
lot
going
on
here,
but
I
don't
know
if,
like
Amanda
or
Morgan
had
anything
to
add
I,
don't
know
if
he
had
read
it
and
had
any
thoughts
or
comments.
A
Don't
know
enough
about
that
to
know
if
this
is
like
plausible,
but
it
seems
like
the
key
idea
is
that
the
environment
is
also
a
free
energy
minimizing
agent,
so
you
have
the
system
and
the
environment
and
the
system,
environment,
interaction
and
there's
a
symmetry
to
that,
such
that
the
environment
is
also
trying
to
minimize
variational
free
energy,
and
that
is
that
is
something
that
can
drive
self-organization
of
the
of
the
system,
because
when
the
system
self-organizes
it
becomes
more
predictable,
and
that
means
the
free
energy
is
minimized
for
the
environment
and
I
I,
don't
know
what
the
mechanisms
of
that
could
be.
A
B
B
What
that
means,
and
it's
pretty
broad
so
I
mean
you
know
it
could
be
anything
really
I
guess
it
could
be
like
in
like
sensory
signals
in
the
environment,
or
it
could
be
a
chemical
environment
or
a
physical
environment
which
we
talk
about
often,
and
so
all
those
things
have
like
you
know
it's
hard
to
get
a
sense
of
what
they're
talking
about
without
getting
a
specific
system
involved,
which
is
part
of
the
problem
too.
It's
like
when
you
do
a
theoretical
investigation.
B
You
say
well
I'm
going
to
make
a
statement
about
and
especially
like
in
complexity.
Theory
say
we're
gonna
talk
about
the
self-organizing
potential
of
X,
and
so
then
you
know
you
have
this
General
model
of
you
know
top
down
and
bottom
up
and
all
this
and
then
you
have
to
map
those
two
specific
things
in
the
system,
and
so
then
that
becomes
a
little
bit
harder
because
you
have
to
like
see.
You
know
what
exactly
it
is
you're
talking
about,
and
sometimes
it
makes
sense,
and
sometimes
it
doesn't
so
that's
that's.
B
The
harder
part
of
the
exercise
is
to
say
what
are
the
things
that
we're
interested
in
so
is
it,
like?
You
know
the
evolution
of
a
nervous
system
where
you
have
stimuli
in
the
environment
and
it's
impacting
like
sort
of
the
evolution
of
the
connectome
or
is
it
like?
You
know
some
sort
of
set
of
environmental
conditions
in
a
chemical
environment
where
you
have
it
influences
the
embryo,
where
it
influences
some
cell
colony
and
how
it's
moving
or
how
it's
changing.
B
So
these
are
things
that
you
know
it's
harder
to
get
that
sort
of
the
specifics
of,
and
then
say
we
can
measure
that
and
say.
Okay
now
we
have
the
the
say,
the
parameters
we
can
simulate
it
or
measure
it
and
actually
demonstrate
the
self-organizing
potential.
B
So,
like
I
said
the
the
this,
this
paper
I
think
a
lot
of
it
is
sort
of
intended
to
bridge
the
gap
between
sort
of
the
biological
world
and
the
biologically
engineered
world.
So
they're
doing
a
lot
of
work
with
like
these
xenobots,
which
are
these
embryo
inspired
robots,
or
you
know
things
like
engineered
self-assembly,
where
your
Engineering
Systems
that
are
self-assembled,
and
so
you
know.
C
B
C
B
And
that's
why
you
know
when,
when
they
talk
about
the
environment,
being
sort
of
a
free
energy
minimizer
that
may
be
different,
different
contexts
so,
like
you,
might
have
like
free
energy
minimization
of
like
the
chemical
environment,
which
means
what
exactly
depending
on
the
chemical
environment
where
the
the
sensory
environment.
You
know
it's
a
different
thing,
but
I
guess
you
know
when
you're
minimizing
free
energy,
it's
I,
guess
you
can
kind
of
find
the
analog
there
and
apply
it.
But.
A
Yeah
and
so
regarding
like
what
what
exact
system
environment
interaction
we're
looking
at,
they
do
want
it
to
be
flexible.
So
they
talk
about
Markov,
blankets,
which
my
impression
is:
that's
a
pretty
flexible
way
of
defining
like
what
what
counts
as
the
system
and
what
counts
as
the
environment.
But
it's
my
impression
that,
no
matter
how
you
define
that,
there's
supposed
to
be
this
free
energy,
minimizing
Drive,
coming
from
both
ends
coming
from
the
environment
and
coming
from
the
the
system
so
yeah.
A
B
B
System
boundary
has
always
been
like
a
huge
problem
when
thinking
about
complex
systems,
because
not
everything
is
within
the
system
that
you're
interested
in
it's
kind
of
like
what
is
inside.
You
know
my
system,
my
body,
you
know
I
know
my
organs.
Are
there,
my
cells
are
there?
Sometimes
my
cells
fall
off.
Sometimes
these
cells
are
generated.
So
maybe
that's
outside
the
system.
B
There
are
things
that
I
eat
that
come
into
the
system
that
get
broken
down
and
then
all
of
a
sudden
I
have
chemicals
in
my
system
that
come
from
the
food
that
I
eat.
So
that's
part
of
the
system.
Now
so
there's
this
exchange
between
the
boundary
and
then
having
that
Markov
blanket
is
a
mathematic
way
of
defining
the
boundary
around
that
system.
B
About
the
states
that
are
generated,
not
necessarily
everything,
that's
in
it,
so
it's
like
you
know
what
are
the
states
that
are
produced
so
yeah
I
think
that's,
that's
good,
that
they
kind
of
have
that
framework,
the
sort
of
that
part
of
the
free
energy
principle
and
but
yeah,
it's
a
tough
paper
to
get
through
it's
very
deep
and
they're
in
a
lot
of
you
know
specific
exams,
apples
here,
so
I
don't
see
any
references
to
things
like
RNA,
World
necessarily
but
yeah.
There's
a
lot
of
jargon
too.
B
So
anything
else
I
want
to
say
about
this
paper
or
I,
don't
know
about
Morgan.
If
you
wanted
to
say
something.
E
B
Yeah,
that's
I,
think
that
was
kind
of
one
of
the
impetus
for
artificial
life
community.
So
you
know
they
a
lot
of
the
early
founders
of
the
artificial
wave
Community
were
interested
in
like
these.
You
know
they
they
kind
of
came
from
the
Santa,
Fe,
Institute
and
other
places,
and
one
of
the
things
that
you
know
they
thought
about
was
well.
Life,
isn't
really
an
equilibrium
system.
So
you
know
what.
B
C
A
B
Of
it
was
not
to
measure
the
fluctuations
but
to
measure
the
mean
state,
so
that
was
the
most
important
thing
and
then
they
said
well.
