The Midterms and the Pro-Human AI Coalition -- Some Suggested Framings

Update 11/13/26: How to Think About Making AI and Social Media Pro-Human is now live on Tech Policy Press and on my new Substack, and will also be posted here soon.
(That is the article previewed in this post, and supersedes it.)

Reinforcement of My Comments at "Can Middleware Save Social Media?"

Thanks to Gabriel Tyler at Knight Institute for a nice summary of this stimulating expert convening on the question, “Can Middleware Save Social Media.” While it captures much of the spirit of this lively discussion very well, there are some points I alluded to that deserve reinforcement. (Full video of these sessions is now online.)

My reinforcements here relate to just what middleware is, how it can be help save social media, and whether it can do that without government intervention to facilitate and mandate interoperability with the dominant platforms.

At the highest level, I want to reinforce my point that there are "two ways you can come at middleware," and that one way is exemplified by Bluesky as “a greenfield development” of a new social media ecosystem. Proponents see this new architecture for social media (based on the innovative generativity of open protocols) as having potential to mature and grow enough user acceptance to replace the dominant platforms, even without government intervention.

There was little discussion of this second path after my comments in the first panel, and admittedly, it remains very unclear whether this kind of ecosystem might succeed at obsoleting the platforms. And of course there is no question that government intervention could greatly facilitate both paths to middleware. But current US policy trends do not augur well for such intervention.

That leaves this glass as arguably half empty, but given the forward-looking work being done in the Bluesky ecosystem that I alluded to -- including the accelerating impetus to build out Eurosky as a counter to the power of the dominant US-based platforms -- I see that glass as half full, even without US regulatory change. If Eurosky development scales successfully, that could spill over to attract demand, and thus supply, for similar services in the US and elsewhere. (The other speakers may be less hopeful about this path than I am, but it would be ironic if it turns out to be a European desire for independence that saves US democracy from social media.)

The other point I would add context to is that, as Gabriel put it, “middleware alone cannot resolve the deeper forces shaping online discourse.” That is true, but we should not view this as a matter of "middleware alone." 

I suggested that “middleware really is a technical enabling facility” – strictly speaking, middleware does nothing more by itself.  What it does is enable services to be built on top of middleware, thus facilitating access to a wide range of  “middleware-enabled services.” My comments referring to my “Three Pillars” framing (see this diagram that I was unable to show at the event) were meant to suggest the unlimited variety of services that could be enabled without permission from any dominant platform -- to make social media much better (whether the platforms like it or not). It is these middleware-enabled services that are built on top of middleware that can save social media.

This bears on the issues discussed in the keynote conversation that followed: concerns that middleware alone cannot save social media -- that instead of a focus on user control, regulators should focus on restriction of addictive or deceptive design practices and commercial surveillance, or supporting other prosocial remedies. The pillars framing emphasizes that middleware (in the narrow sense of the technology enabling interoperation) is not a solution in itself, but it facilitates all manner of solutions -- middleware-enabled services (not just for user control, but also for social mediation and socially-derived reputation, and much more) that can work in concert through an open networked ecosystem. 

This open market ecosystem could offer tools that are less addictive and that do not extract and abuse personal data, and that support greater community roles in social media that would encourage people to demand such tools and choose to use them wisely. Government can certainly facilitate movement toward this more open architecture and to the services that might enable, but user desires for better social media, alone, might drive this shift. That has already brought over 40 million users and growing independent development into the Bluesky's AT protocol ecosystem, even in its highly immature state.

Whichever way we get there, the real promise of middleware is to reduce platform control over the media tools that human discourse depends on. Once fully open interoperation and "exit" is enabled, further sufferance of platforms is not needed to apply a wide range of alternative solution strategies. If you build it, they can come -- and it can work seamlessly (with all of a user’s data and their social network). Then you just have to get them to come, and to find a sustaining revenue model. Ecosystems based on open interoperability and extensibility regularly succeed at that in surprising ways.

