►
From YouTube: DevoWorm (2020, Meeting 40): Elastography/Gigapixels, DevoL Paper/Torch Dreams, Interactions/Agency
Description
Papers on Optical Elastography and Gigapxiels, finishing up DevoLean software paper and demo of Torch Dreams, plus papers on Cell-cell Interactions and Biological Agency. Attendees: Susan Crawford-Young, Richard Gordon, Mayukh Deb, Krishna Katyal, Shruti Rajvanshsingh, Assaf Wodeslavsky, Jesse Parent, and Bradly Alicea.
C
D
C
A
E
F
E
E
E
E
E
H
B
H
A
B
A
E
A
Anyway,
there
are
only
eight
slides,
it's
just
some
sample
images
showing
that
what's
possible,
and
the
idea
is
that
you
don't
need
dye
and
you
don't
need
to
cut
the
tissue
you
can
do
it
live
and
they're
developing
a
device
that
will
do
margin
cancer
margins,
while
in
the
operating
room
with
this.
That's
the
kennedy,
wasinga
group.
B
A
B
You
mean
the
market:
okay,
yeah.
Okay,
a
margin
is
based
on
the
following
assumption:
let's
take
a
skin
cancer.
Okay,
the
skin
cancer
cells
are
assumed
to
be
able
to
move,
and
the
margin
is
a
guess
as
to
how
far
they
can
move.
So
you
cutter
you
not
only
cut
out
the
tumor,
you
cut
out
the
surrounding
tissue
and
that's
called
the
margin.
H
B
Okay
and
then
they
send
the
margin
to
a
pathologist
to
see.
If
there
are
any
cells
near
the
edge
of
the
margin,
if
there
are,
then
they
got
to
cut
more.
The
margin
becomes
bigger,
oh
yeah,
okay,
okay,
so
be
very
nice
to
have
a
direct
measurement
of
what
the
margin
actually
is
right
right
up
till
now,
it's
been
totally
guess.
Work.
H
E
A
Okay,
well,
just
I
had
just
chose
a
couple
of
images
out
of
the
papers
just
to
show
that
it
works
like
they
were
nice,
nice
images.
So
you
can
share
the
slides
if
you
want,
or
just
just
read
the
papers
that
they're
the
one
on
optical
coherence.
Elastography
is
a
little
bit
more
involved,
but
that's
okay,
yeah
hi
yeah.
E
E
E
G
E
E
Yeah,
so
this
is
the
field
of
view
spatial
resolution,
so
this
is
like
how
you
know
when
you
like,
it's
like
microscopy
where,
if
you
have
a
higher
resolution,
you
can
get
to
a
smaller
spatial
resolution
and
then
that's
connected
to
the
field
of
view,
which
is,
if
you
have
a
single
image.
What
what
distance
does
it
cover
across
the
image?
So
you
see
atomic
fork
force
microscopy
is
way
down
here
and
then
oc
is
up.
E
C
B
B
E
B
B
B
H
H
E
Go
ahead
I'll
blow
up
all
right,
yeah
yeah!
So
this
is
let's
see
if
we
can
find
another
image
here,
because
they
just
kind
of
go
through
the
technical
details
of
oce
and
then
some
of
the
math
and
displacement
amplitude
or
strain.
So
they
go
through
sort
of
young's
modulus
and
how
they
calculate
that.
A
E
A
C
A
A
G
A
A
E
E
A
E
A
A
Okay,
I
have
several
interesting
things
happening
with
me,
including
a
dog
barking
and
stuff
anyways.
Then
there
was
then
there's
some
images
that
of
this
skin,
that
there
there
they
go,
there's
all
of
the
different
frequencies
and
you
can
see
at
1100
and
1200
that
you've
got
pretty
clear
images
1300
as
well,
and
they
say
that
it's
not
they
didn't
have
set
it
up
correctly.
For
the
900.
H
A
It's
not
optimized
for
900
nanometers,
but
you
probably
do
better.