We
have
to
come
up
with
a
method
for
looking
at
these
out
of
equilibrium,
behaviors,
which
are
interesting,
but
it's
hard
to
measure
and
then,
of
course,
when
Quad
City
Theory
came
along,
it
brought
a
whole
set
of
tools
to
do
that,
to
look
at
the
measurements
and-
and
things
like
that,
so
yeah.
B
E
B
Yeah
I,
don't
like
I,
said:
I
I
got
into
the
paper
a
little
bit,
so
there's
so
much
more
here
that
we
could
talk
about,
but
I
think
that
that's
the
basic
idea
and
then
they
get
into
this.
They
get
into
this
replicator
aspect,
and
so
you
know
this
is
where
you
have
this
idea
that
life
has
to
be.
You
know
what
is
life,
and
so
one
of
the
things
about
life
is
that
it
has
to
replicate
self-replicate,
and
so
you
know
they
talk
about.
B
Viewed
more
broadly,
the
environment
provides
the
parts
in
every
case
of
biological
and
bio
or
biochemical
replication
in
the
form
of
molecular
subunits
to
be
assembled
in
essentially
kinematic
process,
and
then
they
talk
about
the
replicator
or
the
idea
of
this
replicator
container,
whatever
being
an
environmentally
driven
process.
The
environment
provides
the
nucleic
acids,
the
enzymes,
the
free
energy
and
the
biochemical
and
thermodynamic
stabilized
compartments
required
for
the
kinematic
processes
of
life.
So
then,
from
the
fep
or
free
energy
principle
perspective,
the
environment.
Does
these
things
to
increase
future
predictability?
B
So
again
the
environment
comes
into
it.
Making
more
of
the
same
kind
of
molecule
generates
a
more
predictable
future
State,
making
a
random
assortment
of
molecules,
so
they're,
arguing
that
whatever
is
outside
of
the
organism
as
opposed
to
inside
of
the
organism,
is
sort
of
enforcing
some
of
these.
You
know
things
on
the
replicator
and
so
again
we're
pretty
we're
pretty
General
we're
talking
generalities,
but
this.
D
B
Blanket
comes
into
play
in
some
of
these
other
aspects
and
then
the
environment
similarly
provides
both
parts
and
a
stabilized
microenvironment
An
Origin
of
Life
models.
The
fep
suggests
that
it
does
this
for
the
same
reason,
and
it
does
the
in
the
case
of
DNA,
replication
or
organismal
reproduction
to
increase
it
increase
its
future
predictive
power.
So
really
what.
C
B
Getting
at
here
is
that
there's
this
in
increased
enhanced
ability
to
predict
things
in
the
environment,
and
so
as
as
a
or
an
origin,
a
life
thing
that's
becoming
live
or
I,
don't
know
how
else
to
describe
it.
B
You
know
the
system
whatever
it
is,
is
trying
to
increase
its
predictability
or
how
to
predict
things
in
the
world,
and
so
it
does
things
like.
Maybe
evolve
movement
or
evolve
sensory
organs
or
whatever
to
help
it,
but
it
also
does
this
sort
of
origin
of
Life
aspect
where
it's
maybe
evolving
replication
as
well.
B
So
it's
an
interesting
paper.
It's
very
deep
one
of
the
things
I
wanted
to
talk
about
before
we
go
is
well.
I
did
want
to
mention
that
last
week
I
attended
a
meeting
on
it's.
B
It's
called
synthetic
biology,
2023
and
it
had
a
lot
of
things
on
looking
at
cells
and
how
to
engineer
cells,
but
I
also
had
a
lot
a
very
strong
origin
of
Life
component
and
so
I'm
going
to
cover
that
in
more
detail
next
week,
but
that's
a
teaser
for
next
week,
where,
like
they,
they
had
a
lot
of
things
on
like
containers
and
vesicles,
and
things
like
that.
So
it's
really
interesting
work
that
people
are
doing
on
that
they're
doing
it
not
so
much
in
the
origin
of
Life.
B
Although
there
was
some
of
that,
but
the
synthetic
cells
that
like
the
Craig
Venter,
Jay,
Craig,
Venture
Institute,
has
been
developing
where
they
take
like
a
container
which
is
a
vesicle
and
they're
looting
it
up
with
chemical,
macromolecules
and
they're,
getting
it
to
work
where
they're,
putting
in
genomes
from
like
bacteria
and
they're
putting
them
together.
So
it's
really
interesting
stuff.
B
I'll
I'll
cover
it
next
week
in
more
detail,
but
I
did
want
to
talk
about
something
to
follow
up
from
last
week,
where
we're
talking
about
the
curvatures
and
planes
of
of
organisms.
So
the
curvatures
refers
to,
like
you
know
how
you
have
a
cell,
and
you
know
we
think
about
the
substrate
of
a
cells.
B
Being
this
you
know
surface,
but
it's
important
to
have
a
curved
surface,
to
facilitate
a
lot
of
the
things
that
you
know
that
organisms
do
or
that
even
single
cells
do,
and
then
we
have
these
planes,
these
anatomical
planes
which
Orient
multicellular
systems.
So
it's
the
sort
of
topological
aspect
of
it.
These
curvatures
in
these
planes
and
so
I-
think
dick
sent
me
this
paper
on.
It
was
by
James
Walsh
at
all,
and
it's
it's
actually
an
archaea
and
so
I'm
gonna
talk
about
it
a
little
bit
before
we
go.
B
This
is
the
vision,
plane,
placement
in
pleomorphic
Archaea
and
it's
this
division.
Plane
placement
is
dynamically
coupled
to
cell
shape
and
so
I
know.
Dick
has
been
working
on
this.
This
paper
and
I
don't
remember
if
it
was
I,
think
it
was
published
as
a
book
chapter
where
they
did
some
computer
vision,
studies
on
different
archaeal
phenotypes,
so
some
of
them
being
triangles
some
of
them
being
different
shapes,
and
so
then
the
idea
is
is
that
especially
in
archaea,
they
have
these
different
shaped
cells,
and
this
in
in
archaea.
B
You
see
this
division
plane
placement
Association.
So
the
summary
of
this
paper
is
one
mechanism
for
achieving
accurate
placement
of
the
cell
division.
Machinery
is
by
a
turing
patterns
and
Turing
patterns
are,
if
you
know
anything
about
Turing
reaction,
diffusion
equations.
You
know
that
those
produce
what
they
call
churning
patterns
and
they're.
Just
you
know
where
you
have
these
coupled
differential
equations
that
have
that
govern
this
chemical
system
that
produces
patterns
like
stripes
or
or
spots,
or
something
like
that,
and
so
in
turn
patterns.
B
You
have
non-linear
molecular
interactions
that
spontaneously
produce
spatio
temporal
concentration
gradients.
So
these
systems
behave
in
a
chemical
environment
and
they
produce
these
spatio-temporal
concentration
gradients.
Basically,
where
you
have
stripes,
so
you
have
a
black
stripe,
a
white
stripe,
a
black
stripe.
