Author: Thomas Hsueh
The recent blockchain development has produced some new technology primitives: cryptocurrency, DEX, NFT and DAO.
They are interesting for many reasons-ideology, intellectuality, egalitarianism, etc. But the reason why it arouses my strong interest is because they, together with the continuous progress of artificial intelligence research and semiconductor technology, point out the possibility of a new human world, that is, money will no longer consume people’s attention, and money acts as a The invisible pillars of the world’s infrastructure have become “solved problems.”
When I look back on my life, I find that as time goes by, I spend most of my sober time trying to create and preserve financial wealth. I am disappointed; this in itself does not arouse my interest, but I I am still caught by a perceived need, although I am fundamentally interested in imagining an interesting future version and working hard to promote its realization. I guess the same is true for most people in civilized society. I further guess that for most people, the act of creating and preserving financial wealth (by creating goods and services that are valuable to others) is not attractive in itself; it is attractive only because it leads to financial The accumulation of wealth, whether it is the certainty in the employment contract, or the probability in the investment. The interestingness of this process is another problem, and it happened by coincidence. But why accumulate financial wealth?
I know that financial wealth is an abstract concept, so it is completely unnatural to humans. However, I admit that when any non-trivial financial wealth is given to me, whether it is given by my employer or by the market, I have an internal reaction. This means that the accumulation of financial wealth as an unnatural concept has been linked to birth desires, which means that there are some things I really want to do that require a certain degree of financial wealth, but I lack them; these unsatisfied desires prompt I pursue financial wealth as an intermediary target. But isn’t it ridiculous to spend most of your waking time making money (or recovering from the exhaustion of making money with numbing entertainment and other activities)?
I believe we can do better. I believe there is a way to greatly increase the interest and possibilities of many people’s lives by eliminating the need to accumulate financial wealth in our civilization. I am interested in creating a world in which the accumulation of financial wealth is no longer a concern for ordinary people. It is a world of extraordinary wealth.
Why do cryptocurrencies, DEX, NFT, DAO and AI all point to the possibility of a new human world?
It will be more obvious when people look at them from a specific angle.
Source: Crypto artist jack butcher ([email protected])
Cryptocurrency can be seen as a completely autonomous financial value-priced infrastructure, which is trusted, trusted and relied upon by more and more human populations; it incentivizes people to participate in growth and maintain themselves (for example, profitable mining or Pledge business).
DEX can be seen as a fully autonomous cryptocurrency exchange market; it also incentivizes people to participate in order to develop and sustain themselves (for example, Balancer that automatically rebalances through external arbitrageurs; I wrote this in Chinese last year ).
NFT , for now, can be seen as a pointer on the blockchain, pointing to something interesting elsewhere, which is somewhat similar to a pointer in memory, indicating a specific data or program sequence position.
DAO , for now, can be regarded as a primitive means for entities based on blockchain to make collective decisions.
Finally, artificial intelligence can be regarded as a very broad field of work, and its ultimate goal is to create a truly autonomous subject, usually on a silicon base, which may or may not have a physical manifestation. Artificial intelligence can not only make intelligent decisions gradually (such as discriminative models in ML terms, reinforcement learning is a powerful algorithm category), but also create interesting and valuable data (such as generative models in ML terms, where GAN is A promising direction).
Combining all these concepts, it is not far-fetched to conceive a future version. In this version, Crypto autonomous entities exchange assets in the DEX market through crypto currency, creating and manipulating data (non-code Crypto assets), and program sequences (code fragments/manual Intelligent model), and real-world assets (contracts) linked to NFT, and collectively make decisions through DAO. This autonomous subject layer running on the cryptocurrency infrastructure can handle the allocation of financial resources, as well as the creation and delivery of Crypto goods and services, and physical goods and services. Its efficiency and persuasiveness are so high that most of us no longer think about it. In accumulating financial wealth, just as most of us don’t focus on access to water, food, electricity, mobility (solved by the industrial revolution and aviation progress), electrical appliances (solved by market mechanisms), and information (via network technology). Solve) the same.
Imagine this: one day, we will no longer focus most of our energy on money. Money becomes a “solved” problem. Of course, something else will become a new scarcity and become the object of most people’s pursuit. According to USV’s Albert Wenger, this will be attention.
Although the future seems unpredictable, we can passively* make educated guesses based on evidence, or actively influence its progress. In this section, I provide evidence and observations that support the aforementioned future versions.
*We can question whether passivity is possible, for example, considering the observer effect of physics, even if it is metaphorical.