Yes,
and
if
you
keep
on
going
down-
and
you
can
see
the
there,
these
three
histology
images
compared
to
optical
clearance,
tomography
and
optical
currency
elastography,
I
believe
anyway,
you
can
see
detail
yeah,
quite
decent
detail
using
using
that
method.
A
B
A
E
A
E
A
E
An
idea
a
lot,
you
know
there's
some
of
you
here,
probably
don't
you
know,
maybe
a
little
bit
above
your
head
in
terms
of
the
technique
and
all
that,
but
so
we're
you
know
we're
looking
at
different
ways
to
image
tissue
here
we
have.
You
know
we
have
the
regular
microscopy
that
we
usually
feature
in
the
meetings,
but
that's
usually
on
something.
That's
fixed
or
you
know
a
culture
of
cells
or
of
something
else,
maybe
worms
so
we're
taking.
E
You
know
we
can
take
images
like
that,
but
then
we
can
also
do
histology,
which
is
where
we
take
images
of
tissues.
Sometimes
they
fix
the
tissue
with
a
chemical
solution
and
they
slice
it
up
and
they
put
it
on
a
slide
or
they'll.
Do
it
live,
and
this
is
kind
of
what
we're
talking
about
is
doing
this
sort
of
imaging
live.
E
Where
you
put
it,
you
know
an
imaging
device
on
like
someone's
skin
and
you're,
actually
looking
at
the
skin
live,
and
you
want
to
see
detail
underneath
so
like
they're
looking
for
cancer,
tumors
or
they're
looking
for
other
things
in
the
skin,
you
can
see
them
and
dick
has
done
a
lot
of
work
on
that.
So
susan
is
actually
looking
into
some
of
these
methods.
E
So
this
is
this.
Is
you
know
this
is
all
really
interesting.
I
mean
you
know
if
we
want
to,
I
sent
out
these
two
papers
last
week
there
are
other
papers.
Of
course
you
know,
there's
a.
I
think
that
probably
has
bibliographies
on
this.
If
he's
you
know
willing
to
share
them,
if
you're
interested
you
can
get
in
touch
with
him
sure
yeah.
E
E
Okay,
there's
dick
gordon's
email
here.
If
you
want
to
contact
him,
so
let's
see
the
paper
is
here.
Let's
see,
let
me
share
my
screen
because
we've
got,
I
think
I
was
presenting.
Okay
looks
like
we
have
a
soft
here,
so
he's
been
around
open
worm
a
bit
and
he
you
know
he's.
I've
talked
to
him
before.
How
are
you.
H
J
E
Yeah,
thank
you.
Let's
see,
let
me
I'm
gonna
share
my
screening
in
here.
So
we'll
talk
about
this.
This
is
the
paper
that
kind
of
came
out
of
last
year's
google
summer
of
code,
so
we,
but
we
created
this
divo,
learn
framework,
but
we
also
created
the
diva,
learn
software
package
and
that
was
mayulk's
project.
E
E
We
then
we
have
diva
learn
which
is
the
standalone
software
and
then
c
elegans
diva
learn
which
is
actually
odual's
project
from
last
year,
which
is
taking
a
lot
of
the
resources
that
we've
had
up
to
this
point
and
putting
them
into
a
library
of
machine
learning
tools.
So
we
do.
We
have
machine
learning,
resources,
deep
learning,
resources
here,
but
we
also
have
some
other.
E
Who
are
computer
scientists,
or
you
know,
they've
not
heard
of
some
of
these
organisms
and
so
be
nice
to
give
them
some
information
background
on
some
of
them.
Nothing,
like
you,
know
nothing
too,
comprehensive,
but
just
enough
so
that
they
know,
like
you
know
what
they're
looking
at
if
they
have
a
data
set
that
they're
using.
E
So
this
is
the
joss
paper,
so
this
is
all
sort
of
organized.
Actually,
this
repository
has
some
images
and
some
video
yeah.
So
this
is
where
the
paper
resides.
