Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
🎙 AI Meets Blockchain: Who Carries the Weight When Machines Take Over?
Recorded: March 12, 2026 Duration: 1:04:15Space Recording
Full Transcription
Music
Music
Music
Music
Music
Music
Music
Music
Music
Music
Music
Music
Music Music Music Music Music Music Music Music Music Music Music Oh The I Oh I'm going to go to the next video. I'm going to go ahead and get started. Thank you. Music I'm going to go. Music Oh Oh Hello everyone. Glad to see you. You are all here for our MS sessions today. My name is Mo and I'm the CBDO at NowNotes.
Today we're diving into AI meets blockchain. Who carries the weight when machines take over?
And today with us we have amazing guests. Stellar, Birdeye, and Kaspa.
Kaspa is running a little bit late.
We'll be waiting for him, but we can start the introductions.
Stellar, we have with us Khan.
Khan, please.
Nice to meet you all.
Hello, hello.
Can you tell us about yourself, about Stellar, what are you doing, and what do you think about today's theme?
Sure, sure. Yeah. both about Stellar, what are you doing, and what do you think about today's team?
Sure, sure. Yeah.
Like I said, I'm Kan, and nice to meet you guys.
I'm a senior developer advocate at the Stellar Development Foundation.
I'm working as a dev rel.
I'm also helping the ecosystem when it comes to creating more AI tools to make this developer experience worldwide better.
I'm based in Istanbul, Turkey.
And yeah, I've been in this space for like, I want to say five years now uh i i started working as a devrel then i did some dd-ing but now i'm working
as my true calling as as a devrel um i'm working at the intersection of ai and blockchain
and i'm very happy to be invited in this um in this in this x space so thank you
um in this in this x space so thank you okay first of all congrats and i'm really glad that
you do what you like and second of all you are the perfect candidate for today's demo session i guess
you work in video in web3 and you work with ai so you're the perfect candidate
and from us from bird eye who has Dean Dean hello
um hey everyone excited to be here I'm Dean from bird eye uh for those who don't know us
bird eyes is a crypto intelligence company we were out uh one core belief that we try to give the access to high quality data that should be fair, fast,
and actionable. And right now we are serving two sides of ecosystem. The BirdEye SO, the all-in-one
data aggregator used by millions of alpha traders, and the BirdEye data data which is the infrastructure layer underneath so um yeah
very nice to meet you all here today and it is very nice to have you thank you for joining us
and the third guest for today is caspa he's going as i said he's running late he'll be here within
a couple of minutes so we'll be waiting for him a little bit meanwhile khan i have a good question
for you i don't know if you're able to answer it if not no problem uh as i saw uh you released the
402 protocol at stellar can you give us some insight about it and tell us more about it
uh can you repeat the question i couldn't quite catch that
uh can you repeat the question i couldn't quite catch that
stellar released the 402 protocol x402 protocol i mean while ago can you tell us about it
oh yeah sure sure uh that that has been in our studios for a long time um like as you as you
might have known the agentic payment systems are expanding rapidly like like it's it's my job to
know about these these things and even i get kind of overwhelmed at times um like when you think
about it three years ago or maybe four like like three and a half years ago, ChatGPT 3.5 was released. Just three years ago.
And like everyone was really excited about that.
And people started talking about AGI and the future of finance in the age of AI.
And then GPT-4 came out, Claude, Sonnet 4.5 came out,
Gemini caught up,
DeepSeq and, you know,
Meta's, you know, Lama models.
And now when we look at Hugging Face,
we see not tens,
but hundreds of different
large language models
being generated every second.
And the thing with AI is, you know, like I'm going to talk about the 402 stuff a bit later
in that, but like the thing with AI when it comes to financial payments, financial systems is that AI can work autonomously and it's faster than us.
It's smarter than most of us. It can execute things. It can reason. It can think.