[Minor edits 4/19/26]

Speaking on Social Media Middleware/Interop, Personal Agents, and AGI

My busy month as speaker at interesting workshops and panels:

Governing the AGI Commons: March 4 brought me to Oxford for this one-day workshop organized by Google DeepMind and the Cooperative AI Foundation. I brought a perspective on how the harms of social media might be recapitulated in even worse form in AI, and how multiagent systems that support individual and community agency as loyal and trustworthy advocates in working with counterparty AI systems hold promise to avert that. It was Chatham House Rule, so I can't report on much, but a brief report from the organizers is planned, to be linked from here when available

Interoperability as a Design Principle for the Digital Economy: On March 24, I walked a few blocks for this one-day workshop at NYU Stern Business School co-sponsored by UCL Centre for Law, Economics and Society, speaking on a panel on Designing Interoperability in Social Media. Discussion was catalyzed by an in-depth paper. My understanding is that video and/or decks will become available (to be linked from here). 

Can Middleware Save Social Media?: March 26 brought me to the National Press Club in Washington for this Knight First Amendment Institute full-day convening on this question, speaking on the first panel, Introduction to Middleware. Video of these wide-ranging sessions with leading experts is online and well worth viewing by both newcomers and old hands on this issues.

Conscious Commerce: Delegation, Personal Agents, and the Future of Consumer Power: On April 3, I spoke as an advisor to a Consumer Reports project team of Carnegie Mellon grad students seeking to help specify pro-consumer, privacy-first AI systems that enable consumer agency, trust and conscious choice. This was fun adventure combining my recent work on directions for user agents for social media and AI, with newly emerging opportunities to apply my older work on "post-sale market data" ecosystems for product usage data and ratings, and on the FairPay framework for adaptively win-win relationship commerce.

My thanks to the organizers and participants for these very stimulating events.

[Update 4/15:] This feeds into my efforts to build on my Three Pillars framing and related ideas on personal AI advocate agents, with the idea of an overarching scaffolding for better focusing the combined effect of diverse strategies for alignment and governance of “social” media as it fuses with AI (and moves toward AGI). The objective is to make our media tools intelligently social -- to fully support democratic freedoms and the flourishing of humans as social creatures. This is very much a work in progress -- to workshop it, I have begun sharing my working outline as a discussion draft, and inviting feedback.

Attention Is Power -- Reframing The Future of Free Speech in The Online Era

Versions of this article were first published in The Bedrock Principle, published by The Future of Free Speech think tank at Vanderbilt University (11/4/25) and then in American Purpose on Francis Fukuyama's Persuasion Community on 11/13/25).

____________________________

Debates about free speech online are misframed. Instead of focusing on what content should be removed or censored, we should instead focus on restoring the underlying social and technical structures that enabled free speech to function well in the past. A crucial part of that structure is freedom of attention (which I have previously referred to as freedom of impression, as a complement to freedom of expression). That is the ability of individuals and communities to choose what they listen to, rather than having platforms decide for them.

Humans are deeply social creatures who speak to be heard and understood by others. We make sense of the world, flourish, and find resilience collectively. Free speech is not just a matter of individual expression. It’s a continuous cycle: ideas are formed, expressed, and shaped through social interaction, then fed back into human minds.

The scale and speed of online discourse, and how interactions cascade to influence one another, make it essential to understand that freedom of expression for all speakers, across diverse communities, is feasible and prosocial only when there is freedom of attention for listeners. It’s also only workable when that expression is supported by healthy processes of information sharing and assessment that function as a social mediation ecosystem. Censorship is unnecessary and counterproductive, as it harms the rights of both speakers and listeners, and weakens those processes of collective intelligence.