This
is
the
manuscript
here
and
I
don't
know
if
I
got
to
have
this
or
I
think
I
made
a
mistake
on
the
formatting
here,
but
this
is
the
title
so
pre-trained
deep
learning
models
that
enable
computational
developmental
biology,
research
and
education.
E
It
might
be
a
little
long
but
we'll
I'll
I'll.
Think
about
that
more.
So
it's
my
oak
myself
and
usual-
and
this
is
our
affiliations
this.
So
this
is
the
format
that
they
want.
So
the
journal
of
open
source
software,
they
it's
it's
a
sort
of.
It's
run
on
github,
it's
a
peer
review
journal,
but
they
run
everything
on
github,
so
you're
supposed
to
submit
a
markdown
file
in
a
certain
format,
and
we
need
to
like
finalize
that
format.
I
think
this
is
not
quite
right,
but
I
don't
know
what
it
is.
I
E
I
E
E
I
E
E
Yeah,
I
mean
just
just
the
point
on
this.
You
know
we
want
to
be.
We
want
it's
important
to
be
able
to
cite
where,
where
we
got
the
data
from
some
of
these
open
source
tools
that
we
cite
or
some
of
these
open
source
repositories,
you
know
they.
If
they
want
to
get
funding,
they
have
to
have
citations
of
them
by
other
people.
So
if
you
know
we
can
cite
them
and
then
they
have
those
in
support
of
like
their
further
development.
E
That's
that's
what
that's
that's
a
positive
thing,
so
we
want
to
make
sure
to
cite
them
as
much
as
possible.
So
that's
that's
in
this
part,
so
we
mentioned
like
the
epic
data,
set
the
worm
image
database
and
then
we've
given
a
citation
and
then
that
will
help
them
out.
Then
I
redrew
some
of
these
figures.
So
this
is
the
this.
Is
the
figure
sort
of
like
a
schematic
for
this
software?
E
E
It's
very
data
science
friendly
compatible
with
numpy
and
pandas,
and
then
this
is
the
an
image
of
the
umbrella.
So
then
we
could
start
talking
about
go
transitioning
from
the
divo
learn
program
to
the
divor
divo
learn
framework,
which
is
this
framework
where
you
have
devolver
in
the
program
here
and
then
this
larger
framework
of
things
and
then
that's
where
like
gojoler's
project
is
here
and
then
this
is
some
of
the
existing
infrastructure.
E
And
then
I
kind
of
mentioned
that,
because
you
know
it's
like
people
will
be
able
to
go
to
the
github
organization
and
contribute
and
it'll
drive
everything
forward,
and
then
they
have
some
acknowledgements
here
and
then
the
references
so
the
references,
I
think
that's
good,
but
it
was
well
or
my
yoke
or
as
well.
If
you
want
to
add
in
any
references,
you
can
just
keep
being
mindful
of
this
numbering
scheme,
because
you
know,
if
you
put
in
a
strange
number,
it
throws
everything
off.
E
E
I
E
The
idea
is
that
we'll
submit
this
soon
to
the
journal
of
open
source
software,
they'll
review
it
and
then
it'll
become
like
a
published
thing.
They'll
give
it,
I
think
they
assign
it
to
doi,
and
then
you
know,
that's
it's
a
small
publication.
It
shouldn't
be
any
longer,
I
think,
than
two
or
three
pages,
and
then
you
know
then
we'll
have
this
yeah.
I
think
it's
pretty
prestigious,
but
it's
it's.
The
idea,
I
think,
is
it'll
get
exposure
to
a
larger
audience
of
people.
So
there'll
be
a
lot
of
open
source
software.
E
I'm
not
sure
how
they
deal
with
them.
We
know
that,
like
github
a
lot
of
people
when
they
want
to
publish
something
that
they've
done
on
github
like
a
release,
they'll
often
use
like
a
third
party
like
zenoto,
and
it
generates
what
they
call
a
doi,
and
some
of
the
databases
will
pick
up
like
the
those
sorts
of
things,
but
I'm
not
sure
that
the
web
of
science
picks
up
like
those
sorts
of
citations.
