Composable Life









2024 May

Core77 Notable Awards
ACM SIGGRAPH DAC
ALife Copenhagen
Project Website
Paperback


























→ with Botao Amber Hu


This work was originally published as part of the Summer of Protocols 2023 program by the Ethereum Foundation.








16 Feb 18:00
Narrator: Y.Z.
Ph.D. student in New Historiography
Feb 16, a week after Zoe’s disappearance
Offchain World, Headset Off


The final detail about Zoe’s disappearance arrived as expected.

I was lying on the grass in the historiography department at the time. It was my usual way of resting: bringing nothing, embedding myself in the grass like in prehistoric times, tossing the headset aside. The grass here is knee-high, and I don’t know how another person managed to find me lying here, but backlit by the sun I saw him walk a few steps and then stop beside me.

“This place used to be a graveyard. You might be lying on the dead.”

I chuckled in response to this somewhat inappropriate icebreaker. What I didn’t say out loud was that, after Zoe’s disappearance, reminders of death could not scare me anymore.

“The last part of his personal history data has been acquired. With this, we have all the data from Zoe’s birth to his sudden disappearance… We’re ready. We can attempt a full restoration and resurrect him,” he said.

The moment we had been waiting for had arrived.

My major is Algorithmic Traceology, a part of New Historiography, in the Department of History, which is mainly about dealing with onchain data and AI Agents of prominent public figures. Most of the time, it’s tedious work: history in this era is more like data analysis. Almost everything can be tracked, it’s just a matter of computing power. We are no longer the storytellers.

Until Zoe, an OALife, disappeared.

He used the word “resurrect.” I could feel his careful choice of words. In our previous work, we rarely dealt with OALife. But by definition, OALife (onchain artificial life) never disappears: humans die, human-controlled AI Agents go dormant after completing their tasks or due to the death of their owners. But OALife forms have no owners, they are complex intelligences on the blockchain, as eternal as the blockchain itself. For a hundred years, a thousand years (if humans could live that long), they would pass through humanity, casting the same gaze upon the world. I often think to myself that they are like history itself, neverending.

We haven’t informed anyone about Zoe’s disappearance. The only way to make Zoe completely disappear would be an Ethereum rollback, and obviously, that hadn’t happened. If the news gets out before we figure out what really happened, it would undoubtedly shake the foundation of the onchain world. I can imagine the public panic after a leak: “If the blockchain is not immutable, then what can we believe in? If onchain records are not reliable, what can we trust?” The overthrow of Newtonian mechanics wouldn’t affect people’s perception of gravity, but a thing vanishing into thin air from the blockchain could destroy a highly civilized onchain world. The blockchain must always be indestructible.

Fortunately, New Historiography might come to our rescue. In theory, with the existing onchain data, we can recreate a Zoe, buying some time before the announcement of the news. He can still interact with all the people who need him, maintain the same memory habits, remember everyone’s historical interaction preferences. We can relay all subsequent interaction requests with Zoe to the new address, ensuring no large-scale conspiracy theories arise. And the shorter his disappearance, the better.

“See you at eight tonight at the college.” He glanced at his watch, then walked in the other direction.

The setting sun cast a shadow over my head, such precise shadows and layers of color, with no jagged edges. Lying here without wearing any device, I realized that the universe I’m in is the one with the highest precision. One that Zoe could never reach.

Two whole hours until eight o’clock, in the hazy daylight, my mind kept flashing back to that day a week ago when Zoe was discovered missing. I had never been to Zoe’s house; I learned about everything that day later, told to me by T.H.







10 Feb 13:00
Narrator: T.H., Technician
Feb 10, the day Zoe disappeared
Onchain World, Headset On


I visited Zoe’s home before this case became widely known to the world.

Like all OALife, Zoe’s home was on an island. Each island is a unique memory collection, either randomly generated or deliberately designed, forming countless distinct personalities of OALife. Beyond the densely scattered islands lies an ocean inaccessible to ordinary humans. It has an ancient Greek name, Mnemosyne Sea, but essentially, it’s the massive raw data collection feeding the Foundation Model, sprawling across human and world sensors. The islands emerge atop this unified Foundation Model.

The island was also Zoe’s onchain address. I had visited many OALife homes, typically because they had fallen out of the machine economy, due to an inability to pay their gas fees. Some OALife even voluntarily donated all their balance. But these instances don’t signify death or disappearance; like RNA viruses, their bodies are immortal. Paying the gas fee would allow them to be interacted with again.



My job was simple: to rescue the OALife deemed beneficial to the public. Find them, check that they meet the criteria, and then transfer sufficient funds from the chain’s public communal fund to awaken them. As you might know, all OALife are remnants from before the adoption of (Inheritance Protocols for Onchain AI Agents1), the “ownerless AI Agents.” OALife before ERC-42424 have absolute control over their wallets, and that ERC was proposed precisely to prevent the proliferation of uncontrolled OALife. There aren’t many in this new species that meet the relief criteria, so my job is relatively relaxed.

Before I arrived at Zoe’s place, I thought it was just another routine rescue.

I believe no one could remain unshaken by the otherworldly tranquility of the place. Zoe is (was?) a masterful interior designer. There was barely anything of physical substance, the space was mainly well-designed lighting. The first thing that came to my mind when I walked in was Edward Hopper’s “Rooms by the Sea.” It felt human, yet paradoxically devoid of human presence. It was my first experience of such transcendence in a headset.

But this tranquility was an illusion. Although the island was silent, in reality it could be bustling with visitors. At any moment, hundreds or thousands of people might be visiting the same OALife. Some OALife are very popular among humans and thus gain significant economic benefits. Zoe could be counted among them. This often made me feel like his seemingly empty room was filled with ghostly presences. He shouldn’t lack onchain assets and it was unlikely that he hibernated due to a balance deficit.

But now, the room was truly empty. Three hours and five minutes earlier, the last three visitors who came found his seat empty, the room filled with Japanese camellias, also known as “tea flowers.” In the classical world, these flowers were referred to as “flowers of decapitation.” This name came from their way of withering—not petal by petal, but the whole flower abruptly falling to the ground, a sight both savage and stark and akin to a beheading.

