In Support of the Pro-Human AI Coalition -- Some Suggested Framings

I was very pleased to see the launch of the Pro-Human AI Coalition this week, because I have been informally workshopping proposals for such a coalition for the fusion of AI with SM, based on strong overlap of issues and growing urgency to course-correct. (That included discussions with Jeb Bell of Project Liberty in November of last year and March of this year.)

Since February I have been workshopping a working outline discussion draft for a long form piece (still very rough), and more recently turned my attention to drafting a short-form piece that I hope might run in Tech Policy Press soon.

What follows is an lightly edited extract of key sections from my working draft of that shorter-form piece. I gather that the Pro-Human AI Coalition effort is highly aligned with the framings I propose, and look forward to dialog on how this kind of simple framings can help unify efforts that have so far been fragmented, and hard for non-specialists to understand without getting lost in overwhelming detail.

Key sections from my preliminary draft:

As the fusion of Artificial Intelligence with Social Media (AI/SM) is reshaping humanity, an impressively wide and growing body of work points to a diversity of better ways to design, develop, govern, and apply these systems. But real progress on reversing the current destructive path remains slow and fragmented. To limit the harms we are already seeing – and guide us all toward the benefits we hope for – we need to roughly agree on a few high-level normative and architectural framing principles that are simple, broad, and widely acceptable enough to drive a whole-of-society effort.

[Note that I frame this as applying to "AI/SM," because AI is subsuming the role of social media -- I hope the the Coalition will recognize that fusion, and correct the current failure of the AI hyperscalars to address the pro-human failures of social media that they seem to be recapitulating -- in ways that promise to be even more deeply destructive to democracy and human flourishing.]

High-level framing principles

There has been a growing recognition that first social media and now AI pose serious risks to democracy and human development. Diverse remedies are emerging, but progress has been slow because dominant platform business models and centralized architectures systematically underweight public-good objectives and user rights and agency, while policy responses remain limited and contested.

These are wicked problems, with no simple solution – far too complex, nuanced, and context-dependent for any one person or group to fully understand – thus needing a broadly accepted, diverse, and continuously evolving portfolio approach. 

For the past several years I have been suggesting an increasingly broad reframing, based on a holistic systems view of the processes required:

• A full architecture for governing who controls attention, what sources and reputations matter, how individuals and communities interact (in thought and action), and whether both individuals and their communities are served by loyal advocate agents.
• Shared reliance on simple overarching normative and architectural framing principles for coherently guiding a broad portfolio of efforts – and for marshalling whole-of-society support efforts for that. The key framings (explained below) are:
• Thought as a social process.
• An architecture of Three Pillars enabled to jointly synergize and evolve through open interoperation.
• Personal (and community) AI agents that are loyal advocates for their users.
• A simple, essential, and overarching test question: Who does it serve?

• Recognition that governance of this architecture must harmonize both social media and AI as they fuse into one another:
• AI is increasingly integral to social media -- as a recent major study explains, advocating coordinated development.
• AI is becoming a social medium, as it links humans, communities, institutions, companies – incorporating social media content – and armies of AI agents link with humans and other agents.

• A long view that supports ongoing development and evolution. We are re-engineering and digitizing activities that humans have evolved over millennia, but now are being revolutionized at an unprecedented pace. Nevertheless, time and flexibility for mutual human-AI adaptation must be allowed for – as we shape our tools, and they shape us.

Framing principal 1: Thought as social process

This frames the core human process, the dynamics of the fundamental cycle of thought that enables civilization and human flourishing, as shown in this Figure 1:

Debates rage over expression as a matter of “free speech rights” in isolation – and often “for me but not for thee” – but the real issue is the broader process of how social creatures think. Attention was manageable and social influences helped guide it. Now, under global expressive overload, algorithmic feed attention agents are needed to assist in that role. Platforms have taken over that role, but fail to do it well.

Historically, word of mouth and other social mediation naturally helped us self-regulate both expression and attention. But now platforms control feeds and reputation signaling. Their engagement-optimized systems redirect our social feedback systems toward platform engagement objectives that often diverge from user and community needs, while disintermediating important components of our social mediation ecosystem.

Framing principle 2: The Three Pillars of online discourse, linked by open interoperation

The pillars frame a high level architecture for the core structures of the social mediation process, comprehending the vital synergies of 1) individual, and 2) social processes that guide one another, plus 3) the distilled reputation data that emerge as byproducts of social mediation and help guide those processes – as in Figure 2.

[*The layer labelled "Middleware" is to be broadened to say "Open Interoperability"]

The pillars are:

1. Individual choice / agency – the primal human driving process that steers our inputs through each listener’s attention context, drives individual thought, and outputs expressed ideas to the network.
2. Social mediation ecosystem – the deliberative (or reactive) feedback processes by which others (human+machine) mediate message contexts and flows from speakers to listeners, much as in the earliest kinds of word-of-mouth propagation.
3. Reputation / trust – the distillations from the other pillars – and feedback to them – evaluating speaker and mediator context as worthy of attention (relevant, valid, and desired).

These pillars rest on a foundation of open interoperation (often called “middleware” in the social media policy community), as explained below.

