►
From YouTube: Eric Jonas - Interview with a Neuroscientist
Description
Dr. Jonas is a computer scientist and neuroscientist. He wrote a paper called "Could a Neuroscientist Understand a Microprocessor?" (http://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1005268).
We discuss lots of topics in this interview, including brain lesions, connectomes, and how neurons are like transistors. (Sorry, we did not have time to include the 4th topic I showed, "Brain Data Overload".)
A
B
Know
what's
going
on
exactly-
and
you
know,
Carson's
this
nice
kind
of
small
conference
and
so
and
also
like
bright,
be
at
the
end
of
2012.
The
Obama
administration
had
announced
their
brain
initiative
right
where
they
overboard
from
large
amounts
of
neural
data.
You
know
I'm
starting
off
graduate
school.
Everyone
was
like
don't
work
on
techniques
for
neuroscience.
It's
like
it's
a
dead
end.
No-One
gets
jobs
that
way.
Technology's
all
very
mature,
but
with
the
rise
of
optical
microscopy
is
an
acquisition
of
mentality
right.
B
B
This
was
these
things
were
becoming
very
hot
again
and
there's
this
whole
field
of
connectomics,
where
people
are
increasingly
trying
to
at
a
single
cell
level,
trace
out
these
circuits
right,
yeah
I'm,
generating
all
of
this
incredible
graph
data,
and
no
one
really
knows
right
now
get
what
to
do
with
them
right.
It's.
A
It's
like
the
welcome
dr.
Eric
Jonas.
He
is
here
at
Berkeley.
A
Tec
us
is
a
postdoc
thanks
for
being
with
us
today.
Thank
you.
So,
let's
get
on
with
it,
then
I've
got
four
different
neuroscience
topics
in
that
your
name
on
it
and
I'm
gonna
show
you
and
I
will
talk
over
and
pick
which
one
you
want
to
talk
to.
So,
first
of
all
the
person
one
is
it's
not
a
tumor,
it's
a
legion
guess
that
maybe
about
these
transistors
registers
and
neurons.
Oh,
my
baby,
great
okay.
A
A
A
B
Actually
doing
all
of
this
computation
and
the
brain
appears
to
be
in
Oregon,
where
in
some
sense
all
those
different
parts
at
the
cellular
level
are
actually
doing
very
distinct
things
as
far
as
we
can
tell.
But
historically
you
know,
because
the
brain
is
so
complicated,
I
mean
they're
they're,
eighty
billion
cells
in
your
brain
that
are
wired
up
to
each
other.
B
With
you
know,
let's
say
it's
really
indifferent
in
our
connection,
so
it's
this
massive
kind
of
circuit
right,
this
massive
road
map
and,
historically
the
only
tools
that
we've
had
to
ask
questions
about
how
the
system
works,
have
involved
kind
of
looking
at
it
from
a
very
large-scale
perspective
and
the
only
organism
for
which
we
have
its
full
neural
connectome.
Is
this
tiny
little
worm
called
C
elegans?
Yes,.
B
B
Sorts
of
organisms
the
way
their
neurons
work
is
actually
much
simpler
than
a
million
neurons
works
over
there.
There
was
this
real
interest
in
trying
to
do
kind
of
high
throughput
measurement
of
larger
numbers
of
neurons,
simultaneously
and
figuring
out
how
they're
all
connected
for
here,
but
their
circuits
are
so
there's
been
this
connector
and
push
over
the
past,
let's
say
10
to
15
years,
where
scientists
really
want
to
get
out
that
data
understand.
B
It
accessible
and
I
think
I
think
that
right,
the
underlying
technological
improvements
have
had
been
pretty
substantial,
especially
you
know,
we've
had
for
a
long
time,
nay,
belated
take
you
know,
take
a
section
of
our
mouse
brain
chop.
It
up
really
thin
and
look
at
it
under
an
electron
microscope.
Right,
neurons.
B
In
time,
and
it's
also
like
the
ability
to
do
anything
with
that,
data
also
has
been
a
real
challenge
historically
and.