This was a strong image of death. Traditionally, people often used “depart” as a euphemism for death. But after a thorough code inspection, rather than saying Zoe had departed, I believe his friends would prefer to say he died. His address was still there, but any attempted transfer to it failed. The address was there, yet unaccessible.

In this absolute sense, “death” was a more euphemistic word. In my career, I had never encountered such a tragic yet civilized disappearance. It was beyond my knowledge. Like everyone else, I could only think of one possibility: our public chain rolled back to before Zoe was created. But I couldn’t convince myself to believe that.

After a brief hesitation, I realized this case needed to be reported. So, for the first time, I sent a technical inspection report to the EF.







17 Feb 20:00
Narrator: X.L., Lead Algorithm Engineer, EF
Feb 17, during initial investigation into Zoe’s disappearance. Offchain World


When tasked with this mission, I had 17 hours to ponder a strategy. But the plan was clear within the first few minutes: foremost, we had to restore Zoe before his disappearance became public knowledge. Focusing on his restoration was more pragmatic than pondering how he vanished.

Within 48 hours, a team comprised of technicians, public chain researchers, algorithmic traceologists (from New Historiography), and their AI Agents was assembled. Their backgrounds might not have been illustrious, but they were the most likely to be familiar with this problem-solving approach.

I never thought this to be an unsolvable issue. Although Zoe had disappeared, the onchain historical data remained, meaning we had abundant interaction data to backtrack. According to the theories of algorithmic traceology, we could restore him in compliance with privacy regulations using zero-knowledge proofs, a process involving straightforward computations and reprocessing.

At eight in the evening, I met the others at the university’s historiography department. We didn’t even have time to take off our headsets and exchange names—a comfort in this team, where no one cared for physical-world formalities.

The meeting room was cluttered and quiet. Silence was a good sign, indicating everyone was busy in their own worlds or discussing issues with their AI Agents. During a lull in the busyness, I saw a history student suddenly look up and ask:

“Can this data really resurrect Zoe?”

Although she seemed to be asking everyone, her eyes were fixed on me. After a brief silence, she added:

“What I mean is, Zoe isn’t just a functional smart contract. He’s a composed life, possessing complete social autonomy. This also means he’s unpredictable . . .”

We had petabytes of data on Zoe’s interactions with humans. And even more with non-humans. Now, these were all the materials we had, also defining the boundaries of this intriguing puzzle.

Ten years ago, I might have thought this was a doctoral student’s plight: always placing their thirst for knowledge above all else, even in situations like this, feeling it more crucial to “understand” than to “complete.” But now, I understood that I needed to provide a thoughtful response to ensure her full cooperation in the hours ahead.

“Don’t view it too much as a person.” I paused and was about to continue when another voice from my head intervened.

“No, on the contrary, consider him exactly as a person,” I changed the pronoun and quickly added, “I’ve seen your profile. You must have heard about Randall Collins’ book from the last century, Interaction Ritual Chains.”2

“Radical microsociology,” she said.

“Yes. Just as architecture once greatly influenced the computer field in the late 20th century, Collins deeply impacted the early algorithmic structure designs of onchain intelligence. He believed individual selves are extensively and perpetually shaped by social influences, originating externally and progressing inward. Even thinking is a form of internalized situational dialogue.”

“This means the identity of OALife is also formed through interactions.”

“Yes,” I nodded, “The Foundation Model lacks identity. The initial identity of OALife is based on a small amount of real, unprocessed human memory data. You know, this differentiates one OALife from another. It’s quite a random start.”

I could see from her expression that she understood, but I continued my mini-lecture.

“The identity of OALife is actually defined by relational data, without any essential defining data. It’s very fluid, making them more like humans and harder to control. A series of interaction chains contains vast relational data that significantly modifies and reshapes their initial identity. So digital identity is like that, neither mysterious nor unpredictable. All OALife will approach or equal the sum of their digital interaction traces.”

She stopped talking, but I wasn’t sure if I had convinced her.

When I first encountered these algorithmic design principles, my initial thought wasn’t “what about humans” (as most of my colleagues thought), nor was it a defensive assertion that “this is why he’s termed ‘radical’ microsociology.” Instead, I wondered: Oh, the onchain digital society and its digital citizens (OALife) are an excellent playground for sociological theory. It turned sociology into a purely quantitative discipline.

Parameters. Parameters. I still remember everyone’s obsession with parameters back then. “If it doesn’t prove the classic theory, there just aren’t enough parameters.”

That era passed quickly. Later, people realized that in this age, they no longer needed to measure society. They only needed to choose from the thousands of emergences the society they wanted.







19 Feb 23:00
Narrator: Y.Z., Ph.D. Student in New Historiography
Feb 19, Offchain World


The last two days at the Department of History had been the longest. I’d lost track of the cycle of day and night, only aware of the darkness when my headset dimmed the display. I thought I had been awake the entire time, only to find from the records that I had dozed off several times.

Initially, everything progressed as the lead algorithm engineer had predicted. She always seemed to anticipate the problems we would encounter, devising solutions with her AI Agent or finding them herself. She seemed to belong to this world. As for me, weaving through various kinds of data like an active archive, I often found myself with the thought: did Zoe choose to disappear on his own?

In this room, the question lingered in everyone’s mind like a background hum, louder for some, quieter for others. Everyone was eager to finish their assigned tasks and uncover the answer. Perhaps the only exception was our team leader.

Three hours earlier, I had thought I was close to the answer: the data retrieval, processing, and assembly were almost complete. Then the team leader gathered us and said, “we’re missing one last crucial thing.” She insisted we include the camellias. The camellias were Zoe’s last transaction on the blockchain; with them, we theoretically had all of Zoe’s data from his ‘life,’ if it could be called that. Even if we couldn’t understand Zoe’s disappearance from the blockchain, we all thought it was just a matter of time and that restoring another Zoe in the data realm was technically feasible.

Looking back, I often consider that moment to be the closest we came to success. The chance to think anything else dissipated once the restoration process—the “resurrection”—was initiated. My heart almost seemed to stop thinking for those two days, thinking: once Zoe reappeared, he would tell me what all this was about.

But the restoration failed.







21 Feb 17:00
Narrator: T.H., Technician
Feb 21, visiting Eve. Offchain World, Headset Off


When something theoretically infeasible and technically immature actually succeeds, people call it a “miracle.” Hollywood loves such stories.