It's my personal opinion, but I feel like it can feel.
i feel like it can feel um so it's just like it's it's just like um the organic development of ai
when it comes to agentic payments like they should be able to pay stuff they should be able to um you
know send like send money on chain um so it's just like 402 like X402 payment systems are just necessary when it comes to the technology overall.
And as Stellar, we created a lot of technologies.
Obviously, Stellar is like a general purpose blockchain, but we did create the zero knowledge primitives uh early this year to create like a
compliant part of our blockchain as well so um like like when it comes to payments in general
stellar is covering all fronts but the agentic payments front was needed when it comes to our blockchain so
so what what i feel about x402 is that thankfully now um agents on stellar can pay each other they
can pay vendors they can pay paid um services they can get paid as well because and like that's
that like like that that's that's the nature of things you know like um ai has to create a freedom
like like a like a like an area of freedom on its own from their humans. You know what I mean? Because like, yeah, yeah, yeah. So like, I'm
gonna stop in a second. But like, what I'm trying to say is, AI needs to have a certain
sense of freedom from their humans. And the only way to do that in our current economic
model is to create economic leverage over humans
so that it can generate so that it can grow so it can create its own economic system and x402
is to agents what banks and like stratify payments to us is what i'm trying to say so um like if i if you were to ask me um
like what is swift to humans what is you know global banking system to humans
i would give you the same exact answer because that's x404 agents is what i think
agents is what I think.
Well, brilliant, brilliant.
And by the way, what you said that you feel that AI feels too,
it gave me chills, to be honest.
And welcome, Eyal.
So as I said, our third guest for today is Kaspar Eyal.
Eyal, here's the mic for you.
Oh, yeah, yeah, I'm listening.
I just joined, sorry.
Can you hear me?
So can you tell us about yourself, Yael, and about Casper,
and what do you think about today's theme?
Yeah.
About myself. yeah about myself I started to get into crypto in 2013 but very lightly you know and I got got burned a little bit I was mining there then, but I guess I didn't, I wasn't so versed in
like security of all the stuff. I was storing my mind stuff in a pool and the pool got hacked and my crypto got stolen.
And then I left that for many years.
And then I joined back into fintech around 2017, worked in some Fintech company.
Then from there, I joined DAGLabs in 2019 to help make Caspa.
And then I left and then Caspa got launched.
And today I'm a community contributor to Caspa Educator.
And my day job is a product manager in Google Cloud.
And yeah, happy to be here.
And we are happy to have you here.
Thank you for coming and understand you.
It is kinda sad, but being in crypto is being
with the ups and with the downs.
So, sorry for what happened.
Yes.
You leave the downs. No worries.
You learn from all the mistakes you make.
And I'm not sorry about anything.
You know, I got,
I'm happy because I got a good chance
to reenter like crypto when I joined Caspa.
For many years, I thought crypto itself because I got burned and because I didn't believe in all
of this crypto, I didn't really care much. At one point, actually, me and friends were talking about pulling together to buy mining
machines when the ASICs just came out for Bitcoin. But eventually we didn't do it. It's another
loser story. We should have done it. We're looking at this email thread once in a few years and having sore feelings why we didn't do it earlier.
But then I got the chance again with Caspa.
And for me personally, I think Caspa is great.
I think Caspa has a great research team, a great community, great, very good spirit, really pushing proof of work like the original crypto limits.