All three pillars are distinctly essential – because each one has weaknesses when working alone, but the others can balance those.

• Individual choice / agency is the focus of much reform effort, but there are valid concerns that poorly-mediated individual agency can lead to echo-chambers, polarization, and extremism. It is just this concern that led me to begin workshopping the Three Pillars framing in 2023.
• Social mediation is how society has always moderated individual excesses. But without sufficient individual agency, it can lead to groupthink and tyranny of the majority.
• Reputation and trust draws on and supports both other pillars, as socially-derived metadata that helps guide who and what to consider trustworthy.

The pillars work together in a highly nuanced feedback loop. Agency and social mediation generate collective knowledge and reputation, which guide further attention and interaction.

Open interoperation is the foundation for that interplay. But as computer coding increasingly mediates discourse and cultural evolution, interoperability must be explicitly designed-in to preserve the agency, flexibility, extensibility, and resilience that humans and our social systems historically developed more organically. That is in tension with the rigidity of monolithic, centralized walled gardens, no matter how benevolently intended.

Framing principle 3 – Personal AIs as faithfully loyal advocate agents

AI compounds the new sociotechnical problems that social media have uncovered, but failed to solve. Providers of AI models exert similar, but far deeper and more subtle influence over what we see of the world and how we make sense of it. In addition to parallel concerns of bias, manipulation, and misinformation, the seductive power of AI can threaten human attachment and lead from simple cognitive outsourcing toward cognitive surrender.

As we move toward hybrids of human and AI intelligence, we face the same issues of whether our technology-based agents serve us – or the corporations that build and deploy them. As we move from monolithic AIs to complexes of AI agents, the same pillars and need for open interoperation apply:

• Individual choice / agency – do we each decide what sources are used and what objectives are set?
• Social mediation ecosystem -– do we make full use of the wisdom of other individual humans “in the loop” and of the communities of humans (and hybrids) that we choose?
• Reputation / trust – do we draw wisely on other humans and hybrids, and do we have transparency into who and what we are relying on.

AI developers have made much of the idea of growing agent capability, but not of AI relationship:

• Open AI drew attention to the question of capability as “agenticity” – the ability of an AI to do things on its own.
• Others have emphasized the need to consider a second dimension of AI as “agentiality” – its relationship to those it serves as to whether it can be trusted to loyally serve them.

There is growing realization that given the superhuman complexity and dynamics of AI, our surest protection as individuals and communities is to be able to just say “Have your AI call my AI” – selecting interface agents that can be trusted to represent us as faithfully loyal advocates (and sometimes even fiduciaries with legal duty to serve our interests, not those of their builders/operators or any other party). Technology for ensuring trustworthiness of advocate services is still in early form, but gaining active development.

 
Framing principle 4 – the essential test question is “Who does it serve?”

Instead of reacting to the harms of misaligned technology, we should be building this brave new digital world to support fundamental and broadly supported normative objectives. I suggest that as we design, develop, deploy, and use these new tools – there is one simple and essential principle to keep top of mind:

• The essential test question at every decision point, for every component, is “Who does it serve?”
• All else is commentary. That commentary includes these critical sub-questions:
• Normatively: “Who should it serve?”
• Sociotechnically: “How can we make sure it serves who it should serve?

A loose, big-tent, multidisciplinary, whole-of-society project portfolio

How we govern AI/SM will determine whether it augments or impairs growing challenges to collective intelligence, cooperation, and the workings of democracy. Because no one pillar can stand alone without serious weaknesses, shared framing principles can provide common understanding, objectives, criteria, and motivation for advancing a richly diverse interoperable and adaptable portfolio.

Only through cooperation on shared goals can we develop a coherent portfolio approach to marshal the tools and guide the long-term project of sociotechnical evolution we need to re-engineer democracy for the digital era – to overcome impending threats and enable humanity to flourish. There are organizations supporting work in this direction, but cooperation remains less integrated than needed across the pillars and across AI and SM. A big-tent alliance could make that broader alignment visible and actionable. 

A practical next step would be for existing (or new) organizations to workshop framing principles, seek consensus on objectives, and improve information-sharing across a loosely federated portfolio. That structure could support diverse experimentation and ongoing evolution without imposing a single program. [Hopefully the Pro-Human AI Coalition will catalyze that, for both AI and SM.]

 
Conclusion and call to action 

We are adding AI and social media to institutions that evolved for cooperation, discourse, sense-making, and democratic governance over millennia. Yet development remains piecemeal and concentrated in corporate systems whose incentives do not reliably prioritize long-term public value, and re-engineering remains mostly reactive, with no overarching vision. Better alternatives are emerging, but remain obscure, fragmented, and constrained by platform structures and extreme concentrations of power.

The proposed principles address four questions: 1) how thought remains a social process; 2) how agency, social mediation, and reputation reinforce one another when enabled by open interoperation; 3) how personal advocate AIs can help individuals and communities ensure AIs that serve other parties respect their agency and rights – and 4) above all, how every component must be judged by who it serves.

The goal is to make our media tools intelligently social—supporting democratic freedom and human flourishing through an open, evolving ecosystem. Building it requires a broad alliance that can coordinate this sociotechnical evolution without suppressing diversity and subsidiarity.