I'm
not
now
the
journal
of
open
source
science.
E
E
I
I
Okay,
so
is
my
screen
visible,
yeah,
yeah,
yeah,
okay,
well
quickly,
go
to
the
repository
so
basically
like
what
happens
in
developmental
biology.
Is
that,
like
we
see
patterns
which
emerge
from
basically
like
seemingly
random
things
right,
so
we
can
see
like
similar
things
which
happen
inside
like
which
can
be
of
which
can
be
done
using
neural
networks
so
basically
like,
for
example,
I'll
just
show
an
image
quick
quickly.
I
For
example,
we
have
this
noise
that
this
is
basically
random
noise
right.
But
if
we
feed
this
image
to
a
neural
network,
it
starts
hallucinating.
Certain
kinds
of
patterns
like
different
layers
within
the
model,
hallucinate
different
kinds
of
patterns,
and
if
we
optimize
this
noise
to
further
like
to
further
optimize
for
those
pattern
hallucinations,
we
basically
get
a
lot
of
very
like
fascinating
patterns
within
deep
learning
models
and
they
vary
from
like
model
to
movie.
I
I
That
is
when
you
are
basically
helping
eating
quote
unquote,
so
even
neural
networks,
when
we
feed
them
those
kinds
of
random
images,
they
have
certain
activations
of
animals
or
even
like
castles
or
even
palaces,
and
things
like
that.
So
even
like
we
can.
We
can
optimize
that
input
image
such
that
those
such
that
those
hallucinations
are
they
get
enhanced
further.
So
we
can
see
what
the
neural
network
was
hallucinating
or
seeing
in
that
random
noise.
I
I
I
I
And
then
again
I
optimize
another
layer
and
we
can
see
something
like
grass
or
like
I
don't
know
what
this
is,
but
this
is
some
kind
of
a
pattern
that
seems
that's
emerging
after,
like
like
500
600
iterations
of
optimization,
like
I
am
back
propagating
on
the
image
instead
of
the
deep
learning
model.
That
is
what
is
happening
inside
okay,
so
like
the.
I
If
I
want
to
summarize
it
in
one
statement,
it's
like
if
it
looks
like
a
cat,
make
the
image
look
more
like
a
cat
that
is
basically
a
one-line
summary
of
what's
happening
inside
so
then
I'll
just
and
then
what
I
did
was
that
I
like,
instead
of
optimizing
these
two
channels
individually,
I
optimize
them
together
simultaneously
and
after
doing
that,
I
got
something
like
this,
which
is
a
mix
of
both
and
we
can
even
see
eyes
things
like
eyes
and
stuff.
So
it's
kind
of
fascinating.
E
I
E
Well,
there
there's
actually
yeah,
there
are
people
doing
research
too,
on
like
visual
illusions
and
like
neural
nets,
and
there
I
think,
there's
a
guy
in
japan
who
does
a
lot
of
work
on
this.
Where
they're
looking
at,
I
think
a
long
time
ago
we
did.
I
did
a
presentation
on
on
like
what
is
the
name
of
that
thing.
It's
it's
where
you
perceive
faces
and
things,
and
you
know
that
doesn't
don't
have
faces,
and
so
those
are
things
that
you
know
we
can.
E
I
E
E
D
A
A
F
A
B
Let
me
make
a
general
comment:
the
there
is
a
general
problem
which
has
hardly
been
addressed,
and
that
is
the
detection
of
melanoma,
which
is
a
dark
pigmented
tumor.
It
can
be
started
very
small
in
the
skin
in
dark
skinned
people
when
the
visual
contrast
is
low-
and
maybe
these
optical
coherence
techniques
might
solve
that.
But
it's
still
an
unsolved
problem.
A
F
I
have
one
more
query:
can
we
do
instead
of
using
that?
Can
you
know
find
a
normal
range
for
human
for
specific
tissue
like
taking?