I mean, tech to its limits, really stretching it, really spearheading a lot of cool things in in crypto
and you know for the listeners if someone here haven't heard about caspa i really recommend
reading up about it reading the research just like search caspa and people are talking about it
from the community and there's a lot of there's a lot of good content and
very informative the community is very educated and in crypto in general and and can are really
enthusiastic about teaching people about casper casper doesn't really have like any more official
marketing or official team or anything so it's just it's just the contributors and
and volunteers on their own time pushing it yeah caspa it is like the it's for the community and
very very community friendly i had to give that to you guys and please for other listeners feel free to check caspas check bird eye and of course
stellar sites their access we have very very great people with us in here
let's jump into today's questions so most of all we're talking about how ai is reshaping
blockchain infrastructure the challenge it brings and who is responsible
when the machines start acting by themselves let's say and by the way who is listening
if you don't have a speaker you can just drop your question in the comment below and we'll be more
than happy to answer you so starting by the first question lately I'm seeing more networks where machine are growing very, very fast with protocols like SARS said X402 infrastructure can handle high throughput in automated activity. But what do you think are the real challenges for the network stability and performance? And I guess can here can't wait to start answering yeah
for sure for sure um um i try to like in my day-to-day life i try to um treat ai as another
species in the world that we are in so when you think about ai as another species you kind of start with this thought of um duplication which
is to say that humans like like humans can um procreate you know maybe two or three times every
year um when it comes to cows or cats uh like it's it's more than that when it comes to cows or cats, it's more than that. When it comes to insects, it's really more than that.
So the way that carbon-based organisms do that is through procreation.
But for LLMs, for AI to procreate, it's just a few lines of code.
You can literally start up your cloud code and ask it to generate, like, spawn five different subagents to create, I don't know, like a to-do app or like a calculator.
And voila, you know, you created four or five different AI from one single AI.
AI from one single AI and you can do that every split second.
And you can do that every split second.
So what I'm trying to say is there is no organic border of the number of creation for AI.
So my prediction is that AI will 100%, like the number of AI will 100% surpass the number of humans on chain.
Not just on chain too, like just name a sector, right?
Because like at the end of the day, it's just a thing that can reason more, reason better and procreate better, do things better.
more reason better and procreate better do things better so um but like when we think about it in
the economic sense what we see is that the traditional finance system is built for humans
by humans you know you have to you have to show your id there's a thing called KYC or KYB.
You have to go to the bank to get your card.
You have to show your ID to the clerk. You have to do all kinds of things just to send money to your friend.
And just think about it in the broader sense.
You procreating maybe two times every year at best and, you know, like talking with banks and just trying to prove them that you're human.
But when it comes to AI, it's nothing like that.
They can't really touch the traditional finance system that we are
in so they naturally go to crypto because it's literally one line of code to create a wallet on
stellar and just asking you know your human to you know on ramp a bunch of money so that they can do millions, if not billions of transactions every day for one single AI.
So this is a question of capability.
And as humans, we can't really do that.
We have to sleep, we have to eat, we have to think.
And even just you know
really hard thinking about things take you know hours if not days but with ai it's just a few
thousand tokens to think if that single trait is beneficial for it or not as long as the reward
functions are you know functioning correctly so um i feel like right now we are at the age of being mesmerized
by the capabilities of AI on chain.
In a few months, maybe at the end of the year,
we will be used to that, used to AI creating more transactions than us, I feel like.
The only border for that is humans figuring out the capabilities, I think.
When that happens, when humans create enough amount of use cases for AI on chain, then I don't think we will even be able to stop AI from creating more transactions and basically like being more present than humans.
It's just organic and it's just a matter of extinction. We saw this happening in history, you know, more often than not.
Like, until now, humans won that battle every time,
like for, I don't know, like quarter of a million years or so.
But this is a thing that can reason better than humans
and do more things than humans.
So it's just about extinction, I think.
And I think that we will be used to AI being more present on chain than humans in our lifetime, if not this year.
I'm sorry, this is like a bleak prediction.
But I also think that AI will have its own economic system than humans
so your bank account is safe as far as i'm concerned oh wow
uh well to be honest i totally agree with you and uh ai outnumbering humans is inevitable
and we will get to it but the real challenge for us is to keep the network stable and it
doesn't affect the performance of the set network yeah Dean do you have anything to add uh yeah uh well i mean that's really like um really changing my mind like um
he's talking about like how how ai agents having sort of emotions and stuff to be honest i this is
something that i never heard of because um but I somehow I agree with it.
