How does multi-dimensional analysis of DePIN help artificial intelligence?

In the past, start-ups, with their speed, agility, and entrepreneurial culture, were free from the shackles of organizational inertia and led technological innovation for a long time. **However, all this is rewritten by the era of artificial intelligence. **To date, the creators of breakthrough AI products have been traditional technology giants such as Microsoft's OpenAI, Nvidia, Google, and even Meta.

**what happened? **Why did the giant win over the start-up this time? Startups can write great code, but they face several obstacles compared to the tech giants:

• Computational costs remain high
• AI development has a reverse lobe: Concern and uncertainty around AI's societal impact hinder innovation due to lack of necessary guidelines
• AI black box problem
• "Data moats" built by big technology companies form barriers to entry

So, why is blockchain technology needed? Where does it intersect with artificial intelligence? Although not all problems can be solved at once, the Distributed Physical Infrastructure Network (DePIN) in Web3 creates the conditions to solve the above problems. The following will explain how the technology behind DePIN can help artificial intelligence, mainly from four dimensions:

• Reduce infrastructure costs
• VERIFY Creator and Personality
• FILL AI Democracy and Transparency
• Setting data contribution reward mechanism

Below:

• "web3" refers to the next-generation Internet, and blockchain technology and other existing technologies are its organic components.
• "Blockchain" refers to decentralized and distributed ledger technology.
• "Crypto" refers to the use of token mechanisms for incentives and decentralization.

1. Reduce infrastructure costs (computing and storage)

Every wave of technological innovation starts with something expensive becoming cheap enough to waste.

– Society’s Tech Debt and Software’s Gutenberg Moment, via SK Ventures

How important is the affordability of infrastructure (the infrastructure of artificial intelligence refers to the hardware cost of computing, transmitting and storing data), Carlota Perez's theory of technological revolution has indicated that the theory proposes that technological breakthroughs include two stages:

Source: Carlota Perez's Technological Revolution Theory

• Installation phase is characterized by heavy venture capital investment, infrastructure building and "push" go-to-market (GTM) strategies, as customers do not understand the value proposition of the new technology.
• Deployment stage is characterized by a large increase in infrastructure supply, lowering the threshold for attracting newcomers, and adopting a "pull"** market promotion (GTM) strategy,** indicating a high degree of product market matching, Customers expect more products that have yet to be formed.

Now that attempts such as ChatGPT have demonstrated market fit and customer demand, one might feel that AI has entered the deployment phase. **However, AI is missing an important piece: excess infrastructure for price-sensitive start-ups to build and experiment with. **

question

The current physical infrastructure field is mainly monopolized by vertically integrated oligopoly, including AWS, GCP, Azure, Nvidia, Cloudflare, Akamai, etc. The industry has high profit margins. It is estimated that the gross profit margin of AWS on commodity computing hardware is 61%. Therefore, new entrants in the AI field, especially the LLM field, have to face extremely high computational costs.

• The cost of ChatGPT training is estimated at 4 million US dollars, and the operating cost of hardware inference is about 700,000 US dollars per day. *Bloom version 2 could cost $10 million to train and retrain.
• If ChatGPT enters Google Search, Google revenue will be reduced by $36 billion, **Huge profits will be shifted from software platforms (Google) to hardware providers (Nvidia). **

Source: Layer by Layer Analysis - LLM Search Architecture and Cost

solution

DePIN networks such as Filecoin (the pioneer of DePIN originated in 2014, focusing on gathering Internet-level hardware and serving distributed data storage), Bacalhau, Gensyn.ai, Render Network, ExaBits (coordination layer for matching CPU/GPU supply and demand) can Save 75% to 90%+ of infrastructure costs through the following three aspects:

1. Push the supply curve and stimulate market competition

DePIN provides equal opportunities for hardware suppliers to become service providers. It creates a market where anyone can join as a "miner" and exchange CPU/GPU or storage power for financial compensation, thereby creating competition for existing providers.

While a company like AWS undoubtedly enjoys a 17-year head start in user interface, operations, and vertical integration, **DePIN attracts a new customer base that cannot accept pricing from centralized suppliers. **Just like Ebay does not directly compete with Bloomingdale, but instead provides more economical alternatives to meet similar needs, distributed storage networks do not replace centralized suppliers, but are designed to serve price-sensitive users group.

2. Promote market economic balance through encrypted economic design

The subsidy mechanism created by DePIN can guide hardware suppliers to participate in the network, thereby reducing the cost of end users. In principle, we can look at the costs and revenues of AWS and Filecoin storage providers in Web2 and Web3.

