As Global Chief Innovation Officer at EYJeff Wong helps companies harness disruptive technology and prepare for the future of work.
AI is critical to the operation of the current Web 2.0. It allows internet users to find friends, news and products. It supports the core business model of Web 2.0 targeted advertising. It also drives many of the dysfunctions of the current Internet—disinformation, echo chambers, and amplified extremist content.
Today, Web3 is a high-level concept that describes a future version of the Internet in which power and benefits are shared more equitably thanks to decentralization. The core capabilities of the web will no longer be monopolized by a handful of large technology companies. In this decentralized world, users will own their data, privacy will be preserved, censorship will not exist, and rewards will be shared fairly.
According to the current paradigm, the success of artificial intelligence has been built on the concentration of the web. Now, given the hype surrounding Web3, what role will AI play in this new decentralized world? And how should we go about untangling AI’s centralizing tendencies?
Where We Are Today
Huge advances have been made in using artificial intelligence to understand the meaning of online content and combine this with understanding user needs and intent. Looking ahead requires a look at the critical challenges facing AI today: discovery, matching and filtering.
Discovery: Finding relevant content was easy when we only had three TV channels. The simplest model for addressing today’s major technical challenge of discovering relevant content is to search the web via search engines. However, web search is difficult because it is not just about finding the objectively most relevant content for a search, but also about matching that content to the individual and needs of a particular user. When I search for “Odyssey” to help my daughter with her homework, I’m looking for something other than what I happen to have misspelled Odysee, the blockchain-based video distribution service.
Matching: A generalization of discovery, matching is the challenge of matching content with users, friends with each other, or even ads with buyers. For example, in a social network, one must identify potential links between billions of users, resulting in millions and trillions of possible matches — only a small fraction of which may be relevant. This not only requires complex algorithms and huge training datasets – it also requires massive computing infrastructure.
Filtering: Filtering is the challenge of identifying content that should be removed for a specific reason. It could be due to propaganda, illegal or obscene content, Covid misinformation or copyright theft, to name a few examples. Open web platforms that enable creators to upload content they create create a huge filtering load. On the other hand, users are rightly frustrated if their compliant content is incorrectly labeled as restricted. Even distinguishing between illegal and legitimate videos is an incredibly difficult challenge for AI, requiring massive infrastructure, massive armies of human moderators, massive data sets, and processes to quickly remove offending content that escapes.
Although these applications have dominated the AI literature for more than a decade, the solutions are burdensome in terms of the expertise, infrastructure, data, and costs required to implement them effectively.
While Web3 is basically a high-level concept today, there are some early possibilities that show what it can achieve and the challenges that need to be overcome, as demonstrated by Odysee, NFT markets and Friends with Benefits.
• Odysee is a video sharing service based on the LBRY file sharing service, with both services leveraging blockchain capabilities. The philosophy is that video files can be irreversibly placed on a blockchain and then accessed via BitTorrent. There are several challenges that arise from this approach: How is illegal content removed from this blockchain, and will those hosting the service be held criminally responsible for this content?
• NFT markets are currently the biggest monetization opportunity for Web3 and face the same challenges. As the number of NFTs increases and the customer base expands, the challenges of content discovery, matching customers with relevant content, and filtering out illegal or stolen content arise just as they do in traditional marketplaces.
• Friends with Benefits (FWB) is something of a Web3 analog of a social network. Specifically, it is a decentralized autonomous organization (DAO) and a selective social union. Considered by many to be the future of social networks, these organizations face some of the same challenges that AI faces in existing social networks: discovery, matching and filtering. Which DAO do you participate in? How do you find DAOs that match your interests? This is a classic search/discovery problem.
Overcoming today’s drivers in a decentralized system
Many of the issues that AI faces in Web 2.0 also arise in Web3. However, existing centralized approaches are considered to be the opposite of the promise of Web3. The current high degree of concentration is driven by expensive talent, rapidly evolving technology, and heavy data and infrastructure requirements. The question remains: Can these be overcome in a decentralized system?
A commonly proposed solution is federated learning, where machine learning algorithms partition their learning across different (federated) datasets and infrastructures. These techniques are complex and rapidly evolving. While they are technically interesting, it is unclear whether they meet users’ privacy expectations. The result is that the leaders in federated learning are the existing large technology companies. The opportunity remains for the emergence of truly decentralized AI solutions.
Artificial intelligence and machine learning have been the driving forces behind the centralization of the web, and now the future of Web3 will first require how to decentralize machine learning. It is up to technology leaders to determine how to achieve meaningful interoperability and effectively untangle AI’s centralizing tendencies to create a more private, secure, and fair web.
The views reflected in this article are those of the author and do not necessarily reflect the views of the global EY organization or its member firms.