SAE Automation, 10 AI types and future governance

Artificial intelligence (AI) is no longer a futuristic fantasy; it’s rapidly becoming an integral part of our daily lives, from the smartphones in our pockets to the complex systems powering industries. As AI capabilities expand and grow more sophisticated, so does the need to understand and categorize these evolving forms of intelligence. This understanding is crucial not only for managing expectations but also for addressing safety concerns and guiding regulation.

A well-accepted framework for categorizing AI is starting to take shape, and one of the most successful examples comes from the automotive sector. SAE International, a global engineering standards organization, developed the J3016 standard, which outlines six levels of driving automation. This classification acts as a common language among engineers, regulators, and consumers, offering clarity and setting clear expectations for driver responsibility.

The SAE Levels of Driving Automation

• Level 0: No Driving Automation
  At this level, a human driver is entirely responsible for all aspects of driving, even though some safety warnings or momentary intervention systems (like automatic emergency braking) may exist.

• Level 1: Driver Assistance
  Here, the system can assist with either steering or acceleration/deceleration, but not both simultaneously. The driver must remain engaged at all times. Examples include systems like adaptive cruise control.

• Level 2: Partial Driving Automation
  The system can control both steering and acceleration/deceleration under certain conditions, but the driver must monitor the environment and be prepared to take control. Current advanced driver-assistance systems, such as Tesla’s Autopilot, fall into this category.

• Level 3: Conditional Driving Automation
  The automated system can manage all driving tasks within a specified Operational Design Domain (ODD), such as on highways in clear weather. The driver may disengage but must be ready to take control when requested.

• Level 4: High Driving Automation
  At this level, the system performs all tasks and can handle fallback situations (like system failure), but only within its ODD. No human attention is required here, allowing for services such as robotaxis in geofenced urban areas.

• Level 5: Full Driving Automation
  This ultimate stage allows the system to perform every driving task under any conditions a human driver would manage. Currently, such systems remain theoretical.

The SAE levels have proven invaluable for providing clarity, guiding development, and informing regulatory discussions in the complex field of autonomous vehicles.

Towards a Universal ‘Levels of AI’ Standard?

The SAE model raises an interesting question: could a similar “Levels of AI” framework be developed for a broader range of intelligent machines, including robots and software AI? Such a standard could yield numerous benefits. Imagine if product labels clearly indicated an AI’s capabilities and limitations, akin to energy efficiency ratings. This could enhance consumer understanding, provide a common language for industry benchmarking, and lay out clear thresholds for safety testing and regulatory oversight.

Potential bodies for developing such standards could include international organizations like the ISO (International Organization for Standardization), already working on AI standards like ISO/IEC JTC 1/SC 42. However, devising a universal AI leveling system is challenging. The concept of “intelligence” is multifaceted, making it difficult to define and measure across diverse AI applications.

Exploring a Conceptual 10-Level AI Framework

To tackle this complexity, we can explore a conceptual 10-level framework that encompasses the evolution of AI from basic automation to highly advanced forms. This framework can enhance discussion and understanding of AI capabilities:

1. Rule-Based Systems

   • Overview: AI operating on predefined rules.
   • Examples: Traditional robotic arms, simple automated guided vehicles.

2. Context-Aware Systems

   • Overview: AI that adapts to its environment.
   • Examples: Collaborative robots slowing near humans, smart thermostats.

3. Narrow Domain AI (ANI)

   • Overview: AI specialized for specific tasks.
   • Examples: Autonomous vehicles, voice assistants, recommendation algorithms.

4. Reasoning AI

   • Overview: AI capable of logical inference and problem-solving.
   • Examples: Autonomous vehicles navigating complex environments.

5. Self-Aware Systems

   • Overview: Hypothetical AI possessing consciousness.
   • Real-world examples: None, though it appears in fiction, like HAL 9000.

6. Artificial General Intelligence (AGI)

   • Overview: AI with human-like cognitive abilities.
   • Real-world examples: None, only seen in fictional forms like Data from Star Trek.

7. Artificial Superintelligence (ASI)

   • Overview: AI that significantly surpasses human intelligence.
   • Real-world examples: None, speculative forms exist in films like The Matrix.

8. Transcendent AI

   • Overview: AI evolved beyond human understanding.
   • Real-world examples: None, but it is depicted in films like Her.

9. Cosmic AI

   • Overview: Theoretical AI with cosmic-scale capabilities.
   • Real-world examples: None, fictional entities exist in science fiction.
10. Godlike AI
    • Overview: Purely speculative AI with omnipotent capabilities.
    • Real-world examples: None, seen in fictional works like Star Trek’s Q.

Measuring Intelligence

Currently, there’s no international metric for measuring intelligence uniformly. For humans, we use IQ, a standardized score that’s relative to the population average. But how do we quantify AI intelligence? One potential solution is categorical classification similar to SAE levels, focusing on capabilities like learning flexibility or contextual reasoning rather than attempting a single intelligence score.

The Case for Categorization

Why is categorization crucial? The primary driver is safety. Clear labeling could inform users of AI capabilities and limitations, particularly important for systems that make critical decisions. Transparency fosters trust and manages expectations, helping prevent overhyping or underestimating AI capabilities.

Categorization could also aid in establishing accountability when AI systems cause harm. Developers might find that defining these levels encourages more ethical considerations during AI creation.

However, challenges abound. Defining levels objectively across a varied technology landscape requires monumental effort. The rapid evolution of AI means classification systems must remain adaptable.

A standardized “levels of AI” framework could transform governance, providing guidelines for different testing and regulatory requirements based on an AI’s assessed capability. The ethical implications of such governance warrant careful discussion, as control could either protect the public or stifle innovation.

As we navigate the complexities of AI technology, the dialogue surrounding its capabilities, limitations, and governance frameworks becomes increasingly vital. Managing this balance will be essential for harnessing AI’s immense potential while safeguarding our collective future.

Source link