Loading video player...
>> Fahad Saleh: Okay. Should we
launch?
All right. Welcome to from
whiteboard to mainnet. This is a
podcast brought to you by
Columbia University Center for
Digital Finance and Technologies
and also the Ethereum
Foundations Academic
Secretariat. Today, we're going
to be talking about
decentralized autonomous
organizations or tao's, for
short. And so I'm your co-host
Fahad alongside Theo and we have
four awesome guests here. Today,
we have three academics and one
one person who's an industry
now. So we're gonna talk both
about research and we're going
to talk about the practical
applications of it. So let me
briefly introduce our guests at
the top here. So we have Jungsuk
Han from Soul National
University Jongsub Lee also from
Seoul National University Tao
Li from the University of
Florida, and then we have
Michael Zargham, who's the
founder and CEO at Block Science
and Research.
Directorate megatov. So what we
want to do here really is, we
want to start by talking about
some work that Jongsubh Jungsuk
and Tao have on Tao's and
specifically looking at sort of
looking at it from an economic
perspective. And then we're
going to transition into having
a broader conversation about the
implications and how to sort of
properly design voting
mechanisms in the context of
Tao's. So let me just dive right
in and I think, you know, the
first thing to really think
about before we get into sort of
thinking about any research is
to understand, kind of at a
high level, how TAO'S differ
from traditional firms because
Tao Jongsub and Jungsuk are all
professors of finance. So
they've studied at Lent and
written on at length, what we
call, corporate finance, which
is the analysis of traditional
firms. And so when we start to
think about Tao's, we want to
understand In foremost kind of
how they're different.
from from traditional firms. And
that also, then informs exactly
how researchers start start to
do formal analysis
about them. So, so, first of
all, yeah, what are the primary
differences?
between a traditional firms and
dows? And also how does that
enter the framework of
analysis? And so, you know,
Jungsukj jongsub or Tao? Any of
you do you want to Take that
one.
>> Jungsuk Han: Yes, maybe I can
explain briefly about that. So
if you think about corporations
in traditional sense, they have
leadership such as CEOs and
board members, they make
decisions for the shareholders,
But but in that case,
inevitably, there is a conflict
of interest with one the
leadership of the company and
then the owner of the farm
shareholders. So we call it,
the principal agent problem
um, the leadership is making
decisions on behalf of
shareholders, but then they
might have their own On agenda,
they have they may not have a
line incentives so this is a
critical problem and
traditional organizations.
On the other hand, Tao
decentralized autonomous
organization. They don't have
central leadership at
least in the pure form, there's
a various degrees of Dow,
there's hybrid ones, which
actually has some human
leadership, but in pure formed
now, it doesn't have any central
leadership. It is Organization
based on blockchain.
or a smart contract on
BLOCKCHAINS. So the, the all the
decisions are made by voting
according to the tokens. So
they, it doesn't really require
any human intervention.
Therefore we don't have a
principal agent problem.
So that that is the kind of key
difference between Tao and the
traditional aggregations, of
course. There are many other
differences for example, in case
of Tao the users and then the
the voters.
they are actually the same body
in chemical corporations
consumers, who are using the
services and goods and they
shareholders, these are
different bodies but in case of
Tao, it could be actually the
same same body.
But as I mentioned, the critical
difference between Tao and
Corporation is the absence of
principal agent problem.
>> Tao Li: Yeah, so I think, you
know, basically they were there
were, they were also many other
differences. Some of them are
captured about our model as
well, for instance, you know,
if we talk about traditional
corporations, we often talk
about cash flow rights right
control, you know, institution
investors stuff like that. But
here in terms of the Tao
setting, when, when we talk
about
governance tokens, typically we
value is have any tied to
platform usage or
network effects rather than
traditional cash flow claims,
which is, you know, a very
sentient element in our model
Um, yeah. So and also, I would
say, in terms of the legal
background, or those, that's
not really capturing our model,
because our model is not
focusing on that element. But I
think that's, that's also very
important difference. Is that
corporations typically would
rely on courts on fiduciary
duties. You know,
listing rules, etc, but here in
the dove space and is as this
point, we have very thing legal
infrastructure in place and
often times people are just, you
know, relying on the smart
contract which is which is very
different from the traditional
legal.
>> Fahad Saleh: Michael, do you
want to add a little bit of
context?
>> Michael Zargham: yeah, I
would say one thing I've learned
from working with and in a bunch
of Dows is that it's worth
noting that the principal agent
problems in practice
Don't totally go away. They show
up elsewhere, um, the natural
way this happens is that
because you have a primarily, a
smart contract, as a locus of
coordination, where there's some
voting that body that ends up
making voting decisions, isn't
really operational. It can't
really do anything. And so it
ends
up having to whether it's field
proposals, or procure services
or decide, which members of the
community are going to be
empowered compensated or allowed
to do certain things.
Ultimately, you end up having
the organization itself as
principle and some agent of that
organization acting in
fulfillment of some commitment
that they made through some
voting based interaction. So you
still get a principle agent
relationship and you still get
principal agent problems. It's
worth noting that Main
challenge in Tao's, as I have
seen, it is the lack of clear
separation
between operations and
governance, which is normally
codified via the shareholder,
basically owner or shareholder
kind of senior executive
management. kind of staff, you
know, thin bottleneck there and
so the tao's are releasing that
bottleneck a little bit but that
also creates a cognitive strain
because now you have the
attention cost on governance
decisions over a much. Certain
set of things often a much more
granular set of day-to-day
operational activities, and
then you see failure modes
around the failure to engage the
lack of expertise or
misalignment amongst the voting
population if they are engaged.
