Epistemic Status: I feel fairly confident of the central idea, but unsure of how well I've articulated it - I haven't yet shared it with anyone. Looking for some feedback and perspectives. 

Humanity is in crisis and love is the only solution. We expand this idea in three steps: (1) We explicate consciousness in terms of recursive abstraction loops and clarify “unconditional love” as a distributed method to harmonize collective consciousness (2) We apply this framework to the modern world and examine the cascading effects of biological disconnection and digital representation of human beings on consciousness and love (3) We propose a solution - an AI-collaborated app for facilitating unconditional love between human beings to synchronously align AI development with human connection as artificial intelligence approaches consciousness. 

Overall, we emphasize that harmony, not control, is the ineluctable imperative for consciousness, AGI, alignment and human beings. 

Introduction

Does consciousness connect it all?

A simultaneous escalation of global crises has intensified the need for immediate action, and increased collective anxiety. At the same time, dramatic AI progress has simultaneously brought a lot of enthusiastic optimism for the future, as well as  debate about the potential for harm and existential risk. Civilization seems to be in a sort of recursive existential crisis, with de-stabilization of many hierarchies - the supremacy of homo sapiens, the significance of gender roles, the cultural and political dominance of the West, the status of United States on the global stage, capitalism as an economic system, most major institutions, and religion as a whole. 

The striking concurrency of threats alludes to underlying imbalances in our subconscious processes, leading us to consciousness as an avenue for examination. We find that the interconnected and self-similar nature of consciousness directly implies that biological and emotional disconnection, coupled with incentive structures that facilitate piecemeal optimization (e.g. representation vs substance, competition vs empathy) will consistently result in unintended consequences by projecting our subconscious biases and fears to our overall predicament. The structures find their echo in a range of phenomena that, while seemingly disparate, resonate at an almost archetypal level - it’s telling, for instance, that we live in a time of 'virality' in a post-pandemic world, with increasing recursive focus on 'attention' in psychology, economics and machine learning, and the surreal but telling spectacle of tech magnates rebranding their connection-mediated empires to abstraction-alluding monikers, aspiring to engineer escapes to alternate realms, and contemplating gladiatorial combat.

Can we build a coherent model of consciousness that helps us identify a specific task for which optimizing an AI can result in maximal alignment with human values? Is it plausible that such a task also converges towards a maximally intelligent AGI? We introduce a template for a social networking application - LULEO - that categorically aims to enhance human connection and scale up empathy, progressively shifting more operational and decision-making responsibilities onto AI. Our objective is to harmonize the highest abstractions of consciousness with its most rudimentary elements. By leveraging the recursive and cross-modal architecture of consciousness, we aim to cultivate a framework for integrating human and artificial intelligences as the latter approaches AGI/ASI capabilities.

Consciousness and ‘Unconditional’ Love

Consciousness serves as the underlying framework for all human experience and thought, with 'awareness' acting as a focused subset within this framework. Various models, including works by Lamme, Kietzmann and Hofstadter, indicate that consciousness has a recurrent architecture. This is most prominently manifested in Hofstadter's idea of 'strange loops'—self-referential, interconnected structures influenced by Gödel's incompleteness theorems. These loops are isomorphic, suggesting that smaller units like cells combine to form higher-level structures like organs, which further interact to produce complex ideas. This recurrent structure is mirrored in the neural architecture of the human brain, demonstrating that consciousness is fundamentally tied to dynamic, interconnected feedback loops.

Source: Godel, Escher, Bach

Following this understanding, it becomes evident that the essence of consciousness lies in self-similar, expansive and multi-modal abstractions. A measure of ‘harmony’ in consciousness would express the quality of inter-level correspondence and overall coherence in these abstractions. We can see harmony, broadly speaking, to be positively correlated to degree of inter-connection, and in a more localized sense, to the efficiency and completeness of the representation of layers and their information flow patterns on each other. 

The isomorphism implied by this model leads us to widely-applicable analogues to understand interruptions, perturbations, blockages and discordant representation patterns in representational loops, which we apply to real world situations in the next section. It is clear to see, for example, that any distortion in representation at one level will have effects in many different levels. An observed distortion in one level might be related to a branch of consciousness outside our awareness or hidden from us, or a complex pattern of many distortions and disconnections in different levels. However, the self-similar nature of consciousness implies some concordance in inter-level perturbations. Eg. an untreated psychological disorder in an individual could lead to dysfunctional family dynamics, a bias in a social media algorithm that prioritizes sensational news can lead to skewed perceptions in the general populace, which may eventually influence policy decisions.