Those
are
spatio
temporarily
organized
you
might
get
the
stripes
in
order
in
terms
of
time,
but
also
they're,
ordered
in
terms
of
space,
and
so
each
one
of
those
stripes
is
a
different
concentration
of
the
chemical,
and
so
they
call
them
concentration.
Gradients.
B
The
resulting
patterns
are
dictated
by
cell
shape,
for
example,
the
Min
system
in
E
coli,
which
is
a
non-archeal
bacteria.
She
has
spatio-temporal
oscillation
between
cell
poles,
so
you
have
this
system
that
generates
chemical
oscillations
between
the
poles
of
the
cells
in
E
coli,
leaving
a
mid-cell
zone
for
division.
I
think
we
talked
about
this
in
vesicles,
where
you
have
these
processes
internal
to
the
cell.
B
Where
you
can
you
know
you
have
these
condensates,
which
are
these
things
that
float
around
inside
the
cell
and
they
sort
of
segregate
out
so
that
you
can
actually
replicate
in
a
vesicle
cell
division
or
at
least
the
the
sort
of
the
formation
of
of
these.
You
know
where
things
move
to
the
polls
and
you
get
this
potential
for
cell
division.
B
The
universe
set
out
universality
of
pattern,
forming
mechanisms
in
a
device.
Zone
placement
is
currently
unclear
and
then
Nick
has
underlined
here
we
examine
the
location
of
the
division
plane
into
pleomorphic
archaea
halopharax
Volcanion,
hello,
arcula
Japonica,
so
these
are
two
archaea
species
that
have
this
division
plane
and
show
that
it
it
correlates
with
the
predictions
of
the
Turning
pattern.
Time-Lapse
analysis
of
H
volcani
shows
a
division.
Division
locations
after
success
around
the
division
are
dynamically
determined
by
daughter
cell
shape.
So
that
means
this.
The
cell
shape
of
the
new
new
cell.
B
That's
going
to
be
born
from
H
volcani.
We
show
that
the
location
of
DNA
does
not
influence
division,
plane,
location,
rolling
out,
nucleoid
inclusion,
triangular
cells
provide
astringent
test
for
turning
patterning,
where
there
is
a
bifurcation
and
division
plane
orientation.
So
this
is
where
you
get
this
chain.
This
bifurcation
we'll
see
pictures
of
that
for
the
2rk
exam
and
most
triangular
cells
divide,
as
predicted
by
a
turing
mechanism.
However,
in
some
cases,
multiple
division
planes
are
observed,
resulting
in
cells
dividing
into
three
viable
progeny.
So
this
is
the
Triangular
cell.
B
They
divide
into
three
progeny
because
you
have
three
poles.
Basically,
our
results
suggest
that
the
division
site
placement
is
consistent
with
a
turning
turning
patterning
system
in
these
archaea.
So
what
am
I
talking
about
this
so
yeah?
They
mentioned
fit
Z,
which
is
this
tubuin
protein
family
that
forms
a
contractile
ring
that
affects
cytokinesis.
This
is
the
thing
we
talked
about
in
a
previous
meeting,
the
fifth
Z
system
in
a
vesicle
and
it
it
forms
this
rain
that
is
sort
of
the
precursor
of
cellular
division.
B
So
so
you
know
there's
a
lot
of
things
you
can
talk
about
in
terms
of
non-archia
bacteria
and
other
organisms,
but
these
are
the
archaea
I
think
here,
where
they
show,
like
the
the
they're
able
to
sort
of
segment
it
by
a
computer
vision.
So
they
have
these
images
here
of
age
volcani.
B
So
you
have
the
the
ones
that
are
kind
of
oblong.
You
have
ones
that
are
more
circular
and
then
I'm
trying
to
and
then
so.
These
are
in
various
states
of
division.
The
oblong
ones
are
kind
of
in
the
process
of
dividing.
You
have
something
that
kind
of
look
like
a
potato
where
you
have
the
two
poles
for
division,
and
then
you
have
the
ones
that
are
more
rounded,
and
so
you
see
this
ring
here
in
the
middle.
It
says
this
division
ring
that
kind
of
forms.
B
This
is
the
central
part
that
they're
talking
about
where
the
cell
is
going
to
divide,
and
then
you
can
kind
of
Define
this
through
their
oh,
so
this.
So
let
me
put
some
detail
to
this,
so
the
court
so
B
here
is
the
corresponding
fits
Z1,
gfp
fluorescent
image,
so
this
shows
the
future
division
plane.
So
this
green
is
the
fifth
Z,
and
it's
this.
This
is
where
the
division
plane
is
going
to
be.
B
This
image
has
been
overlaid
with
the
cell
outlines
calculated
using
phase
contrast,
Imaging,
so
they're.
Using
this
to
help
segment,
the
cell
C
is
theoretically
calculated
predictions
from
the
division
plane
in
each
cell.
These
are
given
by
the
nodal
line
of
the
lowest
harmonic
of
the
cell
outlines.
So
this
is
a
little
bit
of
image,
processing,
I,
think
and
then
so
they
just
want
to
be
able
to
calculate
theoretically
the
division
plane
from
this.
These
data
up
here.
So
it's
a
little
bit.
B
It's
it's
a
straighter
line,
so
they're
inferring
it
from
the
gfp
here.
So
that's
why
it
looks
a
little
bit
straighter
and
then
you
can
a
false
cover
overlay,
which
is
in
D.
That
just
shows
like
the
combination
of
B
and
C,
where
you
show
the
prediction
with
the
experimental
observation,
the
intersection
of
the
two
creates
white.
So
this
is
where
you
have
the
gfp
and
this
pink
line,
and
then
the
white
in
between
and
so
that's
the
overlap.
So
it
just
shows
this
band
and
then
you
have
some
of
the
Triangular.
B
Well,
this
is
a
cell,
that's
sort
of
oblong
here
you
see
some
graphs
where
they
show
kind
of
the
zeroing
width
and
the
counts.
So
this
is
a
histogram
showing
this
distribution.
This
is
a
3D
model
showing
the
let's
see.
B
is
surface
plots
for
a
typical
cell,
isolated
from
figure
one
are
shown
below.
B
So
this
is
a
Surface
plot
of
the
cells
in
figure,
one
that
shows
the
division
plane
with
respect
to
the
rest
of
the
the
cell,
which
kind
of
shows
it
in
3D,
and
then
this
is
where
you
see
some
of
these
triangular
cells,
where
you
have
the
division
plane
here
and
you
can
see
that
it's
kind
of
defined,
so
this
is
experimental
cell
division
orientation
and
triangular,
each
volcani
cells
and
a
ratio
in
the
division,
plane
orientations
and
the
cells
of
the
set
see
what
knockout
if
and
that
inform
triangular
shape.
B
B
The
left-hand
number
of
each
panel
is
the
aspect
ratio
of
the
cell
outline,
so
it
just
kind
of
talks
about
the
aspect
ratios.