A
B
B
The
the
cortex,
the
kind
of
outer
part
of
the
brain
is,
is
organized
as
if
it
has
a
bunch
of
tiny
little
modules
called
critical
columns
in
each
of
those
contains
a
fairly
large
number
of
neurons
and
it's
kind
of
weird,
because
we
know
that
all
different
parts
of
cortex
do
different
things
right:
there's
auditory
cortex
and
visual
cortex
and
there's
there's
cortex.
That
just
is
object.
Detection,
there's
prefrontal
cortex,
which
we
think
is
responsible
for
kind
of
higher-order
processes.
B
The
hope
is
yeah
that
it
kind
of
generalizes
in
this
interesting
way
and
but
like
remember,
the
the
connectomics
we're
talking
about
right
now
is
trained
understand.
Every
cell
is
connected.
There
are
other
approaches,
including
like
that
the
the
Allen
Institute
works
on
where
instead
they
try
and
understand
how,
for
example,
at
the
cortical
column
level,
how
these
are
all
interconnected
right
right.
A
A
A
This
be
a
good
time
to
go
to
the
next
card,
which
is
on
the
same
topic
as
your
paper
that
you
just
mentioned.
I
couldn't
narrow
scientists
understand
the
microprocessor,
because
I
thought
it
was
intriguing.
How
will
you
equate
it
a
neuron
to
like
what
what
component
in
the
computer
might
you
equate
and
there
on
to
a
transistor
or
set
of
transistors
I
mean.
B
It's
an
interesting
question
right,
so
neuroscientists
I
mean
you
can
imagine
you
study.
We
study
the
brain
at
a
bunch
of
different
levels.
Right.
You
know
some,
so
neuroscientists
are
very
interested
in
kind
of
the
chemical
mechanisms
by
which
neurons
communicate
with
one
another
really
other
neuroscientists
are
interested
in
kind
of
how
the
whole
brain
reaches
various
names
of
decisions.
I.
B
Formal
you
know,
I
grew
up
is
like
when
I
was
eight
I
was
like
building
circuits
in
my
garage
right.
So
it's
a
very
in
fact.
There's
it
there's
something
very
kind
of
aesthetically,
pleasing
I.
Think
about
this
idea
that
you
know
we
have
small
modules
and
we
glue
them
together.
We
wire
them
together
to
do
better
things
right.
All
of
engineering
is
kind
of
based
on
this
idea
that
you
know,
I
have
small
blocks.
I
can
put
them
together
to
create
function.
B
You
have
missions
of
abstraction
and
composition,
and
this
goes
through
all
of
engineering
and
kind
of
all
of
computer
science,
and
we
look
for
that
in
the
brain
right.
We
think
that
well,
look
different
modules
in
the
brain,
presumably
do
different
things.
They
may
be
different.
Neurons
or
different
types
of
neurons
in
the
brain
are
doing
different
things.
We
don't
really
understand
how
even
basic
neural
systems
work
right.
Let's
see
reason
we
say
we
can
describe
lots
of
kind
of
phenomena
yeah,
but
we
do
understand
how
a
microprocessor
works
right.
B
We
know
that
we
know
that
certain
parts
of
the
brain,
if
you
remove
them,
you
lose
certain
types
of
function,
but
it's
not
clear.
Why
and
it
separately
to
write
the
right
way
to
be
thinking
about
this,
and
so
we
know
that,
whereas
the
what
I
think
is
interesting
in
in
the
microprocessor
case
is
it
you
know,
everyone
who
has
a
computer,
science
or
computer
architecture
degree
understands
how
that
processor
works
all
the
way
from
the
individual
transistor
right
from
the
individual
logic
element
all
the
way
up
to
the
algorithms
and
the.
B
B
That,
like
lower
level
stuff
and
I
love
kind
of
going
between
the
layers,
but
the
nice
thing
about
the
microprocessor
is
that
lets
you
kind
of
side
step
all
of
these
philosophical
questions
about
what
is
understanding?
What
would
it
mean
to
really
understand
a
neural
system?
We
don't
really
know
is
that
having
the
algorithms
is
understanding
the
circuitry
people
arguing
debate
these
things
that
have
been
for
you
know
literally
like
80
years,
it
was
a
microprocessor.
B
You
can
just
we
have
such
a
good
gut
instinct
right
become
no
understanding
when
we
see
it
there,
and
so,
when
we
apply
are
kind
of
common
algorithms
that
we
use
and
neuroscience
to
the
system.
