Our Team

The team primarily comprises Dr. Talib Hussain, who will design the new motivational framework, create the new language structures, and develop the MeTTa components.  His efforts will be supported by 2 student interns - from 3rd year of professional John Abbott College computer science program and from McGill cognitive science program.  The students will support the building out of the motivational framework to include a diversity of drivers as well as support early prototyping.

Project details

Introduction

To achieve AGI, methods are needed to promote and assess a broad spectrum of intelligent skills and behaviors. In prior work (Hussain, 2024), we proposed a novel framework for creating "drivers" of intelligence—motivational mechanisms that support the development of AGI across various AI techniques. This modular, extensible framework operates at multiple levels: assessing individual components of intelligence, guiding an intelligence as it learns, facilitating collaboration between intelligences, and evaluating them to determine overall capability for intelligence. Importantly, these drivers span both subjective, human-like goals (such as ethical behavior) and more objective, task-based criteria. They can be tailored to specific real-world contexts, reinforcing human-like behaviors, or designed for entirely different environments, encouraging more “alien” behaviors.

This proposed research will develop a generalizable, scalable motivational framework for use within the PRIMUS Hyperon infrastructure. The goal is to create a unified, modular system for defining, representing, and applying diverse motivational drivers—agnostic to specific AI techniques or training methods—enabling broad applicability across AGI systems.

Approach

The research will be primarily conceptual, focused on building a representational framework that supports diverse motivational mechanisms (“drivers”) in a way that is interoperable across AGI architectures. The long-term intent of this framework is to support building an ever-expanding shared repository of actionable motivational drivers to enable the development of ever more complex AGI.

The concept of a motivational driver itself will be fleshed out to capture its context of application, interactions with other drivers, how those interactions evolve across situations, and how the motivations themselves adapt over time. The research will concentrate on two key goals:

1. Defining Motivational Drivers: Capturing a robust definition of drivers that encompasses an arbitrarily diverse variety of purposes, contexts and interactions.
2. Defining How to Apply Drivers in AGI Learning: Capturing methods for how these drivers can be used effectively to appropriately influence the learning and adaptation of AGI systems. 

The core challenge is to create an agnostic framework that can work across different AI techniques while ensuring the drivers remain context-sensitive, dynamic, and able to evolve over time.

While primarily conceptual, the project will also involve preliminary prototyping using MeTTa to demonstrate the feasibility of representing and applying diverse drivers in AGI training processes.  This will offer practical insights into implementation and inform future development.

Impact on Hyperon

The proposed motivational framework will enable AGI systems within Hyperon to define, evaluate, and evolve their internal drives over time. Unlike fixed utility functions, each driver will be dynamic – interacting supportively or competitively with others and adapting based on experience. This supports AGI that reasons not just about "how" to act, but "why" to act, enabling adaptive, context-aware decisions.  While the primary focus will be MeTTa-based representations, interactions with other Hyperon components will also be explored.  An early thought experiment below illustrates this potential integration.

Consider the behavioral patterns “practice makes perfect” and “familiarity breeds contempt.” These arise from distinct motivational architectures.

In a human-aligned AGI, core drives such as Recognition (from Maslow’s esteem needs) and Personal Growth (linked to self-actualization) shape behavior. Recognition drives repetition and refinement to earn approval or status, reinforcing “practice makes perfect.” Growth motivates learning, autonomy, and exploration—leading the AGI to disengage from tasks once they no longer provide stimulation, echoing “familiarity breeds contempt” as learning value drops.

By contrast, an alien AGI guided by efficiency and risk reduction behaves differently. It reinforces repetition when it improves precision or reduces cost, favoring efficiency. Yet, to avoid over-specialization, it may engage in controlled exploration—not from curiosity, but as a hedge against long-term fragility.

These motivational dynamics map align with key elements of Hyperon’s architecture. In ECAN, drivers like Recognition, Growth, Efficiency, and Risk Reduction could interact to prioritize goals. A human-aligned AGI may shift attention toward novelty when learning potential declines, while an alien AGI may stick with refinement unless exploration is justified by projected risk.

In DAS, motivational structures, historical evaluations, and relational mappings are encoded and updated over time. This allows AGIs to adapt their internal priorities based on evolving experience — whether reinforcing a learned behavior or generalizing a strategy to avoid overfitting.

The core adaptive capability emerges in MeTTa, where motivational rules are not just expressed but dynamically rewritten. Meta-rules can detect patterns (e.g., "repeated success with Task X") and rewrite the conditions under which a driver activates or competes. This allows the AGI to evolve its own motivational logic, not just its priorities — enabling behaviors like “practice makes perfect” or “familiarity breeds contempt” to arise from deeper principles.

By grounding higher-level motivational patterns in foundational needs—be they human psychological (Recognition and Growth) or alien computational (Efficiency and Risk Reduction)—this framework enables Hyperon to support AGI systems that exhibit strategic, context-sensitive, and purpose-driven behavior, well beyond task execution.

Flexible Motivational Framework

This framework will be modular and scalable, designed to accommodate various AGI architectures and motivations. For human-like AGIs, it could draw on psychological principles (e.g., Maslow's hierarchy), while for alien intelligences, it could incorporate entirely novel motivational constructs.

