Part 2 — What is Artificial General Intelligence? Can AGI Achieve Human Cognition?

In continuation with the previous blog where I theorized what AGI is NOT, in this blog I dive deeper into explaining what it is. Here, I aim to provide a template to understand what AGI is and contrast them with the multifaceted nature of human cognition.

Ok, What is AGI?

Alan Turing, arguably the most brilliant mathematician in history was posed with the profound question of whether machines could ever think independently. He responded affirmatively.

He asserted, ‘Yes. But it does not have to think like humans in order to behave and execute like humans.’ This bold declaration marked the apex of his groundbreaking imitation game concept within the realm of Turing machines, challenging our very perceptions of intelligence and machine capability.

With that context in mind, let’s delve into defining AGI. We’ll draw upon OpenAI’s interpretation of what they consider AGI to be as follows:

AGI — is a highly autonomous system that outperforms humans at most “economically valuable work”.

This offers an excellent starting point. It smartly defines a limited subset of capabilities. A feature to participate in this set should meet the criteria of ‘economically valuable work’ to be included in the set.

Examples of economically valuable work:

• Professional Tasks: Activities in sectors like law, finance, medicine, and engineering, where expertise and decision-making skills are crucial.
• Creative Work: Involving art, music, writing, and other forms of creative expression, where originality and innovation are valued.
• Scientific Research: Work that contributes to scientific discovery and technological advancement.
• Management and Decision-Making: Leadership roles in organizations, involving strategy formulation, resource allocation, and high-level decision-making.
• Manufacturing and Production: Automation of physical labor-intensive jobs in industries such as manufacturing, agriculture, and construction.
• Data Analysis and IT: Roles involving the analysis of large datasets, IT infrastructure management, and software development.
• Customer Service and Sales: Interacting with customers, understanding their needs, and providing solutions or selling products and services.
• Education and Training: Teaching and training roles that require adapting to individual learners’ needs and styles.
• Logistics and Transportation: Planning and management of the movement of goods and people.
• Healthcare Services: Roles in patient care, diagnosis, and treatment planning.

How Does This Compare With Human Cognition?

In Part-1, we established that AGI is not equivalent to AHI, meaning that replicating human cognition in its entirety isn’t a prerequisite for an autonomous agent to be classified as AGI.

Remeber, AGI ≠ AHI. We are not creating ‘Artificial Human Cognition’ or Artificial Human Intelligence. That’s a tall order.

In discussing aspects of human cognition that are not necessarily required for an Artificial General Intelligence (AGI), we can examine various elements of human mental processes and assess their relevance or necessity for AGI functionality.

AGI, aimed at achieving general-purpose intelligence, doesn’t need to replicate all facets of human cognition, especially those that are uniquely human or not ‘directly’ related to problem-solving and learning capabilities.

Emotional Experience and Processing:

• Subjective Feelings: Human emotions, such as joy, sadness, or anger, are subjective experiences deeply intertwined with consciousness. AGI does not need these subjective experiences to function effectively. There are reinforcement objective functions that can effectively provide same learning feedback for AGI.
• Emotional Responses: While AGI might simulate emotional responses for better human interaction, truly experiencing emotions is not necessary for its intelligence.

Consciousness and Self-Awareness:

• Self-Consciousness: The deep self-awareness and introspection that humans experience, which includes understanding one’s existence and pondering abstract concepts like mortality, may not be relevant for AGI. Executing within the framework of moral values and ethics from an alignment perspective is what AGI needs.
• Qualia: The subjective quality of conscious experience (such as ‘redness’ of red or pain) is unique to biological beings and not essential for AGI’s functionality.

Biologically-Driven Behaviors and Needs:

• Survival Instincts: Instinctual behaviors driven by biological needs like hunger, thirst, or sleep do not apply to AGI.
• Reproduction: The biological urge for reproduction and related behaviors are irrelevant for AGI.

Social and Cultural Context:

• Cultural Norms and Practices: While understanding social and cultural contexts can be important for certain applications, AGI does not inherently require adherence to or participation in these human-specific constructs.
• Personal Relationships: The development and maintenance of personal relationships, a key part of human social structure, is not a prerequisite for AGI.

Mental Health and Psychological States:

• Mood Disorders: Human-like psychological conditions such as depression, anxiety, or other mood disorders are not applicable to AGI.
• Cognitive Biases: Human cognition is often subject to biases and irrationalities; AGI can be designed to minimize or avoid these biases, focusing instead on rational analysis.

Creative and Artistic Sensibilities:

• Aesthetic Appreciation: While AGI can be programmed to recognize and even create art, the deep, subjective appreciation and emotional response to art that humans experience are not necessary for AGI.

Physical Sensations and Perceptions:

• Sensory Experiences: The nuances of human sensory experiences (like touch, taste) are specific to biological organisms and not required for AGI.

What Capabilities Should AGI Possess?

To delve deeper into the concept of Artificial General Intelligence (AGI) and its relationship with human cognition, we need to understand that AGI represents a level of artificial intelligence that is versatile and flexible, akin to human intelligence, but it doesn’t necessarily replicate all aspects of human cognition. Key areas highlight this distinction:

Learning and Adaptation:

• Depth: AGI must demonstrate an ability to learn from limited data and experiences, much like humans do, but it may achieve this through different mechanisms, such as advanced algorithms or neural networks.
• Breadth: It should be able to transfer knowledge and skills across various domains, not just improving through repetitive tasks (as seen in many current AI systems) but truly adapting to new and unforeseen scenarios.

Problem Solving and Creativity:

• Complexity: AGI should handle problems of a complex nature that require abstract thinking, reasoning, and the synthesis of disparate pieces of information.
• Creativity: Unlike narrow AI, AGI should be able to approach problems creatively, potentially devising solutions that are novel and not pre-programmed.

Understanding and Perception:

• Contextual Understanding: AGI would need to interpret data and sensory input in a context-aware manner, understanding subtleties and nuances much like humans do.
• Multimodal Perception: This involves integrating information from various sources (visual, auditory, textual) to form a cohesive understanding of its environment.

Decision Making:

• Ethical Considerations: AGI would be expected to make decisions that consider ethical implications, a complex task that involves understanding human values and societal norms.
• Risk Assessment: Weighing potential outcomes and their probabilities in uncertain conditions, much like human strategic thinking.

Generalization and Flexibility:

• Cross-Domain Application: The ability to apply learned skills and knowledge to entirely new domains is a hallmark of human intelligence that AGI aims to replicate.
• Innovative Application: Going beyond pattern recognition and executing tasks, AGI should be capable of innovatively applying its capabilities in unanticipated ways.

Emotional and Social Intelligence:

• While not necessary for all AGI applications, emotional and social intelligence could be crucial for AGIs interacting closely with humans. This involves understanding and responding to emotional cues and social contexts.

Consciousness and Self-Awareness:

• This is a debated topic in AGI. Consciousness, self-awareness, and subjective experience are complex human characteristics. While these are not required for an AGI, the question of whether they could or should be replicated in AGI remains a philosophical and ethical discussion.

Discussion

This exploration into AGI reveals a complex landscape where advanced technology intersects with elements of human cognition. AGI’s journey is not about emulating human consciousness but about mastering a broad spectrum of skills and knowledge, propelling it into realms of functionality that parallel, yet distinctly differ from, human capabilities. As we venture further into this territory, the fusion of technological advancement with ethical and philosophical contemplation becomes increasingly critical.

In the next part, I will cover “AGI; An Advanced Math Perspective”