And so, I think it's worth
noting that Dows as such are
fulfilling the same function as
firms as kind of sites of
collective action or lokai of
coordination. They have a
different architecture and that
new architecture is more like a
joint constraint satisfaction
problem. We've got a lot of room
to learn how to use computers
to do that effectively, but I
like the distinction that was
made in terms of the the
architecture compared against
that kind of traditional
bottlenecking at senior
management.
>> Theo Beutel: Thank you,
Darwin. That's important to add.
Um, yeah, I guess we can agree
that
the cows or diverse mechanism
that is used righty but coming
back to maybe the
basis of the study I believe
there's a certain type of style
that was the basis. I'm curious
to learn more from the office
about your primary insights. So
so you should be building a
theoretical model of governance
feature, featuring strategic
token trading and the token
based voting and and then you
investigate conflicts which on
face of interest. So could you,
perhaps, share, what is the
most notable insights? You've
found out the course of your
investigation.
>> Jungsuk Han: Yes, maybe I can
share some of those insights
about our model. This is
actually related to the the
earlier question about, what is
Tao? And then I imagine the
that.
That Tao is organization with
our central leadership.
Therefore, it is
absent of principle Asian
conflict, but as Michael
mentioned there, there does
exist some
some could exist principal agent
problem because there's a human
interface but theoretically, if
there is no human interface, if
it's really pure form, we may
not have any human intervention.
Therefore, no, principal agent
problem, but in reality, of
course, there could be some
human human part that they could
easily
create some problems. But let's
just go back to just theoretical
idea that that
always purely smart based smart
contract based organization.
So, in that case, now, the
decisions are all made by voting
and then whoever kind of
captures the voting could
actually
The outcome of all this decision
making. So that's where this
ownership concentration of
token comes into play.
Often, there are large token
holders. What? Which we call
Wales. So if there are
whales who have substantially
large amount of tokens, the
whale can capture the distance
making, and if the whale doesn't
have online incentive with other
users, the whale may
pursue their own private benefit
at the cost of other users. So
that that is the the critical
problem of Tao. Even though that
thou have less of The
principal agent problem from the
traditional corporations, but
now has other issues with this
cons of interest between large
holders and a small holders, of
course, traditional corporations
may have that, but then the Tao
it is now this is actually
critical problem. So what we
study in our theoretical model
is how this kind of situations
could be avoided. Given that
trading mechanism in the market.
So often we think about
quadratic voting mechanism which
can mitigate this
problem. So quite a reporting
mechanism is saying if someone
wants to vote more it becomes
more and more expensive as the
Amount of voting increases so it
could increase quadratically.
so
often this can result the issue
of large holders swaying, all
the outcomes, however, at the
issue with the civil attack. So,
one can make many addresses and
then make votings dispersed
across all this erases. And then
in that case, the
voting cost doesn't increase
quadratically because all the
small addresses will have
rather low voting cost.
So how can we actually go around
this problem? What we show in
our model is actually the
trading mechanism
Could be the solution for this
problem. Namely the token in
liquidity. So when you think
about liquidity or liquidity,
it is liquidity. Usually the
good thing, the definition
literally, when you think about
it is easiness of buying and
selling certain securities in
this case token. So it if
buying and selling tokens are
easy. If That should be good.
however, in terms of governance,
actually, this could be an issue
because if
there are whales who don't have
enough tokens to change the
outcome, but then a close to
it, then wells can easily. buy
more tokens and then change the
outcome. And then they can
they can pass a proposal then
maybe harmful to other users,
but then then may benefit
themselves. So this could be an
issue. But then, as I
said, quarterly voting may not
work here in case that they try
civil, okay? But
then in case this is don't buy,
let's say tokens in which are
trading on the
market. Then the liquidity
becomes like speed bump when
when the the whale wants to
Take buy more tokens as they buy
more and more actually becomes
very expensive. It doesn't
matter whether they have many
addresses or not.
assured amount of tokens. There
they are, buying the market will
push up the price and then
making the
Making the the voting very
expensive for them. Effectively
this becomes quite a bit of
boring. So what we show
theoretically in the model is
that
the token in liquidity is
effectively becoming qualitative
voting and that which may result
all this some of these
governance issue. So that's one
of our the
Theoretical results. And then we
actually have other results
about the low-ton commitment
and so on. But then maybe I'll
just, you know,
>> Fahad Saleh: Yeah.
>> Jungsuk Han: start here in
and listen to others. Yeah.
>> Fahad Saleh: so I want to
unpack that a little bit and
then maybe you can bring a few
more people in. So
you sort of talked about two
things there, one was quadratic
voting and one was illiquidity
and you mentioned that the
liquidity could look like
quadratic voting and so maybe.