You
can
say
that
instead
statistical
mean
of
100
000
people,
and
then
you
can
say
I
find
some
general
tendency
of
what
would
be
the
range
and
if
that
range
is
exceeded,
or
you
can
say
intercede,
so
we
can
say
that
there
are
some
abnormality.
E
E
E
All
right,
so
I'm
gonna
as
our
obligatory
paper
cue
here.
I
know
we
presented
susan's
optical
tomography
papers,
but
I
have
a
couple
other
things
to
look
at
here.
So
okay,
here
we
are
so
two
papers
that
I
found
this
week.
One
of
them
is
really
interesting.
I
think,
if
you're
interested
in,
like
cell
cell
signaling
and
morphogenesis,
it's
called
inferring
the
spatial
code
of
cell
cell
interactions
and
communication
across
the
whole
animal
body,
and
so
this
is
actually
something
from
c
elegans.
E
E
However,
the
spatial
code
embedded
in
the
molecular
interactions
that
drive
a
sustained
spatial
organization
and
then
the
organization
that
in
turn
drives
the
intracellular
interactions
across
the
living
animal
remained.
It
remains
to
be
elucidated,
so
here
they're,
looking
at
the
expression
of
ligands
receptor
pairs
obtained
from
a
whole
body,
single
cell
transcriptome
of
c
elegans,
and
this
is
in
the
larvae.
E
So
they
use
a
3d
atlas
of
c
elegans
cells.
They
implement
a
genetic
algorithm
to
select
ligand
receptor
pairs,
most
informative
of
the
spatial
organization
of
cells.
So
from
the
transcriptome
data,
which
we
don't
talk
about
a
lot
in
this
group,
they
have
a
bunch
of
transcription
factors
that
they
measure
and
those
trans.
E
Some
of
those
transcription
factors
represent
these
ligand
receptor
pairs,
and
so
then
you
can
actually
take
the
data
set
and
run
this
algorithm
on
it
to
select
the
ones
that
are
maybe
most
enriched
in
certain
cells,
and
that
tells
you
something
about
like
the
interactions
between
the
cells
validating
the
strategy.
The
selected
ligand
receptor
pairs
are
involved
in
known
cell
migration,
morphogenesis
processes
and
we
confirm
a
negative
correlation
between
cell
cell
distances
and
interactions.
E
Let
me
see
if
there
are
any
images
in
this
paper
that
are
really
interesting,
because
I
mean
going
through
the
methods
might
be
a
bit
much
it's.
You
know,
pretty
intense
and
I'll,
give
you
a
link
to
this
folder
when
we're
done,
but
so
they're
looking
at
this.
So
the
methods
they
have
single
cell
rna,
seq
data,
so
there's
a
method
called
single
cell
rna-seq,
which
is
where
they
take
this.
The
cells
in
the
extract,
rna
and
they're.
E
Looking
at
all
the
rna,
that's
expressed
across
the
genome,
so
there's
a
sequencing
tool
that
they
use
to
align.
What
they
do
is
they
extract
rna
sequences
which
are
like
dna
sequences,
but
a
little
bit
different,
and
then
they
align
them
with
the
genome.
So
you
have
all
these
little
fragments
of
sequence
that
represent
what
they
call
transcripts
and
they
align
them
with
the
genome.
So
they
can
tell
where
these
what
genes
are
being
expressed-
and
they
can
do
this
at
the
single
cell
level,
which
means
each
cell.
E
You
can
extract
its
own
rna-seq
profile,
and
so
they
have
this
data
set
containing
27
cell
types
at
the
larval
l2
stage,
which
is
you
know,
pretty
early.
You
know
moderately
early
in
the
larval
development,
and
so
they
were
able
to
use
that
as
a
transcriptome
and
then
they
were
able
to
align
this
with
a
3d
digital
atlas
of
the
larval
l1
stage.
E
E
E
E
So
you
have
touch
receptor
neurons
across
here
other
interneurons,
and
you
get
this
degree
of
of
the
cci
score,
which
is
a
degree
of
interaction
body,
wall
muscle
versus
germline.