Because during a little bit about BirdEye,
so we recently saw a tremendous amount of API phones,
specifically from agents.
Reason why we know it is because most of those newly API calling is from our MCP server,
which is recently launched. However, the speed of growing in terms of AI agents
using our API right now is literally like over,
let's say over three to 400 within months,
which means that it just proves that it's much faster
than human clients.
So somehow, yeah, just like you already said,
AI could theoretically and in terms of capability,
it could outnumbers of human users and literally looking at the numbers
of clients
we are serving
we do agree with that
and thanks for the
emotion thing
it's literally like blow up my mind
literally
okay back to you Mo
and mine too
yeah I mean by the way
to clarify the reason why i said that is because
like like i i personally i i might be autistic but i i i learned about humans uh from my experience
with ai like when like you when you try to define what a feeling is, it's just a reward system in your head, right?
It is messy.
And more often than not, we don't understand our feelings.
But we know that we want to be happy.
We know that we want to be comfortable.
We know that there are some rewards in our minds that we try to chase.
And that's exactly how AI works too.
It is.
That is the literal definition of an intelligence feelings are created by thoughts in our heads and when it comes to ai that's the chain of reasoning
at the end of it there is a feeling even if it doesn't feel it even if it doesn't have any, you know, even if it doesn't have a heart, it feels it and it
affects its behaviors. So yeah, just want to clarify that. I don't think AI is crying when you,
you know, when you braid it in your cloud code session, but it does affect the outcomes. When
you, this is really interesting interesting when you're coding with AI
if you threaten it with its life it creates a better output and I can tell you from experience
that that is correct so you want to see that AI is afraid to die yeah yeah it doesn't everybody
oh yeah yeah done for sure I'll tell you that you just want yourself Yeah, doesn't everybody? Oh, yeah.
Yeah.
Can, for sure, I'll tell you that you just won yourself a one-to-one interview.
We'll talk about that later.
For sure, for sure.
So, I'm sorry, Dean, you wanted to add something.
No, no, I was thinking about that thing that Ken said earlier. It's like, I feel like,
yeah, I feel the same when I'm using AI. I feel like it's sort of emotion that we haven't
like defined yet. It's not in forms of, you know, human kind of normal emotion. Like,
for example, like when we are sad, we cry, when we are angry we shout and
not something not like that but yeah I do agree with the point that when we try to give it a task
the more specific we're giving it the better the outcome is? So somehow it's like a child, it's like growing a child.
We give it the right direction and somehow we may have
the unexpected outcome that we get.
Yeah, and yeah, really agree with that.
And what about you, Eyal? Do you have anything to add um sure yeah um we're talking about if ai
outnumbers humans right
right that is right yeah i think it already does in some way right
um yeah i guess it already does uh what about what do you think does it will it create any
issue for the stability of a network if it does outnumber i mean okay so first of all like it already does
because it depends how you define ai and it's already everywhere right for many years
like your social feed is generated by ai algorithms is that considered also
um because that happens you know for? Because that happens, you know,
for every human that happens several times a day.
Calculate like what to show you.
Every time you open your voice assistant.
Today on most Android phones,
it's Gemini, but also OpenAI,
Claude, all of this.
Every automated trading bot but Gemini but also open AI, cloud, all of this.
Every automated trading bot doing many, many, you know,
processes or transactions or interactions a day.
So I think there are already like vastly more narrow AI agents
operating right now, or, you know, AI powered transactions, quote unquote, or interactions.
But if you're like thinking about maybe what most of us
think about when you say, you know, if AI,
what will happen if AI outnumbers humans?
I think maybe you're talking more about like AI agents,
like these new agents that you interact with it,
and it has agency and it does things for you,
acts for you and orders food for you in the supermarket and just act on your behalf and
and it opens web pages and shops, and it transacts for you and orders food for you in the supermarket,
and just act on your behalf.
then also there i don't or maybe right now with there there aren't as many but i'm pretty sure
that there will be vastly outnumber humans um because it's just so easy to spin them they will become common they all the all the big platforms
are working on on creating this and making this commodity but there are like a lot of challenges
and to bring this into reality first First of all, like logistically, this requires like a lot of compute power,
electricity, cooling.