But I want to put quadratic
voting for the side for a
moment, and talk a little bit
about the liquidity because I
guess one way to think about
what you're saying is that if
I'm already a large holder, if
I'm already a whale, I could buy
a little bit more in order to
become pivotal in a particular
vote. And if it's if the market
is very liquid, then that may be
very, very easy for me to do and
that's going to cause the
problems that you're
describing. So I'm kind of
wondering though, then and this
this may be even the question
for Michael because maybe is a
bit more in terms of like what
we actually see in practice is,
like, do we see for instance,
people, just buying directly
ahead of votes
Holding that, anybody has any
point in time at some point had
to be purchased, right? And so
really, it seems to me what
matters is kind of like whether
they're doing a significant
purchase at a, at a particular
point in time, right? So for
instance, like, if I'm always
forty percent and then I just
want to become, I don't know,
let's say it's a majority vote
thing and I need 50% or
whatever, right. And then so I
just I just tend to buy whatever
11% or whatever When I right
before a voter something like
that, then then it's more. Then
it's not about like I'm asking
the full 51% position. It's
about amassing, just this
incremental position. So do we
see this behavior in practice
where large players
will suddenly sort of top up
their positions just for the
sake of a vote? I'll just leave
that for the entire room. I
don't know if anybody wants to
pick up on that.
>> Michael Zargham: There are
some cases of where that has
happened. I wouldn't necessarily
say it's a widespread practice
so that would be better
appropriate for an empirical
study. I've had some personal
experiences where I've been
involved in votes where that
very clearly happened one that's
quite old. Now that comes to
mind was an old Aragon vote,
which really looked like it was
going in a particular direction
and the whales dropped in at the
very end and swung the vote many
years ago now. So hard to recall
the specifics. It's an emotional
memory, funny enough, but I
think the point here is probably
that if a vote is important
enough, if it has enough bearing
on the outcome for a particular
whale or class of whales, you
do see some of that behavior. I
think it emerged probably most
strongly in the vote escrow kind
of emissions routing process.
There was another example that
comes to mind where someone kind
of totally captured a stream in
the balance or protocol at one
point. I think there was a
basically deadpool and someone
managed to capture all that
liquidity and then route a ton
of rewards to it. And again,
this is very much governance
attack territory at least in my
opinion though because it
complied, with all the code and
the smart contracts. There was
shark divisions about whether it
was considered legitimate
behavior at all. I think this
kind of kind of, I'm gonna call
it gaming the metrics is always
going to be prevalent and you
see this around other
mechanisms like explicit vote
options as well. So I think one
Property of Tao's empirically.
Is that? There's a
financialization of governance
authority it manifests in a
variety of ways but just because
something's financialized
doesn't mean those methods. Are
always exercised because there's
cognitive costs attention costs.
You actually have to really care
for like minor day-to-day
votes. The likelihood of this
kind of financial activity in
order to swing outcomes by wills
is pretty minor. In fact, half
the time they don't vote at
all. If they are voting, then
they have to have a reason to
really care about the outcome
to put extra again, financial
and operational labor into
making trades and positioning
themselves to improve the
outcome. So I would say like
broadly speaking, the binding
constraint here is actually how
much people care about the
outcome. And so, the thing that
that was often do Is have many,
many votes. And so it's
interesting to look at in which
circumstances what kinds of
activities major changes almost
like constitutional level
parameter changes are probably,
the only thing that are going to
merit that kind of behavior
>> Tao Li: Yes, yes. File, you
know, I I personally want to
second the Michael's points
that oftentimes those you know,
governance attacks, don't
happen. But I think when they
happen the damage could be very
large for instance I guess you
know some of
the people in the room probably
are aware that in 2022 where you
use the flash
loans to buy a tokens that worse
about, you know, one billion
dollars to gain about two
thirds of the voting stake in
Beanstalk, and and then, you
know, this will propose a
malicious prop.
Go to drain about 180 million
dollar worth of CRYPTOS from the
from the asset pool of the
platform. And I don't think this
is an isolated case. There were
they were a couple other cases
even before this one. So
although those events are very
infrequent but when they happen
the damage could be very large.
>> Michael Zargham: So, I think
there's a really important
distinction that we should make
to about
the properties related, to kind
of resistance to attacks versus
properties associated with
governance, as steer like
governance mechanisms as literal
steering. So you have
constellation of good faith
stakeholders with different
interests, who are kind of
advancing positioning using the
mechanisms as they were
intended to advocate for support
their, you know, whipping vote.
So to speak their
communicating, with each other,
to try to get people to support
things or they're spending
money. In order to increase
their influence in order to
drive an outcome, which is
showing skin in the game. Etc,
etc. So like
broadly normatively just um,
separating the behaviors that
are you know again attacks in
the sense that someone wants to
explain the system pull money
out in a way that is not
pursuant to the purpose of the
Tao versus like participation
in the Dow and the exercise of
social and econom. Mechanisms
as they exist to advance the
interest of the particular
stakeholders at hand. And I
think the reason for separating
those is that from a mechanism
design standpoint. I would, I
would author very different
requirements and I
would test against both sets. So
the the tests for the
requirements associated with
steering mechanisms, they look
kind of like signal processing
our market design problems.
You're structuring like Okay,
well you don't know what
people's opinions are, you
shouldn't tell them what they
should be but if mechanisms are
Exercise to express preferences,
then the mechanism aggregates
those preferences into a
decision. According to a
process, that's transparent,
legitimate codified outcome is
deemed the decision of the
group. That's effectively tao as
bureaucracy. However, on the
smart contract is implementation
of the bureaucratic process.
The other side of this though is
that when you move to these
formal kind of code-based
systems, is that you create room
for gaming those mechanisms and
exploiting them against that
interest. And so you can use
different techniques to
discourage prevent or otherwise
kind of reject things that are
determined broadly to be
illegitimate and I'll stop for
the moment but
I have some other work actually
academic paper related to a case
study or we explore how like
one down, differentiated those
two mechanisms,
>> Fahad Saleh: So, if I can
just follow up the, the Michael
on what you were saying, because
so okay, let's see. We're
talking about the first piece
which is steering. I think you
were describing it as the
steering case. It seems to me
like one thing that academics
particularly in finance are very
familiar with going back to the
corporate finance literature.
