DWF 2024 Crypto Expectations: The Potential and Challenges of DAI, Leading the Future of AI through Web3

23-12-20 14:08

Read this article in 15 Minutes

AI summary

View the summary

The original title: "AI in Web3: Navigating AI's Future with Web3"

The original author: DWF Labs Research

The original translation: Sharon, Luccy, BlockBeats

Editor's note: In the past year, the launch of ChatGPT 3.5 has sparked concerns and heated discussions about AI. Vitalik also pointed out in his article that many people are worried about the emergence of a monopolistic version of AI, and therefore tend to delay its progress. DWF Labs Research deeply analyzes the impact of ChatGPT 3.5's breakthrough on AI in the Web3 era, revealing the challenges AI faces and the potential of DAI (decentralized AI). BlockBeats has translated the original article as follows:

As the year draws to a close, we will explore one of the hottest topics of the year - artificial intelligence (AI). Over the past year, AI has become the focus of discussion, stemming from the release of ChatGPT 3.5 by OpenAI. This release demonstrated the enormous economic potential of AI, sparking global discussions about its future, impact, and related risks.

With the growth of optimism, skepticism has also emerged, and the potential consequences have begun to alert regulatory agencies. Due to the rapid rise of AI and the vague regulatory framework, it is reminiscent of the early stages of the cryptocurrency industry. The two industries have been compared, highlighting the decentralized nature of Web3, which seems to complement the potential centralizing power of AI.

Soon, almost every Q1 Web3 venture capital discussion focused on the transformative potential of AI (sometimes I wonder if I'm attending a Web3 event or an AI event). During this year, we also saw some venture capital firms shift towards AI or incorporate it into their investment portfolios.

As the hype around AI gradually subsides, DWF Ventures now hopes to take a fair and impartial look at the AI field. This article briefly outlines the evolution of AI and how it has reached its current level of popularity. However, the narrative of the article is different, as we shift our focus from the traditional concern of how AI affects Web3 to exploring the opposite question - how Web3 affects AI. In this exploration, we delve into how decentralization and Web3 act as catalysts to address the challenges currently facing AI.

AI Overview and Breakthrough of ChatGPT 3.5

Source: Khan, Pasha, & Masud, 2021

Contrary to the recent hype surrounding AI, its history can be traced back to the 1930s. Turing's work in 1950, such as the Turing test, laid the foundation for AI. Although there was optimism about AI in the early days, the "AI winter" of the 1970s was caused by computational barriers and the inability to meet real-time demands. In the 1980s, expert systems revitalized AI by using knowledge databases to simulate human expertise. This era also witnessed the revival of connectionism and the rise of recursive neural networks.

However, expert systems faced challenges in knowledge acquisition and real-time analysis, leading to a decline in the 1990s, and the performance of personal computers also resulted in a gradual decrease in relevance. Over the years, the AI field has developed rapidly, branching out into various technological fields such as machine learning, natural language processing, computer vision, and speech recognition. These developments have enabled AI to move from simple problem-solving to deep learning in complex application areas.

Source: Mukhamediev et al., 2022

During the development process, AI has undergone integration in various sub-fields. Among them, the fields of machine learning and large language models (LLM) have made significant progress in transforming vertical domains. The paper "Attention is All You Need" by Ashish Vaswani and others clearly inspired the Generative Pre-trained Transformer (GPT) model.

Afterwards, a large number of GPT models emerged, such as the bidirectional "BERT" GPT and OpenAI team's GPT. Following ChatGPT, open-source alternatives such as Falcon and LLaMA2 appeared, intensifying competition for the next generation of GPT iterations, potentially closer to artificial general intelligence (AGI).

The hype around GPT has helped to liberate AI from academia and gain the attention of billions of people. Within two months of its release, OpenAI achieved the fastest growth rate with 100 million weekly active users. According to a recent study by McKinsey, approximately 51% of technology industry professionals currently use AI in their work.

AI Reality: Guiding Social Cognition and Its Limitations in Centralized AI

Vitalik Buterin's latest survey in his article shows that many people are concerned about the emergence of monopolistic versions of AI, and therefore tend to delay its progress.

Source: My techno-optimism

The recent surge in concerns about AI can be traced back to the rapid rise of ChatGPT, whose human-like responses are a driving factor. However, most people are not aware that although GPT mimics human interaction, it is not a general AI (AGI).

Every time GPT generates an output, it is variable in statistics, lacking consistency and factual accuracy guarantees. In addition, GPT also faces other limitations, but its most prominent drawback is the inability to perform logical reasoning, especially in mathematics.

Source: "The Limitations of GPT Language Model in Few-Shot Learning"

Given the numerous concerns surrounding AI and the existing challenges in managing large-scale AI models efficiently, exploring the integration of Web3 and AI has become a potential way to alleviate the challenges faced by AI. By leveraging the inherent decentralization and distributed computing principles in Web3, it is hoped that current issues faced by AI systems can be addressed.

DAI (Decentralized AI) Road: Overview, Potential, and Challenges

Due to the concentration of AI capabilities in centralized systems, concerns have been raised about data access, model relevance, and overall sustainability of AI applications. Centralized AI systems face significant obstacles, particularly for proprietary large-scale datasets.

Source: Elon’s tweet

(Note: The content contains HTML tags and a hyperlink, which will not be translated. The Chinese characters will be translated, but industry-specific terms and names will not be translated.)

This led to billing by query, and X set a daily limit on the number of post views. Soon after, the releases of Grok, X GPT allowed users to access X's data in real time. This model created economic barriers and raised concerns about the accessibility and inclusivity of AI benefits.

In addition, due to the rapid obsolescence of published models, there will be significant challenges in maintaining relevance and accuracy without continuous data updates. Currently, the training data for ChatGPT 3.5 includes information up to January 2022. Llama 2 was also trained on data from January 2023 to July 2023.

In response to these challenges, DAI has emerged as a potential solution to the limitations of centralization.

Source: (Janbi et al., 2023)

DAI presents an alternative trajectory to address the inherent challenges of centralized models. A recent meta-analysis paper by Janbi et al. serves as a comprehensive guide, detailing the five main areas of DAI.