AI's limitation

AI的發展迅速, ChatGPT一炮而紅, 用作搜集資料、分析數據、造文章等, 確實用途廣, 省時又方便。上一篇有網友留言舉了印度法庭也用AI來審案, 但「審案」這講法是需要注釋的, 該案的法官是用AI幫助他決定是否批准被告保釋, AI寫了96頁紙的理據, 建議讓被告保釋, 最後法官都不跟從建議, 而把被告收監。所以最終的決定也並非AI的決定。如果是刑事案, 涉及判斷證人是否可信的, 我看不到AI怎樣可以輔助法庭, 若果只涉法律原則、案例等, AI毫無疑問在搜尋功能方面會比一般律師強大, 故此幫助判決, 只限於法律分析的層面。

靠AI搜集資料, 也不是100%準確的, 我舉兩個例子。第一個例子是我早幾天問(openai.com)
香港2023年的公眾假期, 關於4月份的, 竟然給我這答案:

Ching Ming Festival: Thursday, 6 April 2023
Good Friday: Friday, 14 April 2023
Easter Monday: Monday, 17 April 2023

這答案當然很錯, 這AI犯錯的例子卻也沒有足夠代表性。我又舉另一例, 澳洲墨爾本其中一個市長打算控告AI誹謗。先打個岔。澳洲是我印像中全球最多告人誹謗的國家, 多到匪夷所思的程度。前些時被控在國會部長辦公室強姦的被告Bruce Lehrmann, 控方在重審時考慮到受害人在精神上的打擊而決定撤銷檢控後, Bruce Lehrmann隨即控告3間傳媒誹謗, 案件在審訊中。去年爆出因涉嫌在大學時期強姦女同學而辭去律政司長一職的Christian Porter也告ABC誹謗; 在阿富汗獲戰勳被傳媒報導涉及戰爭罪行的Ben Roberts-Smith也告傳媒。十多年前有個原住民出身的裁判官因傳媒報導她偏頗也控告傳媒誹謗。誹謗案在此無日無之。

墨爾本其中一個市長打算控告AI誹謗又究竟是甚麼一回事? 這是ABC的報導:

A regional mayor may take unprecedented legal action to sue the creator of AI chatbot ChatGPT for defamation, after the program falsely claimed he was a guilty party in a foreign bribery scandal.

Brian Hood, the mayor of Hepburn Shire Council, west of Melbourne, was recently alerted to the fact that ChatGPT, a chatbot designed to provide detailed responses to users' questions, had incorrectly described his role in a foreign bribery incident in the early 2000s.

Councillor Hood did previously work for the company involved, Note Printing Australia, but was actually a whistleblower who told authorities about bribe payments being made to foreign officials to win contracts.

Brian Hood是吹哨人, 也是控方證人,  竟然被錯誤描繪成罪犯,  這AI的搜尋功能可謂不知所謂了。不過,  要告AI背後的公司也不容易, jurisdiction是第一個問題, 另一問題是liability, 公司是否vicariously liable for the act of the chatbot? 這提控將會是世界先例, 我期待Brian Hood去馬讓大家開眼界。