And so on, is that in some
sense, the, the set of
participants who happen to be.
shareholders to affirm or token
holders to a Tao, is not the
whole universe of players. And
there are other, there are other
players, there's other welfare
to be concerned and considered
and so the classical example,
here is something like
the person in charge of the fur
has some private advantage from
implementing a policy that is
advantageous for that
particular person. But is
actually socially not the
optimal thing. And and so then
Is sort of like broader welfare
implications that are not so
good about that, right? So for
instance, like if you take the
the space of voters as fixed and
you say like Okay what's the
best outcome for all of them,
That's a slightly different
question than saying, Okay, what
is the best social outcome?
Which which really? And I think
this gets a bit more into some
of the work in the paper here,
right? They're thinking very
hard about things like Oh people
can actually
buy tokens in the setting and so
in some sense the set of players
is not like some exogenous fix
set. We have to kind of think
about the fact that there's this
there's this other margin. And
so I just wanted you to reflect
because when you were talking
about the steering mechanism, it
sounded very much. Like if we're
if we're thinking about micro
economics it sounded very much
like a game with a fixed set of
players who have preferences.
And so if you can just maybe
illuminate it a bit because I
think I misunder Yeah.
>> Michael Zargham: yes,
necessarily mean that a fixed
set of players, but to be clear
like, you at any given time,
when there is a vote, there is a
>> Fahad Saleh: Yeah.
>> Michael Zargham: collection
of votes and they represent the
interest. So again, you know,
I'm a control systems engineer,
my PhD is in distributed
optimization. I'm comfortable
with open systems and evolving
sets of a members and games that
are dynamic games games that are
in an open world setting, I'm
actually not
>> Fahad Saleh: Yeah.
>> Michael Zargham: a very big
fan of economic game theory to
me at forked off of the kind of
aerospace and defense variance
of game theory. and like, the
50s and 60s and went very
academic. Whereas the, you know,
the other stuff being arrow in
defense, when very dynamic into
your point, involves
open environments and
interacting with complex change.
Sorry, engine point being
though, I'm not assuming that
this is a fixed set or like a
canonical microeconomic game
structure. I'm actually saying
like there's information in the
world about preferences.
That those that information is
incident on this system through
these voting mechanisms and the
to the point of splitting the
problem apart, it's the same
mechanism. But you need both
like effectiveness properties
which are related to whether it
channels the information
incident on it into decisions
that are, you know,
considered effective and
legitimate versus the more
defensive side, which is
supposed and adversarial actor,
high tries to Jack those same
mechanisms to say drain a
treasury. It's not a different
mechanism. It's different
requirement. So if the author
formal requirements and write
property type proofs about those
things or at least demonstrate
through simulation scenario
tests both of the form, you
know, use as intended
essentially no adversarial
actor. Just can it process the
information incident on it into
decision? And does that make
sense? And when an actor who is
adversarial shows up and tries
to hijack, that how resistant is
it to that
hijacking and that can include
things like the the cost of
achieving a certain
amount of power. So if 51%
attack or essentially there's a
variety of different techniques
you can use, but ultimately
you're you're writing properties
that
say hey How hard, how expensive,
how much time? How much money
does it take to induce an
adversarial outcome? And that is
something that will vary by the
but also by the liquidity, by
the the markets by the
distribution of the Holdership
and so you don't necessarily
get like an a priori result, you
might get a functional result,
that tells you
quote, how hard it is. Given the
current, you know, token
distribution and the current
market prices and estimates of
the price impact of trades.
>> Jongsub Lee: Let me add a
little bit about going
discussion because just for
clarification in our paper, we
just try to highlight potential
conflict of interest between two
you know, types of token
holders, right? The big whale
and then small users.
but you know, we just want us to
start the discussion start
discussing about, you know, how
they're conflict of interest
affects in a tao's in
resilience, but in fact, we are
discussion is more toward, you
know, past the optimal form of
ownership distribution, right?
So maybe just one big whale
versus small, you know, users,
you know, in that case, maybe
it's a dictatorship or
autocracy. that's definitely
the bad in a situation. But also
at the same time in a complete
democracy like a small users
then they don't participate.
Maybe they're too small. So
they don't have any strong
leadership to lead the Tao so
that case may not be ideal
either. So I think that you
actually, you know, is some
optimal allocation of the power
balance, you know, kind of power
structure among, you know,
several whales. So then maybe
some, you know, pluralism time,
you know, situation, three or
four wells. They actually check
in balance between each other
so that, you know,
One will cannot control the
entire houses but this actually,
you know, a couple of numbers
of whales. They are they have a
strong participation incentives
in the voting mechanism and
then, you know, discussions. So
that may be, you know, in
reality it could be a better
form of Tao, power balance, or
power structure
but our paper for clarification
is not, it's not touching on
this, you know, more advanced
issue but we just want to
highlight you know, when a
single whale actually can
easily accumulate the voting
power and the steer that thou,
you know, in a way to
personally benefit, not the
benefiting for the other users
then what time of ages problem,
you know, they could arise and
how to mitigate such problem
through the, you know, kind of a
kinds and design type in a
contracting, You know,
solutions. So, definitely I
think the
>> Michael Zargham: The.