You
can
see
or
versus
touch
receptor
neurons.
You
can
see
it's
about
the
same
value
and
you
have
this
heat
map,
where
you're,
showing
maybe
a
little
bit
higher
cci
score
for
cholinergic
and
pharyngeal
neurons,
for
example,
or
pharyngeal,
neurons
and
ciliated.
Sensory
neurons,
so
neurons
are
interacting
together
more
strongly.
E
E
And
then
they're
looking
at
these
pairs
selected
in
each
one
of
their
genetic
algorithm,
so
they're
looking
they're
looking
now
at
the
transcripts
and
they're
looking
at
the
abundance
of
these
transcripts,
so
transcripts
occur
at
a
certain
copy
number
in
each
cell,
so
the
more
copy,
the
higher
copy
number,
the
more
the
more
abundant
it
is,
and
that
tells
you
something
about
its
activity
that
it's
higher.
You
know
it's
more
active
in
those
cells,
which
means
that
if
you
have
two
cells
that
have
a
ligand
receptor,
that's
highly
active.
E
So
that's
that's
kind
of
an
interesting
graph
because
they're
using
these
colored
arrows
yeah.
So
I
think
that's
I
mean
this
is
a
pretty
intense
paper.
I
think
if
you
want
to
read
more
I'll
I'll,
give
you
the
link-
and
you
can
read
it,
but
this
is
a
nice
graph
here
of
c
elegans,
as
sort
of
the
phenotype
laid
out
and
showing
like
all
these
different
cell
types,
their
location
and
then
their
degree
of
interaction
here.
E
Okay,
my
oak
had
to
leave,
and
so
did
dick,
but
I'll
go
through
one
more
paper
and
that's
his
biological
research,
and
this
is
something
I
think
jesse
would
be
interested
in.
Specifically,
this
is
life
with
purpose.
This
is
an
aon
article,
it's
a
popular
article
here
and
we
talked
a
couple
weeks
ago
as
well
about
this
new
paper
on
sort
of
non-neuronal,
cognition,
michael
levin,
and,
and
you
know
they
wrote
this
paper
on
non-neuronal
cognition.
That
was
called
cognition
all
the
way
down.
E
The
idea
here
is
that
biologists
bach
at
any
talk
of
goals
or
intentions,
but
a
bold
new
research
agenda
has
put
agency
back
on
the
table,
and
so
this
article
goes
through.
You
know
the
role
of
maybe
like
purpose
or
goals
or
intentions
in
biology,
and
so
some
people
have
suggested
that
you
can
use
a
model
of
cognition
to
represent
goals
and
and
intentions
in
biological
systems
that
don't
have
a
brain,
and
so
this
is
a
little
bit
more
on
this
in
this
direction.
E
E
Other
things
in
the
body
and
they
try
to
minimize
the
damage
to
the
organism,
and
so
there's
this
thing
called
adaptive:
immunity
which
allows
the
immune
system
to
to
do
its
thing
and
and
but
you
know,
then
you
say
well.
How
can
this
be?
If
it
doesn't
have
a
brain?
How
can
you
have
this
kind
of
intentional
system?
E
So
you
know
this
description
of
the
immune
system.
They
ask:
isn't
this
an
absurdly
anthropomorphic
way
of
describing
a
biological
process?
Single
cells?
Don't
have
minds
of
their
own,
so
surely
they
don't
have
goals,
determination,
gusto
when
we
attribute
aims
and
purposes
to
these
primitive
organisms.
E
So
it's
kind
of
a
minimal
image
of
some
sort
of
action
and
ask
people
what's
going
on
and
they
attribute
some
sort
of
thing.
That's
you
know,
there's
a
some
narrative
to
it.
So
these
triangles
and
circle
in
this
box
they're
escaping
the
space,
and
you
know,
whereas
this
could
just
simply
be
a
phage
or
a
macrophage
in
biology
so
where
they
could
just
be
abstract,
geometric
shapes,
but
we
assign
some
sort
of
meaning
to
this.