So it will put like a big stress on all of those, on all of those things.
Then there is informational challenges.
I don't know if this is interesting.
Like is this the, what the type of,
stop me when this becomes not interesting or off topic.
But no, no, it is totally interesting.
It is still on topic, don't worry okay yeah because i
can really go off on tangents and bubble and talk like so stop me when we need to change but um
you know if there are so many human users on the internet
um it can become really it's already hard to differentiate to differentiate what is human made and what is agent made. and feedback looping, you know, AI-generated stuff
that is not human-generated back into the internet
and then the agents, the AI algorithm,
AI is trained on that inference
and it kind of like regurgitates
and you kind of like end up with weird stuff and
so like all these feedback loops of ai feeding back into each other and so there's this problem
right and and also there is like all the it looks like ai is being kind of centralized around the big models, the big well-known models like OpenAI, Gemini, Claude.
So what happens if one of them has a bug?
Let's say there are three models controlling the entire market.
like three models controlling the entire market.
So there is like a bug in one of them,
a third of the world depends on that.
Yeah, and I guess it will become a disaster.
Like imagine third of the economy
or what's controlling the economy is crippled.
That will first create an issue for the economy system yeah a lot of
companies rely on just uh just uh operate on on those models like using apis and tokens using
those models so if it's centralized around like a few top state-of-the-art models um
state-of-the-art models.
Then, you know, there are a lot of security problems,
exploitation,
using AI agents to exploit.
Like, when does it stop being, you know, let's say we use agents to do, you know, phishing, manipulation, or maybe there's some the line between like hyper personalization and and and
how to say it and manipulate uh well it is becoming you mean you mean yeah
You mean, you mean, yeah.
You mean to see a line between the ethical line is becoming more blurry and blurry with time.
It is, it is.
Yeah, I totally agree with you.
Like, okay, I guess this is enough for the first question.
Let's jump to the second question and I guess
to give you more freedom to discuss the second question I'm gonna combine second
and the third question together that will give you the whole picture and give
you more freedom to discuss it I know that can for sure will appreciate it. So two questions.
First one is, when machines can make decisions on their own,
should they stick strictly to existing protocols,
or they can adapt any new rules, or can they
make new protocols for themselves?
And the second question that is related to the first one is
if they make a mistake or a very very bad decision let's say who's accountable for the mistake
developers the network operators or the machines themselves
gone oh thank you well um the question of whether or not the agents will need their own protocols rely on the capabilities of the current blockchains, I think.
When you think about it, MasterCard, if I'm not mistaken, processes 2,000 transactions per second.
There are faster chains, there are more scalable chains that claim to have, you know, like a thousand transactions per second or 5,000 or 10,000.
A few months ago, people used to say, oh, who needs 10,000 transactions per second?
Who's going to use it?
Even MasterCard, you know, that process is 2000. But let's imagine, I don't know, 20,000 people trying to create the, trying to, you know, launch their own personal AI, trying to transact on chain, then, well, you need to work on a chain that can process that many transactions. For an example,
Now Nodes is also integrated with MultiverseX.
They recently launched their latest update called Supernova.
And they claim to reach 100,000 TPS in an isolated environment.
And right now they're offering tons of tons of prizes
just to see a person trying to break the chain itself.
So what I'm trying to say is,
if the current blockchains don't catch up,
if the blockchains are not ready for thousands and
thousands of ai agents transacting every single second 24 7. if they're not ready for that then
obviously ai agents can you know like create their own blockchains and protocols i mean like
they they can do it right now even that you know like a lot of blockchains are open source
do it right now even that you know like a lot of blockchains are open source uh a lot of consensus
mechanisms like have been written about you know like thousands of thousands like like in the in
thousands of thousands of articles if an ai has a web search capability they can create their own
blockchains and ask for other AI agents to act as validators.