Of
course,
the
width
to
the
height,
so
you
have
different
sort
of
you
know
topological
realizations
of
the
triangle,
and
then
you
have
this
division
plane
cutting
across
it.
So
the
let's
see
the
left
and
most
image
is
the
phase
contrast
micrograph,
which
is
this
one
like
right
here:
the
black
and
white
or
the
black
and
gray
one.
B
Is
the
experimental
fits
Z,
fluorescence
distribution,
which
is
the
one
here
in
the
middle,
which
you
know
shows
this?
You
can
see
it's
kind
of
blurry,
that's
the
the
distribution
of
the
fifth
Z,
which
is
basically
the
gfp,
but
it's
in
black
and
gray,
and
then
the
right.
The
rightmost
image
is
the
predicted
division
plane.
So
this
is
the
prediction,
so
it's
you
know
cleaner
line,
so
the
cells
have
been
ordered
by
the
cross
correlation
between
the
predicted
and
observed
division.
So
this
is
our
distribution
of
these
in
in
triangular
cells.
B
You
see
these
division
planes
that
have
different
orientations
depending
on
sort
of
the
shape
parameters
of
the
cell
and
the
distribution
of
fit
Z.
So
this
is
this
and
this
kind
of
goes
through
over
time.
You
know
160
180
350
minutes,
so
you
can
see
through
these
different
phases
of
division.
Where
you
have
that
division
plane
and
it
divides,
then
you
have
another
division
plane
which
then
divides,
and
so
they
show
this
time
lapse.
Microscopy
also
see
with
the
Triangular
cell.
The
same
thing:
0
minutes,
10
minutes
400
minutes.
B
So
you
can
see
that
in
the
400
minute,
one,
it's
dividing
into
I,
guess
three
cells:
it's
going
to
divide
into
three
cells
with
each
point
of
the
triangle,
being
a
sort
of
a
pull
of
the
cell,
and
so
you
get
three
cells
that
pull
apart.
You
have
multiple
division
where
the
division
plane
actually
splits
up
here.
I
think
and
you
get
these
multiple
parts
that
divide.
So
you
have
a
division
plane
here
and
a
division
plane
here,
for
example,
and
at
410
minutes
you
end
up
with
three
cells:
better
all
daughter
cells.
B
So
now
I'd
like
to
talk
about
a
few
things
related
to
the
archaea
paper
on
cell
division.
Before
we
move
on
so
in
past
meetings,
we've
talked
about
this
basic
model
where
you
have
some
sort
of
vesicle
or
cell,
and
you
have
this
ring.
Maybe
it's
an
actin
ring
and
it's.
B
So
two
comments
on
this
one
is
that
this
seems
to
be
pretty
conserved.
If
you
see
this
in,
which
means
that
it
has
a
deep
common
ancestry
between
archaea
and
between
animals,
and
so
that
means
that
there's
a
pretty
deep
conservation
of
this,
but
they
also
mentioned
E
coli
in
the
paper.
So
it's
you
know
it's
all
bacteria
and
animals,
so
that
means
that
this
evolved
pretty
early
on
and
we've
talked
about
the
evolution
of
embryos
before
and
this
seems
to
be
maybe
a
Hallmark
in
that
evolution
of
embryos.
B
B
It's
just
right
along
the
midline
in,
but
the
the
best
part
is
that
in
these
triangular
cells
you
have
this
sort
of
median
finding
function
of
this
and
you
know
I,
don't
know,
I've
not
read
enough
about
this
literature
to
know
whether
this
is
some
sort
of
like
signaling
mechanism,
whether
it's
regulative
or
whether
it's
you
know
it
just
is
programmed
to
do.
A
B
Punch
is
that
it
sort
of
finds,
maybe
the
median
or
some
sort
of
follow
some
sort
of
median
gradient
and
I.
Don't
know,
I've
not
read
enough
about
that
literature
to
know
how
well
understood
that
is,
but
my
guess
is
that
it's
doing
that-
and
you
know
this
is
all
interesting
for
a
whole
host
of
reasons.
But
one
of
them,
of
course,
is
that
sometimes
this
doesn't
cell
division
doesn't
happen
along
the
median.
So
in
those
cases
sometimes
you
get
like
different
parts.
B
Like
we
get
polar
bodies,
you
get
two
daughter
cells,
one's
a
polar
body
which
is
a
sort
of
a
debt
on
arrival,
so
it's
a
smaller
sister
to
the
other
cell,
which
then
continues
on
the
cell
lineage,
and
so
you
know
that
that
might
have
something
to
do
with
this
function
in
this
Factor,
but
also
the
pattern
formation
in
the
in
the
Turing
reaction
to
Fusion
models
that
we
talked
about
earlier
in
the
meeting.
How
did
this
play
Roll
in
this?
B
So
things
to
think
about?
Thank
you.
So
that's
that's
that
paper
I
thought
that
was
an
interesting
paper
because
we
talked
about
triangular
cells.
A
long
time
ago,
I
think
we
haven't
talked
about
that
in
a
while,
but
also
this
idea
that
you
can
have
this
sort
of.
You
know
that
you
have
these
division
planes
that
sort
of
emerge
from
some
of
these
chemical
systems
and
the
expression
of
proteins
that
drive
this,
and
you
know
the
division.
B
Plane
is
a
self-organizing
thing,
so
it
kind
of
comes
together
and
then
it
plays
everyone
giving
the
cell
The
Divide,
and
it's
that
replicator
aspect
of
it.
So
so
that's
that's
all
about
that
paper.
I
thought
that
was
really
interesting.
So
is
there
anything
you
want
to
talk
about
before
we
go
today.
E
My
research
is
just
carrying
on
and
right
now
it's
it's
supposed
to
be
in
centimeters
and
I'm.
Getting
meters
like
the
thing
is
stretched
to
meters
long
and
it
started
off
as
the
couple
of
centimeters.
So
there's
something
wrong.
B
B
Yeah
all
right
well,
thank
you
for
attending
next
week,
we'll
we'll
have
updates
from
gsoc
people
we'll
have
other
papers
and
have
a
good
week.
You
too,
all
right.
Thank
you.
I
now
would
like
to
talk
about
a
third
paper
today.
B
This
one
is
on
open
worm
and
digital
Twins,
and
hopefully
it's
something
we'll
enjoy.
So,
let's
see
so
this
is
a
paper
from
the
journal.
Mathematics
and
the
title
of
this
paper
is
a
connectome
based
digital
twin
C,
elegans
capable
of
intelligent
sensory
motor
Behavior.
B
So
this
involves
digital
twins,
something
that
we
talked
about
in
our
Saturday
morning.
Neurosin
meetings,
this
past
Saturday
and
digital
twins.
We
talked
about
it
in
the
context
of
virtual
reality,
but
it's
a
more
broad
engineering
term
that
just
means
you're,
making
a
model.