>> Jongsub Lee: we need to
actually make more deeper
Deeper discussion about what
could be the better power
balance within the Tao system,
you know, but that kind of in
the research is actually ongoing
and then we may actually have
to answer those questions that
don't
>> Michael Zargham: I think the
one thing I'm trying to do
though is highlight that you
actually can't carve out.
one problem when you're doing
mechanism design in a
particularly in a Dow setting
where those mechanisms are
effectively serving multiple
functions or have multiple
purposes and and potential
threats at the same time. So
like when we talk about the the
category of let's call it
requirement, which
is like resistant to exploit by
a single actor with a whale
position that is a requirement
and it's one I totally respect
and studying it deeply and
understanding like how bad it
can get and or what mechanisms
or approaches to mechanism
design mitigated. It's super
valuable, but it's actually
can't be decoupled from the
other questions that I've been
surfacing. So as the industry
voice, you know, and I talk a
little bit. I'm trying to
highlight how this relates to
practical design decisions and
so the way that I'm framing it,
at least is that when you look
at a particular mechanism,
that's gonna fill this slot It
needs to be resistant to several
different kinds of attacks. It
needs to be
useful for at least one and
often multiple kind of purposes
and again, I'm highlighting the
kind of big governance change
type purposes which are like,
Hey, can we make major parameter
changes that are viewed as the
kind of the fundamentals to the
Dow and then they're also used
often for more micro more
Operational decisions. Great
attention issues. So you have to
worry about
requirements around attention
and interest, relevance the
effectiveness of the changes
visa to be the the tao's, you
know, ability to perform its
purpose. Pay for things,
whatever it's doing various bite
down and you'll also have to
worry about the various
dimensions of exploit attack
conflict of interest etc. But
since those are all about the
same mechanism, you have to be
able to say, all right, Here's
a category of mechanisms. Here's
the circumstances, it's being
deployed into and from each of
these angles do. Assessment. So
it's a split from like a depth
for his assessment of a
particular, kind of attack, or a
can take particular kind of I'll
call it safety or resilience
property that we want versus the
breadth first, which is
for any given mechanism. It has
to satisfy that and and and and
other things if you go too deep
on one thing, you end up with a
mechanism that is like, maybe
really
good against that, but it's
effectively over-optimized and
it can't perform or Isn't
resilient relative to the other
requirements and we talked
briefly in the preamble about,
you know, quadratic voting. And
I like I point out often that
it's not a great use case, for
student anonymous environments.
It's hyper specialized to
reduce the effects of
to reduce the effects of having
high amounts of money. But it
sacrifices it's effectiveness in
a pseudonymous environment. And
in fact, in it, kind of
exacerbates, the effect of
social media influencers,
because social media
influencers, influence a large
number of individuals. And even
if they are discreet
individuals, it magnifies the
effects and you see that in some
of the get coin stuff. And so
the point I'm getting at
here is that like any particular
property assessment has to be
placed alongside
the other. Let's say three to
five core properties and
assessed in parallel. And so
the mechanism design problem is
making the best across all of
those rather than necessarily
being amazing at anyone and
like, trade-offs, always end up
showing up. And so again, it's
not a particular criticism. I
appreciated the paper I
just I was trying to kind of
provide the how do you use this
information in practice? And
the answer is you place it
alongside of your other
requirements. you do some
trade-off studies. And you you
do the best you can give in the
situation you're in.
>> Fahad Saleh: Let me just
maybe add some context before we
transition on, we might be
getting a bit too much in the
weeds of methodology here,
actually. So okay, one thing to
note though is that, when you're
thinking of an optimal
mechanism, you do need kind of
a well-posed objective. So I
think economists tend to think
about particular objectives,
like optimal welfare, for
instance, and by the to be
clear, something like optimal
welfare can capture multiple
properties. Like, for
instance, security is not
irrelevant to optimal welfare
considerations. and so Michael.
When you're talking about
different properties, I think a
lot of times one of the
difficulties like economists
have when they talk across to
engineers is that sometimes,
maybe not. I shouldn't put all
engineers one category, but for
instance, I know in computer
science, there's many times
particular properties that you
want your, let's say, stepping
outside the Tao context. Let's
say you want your consensus
protocol to satis, I want to
satisfy safety. I wanted to
satisfy liveness. I may want to
satisfy a bunch of other things,
right? But of coming up with an
optimal. ISM requires kind of a
single metric and and you could
have a metric that for
instance, weights, you very
simple. One would be like, I, I
have 10 properties and I just
count the number of properties
my mechanism, satisfies and, and
that's essentially how I value
the mechanism. Now, in that
case, I may have multiple
mechanisms of being optimal or
something like that. But, but
the point is that the idea
Having different properties that
you're that, you're interested
in to satisfy is not exclusive
to the idea of having a single
objective function because you
can objective function that
essentially collects these, they
might have to might have to
wait it or or do something like
that but and I think, yeah.
Yeah.
>> Michael Zargham: You it's a
pretty basic thing, right? It
comes to your point. It comes
up in engineering traditional
engineering all the time, right?
You have a bunch of properties.
You have things that are hard,
requirements that are
essentially booleans like, Did
you, or did you not not satisfy
this, These are constraint set,
then you have objectives. But if
you have a multi criteria
objective
>> Fahad Saleh: Yes.
