[論文レビュー] A Survey on Human-AI Collaboration with Large Foundation Models

Vanshika Vats, Marzia Binta Nizam+21 more|arXiv (Cornell University)|Mar 7, 2024

Scientific Computing and Data Management被引用数 12

ひとこと要約

この調査は、Large Pre-trained Models (LPtMs) がHuman-AIチームワークをどのように強化するかを概観し、モデルの改善、共同システム、安全性、分野横断の適用を扱う。方法、課題、倫理的で効果的な協働のための将来の方向性を統合して提示する。

ABSTRACT

As the capabilities of artificial intelligence (AI) continue to expand rapidly, Human-AI (HAI) Collaboration, combining human intellect and AI systems, has become pivotal for advancing problem-solving and decision-making processes. The advent of Large Foundation Models (LFMs) has greatly expanded its potential, offering unprecedented capabilities by leveraging vast amounts of data to understand and predict complex patterns. At the same time, realizing this potential responsibly requires addressing persistent challenges related to safety, fairness, and control. This paper reviews the crucial integration of LFMs with HAI, highlighting both opportunities and risks. We structure our analysis around four areas: human-guided model development, collaborative design principles, ethical and governance frameworks, and applications in high-stakes domains. Our review shows that successful HAI systems are not the automatic result of stronger models but the product of careful, human-centered design. By identifying key open challenges, this survey aims to give insight into current and future research that turns the raw power of LFMs into partnerships that are reliable, trustworthy, and beneficial to society.

研究の動機と目的

• LPtMs がセクターを問わず、Human-AI (HAI) の協働をどのように変革するかを評価する。
• LPtM の訓練と評価に人間の入力を統合する方法を特定する。
• LPtMs によって実現されるHAIシステムの設計、安全性、信頼性の問題を検討する。
• 医療、交通、教育などでLPtMを活用したHAIの適用をマッピングする。

提案手法

• LPtMs を用いたHuman-AIチーミングに関する文献を、Model improvements、Effective HAI systems、Safe and Trustworthy AI、Applications の4つの焦点領域で調査する。
• 従来の HITL アプローチ（Human in the Loop、Active Learning、RLHF、Human Evaluation）を軸に知見を整理する。
• LPtMs を用いたHAI のためのUI/UXおよびシステム設計の考慮点を分類する。
• LPtM-enabled HAI における傾向と課題を示すため、広範な引用研究を参照する。
• レビュー構築には人間とAIの二重の下書きプロセス（著者とAIの支援）を用いる。

(a)

実験結果

リサーチクエスチョン

• RQ1 How do LPtMs reshape model training, evaluation, and deployment in Human-AI teaming?
• RQ2 What are the key challenges (trust, safety, bias, privacy, governance) in LPtM-enabled HAI systems and how can they be mitigated?
• RQ3 What UI/UX and system architectures best support effective, efficient, and ethical HAI collaboration?
• RQ4 What sector-specific applications demonstrate the impact and limitations of LPtM-driven HAI teaming?

主な発見

• LPtMs substantially enhance adaptability, personalization, and conversational capability in HAI teams.
• RLHF, InstructRL, and related feedback methods are central to aligning LPtMs with human preferences and reducing biases, though challenges remain in scaling and consistency.
• Improved UI/UX, intersection of human guidance with AI, and robust evaluation frameworks are critical for trust and effectiveness in HAI systems.
• Applications span healthcare, autonomous vehicles, education, gaming, accessibility, and surveillance, illustrating broad societal implications and deployment considerations.
• Ethical, legal, and policy dimensions (privacy, accountability, labor impact, fairness) are integral to responsible adoption of LPtM-enabled HAI.

(b)