E
And
so
that's
the
idea,
and
so
the
question
is,
is
like:
if
we're
observing
something
at
the
biological
scale
you
know
is
there's,
are
we
just
assigning
narratives
to
it
when
we
come
up
with,
say
a
theory
of
how
something
works
or
even
describing
it
or
you
know,
are
we
justified?
Because
maybe
there
is
something
going
on
in
terms
of
like
you
know,
agency
or
action,
and
so
that's
a
good.
I
think
that's
a
good
way
to
kind
of
think
about
this.
Is
you
know
it
to
be?
E
Mindful
of
you
know,
are
we
just
kind
of
like
adding
in
our
own
story
that
you
know
is
sort
of
anthropomorphic
and
maybe
something
we
would
do
if
these
were
human
agents
instead
of
just
shapes
or
cells
or
whatever
they
are
and
so
yeah?
So
this
paper
said
you
know
they
point
out.
The
author
here
points
out.
One
of
biology's,
most
enduring
dilemmas,
is
how
it
dances
around
the
issue
at
the
core
of
such
a
description
agency,
the
ability
of
living
entities
to
alter
their
environment
with
purpose
to
student
agenda.
E
People
criticize
you
very
much
if
you
talk
about
like
agency
or
sometimes
if
you
talk
about
teleology,
but
you
know
it's,
the
question
is:
is
that
really
justified,
or
maybe
there
is
something
there?
So
you
know,
cells
and
bacteria
aren't
really
trying
to
do
anything
just
as
organisms
don't
evolve
in
order
to
achieve
anything,
that's
the
idea,
and
so
we
want
to
be
careful,
but
we
don't
also
want
to
throw
the
baby
out
with
the
bathwater.
E
So
so
they
they
suggest
that
a
bottom-up
theory
of
agency
could
help
us
interpret
what
we
see
in
life
from
cells
to
societies,
as
well
as
in
some
of
our
smart
machines
and
technologies,
we're
starting
to
wonder
whether
artificial
intelligence
systems
might
themselves
develop
agency.
But
how
would
we
know
if
we
can't
say
what
agency
entails?
E
Only
if
we
derive
complex
behaviors
from
simple
first
principles
says
the
physicist
suzanne
still
the
university
of
hawaii.
Can
we
claim
to
understand
what
it
takes
to
be
an
agent,
and
so
this
is
the
then
they
go
kind
of
go
through
what
an
agent
is,
and
so
you
know
it's
a
pretty
good
article
if
you're
interested
in
that
area,
I
think
but
yeah
there's
a
lot
of
discussion
to
be
had
about
that.
I'm
not
sure
what
the
you
know
I
mean
we
can
apply
to.
E
E
Meetings,
we
can
talk
a
little
bit
more
about
this,
but
I
know
that
jesse
and
I
and
some
other
people
been
having
conversations
about
this,
which
is
why
I
bring
it
up
and
dick
also,
I
think,
has
talked
about
expressed
interest
in
talking
about
the
biology
of
purpose
in
the
past.
I
think
he's
written
some
things
on
this,
so
I'm
not
really
sure
you
know
I
kind
of
brought
this
up
after
he
left,
but
we
will
talk
about
in
future
meetings.
So.
L
L
E
A
M
J
J
E
E
So,
okay!
Well,
thanks
for
attending.
If
you
again,
if
you
have
any
papers
to
recommend
or
if
you
want
to
present
anything
in
a
meeting,
please
let
me
know
in
advance
and
and
we
meet
at
the
same
time
every
week
so
next
week
we
have,
I
think,
an
open
schedule.
So
if
you
want
to
do
something,
we
can
do
it,
otherwise
I
will
send
out
some
maybe
some
follow-up
to
this
meeting
and
hope
to
see
everyone
next
week
have
a
good
week
thanks,
bye,
bye,
thank.