Right.
We
can
ask
this
question.
Well,
is
the
insight
that
it
looks
like
it's
providing
you
know:
does
it
match
up
with
what
I
think
understanding
really
is
in
the
rhythm
and
the
question
you
asked
of
kind
of?
What
is
the
neuron
equivalents
in
a
microprocessor
I
think
is
actually
like
the.
B
B
You're
any
a
and
that's
kind
of
you
know
the
universe.
I
came
from
and
I
can
imagine
treating
kind
of
a
transistor
or
maybe
a
digital
logic.
Gate
like
that
right
yeah.
But
if
you
do,
if
you
go
one
layer
deeper,
a
neuron
is
not
just
a
simple
thing:
right
neurons,
where
the
most
complicated
cells
in
all
of
existence.
It
has
all
this
interesting
physical
structure.
B
B
So
I
think
that
I
mean
one
way.
One
way,
I
explain
is
that
the
actual
processor
we
looked
at
the
6502
could
not
simulate,
could
not
do
a
good
job
of
simulating
a
single
neuron,
really
yeah.
So
if
you
actually
want
to
do
like
a
biophysical
model
like
a
multi
compartmental
model
of
a
single
neuron
like
that
processor
is
really
going
to
struggle.
So
the
level
of
complexity
in
individual
neurons
is
just
so
incredibly
bad.
B
It's
it's
understanding
how
these
starts
assistant
for
I
mean
the
the
thing
that
was
interesting
for
me
is
that
none
of
our
existing
machine
learning
or
AI
technologies
do
a
good
job
of
figuring
out
how
the
microprocessor
works,
and
this
is
like
a
40
year
old
system.
It's
as
I
said
it's
much
simpler
than
really
even
a
single
neuron,
and
yet
we
can't
extract
caning
me
what
we
were
considered
to
be
meaningful
insight
about
the
underlying
computation
from
observing
all
of
that
data
right.
B
A
So
the
cliffnotes
for
that
paper
is
know
just
in
case
you're
wondering
alright
one
one
more
card
here,
it's
not
a
tumor,
it's
a
lesion.
This
is
again
from
the
same
paper.
We
were
just
been
talking
about
because
I
there's
a
section
in
there
about
sort
of
simulating
lesions
in
the
microprocessor
in
different
places.
Maybe
you
can
talk
about
why
lesions
are
important
in
neuroscience
research
and
why
you
tried
to
do
that
in
great.
B
Experiment,
so
you
could
imagine
if,
if
you
were
trying
to
understand
how
your
car
works
for
anybody's
your
car
do
well
use
drive
forward,
you
drive
backwards,
you
drive
to
the
store
one
way
you
might
try
and
understand
of
your
curbs.
If
you
didn't
know
anything,
and
let's
say
you
had
a
lot
of
cars
sitting
around
as
well,
you
might
try
removing
some
part
right.
A
B
Seeing
if
the
car
still
works-
and
so
you
know,
if
you
remove
the
windshield
wipers,
the
car
will
still
work,
maybe
a
more
quite
as
well
in
the
rain,
but
it
will
still
work,
whereas
if,
if
you
remove
the
steering
wheel,
the
car
will
still
try
to
work,
you
can
go
forward
and
backwards.
Wait
so
you'll
see
that
there's
kind
of
this,
this
deficit
of
functionality
right
now,
if
you
remove
the
brakes,
you
can't
stop.
But
if
you
remove
the
wheels
it
won't
go
anywhere.
Yeah.
B
You
know
people
in
neuroscience,
I
mean
have
been
run
kind
of
similar
experiments
for
the
past
kind
of
50
to
80
years,
where
you
know,
we
don't
necessarily
know
how
all
the
different
parts
of
say
about
brain
work.
But
we
know
that
if
we,
if
we
remove
a
certain
part
of
the
mouse
brain
like
it,
loses
its
ability
to
turn
left
or
it
loses
its
ability
to
figure
out
how
to
get
out
of
a
certain
type
of
Maties
sure
and.
A
B
That
starts
telling
us
is
that
those
parts
are
potentially
important
for
figuring
out
or
for
that
function.
