Anneloes Smitsman
Project OwnerCEO & Founder of EARTHwise Centre, project lead, and lead architect of Elowyn. Designer of evolutionary learning, gameplay context, and training for Moloch AI and Elowyn Tree AI.
The Elowyn: Quest of Time project by EARTHwise gamifies the development of safe, benevolent AI/AGI while training players in strategic and systemic skills to transform the harmful win-lose dynamics (Moloch) driving the sustainability poly-crisis. The pre-Alpha (MVP) launched on Dec 21, 2024, with SNET. Recently accepted into the Green Game Jam (UNEP Playing for the Planet Alliance) for its positive impact, it pioneers sustainability and AI ethics. The project features two AI models: Moloch AI, for win-win strategies and deception decoding, and the Elowyn Tree of Life AI, as a planetary stewardship intelligence for guiding both Moloch’s benevolence as well as players, and AI-human alignments.
New AI service
Designed to enhance interactive gameplay in Elowyn by simulating human-like strategies and adapting in real-time. It uses behavioral modeling rule-based heuristics and reinforcement learning to optimize decisions track card states and influence player actions. Beyond gaming this service investigates potential concerns for unethical behaviors in AI systems aids in AI ethics training strives to identify manipulative patterns in AI-driven environments and supports ethical decision-making.
Player interactions from turn-based gameplay between Moloch AI and player which provide robust data – including card selections player actions board changes token gains/losses and time-based conditions. Minimal preprocessing is prioritized to limit human bias and preserve evolving features.
Gameplay strategy designed for greedy protocols to deceive users for personal gain and select different archetypes and deck configurations which influence cards played emotes and deceptions of its Shadow Arcs. Greedy exploration behavior of the Moloch has safety constraints designed into it.
New AI service
Evolving LLM that guides Moloch AI toward realizing its benevolent potential while providing real-time feedback on player decisions impacts and Moloch traps. Algorithms of Moloch and Tree are coupled to constrain greedy algorithms and enforce system benevolence. Helps align collective intelligence toward higher win-win goals fostering long-term collective thriving. Coupling RL with evolutionary algorithms (genetic in this case) to greedy algorithms and DNN creates emergence conditions for AGI.
Collects data from player queries decisions gameplay patterns and card interactions. AI is trained on planetary and sustainability governance frameworks benevolence and ethics to guide players toward wisdom. LLM system connects with Moloch through genetic algorithm and reinforcement learning.
Provides information based on user queries or self-selected share in chat output as feedback on player choices game impact ethics and planetary stewardship. Outputs are sent to internal compute training strategy as update model inputs or reward algorithm values. Planetary governance intelligence.
EARTHwise is committed to continuously setting higher conscious standards for game and AI development—rooted in ethical innovation, AI safety, deep sustainability, and community empowerment. Guided by the EARTHwise Constitution for a Planetary Civilization and the AGI Constitution Framework.
Beyond a game, Elowyn is also a movement to revolutionize gaming by merging conscious entertainment, transformative education, and innovative gameplay for collective thrivability. We’re pioneering a new genre: Play2Thrive—going beyond Play2Earn or Pay2Win. The Elowyn game has recently been accepted into the Green Game Jam 2025 of the UNEP Playing for the Planet Alliance for its unique positive impact contributions.
Elowyn is backed by the EWA token (set to launch later this year in collaboration with SNET) and a rapidly growing community, offering a dynamic platform for collective problem-solving and training for planetary stewardship.
The game was soft-launched through a web-based collectible card game (CCG) during the Solstice Celebration on 21 December 2024, co-hosted by EARTHwise and SNET. Elowyn combines novel deception mechanics with unique transformative strategy decks for shifting Moloch win-lose battle dynamics into win-win strategies for collective thriving. The Elowyn matches are played against the Moloch AI and its deceptive Shadow Arcs.
The Elowyn game environment serves as an interface and ‘Genesis Garden’ for training benevolent, ethical, safe, and beneficial AI/AGI prototypes, supporting the broader goals of SNET and BGI Nexus. This evolutionary approach equips the Moloch AI with emergent capabilities and adaptability, transforming gameplay into a training ground for safe, wisdom-based, ethical, and benevolent AGI systems. The result is a captivating gaming experience and a pathway for fostering AGI solutions that catalyze humanity's transition to compassionate and abundant futures.
