論2024年WHO發表之「人工智慧於健康領域之倫理 與治理」與台灣現況

Ethics and Governance of Artificial Intelligence for Health and Current Situation in Taiwan

現代人工智慧(artificial intelligence，以下簡稱為AI)從數據中學習後並將演算法整合到系統與工具中，使其能夠自動執行任務，讓人類不須再對每一個步驟進行明確的編寫程式。生成式AI為一類AI技術，其演算法被訓練能生成具有原創性的內容，例如文章、圖像或影片等等。世界衛生組織(WHO)首次在2021年6月底公布了一份全面指南：「人工智慧於健康領域之倫理與治理」(Ethics and governance of artificial intelligence for health)。WHO諮詢了20位AI領域頂尖專家並指出了在醫療保健中使用AI的潛在好處和風險，並提出了六項經過共識達成的原則供政府、研究者與醫療提供者在使用AI時參考。這些原則包括：(1)保護人類自主權；(2)促進人類福祉、安全、和公共利益；(3)確保透明度、可解釋性、和可理解性；(4)確立責任歸屬和問責制；(5)確保包容性和公平性；(6)促進具適應性及可持續之人工智慧。

WHO於2024年1月再次更新本指南並探討了生成式AI，即大型多模態模型(Large Multimodal Models, LMMs)，其能接受多種類型的數據輸入並生成多樣的輸出。有研究預測稱LMMs將在醫療保健、科學研究、公共衛生和藥物開發中得到廣泛應用、亦被稱為「通用目的基礎模型」。儘管尚未證明LMMs是否能夠勝任各種程度的工作，不可否認的是其被廣泛應用的速度比歷史上任何工具都更快。LMMs的吸引力在於其透過模仿人類溝通的方式促進了「人機互動」，並在詢問或數據輸入後生成「看似人類」或甚至「權威般」的回應。由於其快速地被大眾接受並使用，以及其潛在顛覆傳統社會服務模式的能力，許多政府機關及企業政府均積極投入發展生成式AI。

WHO發布這份指南，旨在幫助評估使用LMMs在醫療保健方面的權益和挑戰，並制定相對應的政策。本文將簡介這篇指南中關於在醫療領域使用AI的建議與利弊，並討論現今歐盟、美國及台灣在AI領域相關法令以及發展。

 

Modern AI (artificial intelligence) learns from data and automates tasks, while generative AI creates original content like articles, images, or videos. WHO’s comprehensive guideline “Ethics and Governance of Artificial Intelligence for Health” released in June 2021 outlines six principles for AI applications in healthcare, including protecting human autonomy and promoting transparency. The principles aim to guide governments, researchers, and healthcare providers in utilizing AI responsibly.

In January 2024, WHO updated its guidelines, focusing on generative AI, particularly Large Multimodal Models (LMMs), which is capable of producing diverse outputs from various data inputs. Despite uncertainties regarding their capabilities, research suggests that LMMs will have broad applications in healthcare and beyond, known as “general-purpose baseline models.” Their rapid adoption is attributed to their ability to mimic human interaction, potentially disrupting traditional social services, leading to significant investments from governments and enterprises.

WHO releases this guideline to assist in assessing the benefits and challenges of using LMMs in healthcare and to formulate corresponding policies. This article provides an overview of the recommendations and the pros and cons of using AI in the medical field outlined in this guideline, as well as discuss the current EU, US, and Taiwan regulations and developments in the field of AI.