If
you
remove
the
radiator
and
you're
discovered
that
you
can
only
draw
your
car,
for
you
know
a
quarter
of
a
mile
before
it
breaks
that
doesn't
necessarily
tell
you
what
the
radiator
was
actually
doing.
Okay,
you
can
imagine
saying
something
like
well
I
think
the
radiator
is
responsible.
A
B
Would
do
things
like
remove
a
particular
transistor
right
and
then
see?
Could
you
still
play
this
game
and
for
some
of
them
paying
for
the
three
games
we
looked
at.
Some
transistors
only
impacted
a
single
game,
and
so,
if
you
were
a
neuroscientist
said
well,
that
transistor
must
be
intimately
involved
with
heaven.
B
Your
brain
is
kind
of
continually
adapting
and
learning
new
things
and
changing,
and
if
I'm
parts
of
your
brain
have
the
ability
to
kind
of
take
over
roles
from
other
parts
of
your
brain
right.
So
this
plasticity
leads
to
things
like
there
are
people.
Often
you
know
you
hear
these
stories
where
people
with
various
type
of
brain
damage
recover
functionality,
and
that's
because
your
brain
actually
is
kind
of
shocking.
B
Your
neuro
systems
are
shockingly
good
at
recovering
functionality
right
in
the
face
of
various
forms
of
damage,
and
so
what
happens
is
you
can
do
a
lesion
study
and
not
seeing
the
effect
I
can
remove
part
of
the
brain
and
be
like?
Oh,
this
doesn't
appear
to
change
anything
but
there.
What
really
happened
is
that
there
was
this
kind
of
clapping
of
function,
something
else
in
the
brain
kind
of
stepped
up
to
do
to
fulfill
the
role
yeah.
B
A
B
That
the
only
real
path
forward
to
understand
these
really
kind
of
complicated
systems
is
figuring
out
ways
of
making
our
measurement
tools
smarter
right,
having
them
help
us
figure
out
kind
of
what
the
right
questions
are
to
ask,
and
this
is
one
place-
that
kind
of
a
machine
learning
technology
is
getting
better
and
better
at
doing
right
so
about
what
I'd
like
to
be
able
to
do
is
say
kind
of
here's,
my
hypothesis
space.
What
are
the
right
set
of
experiments
for
me
to
run
it
and,
in
fact,
then
run
them
in
an
automated
fashion.
B
B
Out
which
of
these
is
most
likely
to
be
correct,
because
this
kind
of
purely
observational
paradigm,
where
you
know
I,
have
the
thing:
do
a
bunch
of
stuff
I?
Have
the
animal
execute
some
tasks
or
whatever
I
look
at
the
data
and
then
the
algorithms
trying
to
figure
out
what's
going
on,
and
it
seems
to
be
really
unlikely
to
bear
the
kind
of
fruit
that
would
like,
on
the
time,
skills
that
we
care
about.
So.
B
B
B
Really
see
that
genomic
revolution
and
medicine
until
much
more
recently,
and
that's
because
I
think
the
technology
required
the
technology
that
kind
of
sequence,
a
genome,
became
so
cheap
that
what
you
could
do
is
make
some
changes
to
the
system
right,
you
could
genetically,
you
know,
change
some
property
of
some
cells
right
and
then
let
them
do
their
thing
and
then
kind
of
measure
the
resulting
genetic
variation
right.
You
don't
this
part
where
we
could
up
this
kind
of
cheap
sequencing
technology,
enable
us
to
go
from
kind
of
this
passive
observational
model
right.
A
B
And
and
of
course
this
doesn't
sound,
this
doesn't
sound,
especially
a
big
novel
or
deep,
but
we
need
to
remember
that
you
know
for
a
long
time,
we've
had
articles
like
saying
that.
Well,
you
know,
big
data
mean
things.
The
scientific
method
is
obsolete.
Okay-
and
you
know
we're
just
going
to
go
through
all
of
this
data,
but
the
reality
is:
is
that
no
I
think
I
think
more
experiments
are
necessary.
Anything
better
experimental
technologies
are
necessary.
Anything
I'm
using
AI
machine
learning
to
guide
that.
That
is
the
only
way
forward.
Yeah.