Artificial intelligence: angel or devil?

Tieniu Tan, Zhaoxiang Zhang, Zhenan SunScientia 2018 · DOI

Abstract: Artificial intelligence (AI) achieved important breakthroughs under the joint impetus of an unprecedented volume of data and exponentially growing computing capacities, and therefore became a significant focus of competition among the big powers. Meanwhile, AI showed its transformative impacts on human society and caused increasing public concern as well as considerable controversy. This article tries to draw a realistic panorama of AI at the macro level, including basic concepts, development history, current situation, and future trends; to reveal an unbiased profile and growing rules; and to develop a scientific perspective and realistic expectations of social knowledge. Finally, the article proposes an initiative to face the potential safety risks and challenges. This requires the strengthening of forward-looking prevention and guidance on restraint, the minimization of risk, and ensuring the safe, positive, and profitable development of AI.

Summary

The paper argues that while AI has reached a "tipping point" thanks to big data and deep learning, it remains in its infancy, excelling at specialized tasks but failing at general reasoning. The authors advocate for a shift toward human-machine hybrid intelligence and autonomous learning, while emphasizing that AI’s impact—whether "angelic" or "demonic"—depends entirely on our ability to establish proactive legal and ethical frameworks.

Introduction: The Dual Nature of AI

In a strategic overview published in Scientia Sinica Informationis, researchers from the Chinese Academy of Sciences explore the "macro-panorama" of Artificial Intelligence. The discourse surrounding AI is often polarized between the "Angel" camp, which sees it as the engine of the fourth industrial revolution, and the "Devil" camp, which warns of existential threats from figures like Elon Musk and Stephen Hawking. This paper seeks to cut through the hype, providing a scientific perspective on whether AI will ultimately serve as a beneficial force or a destructive one.

The core challenge identified is the gap between public expectations of sentient machines and the reality of current technology—powerful but brittle. The authors advocate for a transition from narrow, data-hungry "Specialized AI" to robust "General AI" that can reason and coexist safely with human society.

A Historical Perspective

The authors contextualize the current boom by dividing AI's 60-year history into six distinct stages: Start, Reflection, Application, Low Point, Steady Growth, and the current Boom. This historical lens reveals that while progress has been non-linear, the current era is uniquely defined by the convergence of big data and deep learning.

The Current State: Specialized AI Dominance

Today, AI excels in narrow domains where it has matched or exceeded human performance. The paper cites several industry-standard benchmarks to quantify this state of the art:

  • ImageNet: Top-5 error rates dropped from 16% in 2012 to approximately 3% by 2017.
  • Labeled Faces in the Wild (LFW): Facial recognition has surpassed human accuracy.
  • Speech Recognition: Microsoft achieved a 5.1% error rate, reaching parity with professional transcribers.

Despite these successes, the authors warn that we are in a "Technical Valley"—transitioning from unusable to usable technology, but still far from "highly effective" general systems.

The Intelligence vs. Wisdom Gap

A pivotal insight of the paper is the distinction between intelligence and wisdom. Current AI possesses "IQ but no EQ"; it can calculate and recognize patterns but lacks the ability to reason or understand "common sense." Most current AI is "Artificial" because it relies heavily on human effort for data labeling. However, systems like AlphaZero signal a shift toward "Autonomous Intelligence," where machines learn through self-play rather than human-curated data.

Strategic Implications and Global Competition

AI has become a "strategic high ground" for national security and global competition. By 2018, 25 countries had signed the EU's Declaration of Cooperation on AI. In China, the core AI industry is projected to exceed 1 trillion RMB by 2030, driving related industries worth over 10 trillion RMB. This economic potential necessitates a shift toward human-machine hybrid intelligence, where human cognitive models are integrated into AI loops to handle complex, low-data scenarios.

Limitations and the Path to Ethical Governance

The paper acknowledges significant hurdles, including the "Black Box" problem of deep learning and the high energy dependency of current systems. To ensure AI remains an "angel," the authors call for proactive adherence to principles like the "Asilomar AI Principles" and international cooperation to prevent malicious uses, such as deepfakes or autonomous weapons.

About the Authors

The strategic overview was led by researchers from the Chinese Academy of Sciences, focusing on the macro-panorama of AI development.

Researcher from the Chinese Academy of Sciences

Conclusion

The impact of AI—whether angelic or demonic—depends entirely on human guidance. By strengthening forward-looking prevention and establishing robust ethical frameworks, society can minimize risks and ensure the safe, positive development of this transformative technology.

Summary updated

Abstract

Artificial intelligence (AI) achieved important breakthroughs under the joint impetus of an unprecedented volume of data and exponentially growing computing capacities, and therefore became a significant focus of competition among the big powers. Meanwhile, AI showed its transformative impacts on human society and caused increasing public concern as well as considerable controversy. This article tries to draw a realistic panorama of AI at the macro level, including basic concepts, development history, current situation, and future trends; to reveal an unbiased profile and growing rules; and to develop a scientific perspective and realistic expectations of social knowledge. Finally, the article proposes an initiative to face the potential safety risks and challenges. This requires the strengthening of forward-looking prevention and guidance on restraint, the minimization of risk, and ensuring the safe, positive, and profitable development of AI.
BibTeX
@inproceedings{tan2018artificial,
  author = {Tieniu Tan and Zhaoxiang ZHANG and Zhenan Sun},
  title = {{Artificial intelligence: angel or devil?}},
  booktitle = {Scientia},
  year = {2018},
  doi = {10.1360/n112018-00142},
}
Figure