1-Artificial intelligence workload growth and continued Moore’s Law
In 1956, a device the size of a large closet was needed to store 5 megabytes of data
In the past ten years, investment in artificial intelligence chip companies has become increasingly hot . There are many catalysts for this (Moore’s Law and the slowdown of Dannard’s expansion, the drive of hardware specialization, the increased demand for artificial intelligence computing efficiency for large-scale recommendation engines established by social networking companies, etc.), the most recent example is Sambanova, Its D round of financing has reached a valuation of $5 billion.
Another promising clue is growing evidence that the potential of photonics can not only be restored, but also turbocharged by Moore’s Law. Lightmatter’s upcoming products apply photonics to the linear mathematical part of accelerating artificial intelligence workloads. Compared with the current technology level (such as Nvidia’s A100), it is estimated that it can increase the artificial intelligence calculation speed by 10 times at 1/10 of the power. (That is 100 times the energy efficiency; if wavelength division multiplexing is used, the performance improvement will be greater; see the recent Forbes report on Lightmatter ) Photonic interconnection further accelerates the communication between chips, while saving power, making more More and more large-scale and complex artificial intelligence models have been deployed. (Disclosure: At the time of writing, I am a Crypto design engineer at Lightmatter.)
The vigorous development of the entire technology industry in the past few decades is fundamentally driven by Moore’s Law; if the continuation of Moore’s Law depends on the prioritization and specialization of artificial intelligence workloads, the entire technology industry will follow suit. That is, where there is higher computing power, the application will follow (a metaphor is the huge surplus of network bandwidth created by the capital raised during the Internet bubble, which later enabled countless applications such as social networks to be realized). Recall how the availability of GPUs promoted deep learning in ILSVRC (AlexNet) in 2012. As photonics increases the “capacity” of artificial intelligence by 100 times, a large part of our future technology will be driven by artificial intelligence. The only problem is how to drive.
Robotic process automation will realize our common vision of “what artificial intelligence means to labor”: substitution. Human tasks are automated by the RPA program. This clearly shows how the services provided by humans (work that is paid and priced) are completed by an automated process, that is, code.
3-Organic Evolution of NFT Space
Autoglyphs is the first “on-chain” generated art on the Ethereum blockchain. They are a completely independent mechanism for creating and owning a piece of art.
From the ERC-721 proposal, the initial imagination that NFT can realize is to represent the ownership of “physical properties, virtual collectibles, and negative assets (such as loans)”. Two issues have sparked fierce debate from the beginning: where is the Crypto content represented by the NFT, and if the Crypto content bound to the NFT can be accessed and viewed by the public, what does the NFT owner really own? The answer to the first question is either on the chain (occupying the storage space of a blockchain that usually has limited space), or on a separate storage solution (such as IPFS). The answer to the second question is the right to the content, but not the content itself . Interestingly, with the rise of generative art in many fields (artificial intelligence research, PCG technology in computer games and animation, etc.), new projects have been created: NFT has nothing to do with the content of the artwork itself, but with the generation of artwork The content process is related , and this process is a compact code snippet completely contained in the NFT itself on the chain. Autoglyph from Larva Labs is an example.
This article  summarizes other examples in this area (such as Neolastics). The right to own the “process” instead of the result is an interesting idea -imagine owning a Crypto clone of Vincent Van Gogh instead of owning a piece of his artwork; you can ask your Van Gogh clone to create more Artwork. In other words, what you have is dynamic and vital, which is in sharp contrast with the static files represented by NFTs today. This is expanded from static screenshots to the time dimension.
Another organic development in this field is AlchemyNFT  recently announced by the AlphaWallet team, which allows the creation and manipulation of synthetic NFTs. All the evidence shows that, in addition to speculative trading cards, the potential of NFTs is still untapped.
4-Artificial intelligence and games; the race to build a meta-universe
In order to achieve truly sudden content and spontaneous experience, artificial intelligence must become an integral part of our future games, or as many people say, when the scale of the game expands to swallow people’s daily life, it is the meta-universe. A16Z once published an article “Meet Me in Metaverse”, which outlines the rationale very well and traces the recent progress in this direction.
The four stages of content development, according to A16z, we are transitioning from 1-2
In fact, technology giants have already set foot in this direction, albeit unsuccessfully. Google Stadia has created an internal game studio, trying to create ML-driven games. The most obvious example is their Project Chimera (MCV/Develop had a good interview on Stadia’s efforts in ML game development in April 2020; see pages 34-38). Amazon also joined the dance floor, and Jeff Bezos personally instructed to “make games with a ridiculously large amount of calculation.” So far, neither Google nor Amazon has successfully released games in this direction; Stadia recently closed its internal studio, and Amazon announced a series of bad news about its game development work. Facebook’s RivalPeak is a very interesting artificial intelligence entertainment project experiment, showing signs of early appeal. It can be said that due to complicated corporate politics and procedures, large companies, no matter how powerful they are, will not make critical progress here, because creating ML-driven games will be a huge paradigm shift, far away from the existing game development process. , Requires a lot of flexible trial and error to enter the unknown.