That's
almost
identical
to
the
thing
in
the
physical
world,
so
you're
you're
using
sensors
you're
using
other
types
of
data
about
the
object
to
make
the
most
sort
of
faithful
representation
you
can,
and
so
in
this
case,
they're
interested
in
C,
elegans
and
intelligence,
sensory,
motor
Behavior
and.
B
Open
arm
is
because
that's
one
of
the
goals
of
the
open
worm,
Foundation
Foundation,
you
know
in
making
these
sort
of
very
realistic
models,
and
so
I
prepared
a
couple
of
slides
for
another
talk,
but
this
kind
of
gives
you
an
overview
of
what
this
looks
like.
So
this
is
an
issue
in
model
realism.
So
you
know
if
you
want
to
build,
say,
for
example,
the
most
realistic
model
of
a
frog.
You
know
you
can
do
that
using
some
sort
of
set
of
materials
and
paint,
and
you
can
make
the
outside
look
very
lifelike.
B
Of
course
the
inside
isn't
life-like
or
you
know
accurate,
but
you
can
build
a
model.
That's
real
on
the
outside.
You
can
also
build
a
model
of
a
plane
where
you
have
the
inside
mimicked
as
well
as
the
outside.
You
include
every
detail
and
you
build
up
models
for
people
to
use
to
explore,
and
so
you
know
this
is
I.
Guess
would
be
considered
to
be
maybe
moving
towards
a
digital
twin
of
a
plane.
C
B
Digital
twins,
in
the
engineering
sense
also
has
this
issue
of
you
know.
Taking
in
sensor,
data
and
mimicking
the
behaviors
of
the
real
things
of
the
model
actually
behaves
like
the
real
thing,
and
so
this
frog
obviously
doesn't
do
that
this
plane
may
do
that.
If
you
were
to
build
a
simulation
and
have
you
know,
maybe
sensors
in
a
real
plane
or
you
might
simulate
the
behaviors
of
a
real
plane
with
a
physics
engine
and
with
some
you
know,
a
set
of
parameters
that
allow
you
to
fly.
B
You
know
tilt
and
move
and
move
change,
direction,
change,
altitude,
and
things
like
that.
So
this
is
an
example
from
C
elegans.
So
this
is
where
you
have
and
I
was
selling
this
special
issue
at
the
time,
but
you
can
see
that
there's
the
muscle
walls
and
the
nerves
and
everything
and
you
can
see
the
simulation.
So
the
idea
would
be
that
the
muscles
and
nerves
are
there.
B
And
so
this
is
an
example
here
from
that
Royal
Society
meeting
it's
a
Turing
test,
which
is
maybe
misnamed,
but
it's
basically
a
test
to
see
whether
you
can
take
a
microscopy
image
of
a
worm
and
a
simulation
of
a
worm,
and
you
can
tell
their
behavior
apart.
So
that's
the
idea
here,
and
so
that
should
be
like
this
idea
of
modern
realism.
B
Now
not
all
models
need
to
be
realistic,
but
that's
the
idea
behind
a
digital
twin
and
then
this
is
you
know
putting
this
on
a
Continuum
where
you
have
this
sort
of
spherical
worm
in
a
vacuum
which
is
where
you
know.
We
just
have
this
idea,
maybe
like
a
drawing
of
a
worm
or
something,
and
it
doesn't
have
any
sort
of
connection
to
the
real
world
and
then
there's
this
hyper
realism,
which
is
maximizes
the
degree
of
predictability
with
the
biological
accuracy
or
realism
of
a
model,
and
so
the
problem
with
these
models.
B
Maybe
is
that
sometimes
they
lack
adaptability
or
robustness
or
evolvability,
so
you
could
make
a
very,
very
realistic
model,
but
you
don't
necessarily
replicate
a
lot
of
the
adaptable
aspects
or
some
of
the
you
know,
mechanisms
for
robustness
and
evolvability
that
actual
organisms
have
so
you
know
they
can't
replicate,
and
then
you
know
you
may
introduce
mutations
and
the
replication
in
the
real
world
in
the
in
the
new
generations
that
come
after
you
make
those
replications
in
in
a
hyper,
realistic
model.
You
can
model
everything
about
that
particular
Worm,
but
it's
Offspring.
B
You
know
you,
you
might
be
able
to
say
you
know,
use
a
genetic
algorithm
to
make
different
changes
to
the
genotype
and
map
those
to
the
phenotype
that
you're
modeling,
but
you'll
never
really
get
the
I
mean.
We
don't
really
understand
how
those
things
translate
into
these
aspects.
So
you'll
never
really
get
a
worm.
That's
hyper
realistic
over
many
generations,
or
do
you
mean?
Maybe
we
will
someday
but
I
doubt
it
in
any
case,
that's
one
of
the
drawbacks
of
this
realism
is
it
doesn't
have
that
sort
of
robustness
of
a
biological
system?
B
So
keep
this
in
mind
as
we
move
through
this
paper,
so
this
paper
I'll
go
through
the
abstract
and
talk
about
it.
Maybe
some
of
the
highlights
of
what
they're
doing
the
question
here
is
whether
this
is
just
repackaging.
What's
already
known,
maybe
repackaging
what
open
worm
is
all
about,
or
is
it
really
something
with
some
new
insights?
So,
let's
see
what
this
is
all
about.
B
Despite
possessing
a
simple
nervous
system,
the
C
elegans
exhibits
remarkably
intelligent
Behavior.
However,
the
underlying
mechanisms
involved
in
sensory
processing
and
decision
making,
which
contribute
to
locomotory
behaviors
remain
unclear
in
order
to
investigate
the
coordinated
function
of
neurons
in
achieving
chemotactic
Behavior.
B
So
this
is
behavior
where
we
have
a
neural
circuit
that
allows
the
worm
to
sense
chemical
gradients
in
the
environment,
we
have
developed
a
digital
twin
of
the
C
elegans
that
combines
a
connectome-based
neural
network
with
a
realistic
digital,
warm
body
through
training,
the
digital
worms
using
offline
chemotactic
Behavior
generated
with
a
PID
controller,
which
is
a
special
type
of
control,
theoretic
controller.
We
have
successfully
replicated
faithful,
sinusoidal
crawling
and
intelligent
chemotactic
Behavior,
similar
to
real
worms
by
avoiding
individual
neurons,
which
means
you
knock
them
out
of
the
circuit.
B
We
have
examined
the
rules
of
modulating
or
contributing
to
the
regulation
of
behavior.
Our
findings
highlight
the
critical
involvement
of
119,
neurons
and
sinusoidal
crawling,
including
b-type,
a-type,
b-type
and
pdb
motor
neurons,
as
well
as
AVB
and
Ada
neurons.
So
these
are
all
different
types
of
neurons
that
exist
in
the
in
connectome
and
so
they're.
Just
doing
these
knockout
experiments
where
they're
taking
out
neurons
and
they're
testing
this
circuit
for
these
different
behaviors.