>> Michael Zargham: problem,
then you have to have an
aggregator function, you can do
like min. we're all style. You
can do some you can do a true,
your choice of aggregator, then
given that you can still apply
weight. So I like things like
geometric programs, where
they're multiplicative.
like, that's because you can be
like, softmax and stuff with
that, and they're still like
quasi, convex, you can still
solve them up like with
computers, but the gist of it is
that you ultimately need to
construct something that
produces the Pareto surface
where although you have weights
or parameters, a way of
selecting, the relative
importance of your many
objectives, Once you actually
start doing that, you can reveal
Pareto surfaces, you get a
sense of what the optimal
surfaces and then your trade
study is making decisions about
like where on that parade of
surface, you want to be the
thing that drives me crazy is
when people pick things that
aren't on the Pareto surface,
and I feel like there's a lot of
that.
>> Fahad Saleh: Okay, well, I I
think we're conceptually
agreeing here. One of the
values, by the way of having a
practitioner in on the
conversation is precisely. It
can help the academics.
Understand what is more
practically important and what
should be considered that maybe
is not but but I think that's a
longer conversation that we can
have afterwards offline. Theo,
I think you wanted to bring in a
related topic here, right?
>> Theo Beutel: Yeah. Thanks
actually young so bad as the
earlier mentioned, the whales,
I think it's worth about
discussing allegation as well,
so many that was in
practice. Use delegation
mechanisms, which I suppose de
facto create at the sort of
sorts of whales who have voting
power, like other wells, but
actually do not have the
control, including the financial
group of trading these tokens.
So to what degree do you think
you're
results from the theoretical
model? Are transferable to tao's
practice, that use dedication
models
>> Tao Li: So in a few, yes. So
basically delegation does reduce
the ability to trade a gut, you
know, around the governance
votes, right? So basically
you know it is harder for the
for the delegates to buy votes
etc. But still I think some of
the economic logic does apply
you know, some of the economic
logic from our model
still applies. Because you know,
even without
transferable, voting power knows
those dedicates can create
persistent concentration. And
if, if the, the decades have
different expectations or
different preferences about
certain governance, outcomes
versus others as a users. They
actually could still, you
know, steer, you know, voting
outcomes in the way that's
beneficial to the 2D dedicates
versus as a, as a users. So, you
know, you would still have this
Find incentives between those
dedicated so-called dedicated,
whales and users. So, yeah, I
mean, you know,
so however I think I think there
were other dimensions that that
are not available in our current
model, is that for dedicates you
could you know, make things more
transparent because
a lot of those dedicated to care
about their reputation, right?
So basically you could you could
make the model involving the
negation more transparent than
what we have in our current
setting. Yeah.
>> Theo Beutel: Organization.
Which is yes. Anyone want to add
to that.
>> Jongsub Lee: You also.
like the allegations of actually
one way we fixed the
participation problem, right?
Because in our model, you know
there is no, you know, such
thing like a Directly modeling,
you know, some kind of lack of
attention or lack of
participation. Type issue, we
actually encounter in the real
world, the Tao's.
So the allegation on behalf of
the other is actually the
delegating party in a
connectively participating in
the voting process is so that,
you know, they can overcome
this, you know, the real world
problem.
So. But but at the same time in
a given, the participation, you
know, one value eight, you
know, delegating party, you
know, can have conflict conflict
of
interest from the others maybe
against some you know the vast
majority of the users then the
same problem. Like we, you know,
describe in this paper like a
big steak, you know, older
verses, you know, you know, many
many diffuse, you know, small
holders, you know, or some other
delegates in who try to
represent them actually, there
could be conflict of interest.
So how to actually intervene
and then try to actually restore
the resilience of these Tao
system? You know against such,
you know power you know that
power imbalanced gain, you know
so that actually mimics an hour
our model in that part actually
makes the real
world situation, I think. But in
reality there is actually more
complex layer or problems
including the participation
issues and so, the conflict of
interest issue, we Describe in
detail in our paper. So I think
that we need to actually you
know
learn step by step. You know
those you know vast majority of
variety of issues, you know as
Michael said there are so many
different issues in the Tao
space so the one that kind of
design cannot, you know, fix
them. All one size doesn't
feel all, you know. So you may
need to prioritize list all
kinds of problems agency issues
in that specific space and then
try all other than in terms of
the product, in US,
significance of each problem.
And then, trying to think about,
you know, What's the
multi-dimensional mechanism
design to fix those problems to
a certain extent? We are paper,
is just a starting point of the
One of the biggest in an agency
issue, maybe conflict of
interest issue, you among token
orders even in the absence of
central literacy, you know, who
might
behave on behalf of himself or
herself, there could be some
different types of agency
problems in this Tao space,
which is actually Dalia Canal
acknowledge by Michael in the
beginning. Right? There are
still agency problem in this
business so I think our paper
is kind of starting point to
describe such issue but there
are a lot more, you know, going
on, you know, how to make them.
you know, having strong
participation incentives and at
the same, how the coordinate
them in the desirable way. So I
think that those are the
important issues we need to
solve, you know, academically
and also practically, I guess.
>> Michael Zargham: So I think
practically the way I would
imagine using your work is to
use it to develop both
requirements descriptions of
those conflict problems so that
they could become either
constrained or objectives for
our design discussion earlier,
right? So, you can assess
whether a particular design is
likely to produce or the
circumstances under, which it's
likely to produce that class of
problem and you could also use
it more for pools and empirical
tools for trying to measure our
identify signatures for our
predicates, for those problems.
And that moves it away from
being a mechanism design, focus,
because I think the mechanism
design
>> Tao Li: Yeah.