Central to this flexibility is the ability to define modular motivational drivers—such as goal orientation, curiosity, social bonding, or ethical behavior—that can be adjusted in priority or structure based on the AGI's evolving needs. These drivers will adapt over time as the AGI encounters new environments or experiences, ensuring continuous learning and adaptation.

Detailed Use Cases

1. Chatbot Systems: In chatbot systems, the framework could guide adaptive dialogue strategies, adjusting responses based on the user's emotional state and intent. For instance, a chatbot might prioritize empathy in sensitive contexts or focus on task completion in practical ones. Over time, it would adjust its motivational priorities based on ongoing interactions, learning to balance between providing information and building rapport.
2. Humanoid Robots: For humanoid robots, this framework could govern behavior in social settings, such as prioritizing human well-being, security, or cooperation. In a healthcare setting, for example, a robot might balance ethical guidelines (e.g., ensuring patient comfort) with practical concerns (e.g., completing tasks). Motivational priorities would adapt based on environmental changes, such as the arrival of new patients, and shift between autonomy (in assisting with routines) and collaboration (working with human staff) as needed.

Scalability and Adaptability

The motivational framework’s scalability will rely on its ability to manage a distributed knowledge structure (e.g., an Atomspace) where data and motivational states are not only recorded but can evolve in response to real-time environmental shifts. This scalability ensures that large, complex networks of AGIs can adapt to the varying demands of different environments, adjusting priorities on-the-fly.  In particular, we will explore context-aware motivation scaling, where an AGI adjusts motivation levels based on internal and external factors like resource availability, task complexity, or the presence of new agents. We will also explore self-organizing principles that enable the AGI to refine its motivational state continuously as it learns from both internal feedback and environmental cues.

For large-scale systems, such as fleets of autonomous robots, this adaptability is crucial. Each robot might prioritize different aspects (e.g., speed, accuracy, resource conservation) depending on the current task complexity, external factors like weather or traffic conditions, and the overall state of the system — ensuring dynamic system-wide adjustment

Ethical Alignment

Ensuring ethical alignment is a core consideration for AGI systems, especially as they gain autonomy. The framework will enable explicit specification of ethical reasoning mechanisms to prioritize human-centered values, such as fairness, safety, and respect for autonomy. These drivers could be based on established ethical models like utilitarianism (maximizing well-being for all agents), deontology (adhering to moral rules), and virtue ethics (promoting moral character and flourishing).

For instance, in a healthcare environment, an AGI might prioritize patient welfare while balancing efficiency and cost-effectiveness. The system would need to resolve conflicts between these drivers and adjust its behavior accordingly, ensuring that its actions remain socially and morally beneficial.

Foundation for Future Research

The proposed driver-based motivational framework will lay the groundwork for future research in AGI. It offers a robust platform for experimenting with various motivational systems and testing how AGIs can adapt to complex, dynamic environments and social contexts. Moreover, its modular design will allow for the continuous integration of new theories, models, and findings from psychology, ethics, and computational neuroscience, ensuring Hyperon remains at the cutting edge of AGI research.

Open Source Licensing

3-clause BSD license will apply to all components related to the software for creating drivers. https://opensource.org/license/bsd-3-clause

For the drivers created using the system, Creative Commons licensing CC BY (https://creativecommons.org/licenses/by/4.0/) or CC BY-SA (https://creativecommons.org/licenses/by-sa/4.0/) will be required for released drivers.  The goal of this project to to create the foundation for encouraging researchers and developers to share their motivational structures.  If a driver is not released publicly or privately (e.g., retained as a trade secret and used to train proprietary systems), then this requirement does not apply.  The trained AI products created using the drivers are unrestricted by any licensing.

Background & Experience

Dr. Hussain has over 20 years of industry research experience in the areas of AI, evolutionary computation and neural networks as well as instructional technology, decision-support systems and more.  He has extensive experience in building research prototypes, supporting deployed systems and in creating new representation languages and authoring systems (see https://talibhussain.com for details).  He uses a broad range of languages and tools, and loves defining new project visions and carrying them through to successful execution.  He recently has focused on investigating what is needed to achieve AGI, with the viewpoint that fundamental changes in our methods and collaborative approaches are needed.  (See https://drive.google.com/file/d/1i71SvtvfXlGP53627JqSj8Qb6XHNv7Ff/view for preprint of his 2024 article).  He leads the Innovation Hub effort at John Abbott college which supports students working on real-world projects, and has led 3rd year interns on researching AI-based tools.

Links and references

Hussain, T.S. (2024). Let’s Evolve Intelligence, Not Solutions. In: Winkler, S., Trujillo, L., Ofria, C., Hu, T. (eds) Genetic Programming Theory and Practice XX. Springer, Singapore.

Hussain, T.S. (2011). "A meta-model perspective and attribute grammar approach to facilitating the development of novel neural network models," in N. Jankowski, W. Duch and K. Gabczewski (Eds.), "Meta-Learning in Computational Intelligence," Springer, p. 245 - 272.