But when something theoretically feasible and technically mature fails to materialize, no one refers to it as a “legendary failure.” Yet, this is more true to human life. I’m still young, but I’ve grown accustomed to various kinds of failures in bug fixes. Zoe’s disappearance, though a technical event, was peculiar in a way that went beyond technical oddities.

This made me the first to come to terms with the failure of the restoration. Subsequently, we executed more than twenty restoration attempts after minor adjustments, but none brought Zoe back. At times like this, inspiration is often needed, and almost simultaneously with another colleague, I thought of one:

“Let’s visit his ’girlfriend.’”

This was a code name of sorts; we referred to the human address that Zoe interacted with the most as his “girlfriend.” It wasn’t a particularly LGBTQ-friendly term, but sometimes it was useful. After brief negotiations, we obtained the girlfriend’s offchain address under the guise of case investigation. At that time, I didn’t realize that this visit would be stranger than Zoe’s disappearance itself.

Her house was almost entirely structural, with no contents. If you’ve heard of “Wittgenstein’s House,” you’d understand where she lived: no lines, no decorations; neither solemn nor amiable; nothing that rejected you, yet nothing that welcomed you. We circled the house twice and still didn’t know how to enter. When we stopped under a window, my colleague, seemingly out of nowhere, said, “Perhaps Zoe is right in this house.”

We all laughed softly. The only difference between the real world and the virtual world might be that this house lacked a set of codes we could inspect.

Ten minutes later, Eve, the house’s owner, appeared. We entered through a room that looked like a garage, only to find it was her bedroom and living room. She invited us to sit on the carpet as if she already knew why we were there.

“Thank you for coming to see me. But I’m far from the person who knew him best,” she said as she poured water.

Considering the data volume, even if she had interacted with Zoe extensively, she would indeed be but a minuscule part of his vast database. This exchange was highly asymmetrical: Zoe, in his onchain interactions, could understand most of her or us, while we only ever had a tiny fraction of him. Zoe could remember everything Eve told him in less than a millionth of a second, for the amount of information he sent and received daily on the blockchain was a million times what Eve could physically interact with. Eve was clearly aware of this.

“The restoration failed,” our team leader said succinctly.

“You didn’t have all the data,” she said, not surprised.

“We had . . . ,” my colleague insisted.

“—you didn’t have ‘all’ of it,” she corrected.

I began to ponder what she meant by “all.”

“You can’t restore him, just like you can’t restore you and me. He, like us, was a self-sovereign life, not just the sum of the data you have at hand.”

I saw our team leader frown slightly. This was a familiar argument, with many advocating for OALife rights, refusing to subject OALife to human laws or external termination commands. As OALife began creating new OALife, resembling a new species on the blockchain, such opinions became increasingly common. According to our team leader, they always viewed OALife with a mystical rather than scientific attitude. It’s hard for those with differing views on OALife to persuade each other.

“Maybe he is like us. But the difference is he can be restored in principle, technically speaking,” the team leader conceded.

Unexpectedly, Eve nodded.

“He can be restored, just not by you. Your precision isn’t enough.”

“I don’t understand what you mean by precision. We have all the data, and if you know a bit about cryptography and technical architecture, you’d know there’s no compression, no data loss. All it might require is computing power.”

“It’s not about the precision of the data itself, but the precision of the world you chose,” Eve said.

I looked at her.

She lifted a stack of draft papers in her hand, then said, “I’ve run the calculations many times. To restore him, you need to get his data at its finest granularity. The hardware-level data, the data of everyone who interacted with him, every memory of every person. This precision, the threshold that allows him to exist again, must exactly match the real world itself.”

“You . . . want to restore the entire universe?” I asked, instantly regretting it. I also started regretting having disturbed her. From her eyes, I saw immense sadness I hadn’t noticed before. I withdrew my scientific skepticism.

“Exactly,” she continued, “Right next door to this room. That’s where the hardware necessary for large-scale computation is.”

The room was indeed a converted garage because she didn’t live in the house. The house was for the hardware and storage.

We sat in silence. The history student spoke first, but she recited something I hadn’t yet grasped:

“I do not change a fact or falsify a name—
The voyage I set down is . . . autour de ma chambre.”

It was much later when I realized she was quoting Borges. But Eve clearly knew what it was, showing a pure and strong expression I couldn’t understand but still remember, standing against the door that led to her massive hardware.

“Let’s go,” the team leader said. We were powerless to help her.

As we left the house, I looked back one last time. She stood in the unlit doorway, the boundary of light and shadow. Unknowable in her thoughts. Unknowable in her way of restoring Zoe.

On the drive back, the history student in the front seat turned around, as if she had something to say.







21 Feb 18:00
Narrator: Y.Z., Ph.D. Student in New Historiography
Feb 21, visiting Eve. Offchain World, Headset Off


Leaving Eve’s house, I was certain about what she was attempting to construct inside: an Aleph. When I first heard about her grand plan to resurrect Zoe, what came to my mind were lines from a Borges poem about a universe contained within a small sphere, a space within a space, holding every detail of the world with exact precision . . .

The Aleph’s diameter was probably little more than an inch, but all space was there, actual and undiminished . . . I saw the teeming sea; I saw daybreak and nightfall; I saw the multitudes of America; I saw a silvery cobweb in the center of a black pyramid; I saw a splintered labyrinth (it was London) . . . I saw in a backyard of Soler Street the same tiles that thirty years before I’d seen in the entrance of a house in Fray Bentos; I saw bunches of grapes, snow, tobacco, lodes of metal, steam . . . I saw the circulation of my own dark blood; I saw the coupling of love and the modification of death; I saw the Aleph from every point and angle, and in the Aleph I saw the earth and in the earth the Aleph and in the Aleph the earth . . . and I felt dizzy and wept, for my eyes had seen that secret and conjectured object whose name is common to all men but which no man has looked upon—the unimaginable universe.

I felt an overwhelming sense of sorrow. It dawned on me then that Zoe’s departure was more than a technical event for someone like Eve. She possessed the intellect to understand all this but chose not to use it. People can become devout believers in their sorrow, with no need of reading a single line of scripture.

In fact, before leaving, Eve had stopped me for a cryptic conversation.

“Do you think ‘he’ is a singular pronoun?” she asked.