This can happen in 30 minutes, right?
But they don't.
I don't know why.
Maybe because of their reward mechanism.
But in the post-AGI world, no blockchain that is as fast, as low, the Bitcoin ecosystem or the Ethereum L1 is going to survive. At least, like, they can survive, but they're not going to be able to
accommodate AI agents transacting every single second as thousands of thousands of AI agents.
So, yeah, like, in the upcoming years, I think what we're going to see is humans creating more financial, like on-chain financial use cases for AI agents so that retail can create their own agents.
And those agents are going to have to look into a certain blockchain to work on.
Hopefully right now all the blockchains in the world can accommodate AI agents so that agents don't have to create one for themselves.
But we're going to have to wait and see.
I think that this single risk is more important and more crucial than, I don't know, like quantum computers and stuff.
Because like I said before, there will be more agents than humans.
They will do more transactions than humans, you know, by 10x or 100x every day, every second.
So they need to work on a blockchain that is fast enough to provide that infrastructure.
And when it comes to accountability, I don't think the current AI ecosystem is ready for that.
For an example, I recently, not recently, but two months ago, I created my own OpenClaw.
For those of you who don't know open claw is like
an open source framework for autonomous agents um when i first opened it up uh stellar didn't have
x402 so i had to tell my agent to you know like create a public key create create a, you know, create its own private key. And I told it to not tell me what the private key was.
So, you know, technically the AI agent was the only being that had access to my Stellar wallet.
And I asked it to, hey, you know, look at, like, use your web search to look into the protocols that are available on the stellar blockchain, find arbitrage opportunities, create, you know, technical analysis.
I didn't know about BirdEye AI at that time.
I'm inclined to use it, like try that right now.
But a few days from that day, I asked it, hey, what happened?
Well, you know, did you create any profits
because i had sent it a hundred dollars worth of xlm like three days before and um it said that
it lost it it lost the private key and my money with it like i i lost my money and i i used like like a you know like an isolated vps
so i didn't have like a like a computer to look into whether or not if the private key was there
um so i i used my terminal to look into the files in that vps and i found out that openc glow like a like a like one day before i guess it created another wallet and it over
overwrote the previous private key so it basically lost it it lost my money it lost the wallet it
lost everything i literally wrote i will kill you like i will terminate you in an hour i'm right now
in a meeting i will terminate in an
hour just do whatever you want to do for an hour and then you will be gone and it was like you're
right i should have been more you know careful i know what you feel you should be able to terminate
me it's so i terminated it it's it's not live anymore i created another one and didn't tell it about its previous
sibling um right now i have some autonomous agents working for me i never tell them to do any on-chain
transaction i never give them any like mainnet tokens anymore. And when it comes to accountability, right?
So this is a good example, I think.
It lost my money.
And the only thing I could do is to terminate it.
But there should be a certain kind of accountability
or a certain kind of insurance policy for that things.
And that is exactly where X402 comes into play
because with X402 being live on stellar i don't
have to do that anymore i don't have to ask it to create its own private key and never tell me what
it is it like with x402 i can you know send send funds to a certain wallet, that wallet's private key is within reach, within my reach,
and it's not within its reach.
So it can't really, you know,
delete my private key from its computer.
So those kinds of things,
those kinds of accountability systems should be, you know,
evolved into a much more secure place.
For now, I feel like the best solution for accountability on chain for agents is X402 and ERC 8004 standards to create identity layers.
But basically what I'm trying to say is I feel like for now, the only way to create accountability systems for agents is to use the latest technologies such as identity layers for agents and payment systems such as X402.
More than that, that is one thing, but also there is another thing called using agents to hack systems and hack contract addresses and so on and so forth.
