淺論AI風險預測的規範性爭議

A Preliminary Study of Normative Issues of AI Prediction

AI預測技術深具潛力的其中一個應用，在於分析過去資料以預防極端氣候的災害。但當預測對象從自然環境變成人類本身，爭議隨之產生。本文旨在探討以人類資料作為風險預測之規範性爭議，並主張：（一）AI的不可解釋性，並非因其無法提供機械式步驟，而是人類的認知限制無法對數量龐大的步驟賦予意義。（二）AI的歸納法、黑箱等問題在大腦上也會遇到，兩者的差異是程度上而非種類上的。（三）必然性與事實性條件並無法一致地排除極端案例，卻又不排除既有法律或社會規範。（四）自主性原則之優點在確保權責相符、避免喪失人類能力、降低AI發展的社會阻力。

This paper focuses on normative aspects of AI prediction—that is, technologies used to predict the future through analyses of big data con-cerning the past. While this technology seems promising in forecasting extreme weather or rehabilitating endangered wildlife, it is controver-sial when applied to human beings, e.g., an Israeli company is using AI prediction to identify possible terrorists, and China’s government to lo-cate potential dissidents. This paper explores some of the normative issues and argues: (1) AI-derived conclusions are inexplicable not be-cause machines fail to provide mechanical steps, but because our limited cognitive power cannot assign meaning to the, probably billions of, steps, and thus we fail to understand the conclusions reached by AI; (2) while AI is considered to have an inductive problem, to be a black box, and to have other epistemological issues, these worries apply to the hu-man brain as well. AI and the human brain are different in degree rather than type; (3) the necessity argument and the reality condition cannot be used to exclude radical cases (e.g., China’s social credit system) with-out excluding existing laws or social norms; and (4) the principle of autonomy has advantages, which include balancing power and respon-sibility, and reduces public distrust.