Centralization and decentralization both have their pros and cons. In this post, I want to explore the opportunities and challenges posed by these two paradigms in two contemporary technological contexts: Social media and Artificial Intelligence (AI).
In the realm of social media, Twitter has become a hotbed of controversy for numerous reasons, primarily due to perceived mismanagement by CEO Elon Musk. However, the Herculean task of moderating a gigantic social media platform with a single set of guidelines is formidable, regardless of who is at the helm. While centralization offers an impressive level of control, it also demands a staggering degree of content moderation, particularly when millions of users are generating and sharing vast amounts of content daily.
Decentralized, or “federated,” alternatives to Twitter’s centralized model, like Mastodon and Bluesky, are gaining traction. The controversies surrounding Musk’s management of Twitter have likely boosted awareness and usage of these relatively new platforms. However, these decentralized platforms present their own challenges. While Mastodon delegates content moderation to individual server managers, it also introduces a greater level of complexity and technical requirements for users, compared to Twitter. This recently prompted Mastodon’s nonprofit parent company to begin onboarding new users on a central server, allowing them to migrate to a different server later. This approach may help Mastodon expand its user base faster, but the unique benefits and challenges of moderation in a federated or decentralized environment persist.
The social media platform Blue Sky, an alternative to Twitter, defaults new user accounts to a comparatively stringent set of content moderation choices. Users can also subscribe to “block lists” maintained by other organizations or individuals. However, the platform’s standards mandate transparency for these block lists, which may stir up drama and controversy as the platform moves out of beta and gains more users.
As an aside, I’ve been using Mastodon intermittently since 2017, and extensively over the past year. I’ve only been using Blue Sky for about a week. I prefer Mastodon due to its larger user base and its support for list-like features, similar to Twitter. Anticipating either the collapse of Twitter or my eventual decision to abandon it, I find myself sharing most of my social media links on up to seven different platforms, including Mastodon, Blue Sky, Twitter, K-12 Leaders, LinkedIn, Facebook, and Instagram. This approach feels somewhat ridiculous and may not be sustainable in the long term.
In addition to these social media dynamics, we also see the potential promises and threats posed by AI, specifically Large Language Models (LLMs), under both centralized and decentralized management. Nearly every expert in AI has publicly called for government oversight, regulation, and “guard rails,” yet it seems we are content to let companies and nation states race toward an AI-dominated future. The exception to this trend is Europe, which leads us to the discussion about challenges posed by decentralized AI models.
The May 15, 2023, article in Stratechery, “Google I/O and the Coming AI Battles,” raises many of these issues and suggests a thought-provoking comparison between our emerging age of AI and the Protestant Reformation and the emergence of the printing press.
The rate of change in artificial intelligence is staggering. Thanks to a November 2022 presentation by an OpenAI leader and former graduate of our school, I learned a bit about DALL-E and Stable Diffusion, which are significant AI models in their own ways.
ChatGPT, developed by OpenAI, continues to receive the most mainstream media attention of all AI platforms as of the first half of 2023. DALL-E was also developed by OpenAI. ChatGPT is a centralized AI platform, meaning that OpenAI independently makes decisions about how to train, develop, release, and share its models and codebase. Microsoft, due to its substantial investments in OpenAI, undoubtedly has a voice in those decisions, but that doesn’t change the fundamentally centralized nature of ChatGPT and its AI approach.
In contrast to the centralized AI model of ChatGPT by OpenAI, Stable Diffusion operates as a decentralized AI platform. The owner and controller Stable Diffusion openly released it in 2022 with the explicit aim of disrupting and accelerating AI development. His stated reasons were to challenge the centralized control and dominance of AI technologies by a handful of big tech companies and nation-states.
Here is the rub, as highlighted in the previously mentioned Stratechery article: AI platforms absolutely require guardrails and limitations. Videos showcasing early versions of ChatGPT, where the AI had no hesitation in answering harmful queries, are alarming. (See “OpenAI’s GPT-4 Discussion with Red Teamer Nathan Labenz and Erik Torenberg.”) While Stable Diffusion has arguably democratized access to AI platforms, it (and other decentralized AI models) also open a Pandora’s box of potential misuse. This includes greater volumes of disinformation and misinformation, and potentially, harmful physical impacts.
In addition to the dangers posed by malicious actors using AI tools, the emergent behaviors of AI systems are a black box that we ignore at our own peril. Sundar Pichai’s recent interview on 60 Minutes serves as an example, where he describes an AI model spontaneously learning a complete foreign language.
Drawing a parallel with the printing press and the Protestant Reformation, as highlighted in the Stratechery article, we see the disruptive influence of technology. The printing press enabled an explosion of ideas in print, much like how the decentralization of AI platforms, fueled by actors like Stable Diffusion, is opening a Pandora’s box of diverse and uncontrolled use of these technologies.
The stakes, I believe, are not just comparable to those in the age of the printing press, but are likely much higher. AI is arguably the most potent technology ever formulated by human beings. To hear commentators like Ezra Klein liken the experience of developing AI’s capabilities to “summoning the demon” is both sobering and disturbing. AI developers are unleashing technologies upon the world that they do not fully understand, and the power of these systems is potentially existential.
This is a long post, and if you get to the end, you have my compliments. I would love your thoughts and feedback. Like my weekly conversations with Jason Neiffer on our EdTech Situation Room webshow and podcast, blog posts like this (which are admittedly much more infrequent these days) provide an opportunity for me to “process out loud” some of my observations and ideas relating to new technologies. My opinions are very much under development and still forming. There are a few things I can say with confidence, however, when it comes to AI, social media, and decentralized disruptions.
We are undoubtedly living in the midst of a highly disruptive and transformative era. AI platforms and technologies are just beginning to have an impact on our lives and shared culture. We need privacy regulation in the United States and globally. The proposed bills in the European Union, aiming to limit the abilities of companies to use AI platforms for any purpose, may significantly impact the growth arc of these CENTRALIZED AI systems. The decentralized platforms, on the other hand, might be able to evade those restrictions, potentially leading to the most disruptive (and hopefully non-catastrophic) changes brought about by AI technologies in the years ahead.
There is much to process and contemplate here, indeed.
AI disclosure: I initially dictated this post on my iPhone in the iOS Notes app, then used ChatGPT 4.0 to lightly edit and polish my prose.
If you enjoyed this post and found it useful, subscribe to Wes’ free newsletter. Check out Wes’ video tutorial library, “Playing with Media.” Information about more ways to learn with Dr. Wesley Fryer are available on wesfryer.com/after.
