談判是生活和工作上場會遇到的事情，過去累積談判經驗需要大量的訓練或實際經驗，而職場新人往往缺乏訓練就被迫參與談判。本研究使用ChatGPT設計一個談判機器人，讓使用者可以透過談判機器人練習談判累積經驗。接著使用談判機器人安排三組實驗，結果發現人對人的談判花費需要最多的次數，但是雙方共識差異較小；機器對機器的隨機變異非常大，使用人工智慧談判代理人要非常謹慎；人對機器可以透過策略和欺騙使ChatGPT讓步獲得談判的優勢，目前的人工智慧對於人際之間的謀略仍然缺乏足夠的認知。本研究結果可以提供日後設計人工智慧代理人和使用人工智慧代理人參考。

Negotiation is an aspect of life and work one is bound to encounter. Accumulating negotiation experience traditionally requires extensive training or practical experience, yet newcomers to the workforce often find themselves compelled to engage in negotiations without adequate training. This study employs ChatGPT to design a negotiation bot, allowing users to practice and accumulate negotiation experience through interactions with the bot. Subsequently, the study arranges three experimental groups using the negotiation bot. The results indicate that human-to-human negotiations require the most attempts, yet the consensus differences between parties are smaller; machine-to-machine variability is significantly high, necessitating cautious use of artificial intelligence negotiation agents; human-to-machine interactions can exploit strategies and deception to make ChatGPT concede, gaining an advantage in negotiations. Current artificial intelligence lacks sufficient understanding of human stratagems in interpersonal interactions. The findings of this study can provide insights for future design and use of AI negotiation agents.