[Paper Review] General Purpose Artificial Intelligence Systems (GPAIS): Properties, Definition, Taxonomy, Societal Implications and Responsible Governance

Isaac Triguero, Daniel Molina+3 more|arXiv (Cornell University)|Jul 26, 2023

Ethics and Social Impacts of AISocial Sciences119 references3 citations

TL;DR

This paper proposes a comprehensive definition and taxonomy of General Purpose Artificial Intelligence Systems (GPAIS), distinguishing between closed-world and open-world GPAIS based on autonomy, adaptability, and self-awareness. It categorizes GPAIS into AI-powered AI and single foundation model approaches, with generative AI and multimodality as key enablers, and advocates for responsible governance and trustworthiness frameworks to ensure safe, controllable deployment of these systems.

ABSTRACT

Most applications of Artificial Intelligence (AI) are designed for a confined and specific task. However, there are many scenarios that call for a more general AI, capable of solving a wide array of tasks without being specifically designed for them. The term General-Purpose Artificial Intelligence Systems (GPAIS) has been defined to refer to these AI systems. To date, the possibility of an Artificial General Intelligence, powerful enough to perform any intellectual task as if it were human, or even improve it, has remained an aspiration, fiction, and considered a risk for our society. Whilst we might still be far from achieving that, GPAIS is a reality and sitting at the forefront of AI research. This work discusses existing definitions for GPAIS and proposes a new definition that allows for a gradual differentiation among types of GPAIS according to their properties and limitations. We distinguish between closed-world and open-world GPAIS, characterising their degree of autonomy and ability based on several factors such as adaptation to new tasks, competence in domains not intentionally trained for, ability to learn from few data, or proactive acknowledgment of their own limitations. We propose a taxonomy of approaches to realise GPAIS, describing research trends such as the use of AI techniques to improve another AI (AI-powered AI) or (single) foundation models. As a prime example, we delve into GenAI, aligning them with the concepts presented in the taxonomy. We explore multi-modality, which involves fusing various types of data sources to expand the capabilities of GPAIS. Through the proposed definition and taxonomy, our aim is to facilitate research collaboration across different areas that are tackling general purpose tasks, as they share many common aspects. Finally, we discuss the state of GPAIS, prospects, societal implications, and the need for regulation and governance.

Motivation & Objective

• To address the lack of a clear, scalable definition for General Purpose AI Systems (GPAIS) that captures varying degrees of autonomy and capability.
• To develop a taxonomy of GPAIS based on underlying mechanisms, distinguishing between AI-powered AI and single foundation models.
• To analyze the role of generative AI and multimodality in advancing GPAIS capabilities.
• To identify open challenges and societal implications of GPAIS, especially regarding safety, control, and trust.
• To advocate for responsible governance and regulatory frameworks to ensure trustworthy and controllable development of GPAIS.

Proposed method

• Proposes a new definition of GPAIS emphasizing properties such as few-shot learning, transfer learning, and proactive self-awareness of limitations.
• Introduces a taxonomy distinguishing closed-world GPAIS (limited adaptation) from open-world GPAIS (high autonomy and self-improvement).
• Categorizes GPAIS into two main families: AI-powered AI (using AI to design or optimize other AI) and single foundation models (monolithic models trained on broad data).
• Analyzes generative AI and multimodal systems as key realizations of GPAIS, emphasizing their role in enabling zero-shot and few-shot generalization.
• Maps GPAIS trustworthiness to the NIST AI Risk Management Framework, identifying 150 trustworthiness properties across seven dimensions (e.g., safety, fairness, explainability).
• Proposes governance mechanisms including legal accountability for developers and safety standards based on model capabilities.

Experimental results

Research questions

• RQ1 How can GPAIS be formally defined in a way that captures varying degrees of autonomy and generalization ability?
• RQ2 What are the key technical approaches to building GPAIS, and how do they differ in architecture and capability?
• RQ3 To what extent do current foundation models and generative AI systems exhibit true generalization beyond their training distribution?
• RQ4 How can GPAIS be made trustworthy, safe, and controllable, especially in open-world scenarios?
• RQ5 What regulatory and governance frameworks are necessary to ensure responsible development and deployment of GPAIS?

Key findings

• The proposed GPAIS definition enables a nuanced distinction between closed-world and open-world systems based on adaptability, self-awareness, and few-shot learning capability.
• The taxonomy identifies two main pathways to GPAIS: AI-powered AI (multi-AI system orchestration) and single foundation models (monolithic, multi-task learning).
• Generative AI and multimodal systems are shown to be central enablers of GPAIS, particularly through zero-shot and few-shot generalization across diverse inputs.
• Current GPAIS, while not achieving AGI, already exhibit emergent capabilities in reasoning, adaptation, and task transfer, though they lack full human-like consciousness or initiative.
• A governance framework is essential, with legal accountability for developers and safety standards based on model capabilities to prevent foreseeable harms.
• Trustworthiness in GPAIS can be systematically addressed by mapping 150 properties to the NIST AI RMF, covering safety, fairness, transparency, and resilience.