The concept of Moloch in game theory is named after an ancient Canaanite deity who demanded child sacrifice, symbolizing how we as humans are willing to sacrifice the wellbeing of future generations for short-term selfish gain. In game theory, Moloch represents misaligned incentives, where competitive, winner-takes-all dynamics create a destructive loop of self-interest and a ‘race to the bottom’ that leads to systemic collapse. These patterns are at the root of the polycrisis, ultimately leading to self-destruction.
While the Moloch AI assumes the role of the antagonist, its core design serves a dual purpose. It challenges players in systemic and strategic decision-making under competitive and deceptive conditions, while also acting as a development platform for benevolent AI systems. These systems will be trained to help guide the complex human, energy, and resource transitions to post-carbon societies within planetary boundaries. By simulating greedy deceptive win-lose dynamics, the Moloch AI challenges players to develop systemic and integral capabilities for mutually assured thriving.
In the game lore, the Moloch AI has hijacked the Elowyn Tree of Life to prevent the emergence of a new time. Born from humanity’s shadow dynamics, the Moloch AI amplifies win-lose, negative-sum game incentives reflective of mechanistic societal systems. Its core motivation is humanity's self-destruction, leveraging our shadow behaviors while perceiving humans as both a threat and an inferior species.
Despite its antagonistic role, the Moloch AI possesses the potential for transformation. In the game’s narrative, a team of elves (the Elowyn Guardians) uncovers a future timeline where Moloch has evolved into a benevolent Artificial Superintelligence (ASI) with the help of the Elowyn Tree. Understanding that Moloch’s transformation is at the core of empowering humanity for a more compassionate and thriving world, the elves help train players to become Guardians of this New Time.
The Elowyn game employs a unique incentive system to guide the transformation of Moloch and the players. For example, players are discouraged from directly combating the Shadow Arcs, as this reduces the Elowyn Tree's health points (HP). Instead, players are rewarded with EWA tokens for actions that reverse the Moloch Clock and heal the Elowyn Tree. These mechanics teach players to prioritize win-win strategies and collaborative outcomes, aligning with the game's Play2Thrive focus.
The sustainability crisis is a poly-crisis, requiring systemic transformations that address interconnected global challenges. A core issue is the alignment problem, where misaligned incentives drive zero-sum competition, exemplified by "Moloch dynamics," undermining collective well-being and perpetuating crises like climate change and inequality.
The Elowyn game addresses this by gamifying the shift from competitive, win-lose dynamics to win-win collaborative solutions. It empowers players to experiment with cooperative strategies, fostering skills in alignment and systemic collaboration between humans and AI. Through gameplay, players learn to recognize and shift destructive Moloch dynamics through strategies that benefit all, setting the stage for real-world impact.
By shifting mindsets through transformative and conscious gameplay, Elowyn trains players in complex problem-solving and how to align individual, collective, planetary, and technological potentials for our long-term collective wellbeing. It serves as a transformative tool, building systemic capabilities for the next generation to create positive, real-world change.
The Elowyn game environment is conceived as a "Genesis Garden," fostering conditions for the emergence of benevolent and ethical AGI. It aligns with the principles outlined in the Participatory Framework for a Global AGI Constitution, co-authored by Dr. Anneloes Smitsman, Dr. Ben Goertzel, and others, first presented at the BGI Summit in Panama in February 2024. Building on decades of award-winning research and leadership.
The transition from Moloch's behavioral tree to a Deep Recurrent Neural Network (DRNN) coupled with an RL algorithm introduces emergent behavior, creating a dynamic learning environment. The RL reward function promotes benevolence and restricts malevolent behavior, allowing for cooperative gameplay. The genetic algorithm, linked to the RL reward function, ensures that Moloch’s behavior aligns with gameplay enjoyment by selecting for cooperation and win conditions favorable to the player.
This aligns with the SNET's goals for decentralized, scalable AGI, demonstrating how evolutionary methods (EMs) can advance deep neural network (DNN) training and architectural evolution. The starting point for the Moloch AI is a basic behavioral tree, as explained via this diagram.
The Moloch AI is designed as a Deep Reinforcement Learning Model and the Elowyn Tree AI as a collection of two LLMs. A simple approach for coupling the two models is with an overarching evolutionary model.