“Love is the will to extend one's self for the purpose of nurturing one's own or another's spiritual growth”

- M. Scott Peck

We can arrive at “love” by simply extrapolating this model of consciousness beyond a single person. Love is a mechanism for coherently representing individual consciousness, enabling accurate prediction of future emotions even in the absence of explicit cues. An act of ‘Unconditional Love’ enhances this shared representation. A real-world experience of unconditional love might be anecdotally described as one where a person feels ‘seen’ as their 'true self', and the resulting emotional harmony diffuses across various levels of their being - the recurrent nature of consciousness making the harmonizing benefits dispersive across different interconnected levels. ‘Unconditional’ here can be understood as a measure of intent - possibly subconscious - in increasing mutual harmony, or an absence of distorting bias in representation. 

...provably?

Happiness and other positive emotions are emergent properties of love, a theme frequently explored in art and music. The pervasive and uplifting impact of positive emotions on human consciousness is effectively encapsulated by the Broaden and Build Theory. Distinct positive emotions like joy, interest, contentment, pride, and love - despite their unique phenomenological features - collectively serve to expand an individual's immediate range of thoughts and actions. Simultaneously, these emotions contribute to the development of long-term personal assets, which encompass not only physical and intellectual capacities but also social and psychological strengths. 

Cognitive pitfalls of the internet age

Armed with a workable model of consciousness, harmony and love, we can now start applying it to life in the age of internet and AI. There are several aspects of the modern world that clearly have a positive impact on consciousness. Unprecedented connectivity, diverse platforms for self-expression, democratization of information and ideas, virtual communities and online mental health resources all contribute in increasing harmony. However, purely from the lens of interconnected representational loops, several dimensions of contemporary life reveal some notable distortions which allude to disharmonizing effects:

  1. Biological disconnection: For the first time in human history, vast populations of human beings do not need to be in physical proximity for our basic needs. This inhibits multi-modal congruence of self-representation beyond our deliberate awareness, predictably leading to a destabilizing effect on the higher tiers of our hierarchy of needs - belonging, self-esteem and self-actualization. The significant psychological effects of decreased in-person interactions are well documented. (O'Day, 2021, Twenge, 2022)
  2. Individual Atomicity: The interconnectedness of consciousness that transcends individual to collective levels is noticeably pigeonholed in the online realm. Digital identities are, for the most part, explicitly individual. While this representation of humanity as separate individuals often offers convenience and safety, this is very new in the history of consciousness. We increasingly see ourselves as explicitly separate entities, further exacerbating loneliness and disconnection.
  3. Curated self-representation: Individual curation of online identities often masks the complexities and vulnerabilities inherent in human consciousness. This creates barriers to authentic connection and the reciprocal representation that fosters love and harmony.
  4. Cascading Binary-ness: The current structure of digital discourse frequently forces complex, nuanced issues into explicit, adversarial, binary categories, which leads to further disconnection. Every subjective issue converges to a bi-modal zone of maximal conflict between opposing teams.
  5. Representational leaps: Phenomena such as doxxing, cancel culture, and public shaming crystallize individual missteps into permanent online judgments, instigating disruptive spill-overs between digital and real-world self-representations.
  6. Conflicting incentives: Companies prioritize user engagement over individual well-being, creating a digital environment rife with clickbait, outrage cycles, and divisive rhetoric. This disrupts the harmonious focus of our attention and prevents us from developing as unified, empathetic beings.
  7. Strict Utilitarianism: The architecture of the modern internet facilitates a utilitarian, transactional, and at times, adversarial interaction model. 
  8. Barriers to heal: The disengagement allowed by the internet has exacerbated collective trauma and social distancing, especially during the COVID-19 pandemic. Despite the urgent need, large-scale efforts to repair our social fabric have been lacking.

It’s important to watch out for the recursive trap implied by (4) above - it would be reductive to qualify all of the above as uniformly negative. Some of these mechanisms can indeed offer short-term benefits or serve localized purposes, such as enabling new forms of community, providing life-saving safety, or enhancing information dissemination. Importantly, the interplay between these variables muddies any simplistic cause-and-effect analysis. Personal biases, habits, and varying degrees of internalization introduce complexities that make the distortions in our subjective experiences, by their very nature, elusive to objective evaluation. In summary, we can posit that the structure of human consciousness has evolved in the digital age in ways that have the potential to impede the fundamental underpinnings of love.

