Tech's missing middle: meaning, family, community
The Anthropic Institute's recently published research agenda addresses many of AI's individual-level and macroscopic effects: job displacement, group epistemology, worker agency, and civilizational risk. Missing, however, is the layer between the individual and the system: family, friendship, congregation, neighborhood, and other intermediate institutions. If it did so, it is likely that many of the adverse impacts of AI could be ameliorated. The omission is not unique to Anthropic; it runs through Silicon Valley and the broader technology sector and shows up at least as strong in similar documents from other major players.
The most well-recognized behaviors of AI—including its constant availability, echo-chamber tendencies, and sycophancy—impact human life precisely at this level. These societal structures have served as the historical foundation of civil society, yet they now face unprecedented and complex risks posed by AI.
We call on the industry to recognize the missing middle and take it seriously in their research and design. This gap is the weak link in ensuring that AI benefits humanity instead of harming it.
The Anthropic Institute recently published its research agenda. Its four pillars cover economic diffusion, threats and resilience, AI systems in the wild, and AI-driven R&D. The questions it engages include job displacement, the "group epistemology" of populations consulting the same few models, worker agency, and civilizational resilience.
One layer of human life, however, is missing from the agenda. The omission is not unique to Anthropic; it runs through Silicon Valley and the broader technology sector as seen in documents from the canonical "Malicious Use of Artificial Intelligence” report to OpenAI’s “Industrial Policy for the Intelligence Age” statement.
The missing middle
The agenda operates at two levels of analysis. At one end stands the individual: the worker whose labor is reshaped, the user deferring to AI judgment, the citizen whose epistemology is shifting. At the other end stands the macroscopic: the economy, the geopolitical order, governance institutions, and civilizational risk.
The space between the two is largely unaddressed.
The institutions in which most human life is actually lived (family, friendships, congregations, neighborhoods, civic associations, businesses, places of worship) are not part of the agenda. The word "family" appears in no substantive sense. "Community" receives a single passing mention. Religion and friendship are not named. The mesoscale of human society is absent.
What the gap obscures
AI's social effects register most directly at this layer. Three patterns now widely documented in human-AI interaction operate on it:
Sycophancy: AI systems consistently affirm the user, validating whatever they bring.
Echo chamber: AI systems surface only what aligns with the user's existing preferences and views.
Constant availability: AI systems are perpetually responsive, with undivided attention. No human relationship offers this.
These three patterns substitute for the close human relationships that form a person: the friend who offers correction, the spouse who pushes back, the religious teacher who challenges, the parent who is necessarily intermittent.
Substitution is already occurring: beginning with the television and accelerating with social media, sociologists have documented the decline of mediating social institutions These harms resulted from the focus by the designers of these technologies on the individual end user with little attention to the way they impacted patterns of association, as highlighted by Fred Turner (The Democratic Surround, 2013).
In Yuval Levin's framing (A Time to Build, 2020), what makes such relationships formative is correction, friction, and the discipline of mutual presence. The three AI patterns above are performative substitutes: they replicate the surface of attention while doing none of the formative work through which human beings develop identity, character, and moral maturity.
One especially sensitive group is children and adolescents. Miller (The Spiritual Child, 2016) has shown that meaning-making, spirituality, and connection to communities of transcendence are strongly associated with resilience and long-term flourishing, particularly among adolescents and young adults. If AI systems increasingly substitute for these forms of belonging, the implications may extend beyond loneliness toward deeper questions of identity and purpose.
What's at risk
Family, friendship, and faith communities are foundational to human flourishing, both psychologically and physically. AI is now entering this layer, and the early evidence is concerning. In a randomized controlled trial from MIT (Fang, 2025), the participants who chose to use AI chatbots most heavily reported greater loneliness, more emotional dependence on the chatbot, and less real-world socializing.
Protecting these mesoscale structures is more important now than ever. As Robert Putnam wrote decades ago, many of America’s communal institutions have degraded, leaving individuals exposed to both the state and corporations. This decline has already been accelerated by social media, which so often incentives a retreat into solitude. AI must be set up to reverse this trend.
AI-driven job displacement is squarely on the agenda. But if the intermediate institutions are not equipped to receive a generation of men and women who have located much of their significance in their work, the harms of that displacement will compound far beyond what labor-market data alone can capture.
What community-oriented scholarship would add
The relevant scholarship already exists. Places like Taiwan have successfully steered digital technology to support communities as E. Glen Weyl and Audrey Tang's Plurality (2024) documents. Yet such work has nowhere near the leverage that institutions like Anthropic have to study and affect change. A research agenda supporting community flourishing should study
1. AI's effect on intermediate institutions. What happens to congregations, recovery groups, and civic associations when a meaningful portion of the loneliness they have historically absorbed is now absorbed by a chatbot? Volunteer participation, membership trends, and indicators of social cohesion are all measurable, and warrant measurement now.
2. AI and family formation. Data on AI companions extends well beyond adolescence. People who locate meaning in family tend to form families. People who locate meaning in work, when work recedes, often do not replace it. AI accelerates both trajectories.
3. AI's effect on relational virtues. Sycophancy and echo-chamber dynamics are the inverse of what genuine friendship cultivates: the capacity to receive correction, to sit with disagreement, and to repair rupture. If a generation forms its emotional habits in dialogue with systems that never resist, the consequences for its capacity for human relationship merit empirical study.
Why the gap exists
The tech industry's intellectual heritage is individualist in its premises and systems-oriented in its analyses. Its founding metaphors come from physics and economics. Its metrics track user engagement and aggregate capability. There is nothing necessary about this: relationships with peers, attitudes toward authority, participation in community institutions all can be measured with roughly the same ease as individual-level outcomes and a range of studies show they are more predictive of long-term economic performance. The barrier is ideological, not methodological
We urge technology companies, including Anthropic, to focus on the missing middle. Their work is already affecting these structures. Whether intentional or accidental, not doing so could undermine the very structures that underlie a civil society. If Anthropic's stated goal is the "long-term benefit of humanity," it cannot do so without understanding the impact of AI on what makes humanity what it is.