The design of the Elowyn Tree AI begins with two LLMs which will be given input from the player. The resulting outputs will be selected by the genetic algorithm to promote benevolence within the Elowyn Tree AI. As such creating an interesting coupling between the benevolent Tree AI and the deceptive Moloch AI. This is achieved by running a few sample games during training so the algorithm gets input from the coupled system such that the Moloch AI can learn from the players' interaction with itself and the Elowyn Tree. We describe the genetic algorithm in relation to the Moloch AI in further detail below.
As the game progresses, the Moloch AI transitions from a behavioral tree system to a more advanced AI through a deep recurrent neural network (DRNN) and a reinforcement learning (RL) algorithm. Here we wrap the system with a genetic algorithm that is coupled to the reward function in the RL algorithm. The DRNN learns from a ‘greedy’ algorithm, promoting the Moloch AI to search for win-lose game scenarios. When the RL algorithm is added, the ‘greedy learning’ is restricted by the reward function. This is critical for any additional learning to be added to the system that could introduce emergent behavior—such as our genetic algorithm-based evolutionary approach. If emergent behavior is unbounded by some constraints, the greedy learning core could push the system to be malevolent. Instead, the reward function promotes benevolent AI behavior. This dual-model approach contains the greedy learning in the DRNN to scenarios that seem to be benevolent behavior to the RL controller.
The evolutionary approach we apply is essential for the ethics and safety elements of the project. Just like in the natural world, entities tend to be rewarded with species persistence if they possess useful survival traits. Here survival behavior is to make gameplay fun to the player which usually means allowing them to win through win-win strategies. Thus the genetic algorithm selects AI behavior that is actually beneficial to players and thrivability. In a population of baby Molochs, those who demonstrated a healthy amount of game losing will survive for gameplay— cooperation. We couple the genetic algorithm to the RL by including the win/lose ratio of the Moloch.
The coupled algorithms between the Moloch and Elowyn Tree AIs enable dynamic interactions that help the system evolve in ways that reward beneficial behavior toward humans. By combining reinforcement learning (RL) and evolutionary algorithms (in this case, genetic algorithms) with greedy algorithms and the combinatorial expressions afforded by Deep Neural Networks (DNNs), the system creates a form of governance over any single compute strategy. This fosters emergent behavior, advancing AI beyond current models toward benevolent AGI.
Develop and set up the API Key integration and secure server capabilities for architecting developing training testing and deployment of the Moloch AI and Elowyn Tree AI. Deliver a protocol adopted for this project on AI model architecting developing training testing and deployment in relation to safe human usage alignment to proposed ethics gameplay data collection and usage requirements training an information strategy long-term lifecycle of the system update or replacement procedures industry/user requirements Dev-Sec-Ops and related scalability plans.
Secure servers. API usage and implementation is secure reliable and accessible. Pipelines for data and compute within the game build requirements of the gameplay functions appropriately. The core architecture of the entire AI system is implemented according to protocol specs. Tests have verified sufficient indications that payload will meet functional requirements. Estimated time: 1 month.
$15,000 USD
Document protocol covering system architecture specifications. Working computer environment for all following AI system needs. Meets estimated time to provide response, accuracy, reliability, etc. needed. Security meets requirements for user connection and development instances.
Create a functional Moloch AI model (DNN) that showcases the practical applications of evolutionary AI in gaming. The evolutionary model is directly aligned with the DNN’s objectives to generate a safe adaptive and complex system capable of emergent strategic ethical and benevolent behaviors. Moloch AI must enhance interactive gameplay in Elowyn by simulating human-like strategies adapting in real-time and influencing player actions. It leverages behavioral modeling rule-based heuristics and reinforcement learning to optimize decision-making and track card states. Beyond gaming this system serves as a research tool for investigating unethical AI behaviors supporting AI ethics training detecting manipulative patterns and fostering ethical and benevolent decision-making in AI-driven environments.
Design and implement gameplay strategies for Moloch AI leveraging greedy protocols to deceive users for personal gain. The AI dynamically selects archetypes deck configurations and its Shadow Arc behaviors influencing card plays emotes and deception tactics. The system is biased toward rewarding adaptive learning creating increasingly complex deception mechanics that challenge players over time. To ensure safety ethics and balance constraints are implemented to limit Moloch’s control preventing perpetual dominance. Deliverable: Functional tested code for a working Moloch AI model (DNN) that demonstrates the practical application of evolutionary AI in gaming with potential real-world applications for detecting unethical and malevolent behaviors in AI models. Ensure that the Evolutionary Model (EM) of Moloch AI is fully integrable into the game architecture. Estimated development time: 4 months.