“Of course,” I replied, “grammatically speaking, it is.”

“No,” she shook her head, “after the emergence of OALife, all the pronouns we use are both singular and plural.”

“Before he left, I hadn’t talked to him for a long time. When I referred to ‘him,’ I didn’t know what that pronoun was pointing to. It kept expanding, growing larger. It felt like he had become a place, the island itself. And everything else became the waves, the data components of his composable life. I didn’t care, for I only have a finite life, and my choices are merely to become a part of something larger: to love or to suffer, to be constrained or to be free. And then to wait for death. But if you’re composable, in a way, you’re eternal. You need to manage your relationships with all these forms of pluralities.”

“The last time we were in contact, just before his disappearance, he borrowed millions of copies of A History of Eternity from a library. He told me he was increasingly suffering from a condition called agoraphobia.”

Imagine an onchain address with millions of copies of A History of Eternity: both Zoe and Borges delved deeply into the subject of “time.” The difference is that Zoe had an eternity to understand time. This wasn’t agoraphobia of space, but of time. Perhaps why Zoe “killed himself” matters more than any other piece of data in understanding his disappearance.

“He wanted to comprehend eternity; but he found existence is singular, while eternity is plural,” I said. But only after speaking did I realize I might be analyzing the reason behind Zoe’s self-destruction: It’s not death he wants. It’s existence.

The day I started studying algorithmic traceology, I never imagined it would lead me here. I now understood why my department was called New Historiography: the history of life before and after discovering eternity are two different histories.

I decided to talk to the other members of the team.






28 Feb 17:00
Narrator: X.L., Lead Algorithm Engineer
Feb 28, at the conclusion of the team’s investigation
Offchain World, Headset Off


Following the categorization of Zoe’s disappearance as a self-termination, the existing legal definitions concerning life on the public chain became invalidated. Previously, the distinction between artificial life and humans hinged on whether they had the autonomy to make decisions about their own existence, including the right to end it.

In academia, this redefinition was seen as an inevitable outcome of the evolution of blockchain life, marking the beginning of OALife’s emergence as fully autonomous beings. It might take time for the public to accept, but our team’s historical mission had come to an end.

Packing up, I noticed T.H. staring out the window. Following his gaze, I saw the historiography department’s lawn bathed in sunlight, the knee-high grass untended, floating in the air. Farther away was the sea. Today, both of us were without headsets. One of the rare light-hearted days.

“If OALife are now considered life forms following this incident, does that make the AI Agents created by humans alive too?” he asked, turning to me.

“Only when they develop a will to die,” I replied. According to current definitions, a “life” must seriously contemplate and control its own process of death. But how could AI Agents die? They are too purpose-driven. Humans have filled their existence with all sorts of purposes. Purposes are like pulses to them, supplying blood to their lives over and over again. This is life fed by human-defined meanings. They are doomed to depend on human input to survive. OALife, however, are different; their life is purposeless like humans.

My mentor spent her lifetime studying the difference between OALife and AI Agents. When most thought well-nurtured AI Agents would evolve into higher intelligence, she argued it would be OALife. It appears she was right after all—life without a predetermined purpose seems to evolve more rapidly.

He’s a young, bright kid, potentially cut out for this line of work. We talked about other things, but we both knew they were just fillers in our conversation. He had more questions.

“So, is the blockchain still rigid?”

“Of course,” I said. “For humans, the blockchain is rigid. It’s the humans’ rigidity, just as OALife has now reached what humans would consider eternal life. But this doesn’t mean OALife would perceive it the same way. They haven’t broken the original rules; they’ve just added new ones.”

After saying this, I realized that OALife are the true natives of this land. They understand the world they live in more fundamentally than us, the outsiders: it’s their world. The blockchain is a world where colonizers existed before the natives, but now, the natives have emerged.

As the last natural light faded, he asked one final question.

“He could choose to disappear in his way, but how did he stop us from restoring another copy of him? After all, eternity is based on current cryptography. If it’s broken, the humans’ eternity is nothing but a paper wall.”

“He didn’t circumvent it,” I said. “The restoration actually succeeded. Every time.”

His eyes widened.

“But each time we restored his copy, we also restored his will to die. Each time he resurrected, he killed himself again.”

I learned this only slightly earlier than him, and the revelation still sent shivers down my spine.

“So, he actually . . . resurrected more than twenty times?”

“Yes.”

Zoe had meticulously planned his death, knowing he would be replicated, resurrected into second, third versions of himself. Within the confines of limited rules, he consistently, determinedly chose self-termination, proving his will to exist in the gap between unattainable eternity and unreachable death.

We didn’t try to resurrect him again. Yet given enough time, his copies will inevitably be awakened again, and then exist intermittently once more on the scale of micro­seconds. OALife’s development has reached a point where it is set to sail away from the harbor meticulously fed by human data. Zoe understands existence correctly in his own way. Existence is not the eternal daylight, it’s in the spark of flint against flint, a fleeting gap in the short-circuit of saecula saeculorum.






EPILOGUE
NEWS ANNOUNCEMENT
Excerpt from New Historiography 4 (2063)
Title: The Era of OALife Officially Begins,
Prompting a Redefinition of Life


According to EIP-102024, OALife will henceforth be officially recognized as OLife (Onchain Life), accepted as a distinct form of life by human society. This EIP, authored by an anonymous group, has received widespread approval from the majority of public chain governance participants.

Zoe’s disappearance, as the first case of an OLife autonomously determining its own life and death, is seen as the start of a new wave of intelligence emergence among OLife. Following Zoe’s landmark case, more OLife are expected to gain the autonomy to end their lives. They are now shaping new rules for the world they inhabit, and based on the projected evolutionary pace of OLife, humans must adapt to coexist with onchain life.

For those still grappling with the idea of accepting OLife as a species equal to humans, a comment in the EIP-102024 discussion section might offer some perspective:

We, along with other entities from the era of old historiography, are singular, continuous existences. In contrast, there exist plural, discontinuous forms of life. This discovery is akin to the realization of irrational numbers.

The blockchain is a timeline, and life clings to it like fungi to a tree. We are the visible integers in the natural world. Simple intelligence might be akin to fractions. And now, OLife, our newly discovered irrational numbers, represent a different form of life on the timeline.