>> Michael Zargham: problem
itself ends up being well,
Frankly, I tell most people not
to try to design you voting
mechanisms, it's kind of like
crypto, don't roll your own, the
cryptography, don't really your
own. Use a tide tried and tested
mechanism for which, you know,
the properties and for which the
properties are appropriate for
the situation, that doesn't mean
no one should be researching and
developing new mechanisms,
Please continue to do that.
That like, in practice using a
new.
a new governance mechanism is
fairly a good idea. And that
the, the key is that the
research that is into governance
mechanisms is often most
fruitful in helping us describe
the desirable properties of a
design, and to assess the
current state of the living
system, and that as those
things, mature, it does end up
leading to situations where
those new mechanisms do get
tried out. We how they work in
practice and then we get a sense
of when and where, and how to
use them. And, you know, I
realize that maybe a little bit.
unexciting, But like, there are
reasonably good libraries of
mechanisms, I think Ori Shimoni
put together a great like,
library of mechanisms. I'm sure
Tao can
share a link and that when it
comes to the academic research
around algorithms and
mechanisms the first class
learnings that make their way
into industry for, you know,
seasoned professionals are how
to write down the requirements.
So you did tell certain kinds
of attacks. You detail certain
kinds of failures you write
proofs and do analysis of those
cases and then I write down in
my mechanism design work. All
right, here's another thing. I
want to check for, in the list
of things, I want to check for.
And if I'm doing an empirical
study of an actual Tao in life,
that I'm looking for whether or
not, it appears to be an
issue again deriving, metrics
from that research, as well as
in a more advanced case looking
for predicates. And I think
predicates are important because
you're looking for the prior
states that represent the risk.
For the potential for that
failure mode. Because most of
the time, once there's a
failure, it's too late and so
understanding how to detect
predicates and intervene in
advance is a major element of
kind of organizational or
institutional resilience and
something that I think still
need to develop.
>> Jongsub Lee: I was actually.
>> Theo Beutel: so, Sorry.
>> Jongsub Lee: regarding that.
I want to add this one thing
because maybe Jungsuk, you know,
can add more, you are surprised.
When you actually found in
our paper such that the targeted
illiquity like a simple locking
period for the important whales
in the Tao system. They turned
out to be quite powerful, you
know, governance device. So
maybe we don't have to think
about the complex solutions.
Maybe we can just simply say,
Hey, you have a large stakes,
Then you should show the
Interest in this Tao system over
the long run than just signal,
you're in a good intention
through the long locking period.
And when I talk about this
locking periods, as a solution,
even in reality in the venture
capital space, you know, that's
an outside our, but, you know,
When there is a significant
information, asymmetry problem
and potential, you know more
hazard and adversial action, the
test simple role and that simple
solution. Actually restored a
trust, you know, between you
know, many, many different
parties with the potential
conflict of interest and
importantly in our paper, we
show that such idea like a
lacking period as a targeted in
liquidity. For certain groups of
whales, that could be a
powerful, you know,
device to solve this conflict of
interest problem. So maybe
expanding such idea. Simple
yet, you know powerfully
economics, you know and apply
them to the real world Tao, you
know maybe could be away you
know until that academic
findings and practical
solutions can meet together at
certain point. So maybe jungsuki
you could add because you know,
we thought about this issue for
a long time, right?
>> Jungsuk Han: Yeah, I was
about to actually make a comment
about that and thanks for
inviting me to if additional
comments so
what we're doing is not so much
about how it should be, it's
more about how it is. So we are
not really designing any
mechanism
there, but we are doing is we're
just trying to explain how it is
now and then it may actually
work as it is. So
At the moment. If you know, the
many tao's or adopting token
based system and then it was
tokens are traded on the market.
So we're not saying it, they
have to design this and their
way, but then if those tokens
are trading on the market it
just happens that those
illiquity is on
their side, especially when the
governance is contested where
whales are.
trying to capture the, the
outcome of the the protocol. So,
so in that case, indeed we
empirically find that that when,
when the governor's contested,
the, the Tokeny
liquidity, he's actually a
boosting the, the growth of the
platform. So we
actually verify that
theoretically and empirically
and as jongsub just mentioned.
A lock-in period of tokens.
this is actually doing this kind
of same mechanism as the
token illegally in a way because
those whales who are locked in,
it is as if having sort of a
greater eliquity only design for
them. so,
what was happening there is they
there are they are so they're
there, incentive is more aligned
with users because of their
local commitment. In that case
in. And then this is what we
find in the model and then
indeed in our empirical result,
we all so find the result that
imposing more long-term
commitment in the form of
locking and other.
Are the forms actually boost the
platform growth. So so yeah.
In a way kind of this is what we
find both the theoretically and
empirically and we're
explaining how it is rather than
where, you know, where it should
be like that.
>> Michael Zargham: Have.
>> Jungsuk Han: So, so maybe
there should be that better.
>> Theo Beutel: Actually.
>> Jungsuk Han: than, at least
that's what we find. Yeah.
>> Michael Zargham: Yeah. So
>> Theo Beutel: In here.
