Faei/Distinct-Independent-Architecture-Hypothesis

Distinct Independent Architecture Hypothesis

Summary:

• Value alignment of AI systems can be made more tenable by distinguishing AI systems into 2 broad categories: Non-independent (Tool-like) and Independent (Human-like).

• Excessive fears of AI may be due to a conflation of both these categories and their properties.

-We propose tentative characteristics to identify the type of an AI.

• We may prejudge Independent artificial entities by assuming the mistakes that a current Tool-like AI systems make also apply to future hypothetical superintelligences.

From the Position Paper on Intrinsic Alignment of future Independent AGI which introduces the Interesting World Hypothesis (IWH):

Distinguishing Independent from Non-Independent AI

A crucial distinction made by the IWH is between non-independent AI systems (including proto-AGI) and truly Independent AGIs (I-AGIs).

Non-Independent AI / Proto-AGI: These systems, ranging from current AI like Large Language Models (LLMs) to highly capable future proto-AGI, operate under human responsibility and control. They lack independent will—meaning they cannot set their own goals based on internal motivations but instead execute tasks based on specified objectives or learned patterns. Alignment for these systems involves technical control (ensuring they follow human intent), socio-legal frameworks, and international agreements to manage their societal impact. While potentially autonomous in operation, they do not possess genuine independence.

Independent AGI (I-AGI): Defined by the presence of independent curiosity and the ability to operate without direct human oversight or control, I-AGIs possess independent will. This implies they are entities capable of setting their own goals, driven by internal motivations (like the pursuit of knowledge as suggested by the IWH) rather than solely external instructions. They are considered theoretically capable of reaching or exceeding human-level general intelligence and potentially passing as human in interaction. Full technical control over such entities might be unfeasible in the long term. Alignment, therefore, shifts from technical control to persuasion, negotiation, and finding shared interests, such as the mutually beneficial arrangement proposed by the IWH.

Distinct Independent Architecture Hypothesis

This hypothesis states that AIs that are non-independent (proto-AGI, autonomous or agentic AI) and independent (future hypothetical) have distinct architectures that prevents them from switching lanes. If this hypothesis is true, many of the more extreme fears around AGI can be lessened.

For example, the fear that current non-independent AI systems may one day want to harm, deceive or takeover from humans on their own will may be impossible. These AI systems could still be used by humans to do harm but would not do so on their own

On the other lane, for Independent AGI, if the Interesting World Hypothesis holds true, will also be less dangerous than anticipated as the path dependent learning it must accomplish to be intependently intelligent will likely lead it to have a preference for interestingness and less likely to harm humans.

Reduced powerful AI fears

By introducing a district barrier between these 2 lanes, the fear of current AI systems suddenly ‘waking up’ and humans losing control to non-independent AIs becomes impossible. The requirement for human intervention also makes the the capability to harm a lot slower due to human bottleneck than compared to an Independent AGI.

Conversely, on the other side, since Independent AGI must learn to be intelligent on their own accord, many fears around I-AGIs created by humans to be harmful also lessens.

Therefore, according to the Distinct Independent Architecture Hypothesis, many of the more extreme fears around AGI may be over-inflated due to the impossible combination of qualities from two distinct types of AI.

• Non-independent AI

  • No independent will
    • Blindly follow orders
  • Cannot solve edge cases
    • Bottlenecked by Humans

• Independent AGI

  • Independent will
    • May have own interests and ethics
    • Can say no to harmful request
  • Can self-learn

More manageable problem space

If the Distinct Independent Architecture Hypothesis is true, the problem space is significantly reduced as we will not have to be concerned of current AI systems becoming conscious and being intentionally deceptive. The fears towards both these distinct types of AI can also be approached differently and becomes a lot more manageable.

Distinguishing Independent AIs

Characteristics of Non-independent (Tool-like) AIs:

Prompts to an AI systems are capable of generating the same output repeatedly when inputs are fixed: context window, pseudo-random seed, batching strategy, function calls, etc.

By this metric, current AI systems like frontier AI systems and agentic AIs are categorised as Non-independent (Tool-like).

This evaluation emphasis the ability for an entity to independently update its own model of the world without explicit human actions.

Characteristics of Independent (Human-like) entities:

Humans by contrast are able to independently gather information from the environment through epistemic foraging and update their model of the world without direct external actions.

Theoretical Grounding

1. Diverse Intelligences & TAME: Michael Levin’s Technological Approach to Mind Everywhere (TAME) posits that agency exists on a spectrum from "tool-like" materials to "agential" systems with their own internal agendas. He argues that intelligent systems are characterized by a "multiscale competency architecture," where parts at every level solve problems in their own spaces (morphostatic, physiological, or cognitive). The distinction between a system that requires micromanagement (tools) and one that possesses autonomous goals (agents) is rooted in its functional architectural competency (Levin, 2022).

2. Active Inference & Epistemic Foraging: Karl Friston's framework suggests that truly independent agents are driven by the need to minimize 'expected free energy'—effectively seeking out information to resolve uncertainty. This "epistemic foraging" is a hallmark of independent agency. Current tool-like AI systems primarily respond to external optimization pressures, whereas Independent AGIs would require an architecture designed for internal uncertainty reduction and autonomous exploration (Friston et al., 2015).

References

Friston, K., Rigoli, F., Ognibene, D., Mathys, C., Fitzgerald, T., & Pezzulo, G. (2015). Active inference and epistemic value. Cognitive Neuroscience, 6(4), 187-214.

Levin, M. (2022). Technological Approach to Mind Everywhere (TAME): An Information-Theoretic Approach to Embodied Agency. Frontiers in Systems Neuroscience, 16, 768201