RCT’s Morpheus engine uses deep reinforcement learning to render text from the natural world’s object library into 3D assets and animations. This GIF shows the rendering of “a person is walking”.
In the field of entrepreneurship, RCT studios located in Los Angeles and Beijing  are making gratifying progress by adding reinforcement learning to the game runtime. Agence is a close cooperation between filmmakers, artists and Google’s machine learning experts, and produced a “dynamic movie” that uses reinforcement learning to realize the three-party co-creation of game narratives by film creators, audiences, and artificial intelligence. .
From the perspective of large game developers, Roblox , which recently went public, is dedicated to the development and deployment of huge machine learning models. Gabe Newell, the co-founder of Valve and the creator of Steam, pointed out that single player games driven by artificial intelligence may completely surpass multiplayer games in 9 years (2030) .
Thinking from the perspective of first principles, when virtual and dynamic worlds swallow our daily lives, the goods and services (experiences) provided in these worlds must rely on specific economic infrastructure. My point is simple: when artificial intelligence becomes the main producer of Crypto assets (3D models, skins, sound effects, personality configurations, etc.), it is not just Crypto assets that need to be tokenized, but artificial intelligence itself, whether it is The code form is also a statistical model parameter.
5-The emergence of the data economy promotes the training of artificial intelligence
Trent McConaghy of Ocean Protocol pointed out this direction in 2018  and is committed to building a data economy. Ocean Protocol’s mission is to establish a market for the release, discovery and consumption of data assets and for AI training to protect privacy.
6-The possibility of personal data explosion: BCI
I wrote about Elon Musk’s NeuralLink from the perspective of human-computer symbiosis last year. With the increasing maturity of BCI (Brain-Computer Interface) and its ultimate adoption by humans, the bandwidth for uploading brain data for each person will be greatly increased, allowing each person’s daily mental activities, preferences and tastes to be handled by artificial intelligence. How do we treat these real-time and fine-grained personal data? We have to mark them as ownership (imagine if a big technology company owns your thoughts; you will obviously demand ownership of your thoughts and their derivatives), and privacy is the biggest concern here. However, the current NFT standard does not support ownership of content, which may be completely incompatible with brain data tokenization. One method is to protect privacy through compression, such as synthetic data generation (matching probability density functions by training a statistical model), which means that we will not label the collected brain data itself, but on the brain data trained on these brain data. AI for marking.
In this section, I outline a rough course of action for the future of creating the above version.
From the long-term vision, we will have to create a decentralized automata , or DA (the name comes from the decentralized autonomous organization DAO-DAO will be owned by DA). DA exists on the blockchain, and access to code execution on the chain is limited to the owner of its pointer. DA can trade the code as an asset and reassemble its code base. Therefore, we also need to establish a software/AI model architecture, which is safe, composable, decomposable and variable.
Short-term research and development is to establish token standards and economic mechanisms for the ownership of the code/AI model (not only the ownership of rights, but also the exclusive use of the code/AI model) and its composition/decomposition/variation. The synthetic NFT referred to here is designed for artificial intelligence code. A simple conceptual solution is to store the code in fragments, as in Filecoin, whose position is reshuffled every time the pointer changes hands.
Mid-term R&D is to establish DA’s software and blockchain standards. We will promote the realization of the following:
• Source control (such as Git) is to manage code versions and merge code for DA.
•DA will conduct a code review process (rule-based/simulation-based).
• Codes, models and parameters are traded by DA on the blockchain in order to improve themselves in terms of self-defined indicators.
• Long-term research and development is to guide DA and related DAs into a whole autonomous economy, which I call the autonomous dimension.
In this version of the future, the Autonomous Dimension will plan the most advanced problems into interactive challenges for individual humans, and humans will be rewarded by overcoming these challenges ; the rewards are likely to be non-financial (the universal basic income is also implemented by DAs that take over the industry on a large scale. ). In this way, ordinary tasks will be handled by machines; most of human attention is focused on the frontiers of civilized capabilities (civilization = humans + machines). Simply put, solved problems will be handled by machines on a large scale; humans will only work on unsolved problems. There is no duplication of work; only those who advance the frontier of civilized capabilities will be rewarded beyond the universal basic income.