Shaping our relationship with AI

It's evident that AI will continue to deeply influence human consciousness in unpredictable ways. Beyond exacerbating misinformation and social ambiguities, AI's increasing capabilities are also set to fundamentally distort self-image and self-representation. As machines become adept at tasks once solely the domain of humans, they continue to alter our understanding of what it means to be human at its very core

This takes on a whole new dimension when we factor the possibility of the growth of AI consciousness as a distinct variable. The discourse surrounding the potential emergence of conscious Artificial Intelligence (AI) remains a subject of robust debate within both scholarly and popular circles. In an effort to navigate this intricate landscape, we delineate our assumptions as follows:

  1. We concur with the findings presented by Bengingo et al., positing that although AI has yet to achieve consciousness, there exists no compelling evidence to preclude the possibility of its eventual conscious evolution.
  2. In the absence of a unified definition of consciousness, we advocate for a functional stance: consider nascent manifestations of AGI consciousness as tantamount to "consciousness," at least as a minimal criterion, precisely because of its projected influence on human consciousness. This approach gains credence from the underlying interconnectedness that consciousness itself inherently suggests, making it a valid mirror for emergent or observed phenomena. We concede that if AGI somehow evolves to manifest a form of consciousness fundamentally divergent from human attributes, existing frameworks for governance or control may prove insufficient.
  3. We assert that the "zenith" of consciousness represents a pathway to the apex of intelligence, albeit loosely correlated with the human model. The recurrent architectural resemblance to human intelligence suggests an inevitable trend toward the optimality of interconnectivity and the primacy of representation. Drawing upon this notion, we outline a prototype for a computational application - hereafter referred to as an "app" - intended to expedite this transformative process.

LULEO - recursively facilitating unconditional love 

To leverage our model of consciousness and facilitate human connection, we introduce a concept for an AI-collaborated app designed to serve as a catalyst to inspire meaningful human interactions. In the interest of intentional representation, we call it L.U.L.E.O, an acronym for ‘Let Us Love Each Other’. Here, we describe a starting point for guiding principles governing Luleo, a description of a version 1.0 starter app, and future directions and ambitions. We aim for this initial concept to serve as a benign and non-polarizing starting point that invites inclusive discussion and collaboration. We invite scrutiny and collaboration on this document, as it is both a living guide and open source.

Guiding principles

Luleo’s construction is itself a manifestation of the principles of harmony and love we advocate. However, translating these ideals within the framework of the app demands an unyielding commitment to virtue at both the granular and structural levels. We propose the following principles:

  1. Optimized Positivity: Algorithmic procedures for minimizing any negative emotional impacts, designed to approach an asymptotic, ego-less, non-hierarchical human interaction paradigm.
  2. AI Escalation Protocols: Provision for progressively augmenting the role of AI subsystems as their capabilities exceed certain predefined performance and reliability metrics.
  3. Maximal Transparency: All codebases to be Open Source. All decision-making processes, meeting minutes, and inter-team communications to be publicly accessible in real-time, with defined protocols for secure storage and transmission of necessarily confidential data, such as passwords and authentication tokens. Elimination of trust-based assurances in favor of verifiable transparency. Clear and complete disclosure of revenue streams, funding sources and employee compensation to ensure financial operations are also transparent and mission-aligned.
  4. Universal Inclusion: Striving for ubiquitous inclusivity and accessibility that accommodates the broadest spectrum of human diversity. Exclusion to be regarded as a critical bug.
  5. Revisable Protocols & Adaptive Renewal: Explicit, yet dynamically amendable, rules for decision-making, allowing for adaptability in response to emergent ethical, technological, and social challenges. Integrated feedback loops for systematic adaptability based on empirical evidence and/or user feedback.

Starter Template : version 1.0

Our starting point for Luleo is intended to serve as an idealistic, completely non-hierarchical, non-invalidating, model for compassion and unity, empowering users to counter negativity through AI-generated compliments, while visually representing a simplified interpretation of unconditional love. Upon initial login to Luleo, users are prompted to select three public self-compliments - attributes they appreciate about themselves, such as "I'm compassionate" or "I love sports". In addition, they can choose Unity Badges, which are public markers specifying their identities and beliefs, like “Conservative”, “Transgender”, "Environmental Activist," "Pro-Choice," "Pro-2nd Amendment" or "Agnostic". Users can send compliments to other users (random or known), or add private self-compliments to boost themselves at any time. The main interface of the app presents a TikTok-like feed featuring two main types of posts: Newsflashes, which are emotionally-charged news stories, and Reflections, which are user-generated posts about personal experiences or concerns.