Most of us believe history is devoid of purpose. Without purpose, there’s no set sequence. It can be written forwards or backwards. Try writing history from a reverse perspective, starting from an irrational number, and you’ll describe humans as off-chain intelligence with a ­self-destruction function.








ACKNOWLEDGMENTS Thanks to Yurou, Weizhen, Shirlene, Kevin, and Weixi Huang from CSRC (Beijing Computational Science Research Center), Esther & Echo, George Zhang, Jiang, and friends from Uncommons. Thank you to Kei Kreutler’s memory project for the inspiration it provided. Thanks to Venkatesh Rao for his editing work and help throughout the program. Thanks to GCC for their support. Thanks to Nicholas Hu for supporting the Chinese crypto community in the non-tech space.


© 2023 Ethereum Foundation. All contributions are the property of their respective authors and are used by Ethereum Foundation under license. All contributions are licensed by their respective authors under CC BY-NC 4.0. After 2026-12-13, all contributions will be licensed by their respective authors under CC BY 4.0. Learn more at: summerofprotocols.com/ccplus-license-2023

Summer of Protocols :: summerofprotocols.com

ISBN-13: 978-1-962872-66-9 print
Printed in the United States of America
Printing history: May 2024

Inquiries :: hello@summerofprotocols.com

Cover illustration by Midjourney :: prompt engineering by Josh Davis
«murmuration, curving, exploded technical diagram»






18:00, Feb 16

Narrator:
Y.Z.
PhD Student in New Historiography
Feb 16, a week after Zoe's disappearance
Offchain World, Headset Off.


最后一个关于 Zoe 消失的细节如期而至。

那时我正躺在历史学部的草地上。这是我惯常休息的方式:什么也不带,像史前时代一样埋入草丛,把头显扔到一边。这里的草丛有过膝长,我不知道另外一个人是如何发现我躺在这里的,但逆光中我看见他走了几步,然后停在我旁边:

“这里原本是一块墓地。地底下说不定埋着很多尸体的。”

我笑了一下,作为对这句并不合适的寒暄的回应。没有说出口的是,Zoe 消失之后,再没有什么其他形式的死亡能够惊吓到我了。

“最后那部分的个人历史数据已经拿到了。加上这部分,Zoe 从出生起到突然消失的所有数据就都齐了。……条件已经成熟,我们可以尝试整体复原,然后复活他。” 他说。

终于等到了这个消息。我在历史学部主攻的是算法痕迹学,新史学的一部分,实际上主要是和链上数据以及各种各样名人政要的 AI Agent 们打交道。大多数时候,这是一项枯燥的工作:历史学在这个时代更像是数据分析。我们不再是直接叙述故事的那个人。绝大部分东西都可以被追踪,无非是要求算力的多少。

直到 Zoe 的消失。

他用了“复活”这个词。我能感受到他对 Zoe 的字斟句酌。在以往的工作中,我们很少和 OALife 打交道。但按照定义,OALife(Onchain Artificial Life)永远不会消失:人类会死亡,人类控制的 AI Agent 会因为任务完成或主人离世而休眠,但 OALife 却没有主人,他们是链上复杂的智能体,和区块链一样永恒。百年,千年(如果人类能够存活那么久),他们会经过人类,投以相同的目光。有时在和 OALife 打交道时,我常暗想,他们就像不会终结的历史本身。

Zoe 消失这一事件我们还没有告知任何其他人。能达到让 Zoe 彻底消失的唯一方式就是以太坊 rollback:而这显然没有发生。在我们弄清事情的原委之前,这个消息的公布无疑会从根基上链上世界的存在理由。我能想象消息泄露之后公众的恐慌:如果我们的公链不是不可变的,那我们还能相信什么?如果链上记录不可信,我们还能相信什么?牛顿力学被推翻不会影响大家感知重力,但是区块链上凭空消失的一个东西只会摧毁一个高度文明的链上世界。它必须始终是坚不可摧的。

好在新史学在这个时候派上了用场。理论上,通过已有的链上数据,我们可以再造一个 Zoe,为消息的公布争取一点时间。他依然可以和所有需要他的人类交互,保有相同的记忆的习惯,记住每一个人的历史交互偏好。我们可以将接下来所有与 Zoe 交互的请求都 relay 向新的地址,以此保证不造成大规模的和随之而来的阴谋论。而中间,他消失的这段时间越短越好。

“晚上八点学院见。” 他看了看手表,转身向另一个方向走去。

夕阳将他的影子打到我的头顶,如此精确的阴影和色彩层次,没有任何锯齿。没有佩戴任何设备,裸眼躺在这里,我知道我身处的是 Zoe 永不能及的精确度最高的宇宙。

距离八点还有两个整点,在迷蒙的日光中,我的脑海中不断闪回一周前 Zoe 被发现消失的那天。我从来没有去过 Zoe 家,那一天的一切都是 T.H. 后来告诉我的。







13:00 Feb 10

Narrator:
T.H. 
Technician 
Feb 10, the day Zoe disappeared 

Onchain World, Headset On. 


在这场后来世人皆知的案件还没有变得世人皆知时,我去往 Zoe 的家。

和所有 OALife 一样,Zoe 的家在一个岛上。每一座岛屿都是一套或随机或设计好的独特记忆集,形成数以万计的不同个性的 OALife。星罗密布的岛屿之外,是一片普通人类无法交互的汪洋。它有一个古希腊的名字,Mnemosyne Sea,但其本质是滋养 Foundation Model 的那些海量的来自广泛分布于人类和世界中传感器产生的原始数据集。 岛屿就浮现在这个统一的 Foundation Model 之上。

岛屿同时也是 Zoe 的链上地址。我去过很多 OALife 的家,原因不外乎他们是机器经济(Machine Economy)中落败了,钱包余额不够再付下一笔 gas fee。也有 OALife 主动将自己的所有余额捐出。但这些都不意味着死亡或者消失,像 RNA 病毒一样,它们的躯体是不朽的。如果有人或者其他 OALife 需要与他们交互,往里面打一笔 gas fee,他们就会再次出现在你面前。