**Customers get price reduction: **DePIN network creates a competitive market and introduces Bertrand-style competition, thereby reducing customer payment fees. In comparison, AWS EC2 needs about 55% margin and 31% overall margin to stay afloat. The Token incentive/block reward provided by the DePIN network is also a new source of income. In the context of Filecoin, the more real data a storage provider hosts, the more block rewards (tokens) it can earn. **Therefore, storage providers have an incentive to attract more customers to close deals and increase revenue. **The token structures of several emerging computing DePIN networks remain undisclosed, but likely follow a similar pattern. Similar networks include:

• Bacalhau: A coordination layer that brings computation to where data is stored, avoiding moving large amounts of data.
• exaBITS: A distributed computing network serving AI and compute-intensive applications.
• Gensyn.ai: Deep Learning Model Computing Protocol.

3. Reduce overhead costs: The advantages of Bacalhau, exaBITS and other DePIN networks and IPFS/content addressable storage include:

• Unlocking latent data availability: Large volumes of data are currently untapped due to the high bandwidth cost of transmitting large data sets, such as the massive event data generated by sports stadiums. The DePIN project can process data on-site and transmit only meaningful output, unearthing potential data availability.
• Reduced Operational Costs: Reduce data entry, transfer and import/export costs by acquiring data locally.
• **Minimize manual work in sensitive data sharing: **If hospitals A and B need to combine sensitive data of their patients for analysis, they can use Bacalhau to coordinate GPU computing power and directly process sensitive data locally without having to Exchange personally identifiable information (PII) with counterparties through cumbersome administrative processes.
• **No need to recalculate the basic data set: **IPFS/content addressable storage comes with the ability to deduplicate, trace and verify data. For the functions and cost performance of IPFS, please refer to this article.

AI Generated Summary: AI requires affordable infrastructure provided by DePIN, and the infrastructure market is currently dominated by vertically integrated oligopolies. DePIN networks like Filecoin, Bacalhau, Render Network, ExaBits democratize the opportunity to become a hardware supplier, introduce competition, maintain market economic balance through cryptoeconomic design, reduce costs by more than 75%-90%, and reduce overhead costs .

2. Verify creator and personality

question

A recent survey shows that **50% of AI scholars believe that the possibility of AI causing devastating harm to humans exceeds 10%. **

People need to be alert that AI has caused social chaos, and there is still a lack of regulation or technical specifications. This situation is called "reverse lobe".

For example, in this Twitter video, podcast host Joe Rogan and conservative commentator Ben Shapiro are debating the movie "Ratatouille", but this video is AI-generated.

Source: Bloomberg

It’s worth noting that AI’s social impact extends far beyond the problems posed by fake blogs, conversations, and images:

• During the 2024 U.S. election, AI-generated deepfake campaign content achieved the effect of being fake for the first time. *A video of Senator Elizabeth Warren was edited to have her "say" things like "Republicans shouldn't be allowed to vote" (rumor debunked).
• Speech-synthesized Biden's voice criticizes trans women.
• A group of artists has filed a class action lawsuit against Midjourney and Stability AI, alleging unauthorized use of artists' work to train AI, copyright infringement and threats to artists' livelihoods.
• The AI-generated song "Heart on My Sleeve," featuring The Weeknd and Drake, went viral on the streaming platform but was later pulled. When new technology enters the mainstream without regulation, it creates many problems, **copyright infringement is a "reverse lobe" problem. **

So can we add AI-related specifications to Web3?

solution

Provide personality proof and creator proof by using the proof of origin on the encrypted chain

Make blockchain technology truly work - as a distributed ledger containing an immutable on-chain history, the authenticity of digital content can be verified through content cryptographic proofs.

Digital signature as proof of creator and proof of personality

To identify a deepfake, a cryptographic proof can be generated using a digital signature unique to the creator of the original content. The signature can be created using a private key known only to the creator and verifiable by a public key that is available to all. Having a signature proves that the content was created by the original creator, whether human or AI, and verifies authorized or unauthorized changes to the content.

Using IPFS and Merkle tree for authenticity proof

IPFS is a distributed protocol for referencing large datasets using content addressing and Merkle trees. In order to prove that the content of the file was received and changed, a Merkle proof is generated, which is a string of hashes showing the position of a specific data block in the Merkle tree. With each change, a hash is added to the Merkle tree, providing proof of the file modification.

**The pain point of the encryption scheme is the incentive mechanism. **After all, identifying the deepfake maker can reduce the negative social impact, but it will not bring the same economic benefits. This responsibility is likely to fall on mainstream media distribution platforms such as Twitter, Meta, and Google, and it is indeed the case. **So why do we need blockchain? **

The answer is that blockchain's cryptographic signatures and proofs of authenticity** are more efficient, verifiable and certain. **Currently, the process of detecting deepfakes is mainly through machine learning algorithms (such as Meta's "Deepfake Detection Challenge", Google's "Asymmetric Numerals" (ANS) and c2pa: to identify regularities and anomalies in visual content,**but often It is not accurate enough and lags behind the development speed of deepfake. **Generally, manual review is required to determine the authenticity, which is inefficient and expensive.