Censorship is Not the Answer

We are now engaged in fierce debates over how to address the aspects of free speech that are not in need of repair. Demands for “freedom of expression” have devolved to zero-sum battles over which side gets censored. Most people want platforms to take some responsibility for reducing harm, but when content policy teams focus on removing “lawful but awful” content from the public sphere, it is neither effective nor desirable. Instead, it entrenches the inherent dilemmas of removal as a form of censorship:

• The problem is an intrinsic matter of viewpoint: “free speech for me, but not for thee.” Who decides what viewpoint is true or good?
  
  
• It is also a practical matter of power: how can we find nuance and balance between top-down control (rules imposed by governments and platforms) and bottom-up freedoms (users and communities setting their own norms and practices)?

There are more effective, win-win remedies than removal, as many have argued.* We forget that we traditionally protected our attention by walking away or changing the channel. Speech may, indeed, have social consequences, but that is no justification for deplatforming. 

Thought as a Social Process 

Online technology surfaces a neglected question: “Who controls your attention?” Before the digital age, people naturally had control over what they paid attention to. We were free to walk away from speakers; free to associate or not. Now, platforms control what appears in our feeds and from whom. Instead of choosing for ourselves, we are fighting over how these centralized platforms make those choices for everyone—because they do it badly.

Consider this diagram of Thought as a Social Process.

• Thought and Expression happen at the individual level. Current battles are for or against censorship of expression, but expression is just the starting point of the full cycle of thought as a social process.
  
  
• Social Mediation is the rich, nuanced stage where others process and interpret what’s been expressed, pass it on, and feed it back (or not). When expression is censored, it nips this cycle in the bud, as governments, platforms, or institutions prevent certain expressions from ever reaching the open social mediation layer. Historically, social mediation occurred through the natural dialogue of communities, institutions, and shared cultural spaces, often guided by trusted intermediaries such as clergy, editors, local leaders, and other influential figures. This layer introduced natural forms of friction through reflective feedback, social norms, and editorial choices over the use of limited channels. These processes encouraged both individual and collective reflection, allowing reputations to form and circulate, thereby slowing the spread of raw information and adding context before ideas reached wider audiences.
  
  
• Attention is our selective awareness of what comes back to us: which of those mediated ideas are impressed on our attention to shape new thoughts. This is where freedom of attention comes in. This is the listener’s right to decide what should feed into their attention. Historically, people exercised this simply by choosing who to listen to or what to tune out. That was how word of mouth had worked for millennia—expression to one or a few, then propagated more widely in successive cascades of cycles (if deemed worthy by enough listeners).

Broadcast technology and the emergence of mass media changed that dynamic. Control of social mediation shifted from the soft power of influential intermediaries to the imperious gatekeepers of costly and limited channels, which constrained freedom of expression (as only partially balanced by policies such as the Fairness Doctrine in the United States and impartiality rules in the UK, which required broadcasters to present contrasting viewpoints).

But now online social media shifted dialogue back to a word-of-mouth style of person-to-person propagation, driven by individual posts, likes, shares, and comments—but algorithmically, at dramatically increased speed and scale, connecting anyone to anyone globally.

The problem became that “a wealth of information creates a poverty of attention.” Online feeds became a “firehose” of information, rushing far too fast to drink from. As a result, social media feeds now function as attention agents to help listeners manage what they see by ranking and filtering content to allocate their attention.

The “original sin” of the ad model incentivized platforms to build attention agents that compose and optimize feeds, not to serve users but to maximize engagement for ad revenue. That oligopolistic capture of our attention now threatens democracy and our very sanity. Dominant platforms have concentrated power in a few opaque algorithms that are easily manipulated to amplify chaos rather than distill collective wisdom. This also bypasses and drains vitality from the social mediation ecosystem that previously constructed “social trust,” which is essential to democracy and is now being lost.

Compounding this, advanced AI is merging with social media, making control over our attention an even more opaque problem.

We can no longer maintain freedom of speech unless our attention agents protect our freedom of attention. This is an aspect of the freedom of assembly or association recognized in the First Amendment and similar human rights formulations. It works because “harmful” speech does little harm if not propagated widely—and without human social processes, we cannot determine which speech we should propagate.

