Intelligence Quotient and Intelligence Grade of Artificial Intelligence (1709.10242v2)

Abstract: Although artificial intelligence is currently one of the most interesting areas in scientific research, the potential threats posed by emerging AI systems remain a source of persistent controversy. To address the issue of AI threat, this study proposes a standard intelligence model that unifies AI and human characteristics in terms of four aspects of knowledge, i.e., input, output, mastery, and creation. Using this model, we observe three challenges, namely, expanding of the von Neumann architecture; testing and ranking the intelligence quotient of naturally and artificially intelligent systems, including humans, Google, Bing, Baidu, and Siri; and finally, the dividing of artificially intelligent systems into seven grades from robots to Google Brain. Based on this, we conclude that AlphaGo belongs to the third grade.

• The paper proposes a standard intelligence model and an AI IQ formula based on capabilities in information acquisition, output, storage, and creation.
• It introduces a hierarchical grading system (zero to six) to classify and compare intelligence levels of both AI and human systems.
• Practical application shows current AI systems like Google and Baidu have IQs below a typical human child but demonstrate significant progressive capability.

Standard Intelligence Model and AI IQ Evaluation

The paper by Feng Liu, Yong Shi, and Ying Liu offers a rigorous framework for assessing the intelligence quotient (IQ) of AI systems through a unified model, the "standard intelligence model." The model synthesized traditional concepts of intelligence, primarily drawing inspiration from the von Neumann architecture, David Wechsler’s multi-factor human intelligence model, and the Data-Information-Knowledge-Wisdom (DIKW) paradigm. This model functions on four primary axes: input, output, information mastery, and creative capability.

Standard Intelligence Model

The standard intelligence model comprises four definitive characteristics that frame a system's intelligence: the ability to absorb external data, transform it into internal knowledge, innovatively utilize this knowledge, and articulate this knowledge back to the external environment. Inspired by the von Neumann architecture, the model extends the paradigm by introducing creative and external knowledge storage components. This provides a semblance of 'aliveness' and adaptability to both human and machine intelligence systems.

Quantifying AI Intelligence

The authors propose an artificial intelligence IQ (AI IQ) metric, measured by analyzing an AI system's ability to acquire, master, create, and feedback knowledge. The AI IQ formula is as follows:

$$Q = f(I, O, S, C) = a \cdot f(I) + b \cdot f(O) + c \cdot f(S) + d \cdot f(C)$$

where $I$, $O$, $S$, and $C$ represent the system’s capabilities of information acquisition, output, storage, and creation, respectively, and the coefficients are their assigned weights.

The research delineates a stratified hierarchy of intelligence grades, from zero to six, which allows evaluation, ranking, and comparison of both human and AI systems. The placement at different intelligence grades facilitates distinguishing an AI system’s capability, from basic I/O functions to complex knowledge sharing and innovation.

Practical Implications

Table 2 in the paper illustrates the advancement of AI IQ from 2014 through tests conducted on search engines and intelligent systems. The notable progress of Google and Baidu systems showcases a significant leap, positioning them yet below the expected human intelligence benchmark (a six-year-old human child). This highlights both an encouraging development trajectory for AI systems and a remaining gap in reaching higher levels of self-directed innovation.

Applications and Future Directions

The methodology proposed here provides a systematic method for grading AI and offers a lens to track AI progress quantitatively versus human intellect. This grading model signifies a structured approach to monitor AI systems, aiding in predicting growth vectors and orienting AI advancements within predictable bounds. Furthermore, the model and AI IQ metric could aid in developing safety and ethical guidelines, especially when considering the implications of advanced AI systems potentially surpassing human-equivalent intelligence.

The authors envision future AI progression along one of two paths. AI may eventually meet or surpass human intelligence or remain asymptotically close. Continual assessment using the AI IQ framework may validate either scenario.

Conclusion

The paper contributes a comprehensive quantitative intelligence evaluation system critical for AI development assessment. By introducing a standardized intelligence model, defining an AI IQ, and lending insights through hierarchical grading, the authors potentially lay the groundwork for structured, progressive AI benchmarking that could influence research directions, policy-making, and AI deployment strategies, holding considerable weight in AI's progressive narrative.