AI, Help Me Think$\unicode{x2014}$but for Myself: Assisting People in Complex Decision-Making by Providing Different Kinds of Cognitive Support

Abstract: How can we design AI tools that effectively support human decision-making by complementing and enhancing users' reasoning processes? Common recommendation-centric approaches face challenges such as inappropriate reliance or a lack of integration with users' decision-making processes. Here, we explore an alternative interaction model in which the AI outputs build upon users' own decision-making rationales. We compare this approach, which we call ExtendAI, with a recommendation-based AI. Participants in our mixed-methods user study interacted with both AIs as part of an investment decision-making task. We found that the AIs had different impacts, with ExtendAI integrating better into the decision-making process and people's own thinking and leading to slightly better outcomes. RecommendAI was able to provide more novel insights while requiring less cognitive effort. We discuss the implications of these and other findings along with three tensions of AI-assisted decision-making which our study revealed.

• The paper compares two AI decision support types, finding ExtendAI improves financial diversification by integrating feedback into user reasoning, despite increasing cognitive load.
• The study reveals a split in user preference between the ease of direct recommendations and the value placed on AI that maintains cognitive involvement.
• While ExtendAI increased cognitive effort due to requiring rationalization, users reported higher satisfaction and a stronger sense of agency with this process.

Analyzing AI's Role in Decision-Making: Insights from the Study on ExtendAI and RecommendAI

The paper "AI, Help Me Thinkbut for Myself: Assisting People in Complex Decision-Making by Providing Different Kinds of Cognitive Support," investigates an alternative framework for AI-assisted decision-making that moves beyond traditional recommendation systems. The authors introduce two AI modalities: ExtendAI, which integrates AI-generated feedback into the user's decision-making rationale, and RecommendAI, which generates direct recommendations. The study evaluates the efficacy, cognitive impact, and user perceptions of these approaches within the context of financial investment decisions.

A mixed-methods user study enlisted participants with varying degrees of familiarity with exchange-traded funds (ETFs) to engage with both AIs during an investment simulation task. The study sought to unveil which AI paradigm more effectively secures integration into the decision-making process without undermining the user's cognitive engagement.

Key Findings and Numerical Results

1. Cognitive Engagement and Decision-Making: ExtendAI prompted users to reflect critically on their decisions by assimilating the AI's feedback into their reasoning, which often resulted in improved portfolio diversification. Participants' satisfaction with ExtendAI highlights the method's effectiveness in fostering a more analytical approach, which correlates with enhanced decision outcomes.
2. User Preferences and AI Interaction: The study illuminates a split in user preferences, with approximately half of the participants favoring the directness and ease of use associated with RecommendAI. However, some participants perceived RecommendAI's approach as potentially diminishing their cognitive involvement, which led them to appreciate the more integrative feedback from ExtendAI.
3. Cognitive Load: A notable result was the higher cognitive effort reported by users of ExtendAI. This additional burden arises from the need to formulate a detailed rationale before receiving feedback. Despite this, participants reported a strong sense of agency and satisfaction with the decision-making process.
4. Impact on Portfolio Diversification: ExtendAI users showed a greater increase in diversification across sectors and country allocations within their investment portfolios. This suggests that the nuanced feedback provided by ExtendAI was more effective in encouraging users to consider broader diversification strategies as opposed to merely aligning with conventional investment heuristics.

Implications and Future Directions in AI Research

The implications of this research are multifaceted, offering insights into designing AI systems that complement rather than supplant human intuition in decision-making processes. The deployment of AI systems like ExtendAI could be particularly beneficial in domains demanding nuanced judgements, such as financial planning, healthcare, and strategic management. The study also prompts further investigation into the balance between cognitive load and the quality of decision outcomes.

Future development in AI systems should aim to optimize the balance between providing actionable insights and preserving user engagement. Investigating adaptive systems that can modulate advice strength based on user expertise levels or task complexity could enhance user satisfaction and decision quality. Additionally, exploring multi-turn interaction paradigms could help maintain a balance between cognitive load and user engagement across varied decision-making contexts.

In conclusion, this paper advances the discourse on human-AI synergy by revealing how AI can augment human reasoning without overwhelming or bypassing user intellect. It provides a foundation for future research seeking to harness the full potential of AI in enhancing complex decision-making tasks.