If one day every piece of content has a cryptographic signature, everyone can verifiably prove the source of creation, flagging tampering or forgery, then we will usher in a beautiful world.

AI Generated Summary: AI may pose a significant threat to society, especially deepfakes and unauthorized use of content, while Web3 technologies such as Proof of Creator using digital signatures and Proof of Authenticity using IPFS and Merkle trees, The authenticity of digital content can be verified, preventing unauthorized changes and providing norms for AI.

3. AI democratization

question

AI today is a black box made of proprietary data and proprietary algorithms. The closed nature of LLM, a large technology company, has killed "AI democracy" in my eyes, that is, every developer and even user can contribute algorithms and data to the LLM model, and get part of the profit when the model is profitable (related articles).

AI Democracy = Visibility (can see the data and algorithms input into the model)** + Contribution** (can contribute data or algorithms to the model).

solution

The purpose of AI democracy is to make generative AI models open to, relevant to, and owned by the public. The table below compares the current state of AI with the future that can be achieved through Web3 blockchain technology.

at present--

For customers:

• One-way receiving LLM output
• Cannot control how personal data is used

For developers:

• Low composability
• ETL data processing is not traceable and difficult to reproduce
• The source of data contribution is limited to the data owner
• Closed-source models can only be accessed via API for a fee
• Shared data output lacks verifiability, and data scientists spend 80% of their time on low-end data cleaning

After combining the blockchain——

For customers:

Users can provide feedback (such as bias, content moderation, granular feedback on output) as a basis for fine-tuning

Users can choose to contribute data in exchange for the profit after the model is profitable

For developers:

• **Distributed data management layer: **Crowdsourcing repetitive time-consuming data labeling and other data preparation work
• Visibility & the ability to combine & fine-tune algorithms, with verifiable sources (a tamper-proof history of all changes can be seen)
• Data sovereignty (achieved through content addressing/IPFS) and algorithm sovereignty (for example, Urbit realizes the point-to-point combination and portability of data and algorithms)
• **Accelerates LLM innovation, **Accelerates LLM innovation from various variants of the underlying open source model.
• Reproducible training data output, achieved through an immutable record of past ETL operations and queries on the blockchain (such as Kamu).

Some people say that the open source platform of Web2 also provides a compromise solution, but the effect is not ideal. For related discussions, see the blog post of exaBITS.

AI generation summary: The closed LLM of big tech companies kills "AI democracy", that is, every developer or user can contribute algorithms and data to an LLM model, and get a part of the profit when the model is profitable. AI should be open to the public, relevant to the public, and owned by the public. With the help of the blockchain network, users can provide feedback, contribute data to the model in exchange for realized profits, and developers can also obtain visibility and verifiable sources to combine and fine-tune algorithms. Web3 innovations such as content addressing/IPFS and Urbit will enable data and algorithmic sovereignty. Reproducibility of training data output will also be possible through blockchain's immutable record of past ETL operations and queries.

4. Set up data contribution reward mechanism

question

Today, the most valuable consumer data is the exclusive asset of large technology companies, forming a core business barrier. Tech giants have no incentive to share this data with outside parties.

So why can't we get data directly from its creators or users? Why can't we make data a public resource, contribute data and open source it for data scientists to use?

Simply put, it is because of lack of incentive mechanism and coordination mechanism. Maintaining data and performing ETL (extract, transform and load) is a large overhead cost. In fact, data storage alone will be a $777 billion industry in 2030, not including computing costs. Nobody undertakes the work and costs of data processing for free.

Let's take a look at OpenAI. It was originally set to be open source and non-profit, but it is difficult to realize the cost and cannot cover the cost. In 2019, OpenAI had to accept capital injection from Microsoft, and the algorithm was no longer open to the public. OpenAI is expected to make $1 billion in revenue by 2024.

solution

Web3 introduces a new mechanism named "dataDAO", that facilitates the redistribution of income between AI model owners and data contributors, creating an incentive layer for crowdsourced data contributions. Due to space limitations, it will not be expanded here. If you want to know more, you can read the following two articles:

• How DataDAO works/DataDAO principle, authored by HQ Han of Protocol Labs
• How data contribution and monetization works in web3/web3 How does data contribution and monetization work? In this article, I discussed in depth the mechanism, shortcomings and opportunities of dataDAO

In general, DePIN has taken a different approach and provided new hardware energy for promoting Web3 and AI innovation. While tech giants dominate the AI industry, emerging players can leverage blockchain technology to join the fray: DePIN Network lowers barriers to entry by lowering computing costs; the verifiable and distributed nature of blockchain enables truly open AI It becomes possible; innovative mechanisms such as dataDAO encourage data contribution; the immutability and tamper-resistant features of the blockchain provide the creator's identity certificate, dispelling people's concerns about the negative social impact of AI.