Taking Back Control of Our Attention

Barriers to reform include the fact that platforms have little incentive to yield the control that is so profitable, or to serve smaller or demanding communities—and that few people have the skills and patience to manage their feeds directly. What is needed are trusted and prosocial intermediary attention agent services that we can independently select and combine for ourselves, to support the needs of our communities. Open interoperation protocols, or “middleware,” can make this possible by sitting between users and social networking platforms.

The obvious pathway to regain control of our own attention is to open up today’s dominant platforms, for example by expanding on the proposed ACCESS Act, to “[a]llow users to delegate trusted custodial services … required to act in a user’s best interests through a strong duty of care, with the task of managing their account settings, content, and online interactions.” Currently, that path seems to have been stalled by platform resistance and government dysfunction.

Another way to regain control of our attention is through a growing grassroots movement to build open social media ecosystems. This emerged with the Mastodon-protocol-based “Fediverse” and is now gaining momentum with the more deeply interoperable Bluesky-protocol-based “ATmosphere” that already serves some 40 million users.

What we see now is just embryonic—green shoots of a revitalizing sociotechnical ecosystem. The power lies in the open ecosystem of attention agents, built on these protocols, that will enable people (and their communities) to filter and shape their feeds for themselves, rather than relying on platforms to do so. Some misunderstand Bluesky as simply “lib Twitter,” but that is just part of a broader shift toward user-controlled filtering and community-shaped information flows that can enable a user-selected diversity of “vibes.”

One early example is BlackSky, a community-mediated feed service drawing on human and machine judgment to filter Bluesky feeds that are protective of black people. Grassroots initiatives like Free Our Feeds, the Social Web Foundation, Project Liberty, and Protocols for Publishers, as well as feed service tooling providers like Graze, are working to catalyze development of this open “pluriverse.” Meanwhile, reactions to the dominance of American platforms are fueling the EuroStack/EuroSky initiatives to create new and more open infrastructures to serve other communities.

Either way, a key objective is not only to enable our freedom of attention, but to protect it from “enshittification”—the extractive practices that arise when users are captives of entities that control not only their attention, but their connections and data. To prevent that, these new initiatives aim to make “exit” (switching to alternative services) easy, thus keeping real competition in play.

Shifting moderation from top-down removal to more bottom-up, community-mediated management of attention raises concerns, such as the creation of echo chambers and the responsibilities to protect vulnerable users. As I will address in a companion piece, restoring reliance on community-level social mediation, supported by “least-restrictive” ways of adding “friction,” can make those issues manageable in democratic ways that support the fullest practical levels of free speech.

Beyond Technological Solutions

We urgently need a whole-of-society effort to reframe how we understand free speech in this new digital era, and to reshape social media and AI so they support and augment both individual freedom of attention and healthy prosocial mediation. This is not just technology, but an evolving sociotechnical process that includes:

• Improving public understanding of how and why free speech works in more positive ways, without the need for censorship, so society can find a stable path between tyranny and chaos.
  
  
• Reducing censorship, removal, and deplatforming as rapidly as is practical.
  
  
• Creating space for innovation by opening markets for supportive and open media-tech infrastructure, regardless of whether dominant platforms participate.
  
  
• Supporting the growth of alternative attention agent tools and services like the Bluesky and Fediverse ecosystems that are open, flexible, and extensible, even if they’re still developing.
  
  
• Fostering and nurturing the essential interplay of both individual freedom and reinvigorated communities and institutions in online discourse.
  

____________________________

My thanks to the many people who have helped me formulate these ideas, and to Jacob Mchangama and Francis Fukuyama for encouraging me to write this and a planned companion piece to go deeper on revitalizing humanity’s social mediation ecosystem (and for invitations to help organize a major project on middleware, and to the Global Free Speech Summit). And thanks to Audrey Hingle for editorial support.
____________________________

*For example, Goldman, Gillespie, Stray, and Riley and me