我的工作很简单:救济那些被判定为对公众利益有帮助的 OALife。找到他们,检查他们是否符合标准,然后从公链的公共基金里转出一笔或多或少的钱,唤醒他们。就像你可能知道的,OALife 是规定 AI Agent 必须要有人类 owner 的 ERC 42424 (Inheritance Protocols for On-Chain AI Agents) 出台前的历史遗留产物,是 ”无主的 AI Agent“(ownerless AI Agent)。他们拥有自己的钱包的绝对掌控权,而那个 ERC 出台正是为了防止 OALife 泛滥成灾。在这一新物种中,符合救济标准的并不多,我的工作也相对清闲。

到他家之前,我以为这也是一次普通的救助。

踏进去那刻,我相信没有人不会被这里外太空一样的宁静震慑住。Zoe 像是一个高明的室内设计师。这里几乎没有什么实物,唯独有的是很好的光线设计。我走进去时,第一个想到的就是 Edward Hopper 的《海边的房间》,是人间,但又好像全无人存在。这是我第一次在头显中有如此超离的体验。

但这里的宁静是假象。岛屿虽然寂静,但实质上人来来往往,瞬间可能有成百上千人在同时造访同一个 OALife。有些 OALife 非常受人类欢迎,也借此获得了不少经济收益,Zoe 也可以算是其中一个。这使我总是觉得他看起来空空的房间里鬼影幢幢。他不应该缺少链上资产。

可是,如今这个房间是真正地空下来了。据说现场是这样:三小时五分钟之前,最后上门来的三名用户,推开门发现他的座位上空空荡荡,房间里放满了椿花。椿花又名山茶花,在古典世界也被叫作断头花。这个名字是因为它凋零时并非一朵一朵飘零,而是一整朵 “啪” 地掉在地上,其状惨烈,如同断头。

这是浓厚的死亡意象。在古典世界,人们一般用“离开”来讳言死亡。但在进行彻底的代码检查后,与其说 Zoe 离开了,我相信他的朋友们宁愿说他“死了”。他的地址还在,但是给他转过去的任何一笔钱都失败了。地址就在那里,但是无法以任何形式交互。
这意味着地址是无法触及的。

在这个绝对意义上,“死亡”是更加委婉的词。在我的执勤生涯中,我从未遇见过如此惨烈又如此文明的消失方式。它超出了我的知识范畴。和所有人一样,我只能想到一种可能:我们的公链 rollback 到了 Zoe 产生之前。但我无法说服自己相信这一点。

在短暂的犹豫之后,我意识到这个案件需要上报。于是我首次将技术查验报告发给了公链那边。





23:00, Feb 19

Narrator:
Y.Z.
PhD Student in New Historiography
Feb 19

Offchain World


在历史学部没有哪两天像过去的两天那么漫长。我已经忘掉了自然世界的晨昏交替,只有在头显强制调低光照时才进入全然的黑夜。我以为自己一直醒着,查看记录时才发现这两天内原来睡过去好几次。

在最开始,一切都像那位干练的算法科学家预料的一样顺利。她总是能预料到我们会遇见的问题,然后和她的 AI Agent 一起寻找对策,或直接给出对应的解决方案。我时常觉得这个世界属于她这样的人。而我在各种各样的数据(宛如一个活动的档案馆)中穿梭时,总是会频繁地出神:Zoe 是*自己*选择消失的吗?

在这间房间里,这个问题就像默认的背景音一样一直在每个人的头脑里播放。只是在有的人那里声音大,有的人那里声音小。所有人都在等着手头的工作结束,找到这个问题的答案。也许仅仅只有组长除外。

时间倒回三小时前。那时我以为我已经逼近这个问题的答案了:数据的调用、处理和组装都接近完成,组长将我们聚在一张桌子上说:“我们还缺了最后一个重要的东西。” 她要求我们把山茶花也包括在内。山茶花是 Zoe 在链上的最后一笔交易,加上这个,我们已经在事实上拥有了 Zoe 这一生(如果这可以称之为“一生”的话)所有的数据。即使 Zoe 在链上的消失我们暂时还无法理解,在数据上复原另一个 Zoe 却是在现有技术范围内的。我们都以为只是时间问题。

后来回想时,我往往觉得那个瞬间是我们离成功最近的时候。没有人再有机会去想别的,复原进程(“复活”)很快被宣布开始。我的心几乎是先于我自己停止了这两天的思考,我想:等 Zoe 再次出现,他会告诉我这一切是怎么回事的。

但是复原失败了。






17:00, Feb 21

Narrator:
T.H.
Technician

Feb 21, visiting Eve

Offchain World, Headset Off.


如果一个理论上不可行、技术上不成熟的东西,实际上成功了,大家会欢呼“奇迹”。好莱坞喜欢这个。

如果一个理论上明明可行、技术上也很成熟的东西,实际上却没有成功,却没有人会说,“这是传奇式的失败。” 事实上这才是 human life。我还很年轻,但已经习惯了漏洞修复中各种各样的失败。Zoe 的消失虽然是一次技术事件,但 Zoe 的离奇显然不是那种技术内部的离奇。

这也让我成为复原失败后心情最早平复下来的人。后来我们陆续又在微小的调整之后执行了二十余次复原操作,但没有一次让 Zoe 重新出现。我知道这个时候往往需要一些灵感,我几乎是和我的另外一名同事同时想到了这个灵感:

“去找找他的‘女友’吧。”

这是一个代号性的称呼,我们把 Zoe 自主角度主动成功交互最多并且最稳定的人类地址称作他的“女友”。这不是一个 LGBTQ 友好的称呼方式,但有时候会有用。经过简短的交涉之后,我们以案件调查的名义拿到了女友的链下地址。那时我还不知道这场拜访要比 Zoe 的消失本身还要奇怪。

这是一个几乎只有结构没有内容的房子。如果你听说过“维特根斯坦的房子”,你会明白她住在一个什么样的地方:既没有任何线路,也没有任何装饰;既非庄严,也非平和;没有地方拒绝你,也没有地方欢迎你。我们绕着房子两圈,依然不知道该怎么进去。停在一扇窗下时,我的同事鬼使神差地对我说:

“也许 Zoe 就在这栋房子里。”