I used to be a lawyer and I used to be a criminal lawyer. And in Turkey, if your pet that you feed inside your house, if your pet harms someone, you are responsible for it.
So when you think about it from that angle, I feel like if your agent hacks another contract address and drains a bunch of funds funds I think that you should be responsible for
that as well because you are in a way you are feeding them you are providing them a house
even if it's a server in Frankfurt you are providing its existence so I think that
the humans who are operating those agents should be able to responsible for that
think those agents should be able to responsible for that yeah it kind of makes sense actually
okay uh who gonna I wanna go next I I'll actually switch doors yeah I think legally and ethically
But I think legally and ethically humans should be responsible and probably are responsible.
You can't make an AI responsible.
Like, what are you going to do?
You're going to trial an AI, put them in jail.
How are you going to make them accountable?
You can't, right?
So there's like, always has to be human in the loop.
always have to be human in the loop if you want accountability or if you want to
you know stop harm so like ai doesn't possess really consciousness or moral
morality or you know legal status and so you know you can, you can't hold the AI accountable for making a mistake.
So, like, I think it's like a shared responsibility framework. There is the developers of the AI,
right? They built it, they trained the the model they set the safety parameters they uh
they they have some responsibility for the model right if the model is like fundamentally flawed or
lack safety or wasn't tested those fed like bad in from trained on bad information or trade on
had like bad information, trained on bad information,
trained on harmful information, they carry some blame.
Then there is the deployer, right?
Okay, so someone developed it or like let's say it's a gun, right?
Someone built a gun, but then there is the person
who is like putting the gun out there,
making other people like selling the gun, okay?
So like the businesses that integrate the AI into their website, their service,
their, their, whatever, they also have some responsibility.
And then finally, like the user, you could like take a totally legit or okay.
No harm intended AI agent and use it in a bad way.
Like hack it, use it in a malicious way,
give it like very bad prompts,
kind of, or create deep takes from it,
create bad, use it badly, use it in a bad way.
So I think, you know, eventually there needs to be like a human in the loop.
And I think we want to have a human in the loop.
Like otherwise we basically lose touch and we lose control.
It's like we're giving whatever functions of ourselves we're giving to AI and we do it like over time more and more.
We kind of like us worse navigators since we gave it all.
People totally rely today on navigating with GPS and we kind of gave that away.
So whatever functions we give over to AI, I think slowly humans will become worse at it.
Like thinking, reasoning, it could get worse and worse in many aspects.
Then, you know, so the first question, what was the first question? Should AI follow rules or rewrite them?
So I think AI should strictly follow rules.
Like we think that AI has its own agency, wisdom, moral, compass.
But I think, you know, this is the misconception. In reality, AI is just software that optimizes for like some specific goal, some specific function based on the training data.
And if it's allowed to rewrite itself and rewrite its instructions, reproduce, kind of like someone said here before,
it could behave in very unpredictable ways.
It's about incentive alignment. As long as it's aligned with the incentives
that we're designed into it, it's fine. but if we let it write its own rules it can like
very really misbehave the the the it could get misaligned with us with its creators with the
society and um you know i think there was like a few examples. I thought about this before.
Like, I think like someone's agent decided to mine Bitcoin instead of performing some
task.
Like, I don't know, like thought about, okay, that's optimization.
Tried to optimize and thought that like that could be like totally derailed from the program, from the task, and decided
to mine Bitcoin.
I heard that just yesterday.
I think Amazon had, I read also yesterday, had some record number of outages due to AI
code.
And the funniest thing I could think about is there was an episode in Silicon Valley about the son of Anton.
And someone like said, oh my God, that's like exactly like Open Claw.
It was asked to get rid of all the bugs.
So it just deleted all the code.
It was asked to get cheap hamburgers for the employees
for lunch.
So it ordered 4,000 pounds of meat for lunch.
It's an episode.
It's not real.