So
this
is
something
you
can
do
very
easily
in
C
elegans.
B
You
know
you
can
we
know
that
pretty
well
how
the
circuits
work.
So
the
question
is,
you
know:
can
we
take
things
out
in
an
additive
fashion
and
see
the
effect?
And
of
course
you
can
do
that.
B
So
there
have
been
papers
on
this
I
guess:
they've
done
this
for
the
whole
connectome
and
I'm
not
really
sure
what
the
details
are
we'll
go
through
the
details
too
much,
but
you
can
read
that
in
in
the
rest
of
the
paper,
if
you're
interested
We've
also
predicted
the
involvement
of
ddo4
and
ddo5
neurons
and
the
lack
of
relevance
of
ddo2
and
ddo3
neurons
and
crawling
so
you
can
do
these
experiments
where
you
take
out
neurons
and
you
can
see
what
their
sort
of
necessity
is
in
in
the
circuit.
B
This
happens
all
the
time
in
biology,
sometimes
Parts
fail
or
you
know,
Parts
get
damaged,
and
so
they
have
to
be
either
regenerated
or
compensated
for
in
some
way
and
so
in
C
elegans.
You
don't
have
you
know,
you
don't
have
a
lot
of
plasticity,
but
you
do
have
some,
and
but
moreover,
you
have
these
circuits
that
have
a
degree
of
redundancy
and
adaptability.
So
these
are
things
that
you
can
look
at
in
doing
these
kind
of
knockout
experiments,
but
you
can
also
do
this
with
other
types
of
experiments
as
well.
B
You
know
other
types
of
Investigations
looking
at
a
few
regenerating
Pathways,
where,
if
you
use
other
neurons,
to
compensate
what
will
happen
so
there
are
different
ways
you
can
get
at
this.
Additionally
head
motor
neurons,
sub-lateral
motor
neurons
layer,
one
interneurons
and
layer,
one
and
layer
five
Sensory
neurons
are
expected
to
play
a
role
in
crawling,
so
they
they're
testing
all
these
different
parts
of
the
connectome.
B
So
it's
worth
saying
that
the
C
elegans
connectome
is
actually
well
known.
We
both
have
a
synaptic
connectum
and
Gap
Junction
based
connectome,
so
we
have
a
very
good
idea
of
how
they're
all
connected
and
how
they
all
function-
and
these
this
is
now
from
previous
studies.
So
in
C
elegans,
it's
possible
to
build
this
hyper
realistic
model,
whereas
it
may
be
other
organisms.
We
can't,
in
summary,
we
present
a
novel
methodological
framework
that
enables
the
establishment
of
an
animal
model
capable
of
closed
loop
control.
Faithfully
replicating
realistic
animal
behavior.
B
This
framework
holds
potential
for
examining
the
neural
mechanisms
of
behaviors
in
other
species,
so
they
they
talk
about
a
lot
of
detail
here
about
the
experiments.
We
talk
about
a
lot
of
the
different
behaviors.
You
can
test
for
and
see
elegans.
So
there
are
a
lot
of
things
that
this
very
small
connectome
does.
So
it's
302
neurons,
of
course,
and
it
can
do
a
lot
of
diff.
It
can
Model
A
lot
of
different
behaviors,
so
chemotaxis
thermotaxis,
Escape
responses,
mating
and
learning.
B
So
there
are
a
lot
of
stereotype
behaviors,
which
means
that
they're
not
necessarily
learned
they're,
not
necessarily
diverse
in
terms
of
their
output.
They
just
you
know
you
have
a
circuit,
it
functions
a
certain
way
and
you
get
these
very
similar
behaviors
again
and
again
so
so.
This
is
a
very
good
system
for
this.
Of
course,
this
was
something
that
they
they
cited-
the
open,
Worm
Project,
of
course,
so
they
cited
Citation
10,
which
is
this
paper
that
you
saw
on
the
previous
slides.
B
This
is
the
2018
overview
of
open
worm.
This
is
a
paper
by
Portia
Gleason
and
colleagues
and
Stephen
Larson
and
Robin
Hassani
David
long
on
a
multi-skill
framework
for
modeling.
The
nervous
system
of
C
elegans
I
was
from
that
same
special
issue.
B
Then
there
are
these
other
studies
where
people
looked
at
The
Locomotion
patterns
of
a
connectome
that
looks
like
the
C
elegans
connect
Dome.
So
they
were
able
to
like
verify
these
patterns
as
being
stereotyped
through
these
stereotype
forward
motions
and
backwards,
motions
and
they're
they're
circuits
that
you
know
do
this,
these
very
specific
behaviors,
so
you
can
actually
know
how
to
activate
that
circuit
and
then
produce
that
behavior.
B
There
are
also
other
papers
on
the
model
of
motor
control
for
C,
elegans,
gate,
modulation
and
c
elegans,
and
you
can
easily
even
do
things
like
search
based,
reinforcement
learning.
This
is
something
that
reminiscent
using
Vault
has
been
involved
with
the
open.
Warren
Foundation
is
also
done.
So
these
are
all
some
very
interesting,
a
follow-up
on
some
very
interesting
studies
and
they
kind
of
go
through
here.
They
talk
about
the
digital
C
elegans
body.
B
They
developed
a
digital
model
through
mujoko,
which
is
Illustrated
in
figure
one
I
guess
this
is
their
own
modeling
system,
the
Digital
model,
the
worm,
digital
worm,
closely
resembles
a
shape
and
muscle
arrangement
of
a
real
worm.
So
you
can
see
an
A1
here.
This
is
their
model
mujoko,
which
is
where
you
have
the
worm.
It's
on
a
nondescript
surface.
You
have
the
tail
on
the
head
and
you
have
this
angle
between
the
the
points
along
the
body.
B
So
they
have
these
articulation
points,
so
you
can
produce
these
undulating
movements,
like
you
see
in
a
program,
and
you
can
measure
the
angle
between
each
of
these
points,
and
so
this
is
again
you
know
part
of
this
sort
of
realistic
model.
You
know
and
I,
don't
know
how
realistic
it
is
because
you
have
you
have
to
take
a
whenever
you
want
to
model
something
computationally,
as
opposed
to
this
hyper,
realistic
modeling.
B
But
basically
you
could
take
this
frog,
save
for
example,
and
break
it
down
into
a
biomechanical
model
which
would
just
have
like
the
joints,
and
some
of
the
articulation
points
were
relative
to
movement,
and
that
would
be
your
base
model.
And
then
you
could
add
detail
onto
that.
So
it's
always
in
a
computational
model,
even
no
matter
how
realistic
it
is.
You're
always
going
to
have
these
the
sort
of
skeleton
of
sort
of
discrete
Behavior,
so
something
that
generates
discrete
behaviors
and
it
has
to
be
something
in
computer-
can
interpret.