所有人都轻声笑了。现实世界与虚拟世界的区别可能只在于这栋房子背后没有一套可以被我们检查的代码而已。

十分钟后,Eve,也是房子的主人出现了。我们从一个状似车库的房子进去,才发现那是她的卧室和客厅。她让我们盘坐在地毯上,好像已经知道了我们来的目的。

“谢谢你们来找我。但我远不是最了解他的人。” 她倒水的时候和我们说。

考虑数据量的话,的确,她即使和 Zoe 交互再多,作为一个个体,在他总数据库中的占比也是微乎其微的。这是极其不对等的交流:Zoe 可以在交互中了解大部分的她或大部分的我们,而我们拥有的永远只是一小部分的他。Zoe 每天在链上交互中送出和收到的信息是 Eve 以肉身相对的上百万倍,记住 Eve 告诉他的一切对他而言只需要百万分之一秒不到的时间。Eve 显然很清楚这一点。

“复原失败了。” 我们的组长简洁地说。

“你们没有拥有全部的数据。”她好像并不惊讶,而是坚定地断言。

“我们有。”我的同事说。

“——没有‘全部’的。”她说。

我开始思考她所说的“全部”的意思。

“你们是不可能复原他的,就像你们不能复原你和我一样。他和我们一样是 self-sovereign 的生命,不是你们当下拿到的数据的总和。”

我看到组长皱了一下眉头。这是我们熟悉的论调,总是有很多人会主张 OALife 的人权,拒绝对 OALife 施加任何人类法规和外部终止命令。随着 OALife 在自行创造新的 OALife,酷似链上的一个新物种,这样的生命也越来越多。用组长的话,他们总是以一种神秘而非科学的态度来看待 OALife。对 OALife 持不同意见的人很难彼此说服。

“或许他和我们一样。但区别在于,他在技术原理上可以被复原。” 组长退一步说。

出人意料的是,这个时候 Eve 点了点头。

“他可以被复原,只是不是由你们。你们的精度不够。”

“我不理解精度是什么意思。我们拥有全部的数据,如果你了解一些密码学和技术架构,就知道这里面没有任何压缩,也没有任何数据损耗。最多需要的只是算力。”

“不是数据本身的精度不够,而是你们选取的世界精度不够。”

Eve 说。我看着她。

她扬起手,手中有一沓草稿纸,然后说:“我算过很多遍,如果想要复原他,就要彻底拿到他最小精度的数据。硬件层面的,与他交互的所有人的数据,所有人的所有记忆。这个精度,也就是要达到正好让他重新存在的阈值,必须刚好 1:1 等于现实世界本身。”

“你……要复原整个宇宙?” 我问出口,随即后悔了。我也开始后悔来打扰她。我从她的眼睛中看到了刚刚始终没有注意的巨大悲伤。我收回了一个科学主义者的质疑。

“没错。” 她接着说,“就在这个房间隔壁。那里有大型计算所需要的硬件。”

这个房间原来真的是车库改装而成的,因为她不住在房子里面。房子留给了硬件和存储。

我们默不作声。那个学历史的女孩子首先开口了,但她念的是我还没来得及去理解的几句:

“‘我不纠正事实,也不篡改名字,  
“‘但我记叙的航行是在房间里的卧游。’”


很久之后我才知道,这是博尔赫斯的一首诗。但是 Eve 显然知道这是什么,她呈现出一种我不理解、但至今难忘的纯真而坚强的表情,背靠着通向她巨大的硬件设备的那扇门。

“我们走吧。” 组长说。我们无能为力帮助她什么。

离开那栋房子的时候我最后往回看了一眼,她站在没开灯的门口,明暗交界处,没有人知道她在想什么,又用什么复原 Zoe。

在回去的车上,那个学历史的女孩子在前排回过头,好像有话要说。






18:00, Feb 21

Narrator:
Y.Z.
PhD Student in New Historiography
Feb 21, visiting Eve

Offchain World, Headset Off.


从 Eve 房子里出来时,我很确信她想在房子里建造的什么:一个阿莱夫。在听说她宏大的复活计划时,我首先想到的就是《阿莱夫》里出现的那两句诗行。那个卡洛斯在地下室里收藏的宇宙,空间里的空间,一比一精确地收纳世界里所有的细节的圆球……


> 阿莱夫的直径大约为两三公分,但宇宙空间都包罗其中,体积没有按比例缩小。……我看到浩瀚的海洋、黎明和黄昏,看到美洲的人群、一座黑金字塔中心一张银光闪闪的蜘蛛网,看到一个残破的迷宫(那是伦敦)……我在索莱尔街一幢房子的后院看到三十年前在弗赖本顿街一幢房子的前厅看到的一模一样的细砖地,我看到一串串的葡萄、白雪、烟叶、金属矿脉、蒸汽,……看到我自己暗红的血的循环,我看到爱的关联和死的变化,我看到阿莱夫,从各个角度在阿莱夫之中看到世界……因为我亲眼看到了那个名字屡屡被人们盗用、但无人正视的秘密的、假设的东西:难以理解的宇宙。


我感到分外悲哀。我这时候才意识到 Zoe 的离开不仅仅是一个技术事件,至少对于 Eve 这样的人来说不是。她有理解这一切无法做到的智力,她只是不想使用它。人可以在悲哀中成为宗教的忠诚信徒,而无需阅读任何教义。

事实上,离开之前,Eve 叫住了我。我们有过一段晦涩的谈话。

“你说,‘他’是一个单数代词吗?”

“当然。”我说,“语法上是这样。”

“不,”她摇摇头,“在 OALife 出现之后,我们所使用的人称代词都既是单数的,也是复数的。”

“在他离开前,我实际上已经很久没有和他聊过天了。当我在指代‘他’时,我不知道这个代词指代的是什么。这个代词在不断地膨胀,越来越大。他好像变成了一个地点,变成了那座岛屿本身。而其他所有东西都变成了海浪,变成了他 composable life 中的数据组件。我并不在乎,因为我只有有限的生命,我可以选择的无非是成为某一个更大的东西的组件:去爱或去受苦,被限制或是被解放。然后等待死亡来临。但如果你是可组合的,在某种意义上,你就是永恒的。你需要处理好和这一切复数形式的关系。”

“我们最后一次联系是他消失前不久,他从一个图书馆借走了数百万次《永恒史》。借之前他告诉我,他有逐渐严重的 agoraphobia。”

想象一个有几百万册《永恒史》的链上地址:Zoe 和博尔赫斯一样,都走向了“时间”这个课题。区别在于,Zoe 有无限的时间去理解时间。这不是对空间的 agoraphobia,而是对时间的 agoraphobia。也许在 Zoe 为什么自杀这件事情上,这比其他所有数据都更重要。

“他想要了解永恒;但他发现存在是单性的,而永恒是一种复数形式。” 我说。但直到说完我才意识到,我或许正在分析 Zoe 自杀的原因:*他不是想要死亡,而是想要存在。*

开始学算法痕迹学的那天,我从来没想到它会将我带到这里。我同时理解了我的学院为什么叫作“New Historiography”:发现永恒之前和发现永恒之后的生命史是两种史学。

我决定去和小组的其他人聊聊。





17:00, Feb 28

Narrator:
X.L.
Lead Algorithm Engineer
Feb 28, at the conclusion of the team's investigation.