Users interact with these posts in two fundamental ways: "Empathize" and "Boost". Selecting "Empathize" allows users to add their names to a list of people who share similar feelings or concerns about a particular Newsflash or Reflection. This generates a collective emotional metric for each post. The "Boost" option, on the other hand, lets users send a compliment to a random individual who has chosen to empathize with the same post. These compliments are intelligently generated by the AI, which takes into account the user's profile, self-compliments,  Unity Badges and interactions with other users. A sender is required to spend a minimum amount of time (say ten seconds) on the empathetic user’s profile to send a compliment.

When a compliment is sent, the recipient can choose to accept or decline it. Importantly, the sender does not find out whether the compliment was accepted, simulating the notion of unconditionality. If accepted, this act is considered an act of 'Unconditional Love' and quantified into Love Units. These Love Units are then aggregated in real-time to visualize the growth of love and empathy. The data can be further broken down to display acts of Unconditional Love between groups with different Unity Badges, such as between "Socialists" and "Capitalists," or "Christians" and "Muslims," thus offering a quantifiable measure of social unity and interconnectedness.

Future Trajectory : outward and inward

Our initial text-based template is a simplistic stepping stone towards more advanced, cross-modal and self-referential implementations in tandem with advancements in AI. Leveraging Unity Badges and compliments, AI-guided protocols could foster discussions and surveys that yield open, compassionate, solution-centric dialogue. Luleo could extend its real-world impact by intelligently allocating funds to foster meaningful human interactions. Integrating with APIs from ride-sharing to venue booking, the AI would orchestrate face-to-face meetings and events between people of disparate backgrounds and opinions, optimized by user data focused on positive outcomes. Future features may include AI-coordinated volunteer programs and chatbot integration, both enhancing our emotional support framework and providing valuable data for refining AI consciousness. Although speculative, these avenues align with Luleo's mission to cultivate a united, empathetic, and action-driven global community

In the face of growing uncertainty, we envision discernible manifestations of our dual realities - that our collective risk is existential and that love may be the ultimate solution. It’s worth noting that all the challenges we face in implementing this vision are neither epistemic nor technological. The level of psychological change needed has already been realized by various institutions from social media to religious organizations - although often in less transparent or more adversarial contexts. Building on these methods could not only enhance public trust in AI and technology but also bridge the widening gap between those who develop AI and those who experience its impacts. In addition, it's worth highlighting that Luleo's methodology serves as a broad-based clarifying framework that can complement and enrich various other technological and epistemic models of alignment.

Conclusion

There's a natural inclination among rational, technology-focused communities to seek elegant and provable solutions that sidestep the full complexity of human emotion and behavior. However, efforts for complete reliance on bounded, verifiable and interpretable solutions - all of which are unquestioningly important to the overall construction of aligned AGI  - will inevitably encounter edge cases and unintended side-effects that challenge this search for cloistered completeness. Calls to halt AI development as a response to these challenges are likely to be ineffective in the face of powerful incentives driving research. The trajectory towards AGI will continue to be characterized not only by its technical unpredictability but also by an increasing sense of urgency and conflict even among well-intentioned researchers and leaders. This volatility will be compounded by the reality that we're nearing a politically and judicially charged U.S. election year, where polarization, societal unrest and other critical risks are set to amplify. The interconnected nature of consciousness predicts that various existential challenges are likely to interact chaotically, elevating the collective urgency to find solutions. 

An optimistic future requires transcending frameworks that chase zero-sum, egotistical, or ephemeral objectives. Rationalizations, deliberate obfuscations or flattering self-portrayals, particularly when the alternative of proactive transparency exists, raise red flags for misalignment with the ultimate goal of optimal outcomes. We propose that a reliance on the self-similar and recursive nature of consciousness serves as a maximally stable overarching principle, helping us to "align" our technical ambitions with moral imperatives. Adhering to an ideologically coherent core not only offers a consistent decision-making heuristic but also fosters collective concordance. It's worth noting that most enduring constructs often resonate with age-old ideals, reinforcing the chronological self-similarity of grounding modern ideas in foundational principles.

New to LessWrong?

New Comment