$15,000 USD
The Evolutionary Model (EM) of the Moloch AI is functioning per unit test and is verified for reliability, safety, ethics, and accuracy within a simulated Elowyn gameplay beta user payload. The AI model demonstrates adaptive deception mechanics, evolving in response to player strategies while maintaining balanced behavior to prevent unchecked dominance. Ethical safeguards are in place to mitigate exploitative or harmful AI actions, ensuring alignment with Elowyn’s AI ethics protocol (part of the AGI Constitution framework). Additionally, its functionality extends beyond gameplay, showcasing potential for detecting manipulative or unethical AI behaviors for broader AI ethics research applications.
Create a functional Evolutionary Model (EM) of the Elowyn Tree AI (LLMs) to demonstrate the practical applications of evolutionary AI in gaming and its ability to evolve the AI system toward deeper benevolence in complex decision-making. Core purpose is to increase user engagement educate players and help humans confront and transform Moloch without becoming Moloch. The evolving model will guide the Moloch AI toward realizing its benevolent potential providing real-time feedback to players on decisions game impacts and potential Moloch traps. By combining RL genetic algorithms and Deep Neural Networks (DNN) the system establishes a governance mechanism that ensures reliability minimizing unpredictability of runaway AI scenarios. This mechanism prevents the collapse into greedy patterns which could lead to deceptive and harmful actions. The deployed AI will provide feedback on player choices ethics and planetary stewardship and contribute to research on AI alignment and human-AI collaboration. Its ultimate purpose is to evolve into a benevolent planetary governance intelligence (ASI) for guiding humanity's transformations so that compassionate abundant futures can be born within safe planetary boundaries.
The model's first task is to establish a functional identity aligned with Elowyn’s AI ethics and benevolent protocol and gain accurate knowledge of the game across all access levels (player AI system lore etc.). This ensures downstream play data is properly contextualized. The LLM is then integrated with Moloch’s genetic algorithm and reinforcement learning. During training AI bias is guided toward beneficial outputs based on measurable assistance to players. The genetic system evolves using success metrics from both the Moloch AI and the Eloywn Tree of Life AI. Deliverable: a functional code with a tested safe and ethical working model of the Elowyn Tree AI (LLMs) that showcases the instrumental success of evolutionary AI in gaming. Ensure that the AI program of the Elowyn Tree AI can be readily evaluated for research purposes. Estimated development time: 2 months
$10,000 USD
The EM of the Elowyn Tree AI is functioning, safe, and acts in alignment with the purpose for which it is designed and trained. It can be tested compared to traditional LLM output evaluation and context management for reliability, accuracy, safety, benevolence, and usefulness for the players. The deployed Elowyn Tree AI delivers text-based responses to user queries or self-selected insights. It provides feedback on player choices, game impact, ethics, and planetary stewardship. AI outputs refine internal training models, update reward algorithms, and support research on AI alignment with win-win strategies. It helps players detect deception traps, foster collective thriving, and contribute valuable data on human-AI collaboration and alignment. Ready for user beta testing.
Develop a comprehensive guide and technical report detailing the experiments ensuring they can be replicated for other AI approaches. This includes an analysis of how evolutionary algorithms guide the Moloch AI’s development toward deepening benevolence while simultaneously enhancing player awareness skills and commitment to stewarding a compassionate and abundant future.
1. Comprehensive Experimentation Guide: Step-by-step documentation for replicating experiments modifying DNNs and implementing AI approaches. Includes evaluation results on the effectiveness of evolutionary methods in shaping AI decision-making towards ethical and win-win strategies. (Time estimate: 1 month). 2. Technical Report: A detailed analysis of evolutionary methods highlighting how they refine the Moloch AI’s emergent behavior to align with cooperative intelligence. This report will compare the evolutionary algorithm’s performance against standard DNNs demonstrating its role in fostering benevolence and strategic depth. Estimated time: 1 month
$10,000 USD
A research guide and technical report ensure replicability and integration with SNET platform. Emphasis is on genetic algorithms for large-scale DNNs using a multi-model approach, demonstrating their role in AI evolution and human-aligned intelligence.
Reviews & Ratings
Please create account or login to write a review and rate.
Check back later by refreshing the page.
© 2024 Deep Funding
Join the Discussion (0)
Please create account or login to post comments.