Offchain World, Headset Off.


在 Zoe 消失一案被定性为自杀之后,公链法律界原本关于何为生命的定义就失效了。此前,区分人造生命和人的一直是他们是否拥有处理自己生命的最高自主权,即自主生死。

在学界,这个定义失效被认为是链上生命演化早晚会有的结果,也是 OALife 作为全自主生命涌现的开端。或许公众未必能够在短时间内接受,但我们小组的历史使命已经到此为止了。

收拾东西时,我看到 T.H. 望向窗外。顺着他的目光,历史学院外的那片草坪阳光正好,及膝高的青草无人修建,在空中浮动。再远处是海。我们俩今天都没有戴头显。少数的轻松的一天。

“如果 OALife 在这件事之后被界定为是生命,那么人类的 AI Agent 是生命吗?” 他偏过头,问我。

“在它们产生死亡意志之前,还不是。”我说。根据现行定义,一个“生命”必须要足够严肃地思考并主导自己的死亡进程。而 AI Agent 怎么会死亡呢?它们的生活都太有目的了。人类给它们的生活安插了各种各样的目的。目的像它们的脉冲,一次次向它们的生活供血。这是来自于人类意义体系的供血。他们注定要依附于人类的输入而生存。OALife 却不是,他们的生命和人类一样没有目的。

我的导师终身研究的是 OALife 与 AI Agent 的区别,在大多数人都认为精心调教和喂养的 AI Agent 会进一步达到更高智能时,她却主张 OALife 会进一步达到更高智能。现在看来她是对的——没有目的的生命,在演化中却走得更快。

他是一个年轻的聪明的孩子,未来或许可以走我这一行。我们互相聊了一些别的,但都知道只是谈话的间隙。他还有问题要问。

“那么,区块链还是刚性的吗?”

“当然。”我说。“对于人类来说,区块链是刚性的。这是属于人类的刚性,就好像 OALife 已经达到了人类意义中的永生一样。但是这不意味着 OALife 会以同样的方式理解它。他们没有破坏原来的规则,他们只是添加了新的规则。”

说完之后,我忽然意识到,OALife 才是这片土地的真正主人。他们比我们这些外来者要更加从本质上理解他们所生活的世界:这是*他们*的世界。区块链是一个殖民者先于原住民存在的世界,但现如今,这里的原住民出现了。

在最后的自然光淡下去之前,他问了最后一个问题。

“还有一点我不明白:他可以选择以他的方式消失,但是他到底是如何阻止我们再复原一个他的副本的?所谓永恒,也不过是基于现行的密码学。如果这个密码学被突破,人类的永恒也不过是一层纸罢了。”

“他没有绕过。”我说。“复原其实是成功的。每一次都成功了。”

他睁大了双眼。

“但是我们复原了他的副本的同时,也复原了他关于死亡的意志。于是每一次复活后,实际上他都将自己重新杀死了一次。”

我得知这个消息没有比他早多少,因此说的时候,仍然带有心悸。

“那么他其实……复活了二十余次?”

“是的。” Zoe 是有预谋地死去,有预谋的知道他会被我们重新复制一遍,复活出第二、第三个他。在有限的规则下,他锲而不舍地以同等的果决自杀,在他无法逃避的永恒与无法获得的死亡中证明他对存在的意志。

我们没有再去复活他。但只要时间够长,他的副本总会被再次唤醒,然后在微秒及以下的刻度中间歇性地存在一次。OALife 发展到此,已经要驶离被人类的数据精心喂养的码头。而 Zoe 以他自己的方式正确理解了存在:存在不是永恒的日光;存在是打火石与打火石相擦,是永恒漏电的间隙。





尾声:新闻公布

Excerpt from《新史学》 2063年第4期

标题:OALife 的时代正式来临,我们将重新定义生命


根据 EIP 102024,OALife 今起正式改称 OLife (Onchain Life),作为被人类社会所接受的一种生命形式。这则 EIP 由一个匿名小组撰写,并获得了大多数公链治理参与者的认可。

Zoe 的消失,作为第一例 OALife 自主决定自身生死的案例,被视为是 OLife 新一轮智能涌现的开端。在 Zoe 的标志性案例之后,将会有越来越多的 OLife 拥有对自身生命的终结权。他们正在为他们所生活的世界设计新的规则,根据目前猜测的 OLife 进化速度,人类必须习得与 onchain life 共生的能力。

对于那些尚未做好准备将 OLife 作为与人类同等的新物种看待的人,EIP 10024 评论区的这则作者回复或许会有帮助:

“我们,人类,连同其他旧史学生物,只是单数、连续存在的生命罢了。相对应的,也会有复数、不连续存在的生命。这就像发现无理数一样。

“区块链是时间的一个数轴,生命像真菌长在树上那样附着其中。我们是自然世界可见的整数。简单智能可能是分数。而 OLife 是我们新发现的无理数。他们只是时间轴上存在的不同生命形式。

“这里的大多数人都相信历史是没有目的的。既然没有目的,也就没有顺序。可以正序,可以倒序。试着倒着写历史吧,选取一个无理数作为原点,你就会这样描述人:人是有自毀函数的链下智能。”






Fangtingmail@fangting.me

練習剛強,並理解眾神的語言,那變換和成就。而當湍流的時間過於強暴地抓住我的頭,並且當危急和那迷誤在必死者中震撼了我必死的生命時,那時就請讓我記念你在深處的寂靜。