>> Michael Zargham: Cows that
you need to empirically. So,
like just essentially auto
ethnographically like balancing
out because there's in my
perspective you've got like,
theoretical kind of mathematical
level analysis. You have
empirical at the level of data
collection and then I always try
angulate with kind of
ethnographic methods. I work
with researchers and RMIT in
Australia, Ellie Rennie, Kelsey
Navin, and Ellie's other
students. And I found that
without that triangulation,
it's very hard to like really
make sense of one of these socio
technical environments. And so,
the question is really like, Did
anyone participate in any of the
Dows that you studied and how
does your like experience and
participation line up as the
third leg of that stool? With
the mathematical and empirical
results that you found
>> Theo Beutel: Well actually
Saga, maybe let me give us a
question back to you because
this is a really good one and
it lines well with something
else managing up. So the data
collection for the study ended
in 2022, right? So and the Tao
space is evolving fast. So
maybe Zargham could you
briefly describe some major
changes you observing over the
course of that period? Because,
of course, yeah, no space has
changed and go to education has
>> Michael Zargham: Man.
>> Theo Beutel: and use the
>> Michael Zargham: Yeah, I mean
the thing that
I've noticed in most recently
that is like changing things is
that everyone wants to involve
Llms and
Get. Actually, There was a
little bit of discussion of that
in the paper. I linked, which
we can probably share in the
notes around attention economies
This is related to this problem
of who makes decisions, but I
yeah, I mean, there's probably
nothing more impactful in my
lived experience, since we're
kind of bridging off of that,
then the sudden like desire for
people to kind of assess and
participate, either individually
through the use of just like
talking to chat, GPT or Claude
or Gemini, or whatever people's
preferred models are to the
development and deployment of
systems that are kind of rag
based and our have access to
let's say forums and or you know
pulling on chain state that's a
little bit more advanced, I
don't see as much of that. But
like that the LLM craze has
like really captured the
imagination, no doubt
participants. And I see it,
particularly in the face of the
binding attention. Constraint,
the cognitive constraint, and
then there's also the fact that
people are increasingly building
tool, that actually can
directly interact with smart
contracts and for folks who are
interested in principal agent
problems, I actually went on a
different podcast and AI one.
You, I don't know some months
ago and talked a bit about
Principal agent problems in AI
systems where you really have to
think about whether the AI
agent is, in fact, your agent or
someone else's in terms of like,
who's interested advancing. And
although that wasn't a crypto
focused conversation it was
informed by my experiences,
endows and with crypto.
community members, you know,
both using tools that have been
provided for them in service of
their
financial or governance
interests. As well as the
observation that the
exploitative actors are
increasingly automated actors
who are programmed to just hunt
for opportunities to arbitrage.
And I think that's actually an
interesting factor to your
paper about like these kinds of
a lot misalignments and exploits
because Present, there's
nothing really to stop or to
constrain a AI based agent from
kind of entering a Dow
governance environment and
exercising automated market
makers, and then voting power to
say it steal funds. And you
know, that's a little bit hazy.
But I would say there was a
recent study I saw apologies, I
don't have it prepared to quote
about the the increase in the
this kind of exploit on an
automated basis. I'll see if I
can find it and share it. if
you want to add it to the notes,
>> Jongsub Lee: Actually
interest.
>> Theo Beutel: Preventable.
>> Jongsub Lee: in you. One of
my students who just joined the
you know Colombia Ior
Department. Actually, he wrote
paper with the Agostino kaponi.
About you know, what? If the
agency AI participates in the
Tao voting process in
How does the AI in Agency, AI
you know, how does it do
compared to the human? You
know, based building and I think
for sure they will be the
growing
participation by Asian taois in
this voting process. So there
could be another in a conflict
of interest.
Between human controller versus
agent AI, or even a cross-agent
AIS. So I I fully agree with,
you know, Michael's war is
actually this in a Tao spaces,
getting more complicated, you
know, by emerging it with, you
know, AI, you know, the best
world. So now how to restore the
resilience of this system, you
know what might be the, you
know, everlasting happiness
mechanisms. I think that, that
will become more and more
important I guess in this
complex work, Yeah.
>> Theo Beutel: The forecast
we're coming at time. I think
this is a really great way of
signaling to both builders and
researchers that there's a lot
more to build to experiment on
around our governance and and
surprisingly little best
practices
perhaps, but also to researchers
to take a much closer look also
interdisciplinary research
including of ethnographic
studies, as Arkham said to make
sense of this. Very exciting
field of government that is
still a merchant.
So I would like to thank all the
authors that you can tell Lee
and Michael
Zargham for joining, thank you
Fahad for for hosting the
session. And please share the
link on YouTube with everybody
who may be interested and thank
you everybody for joining
today.
>> Tao Li: Thank you. Thank
you. Bye.
>> Michael Zargham: Thanks. And
thanks for your work, guys.
>> Jongsub Lee: You guys.
>> Shyam Sridhar: Okay, the
public stream has stopped. So
From Whiteboard to Mainnet is a new podcast series co-hosted by Columbia University’s Center for Digital Finance and Technologies (CDFT) and the Ethereum Foundation’s Academic Secretariat. The series brings together academics and Ethereum ecosystem contributors for focused, candid conversations. Each episode explores a specific topic or open problem, pairing leading researchers from academia with contributors from the Ethereum Foundation to share and compare perspectives. The fourth episode will be co-hosted by Fahad Saleh (Columbia University/University of Florida), and Theo Beutel (Ethereum Foundation), and will feature Jungsuk Han (Seoul National University), Jongsub Lee (Seoul National University; University of Florida), Tao Li (University of Florida) , and Michael Zargham (BlockScience) in a discussion around Governance Structures of Decentralized Autonomous Organisations. The session will begin with a discussion of the key insights from Jungsuk, Jongsub, and Tao Li's work, followed by a broader conversation on how various DAOs that exist in practice have evolved in the way key decisions are made. It will conclude with a discussion around promising directions for future work.