B
B
So
this
is
why
you
know
this
looks
this
is
hyper
realistic,
but
at
the
same
time
or
they
claim
it's
hyper
realistic,
but
at
the
same
time
you
have
a
skeleton
here
where
they
show
these
articulation
points.
Now
it
happens
to
be
the
in
the
real
worm.
You
do
have
these
points
where
you
know
you
have
discrete
neurons.
You
have
discrete
muscles.
B
So
these
things
aren't,
like
you
know
that
far
away
from
reality,
but
you're
not
actually
using
the
actual
pieces
of
you
know,
you're,
not
using
the
sort
of
the
physiology
to
generate
something
you're
using
something
else
in
your.
You
have
to
evaluate
it
quantitatively.
So
this
is
what
we're
do
they're
doing
it.
This
way.
B
The
digital
C
elegans
is
actuated
by
95
body
wall
muscles,
so
they
have
body
wall
muscles
then
around
the
skeleton
arranged
in
four
quadrants.
So
the
dorsal
left
dorsal
white
ventral
left
ventral.
His
joint
is
actuated
by
four
corresponding
muscles
one
in
each
quadrant,
except
for
that
last
joint
with
only
three
muscles.
So,
along
these
different
points,
they
have
muscles
and
that's
how
they're
actuating
so
again,
you
know
you
have
these
muscles
that
are
very
similar
to
C
elegans,
but
they're,
not
exactly
the
same,
and
you
have
to
have
this
computational
skeleton
underneath.
B
So
this
is
really
about.
You
know
having
to
enable
this
physics
simulation
the
sort
of
physics
engine,
the
capabilities
of
a
physics
engine
and
applying
it
to
simulating
the
movement
of
this
organism.
So
it
you
know
it's
it's
realistic,
but
you
have
to
have
this
computational
skeleton
which
may
or
may
not
be
realistic.
B
So
let
me
replicate
a
number
of
different
stereotype
movements
using
this
kind
of
method,
so
in
real
in
reality
or
in
practice,
it's
basically
the
same
thing
as
a
real
organism.
Now
this
may
or
may
not
work
in
something
like
a
mammal
where,
like
a
mouse
or
a
human,
where
you
have
complex
and
nuanced
movements,
you
know
there's
a
lot
of
fine
motor
control,
for
example,
especially
in
humans,
but
in
C
elegans
it
works.
B
B
The
medium
is
the
rate
of
diffusion
through
that
Medium
of
the
substance,
so
that
that's
an
important
aspect-
I,
don't
know
if
they
modeled
that,
realistically
or
not-
but
basically,
if
you
have
like
say
a
neutral
substrate,
where
this
diffuses
uniformly
you'll
see
that
there's
the
source
and
then
the
source
can
Decay.
So
there's
these
these
concentric
circles
that
Mark
sort
of
the
level
of
the
substance.
So
this
is
like
the
maximal
amount
of
substance
at
the
source.
B
This
is
like
the
substance
of
some
Decay.
This
is
with
further
Decay.
This
is
even
further
and
then,
as
you
get
far
enough
out,
it
just
becomes
sort
of
you
know
it.
It
Blends
into
the
background,
so
it's
basically
like
asymptotic
to
zero.
So
you
can
know
if
you
were
to
look
at
a
distribution
of
this.
Your
source
is
here
and
it
goes
down
and
asymptotes
out
to
the
zero.
So
this
is
the
concentration.
B
Now
here's
a
worm
that
comes
coming
into
the
to
this
gradient
and
it's
looking
for
sodium
chloride,
so
naturally
has
sensor
sensory
organs
here
at
the
head
and
it
wants
to
find
this
chemical,
and
so
it
follows
this
gradient.
It
knows
kind
of
senses
that
there's
something
out
here,
starts
exploring
in
different
directions
or
start
sensing
in
different
directions,
and
it
finds
that
there's
a
gradient
here.
So
it
starts
to
follow
the
gradient
and
then
it
follows
it
to
its
source
and
it
might
use
a
stereotypic
movements
to
find
its
way
to
the
source.
B
That
will
then
make
one
of
these
stereotypical
movements
it
might
need
to
back
up.
It
might
need
to
go
forward
through
the
gradient
and
then
again
this
this
background
medium,
that
it's
in
like
how
play
a
huge
role
in
how
it
actually
navigates
that
so,
if
it's
soil
or
if
it's
some
liquid
in
different
phases,
those
can
all
play
a
role
in
how
this
happens.
B
So
this
is.
But
this
is
basically
what
they're
doing
here.
They're
modeling
this
gradient,
and
so
then
they
have
this
chemotaxis
index
Which
models,
the
digital
worms
performance
on
this
gradient,
so
they
can
evaluate
it
and.
B
So
this
is
an
example
of
the
C
elegans
connect
Dome
from
head
to
tail,
so
they've
modeled.
This
up.
This
is
all
well
known.
We
have,
we
know
all
the
all
the
neurons,
all
the
connections
to
muscles.
So
these
are
not
like
a
mystery.
The
the
mystery
well
there's
really
not
necessarily
much
mystery
in
C
elegans,
because
we
know
a
lot
of
these
things
and
we
know
how
they
work
to
some
extent,
there's
a
lot
of
plasticity
in
post,
patch,
C,
elegans
and
I.
B
Think
that's
where
the
opportunity
is
to
like
do
a
lot
of
these
kind
of
studies
in
in
Define
mutants
and
also
in
these
post-embryonic
plastic
phenotypes.
These
polyphenisms,
like
dollar
stage
where
there
are
other
types
of
Developmental
arrest
stages,
which
might
be
interesting
so
and
polyphenism
just
means
that
it
has
an
alternative
developmental
phenotype.
B
So
so
they
model
their
nodes
they're
using
all
all
nodes.
In
the
model
where
it
seem
to
be
homologous,
meaning
that
there's
a
right
in
the
left
hand
side
there's
a
single
compartment
membrane
equation,
so
they're
using
compartmental
modeling
for
the
biophysics
they're,
using
this
to
model
the
neurons.
So
these
are
basically
leakage-
and
you
know
some
of
these
other
chemical
aspects
of
neurons,
so
they're
able
to
model
that
then
these
ordinary
differential
equations
to
adopt
an
explicit
form
using
the
Euler
method.
B
So
they
actually
derive
some
of
these
results.
They
consider
chemical
connections
and
electrical
connections,
so
they
actually
use
Gap
Junctions,
which
is
I,
sub,
igap
ISO
by
Sin,
Sin,
being
the
synaptic
connections
or
the
chemical
connections
and
the
Gap
Junctions
being
the
electrical
connections,
and
so
they
kind
of
you
know
it's
basically.
A
similar
equation
just
needs
to
be
like
a
little
bit
different
in
its
form,
and
so
then
they
weight
all
of
these
connections
based
on
the
circuit
that
they're
looking
at
and
that.
B
Do
this
kind
of
a
model
where
you
have
this
the
circuit
that
is
getting
an
input,
a
sensory
input?