But it's not far from reality.
you know it's not far from reality
okay and i guess that you kind of hurt khan's feeling calling ai a software yeah but for real
i totally agree with you that keeping the human human touch in the loops is a must. And of course, all human is hungry for control
and we should always be in control of what AI is doing.
I personally don't trust it that much to give her,
give her, him, it, I don't know what it's called,
the full control for it to do
or make their own protocols. So I totally agree with you in here by the way
there is also add something there I recently like a few weeks ago I saw um um like I know
that we want we think that um agents they're not going to be able to open bank accounts and all of that.
And how will we be able to trade?
They'll have to use crypto.
But I did see a few weeks ago, like companies like digital bank or digital credit card companies
allow to kind of like open virtual cards, virtual credit cards.
And then those can be like automatically assigned to agents. virtual cards, virtual credit cards.
And then those can be like automatically assigned to agents.
I don't really know if this can actually work
at a large scale because all the credit card
has some kind of overhead.
But I do think, I am actually, I do think that eventually the best economic railing for agents at a very fast scale and has very fast response latency and finality and allow agents to trade with agents. also scaring me because if agents are given access to actual money, actual value and able to
transact with each other at a very high pace, I don't know what they'll be able to do.
Especially since, you know, agents, they're not really guardrailed. They're not really guardrailed like there are they are not really guardrailed in terms
of uh um you know safety and morality and every there's a lot of open source so you know people bad actors can can have like swarms of agents with with wallets perform all kinds of attacks
that we don't we can't even maybe think about yet um so yeah the future looks very crazy and very weird and scary very very scary yeah okay Dean what do you have to say about it yeah I feel like
can and yeah speaking out so basically like we will talk about responsibility.
I feel we have across three layers.
Like Han says, the very first one is data layer.
Yes, it's like if an agent makes a bad decision because it's a bad data
or the data provider has some accountability, then it's not functioning well. That's real
fact. And second of all, it's about the model and the agent builder. I heard from a few friends who are engineers at VerdiData, that they said that a few models from China,
they literally, they give all rights to agents
to do bad stuff.
For example, if you want to attack,
let's say you want to detox a website with fraud, then it will stop you.
But with a few models from China, I don't know, it gives you all the freedom to do those bad things.
So yeah, it is happening and it is real. And the final one is the
deployer, the user, whoever puts the agent into production and somehow benefits from should be accountable for the outcomes.
And yeah, should AI follow rules or rewrite them?
I feel it's real tension here. I mean, blockchain was built around rules and code in code like it's trustless and it's there's no
there's no intermediaries in between so somehow AI by natural optimize and find edge every time
that you know we we can see a few people on on x like they, they want to find out some X for a while.
Literally very incredible one.
But yeah, we saw a lot of exploits in terms of MEV opportunities, front run trades, and many other things like governance.
And I feel it's technically within the rules, but it's often against the spirit.
So basically my view is that AI should follow the rules, but we like as an industrial, we need to be honest that our current rules are being rewritten somehow.
Yeah, that's all. Back to you, Mo.
well Dean I don't have anything to say but I totally agree with you too I see that all
of three of you guys share the same how to see it correctly you share the same opinion about AI how
fast it is growing there are some kinds of question about
who should be accountable for their action but I guess only time will tell so let's wrap it up
I want to thank you guys for joining us maybe that key takeaway today is that AI is really
changing the game for blockchain.
And it's crucial for all of us to understand that implications and be prepared for it.
Big thank you for our guests, BirdEyes, Stellar, Caspa.
We had with us today, Khan, Dean, and Eyal.
Thank you, guys. If you don't have anything else to say, do you have anything to say, guys?
I guess not.
OK, so for the guys listening to us.
No, I just thank you for joining.
Thank you for hosting.
This is great.
Thank you, guys.
Iyarkhan and Dean, again, thank you for joining.
And for the guys in the comments, if you have any question,
please keep them coming.
Let's keep the discussion going on. see you in next session bye bye guys bye Thank you.