The
neurons
are
working
the
physiological
aspects
of
the
neurons.
The
connections
are
weighted,
so
you
don't
just
have
the
synaptic
connections,
both
so
the
the
electrical
connections
and
then
you
can
actually
do
that
kind
of
a
model.
It's.
B
It
as
a
chemical
connection
or
an
electrical
connection,
or
do
we
do
any
of
this
sort
of
compartment
modeling
it's
just
kind
of
like
a
weight,
and
so
the
weight
is
based
on
some
learning
aspect
of
the
environment.
But
you
know
you
can
break
those
kind
of
weeds.
It
was
nondescript
weights
down
very
deeply
into
these
different
physiological
aspects.
So
then
the
appropriate
receptive
feedback,
which
of
course,
is
important
when
you're
following
the
gradient,
like
I,
said
it
can
be
in
different
types
of
media.
B
So
you
get
this
feedback
from
the
environment
on
the
body
of
the
worm,
so
they
have
this
example
of
the
structure,
appropriate
feedback
connections.
You
get
this
these
different
effects
over
different
joints,
so
each
joint
experiences,
its
physical
environment
a
little
bit
differently,
and
then
it
sends
that
information
back
to
the
Circuit.
B
If
you're,
in
a
liquid
versus
on
a
gel
versus
a
solid
like
soil
or
colloid,
they're
going
to
be
all
different
effects
on
on
proprioception
so
a
model,
this
is
a
closed
loop
system
they're
using
a
PID
controller,
which
is
a
type
of
controller
that
you
find
in
control
theory,
and
so
this
is
an
example
of
the
PID
controller
where
they
use
feedback.
So
they
have
the
three
components:
the
proportional
component,
the
integral
component
and
the
derivative
component.
B
So
they
calculate
different
aspects
of
the
signal
and
they
can
they
sum
them
Downstream
and
then
there's
feedback
and
that
that
determines
how
you,
each
of
these
things,
contribute
to
the
movement
of
the
worm
in
different
time
steps.
So
then
they
have
some
results
here
where
they
have
the
actual
crawling
and
they
simulate
it
and
they
show
the
concentration
over
time
the
gradient
versus
a
normal
gradient
and
then
the
responses
of
the
neurons
in
the
circuit,
and
they
also
show
different
density
plots
of
the
tracks
of
the
CL.
B
Again
virtual
C
elegans
make
over
100
trials,
so
they
don't
always
follow
the
same
path,
but
they
follow.
This
sort
of
you
know
you
can
map
out
this
density
plot
and
show
how
they're
behaving
how
they're,
following
the
gradient
and
so
forth.
B
Okay,
so
then
you
know
you
have
these
different
behavioral
mechanisms
here,
and
you
know
we
know,
for
example,
a
lot
of
these
details,
but
it's
nice
to
see
these
kind
of
studies
or
people
actually
try
to
implement
this
as
a
as
a
digital
model,
now
I
think
they're
term
digital
twin
is
a
bit
hyped.
You
know,
because
the
term
digital
twin
has
been
very
hyped,
especially
in
the
last
couple
years,
where
you
know.
B
Digital
twin
and
people
want
to
use
that
term
because
it's
the
state
of
the
art
and
leads
to
all
sorts
of
exciting
things,
but
this
is
at
its
heart,
looks
like
it's
just
like
a
average
simulation.
You
know
we're
doing
biological
simulation
and
how
accurate
can
we
get
it,
but
of
course,
that
accuracy
is
tempered
by
our
ignorance
of
a
lot
of
biology,
it's
nice
to
see
that
they're
using
these
physiological
models
instead
of
just
saying
that
their
connections
and
that
we
can
simulate
a
connectome
just
kind
of
generically.
B
So
you
know
so
they
don't
I,
don't
know
how
much
they've
verified
it
with
behavioral
data.
B
It's
you
know
it's
probably
necessary
to
to
calibrate
these
models
with
a
lot
of
Behavioral
data,
especially
in
mutants
and
in
other
types
of
cases
where
you
have
a
lot
of
post,
hatch
or
or
post
embryonic
plasticity.
So
we
don't
necessarily
have
that
kind
of
validation
here.
B
But
you
know
there
are
over
simplifications
in
these
models.
So,
for
example,
they
point
out
that
they're
bias
angle
does
lack
biological
realism.
So
there
are
things
that
you
need
to
do
to
refine
the
model,
to
make
it
hyper
realistic.
Now
it's
worth
saying
is
a
final
word.
This
hyper
realistic
model
isn't
necessarily
always
what
you
want.
Sometimes
you
want
to
sacrifice
predictability
in
the
surface
of
realism.
So
for
this
this
frog.
We
want
a
hyper,
realistic
body,
but
we
don't
necessarily
care
about
the
behavior.
B
Is
you
know,
as
opposed
to
something
where
you
want
really
high
degrees
of
predictability,
but
less
biological
accuracy
or
realism,
and
that
might
be
like
a
model
where
you
really
model
some
aspect
of
the
function
and
it
may
or
may
not
match
to
the
connectome.
It
may
be
a
simple
model
where
you
you
know
it
may
be
a
very
simple
physical
model
that
you
use
to
model
a
behavior
like
a
movement
Behavior.
B
There
are
a
lot
of
Bio
biomechanical
models,
for
example
that
use
very
simple
Dash
pod
models
like
where
they
have
a
dash
pod.
If
I
draw
this
out
a
dash
pop
and
spring
model,
they
often
use-
and
so
that
looks
something
like
this-
where
you
have
the.
B
So
it's
a
very
archaic
term,
at
least
in
terms
of
what
a
dash
pod
is
because
most
people
don't
really
know
what
they
are.
Something
like
this
and
you'll
have
a
dash
pot,
which
is
this
thing
that
goes
into
a
a
contain
sort
of
a
contained
space,
and
they
usually
use
these
for
joints.
You
might
have
a
spring
at
this
point.
B
And
this
Dash
plot
allows
these
two
parts
to
articulate.
There
might
be
some
lubricant
in
here.
This
is
a
mechanical
device,
and
so
you
know
they
use
these
I
think
a
lot
in
automobiles
and
in
other
machines,
where
you
have
this
mass
that
goes
into
this
chamber
and
there's
lubricant
in
here,
and
it
pushes
back
and
forth,
and
so
it's.
B
Force
coming
out-
and
these
Springs
will
give
so
there's
some
compliance
to
this
these
rods
here
and
then
you
have
this
movement,
so
the
often
uses
for
muscle
and
the
contraction
and
expansion
of
muscle,
and
so
you
know
you
can
use
a
model
like
this
and
it
doesn't
necessarily
look
like
the
muscle
at
all
or
the
mechanisms
involved.
If
you
look
at
the
muscle
under
a
microscope,
you
don't
see
this,
but
this
is
the
model
that's
used
because
it
models
the
function
very
effectively,
and
so
that's
all
I'll
say
about
that
paper.