學者:AI音樂突破關鍵 在「非線性關係

Press/Media

Description

The independent band Nanyang Party NYPD is on the frontline of music, taking the lead in using AI to create music. How does the academic community view it?

Roberto Alonso Trillo, assistant professor at the School of Music of Hong Kong Baptist University, one of whose research interests is "music and AI", accepted an exclusive interview with this newspaper to discuss his views on AI and music creation (Q&A translated and excerpted).

Answer: Roberto Alonso Trillo Question: Yan Jiabao

Q: AI should logically do some monotonous and repetitive tasks to release human creativity. However, today the trend is in the opposite direction, and AI is creating new music. What do you think of this trend?

Answer: Human intelligence is already a simplified model, map, and reflective self-narration, and AI can only be a model-of-a-model. However, as a memory machine, AI digitizes and condenses human experience. If it can move towards a more powerful AI paradigm and go beyond simple statistical inference and imitation without wisdom, it may deviate from the human wisdom we are trying to express and reveal and become a kind of human intelligence. Alien powers that originate within us but transcend our own ability to model and understand the world.

Q: Some musicians use generative AI to create music. Do you think generative AI is a form of creativity? Do you think music created using AI can still be called their work?

A: The term (creativity) retains many meanings - it is British cognitive scientist Margaret A. Boden's evaluation criteria, such as novelty, value, surprise, etc.; or it is the driving force affecting the capitalist free economy, which is most important to us. It's important. If the question is, when content is generated within the realm of art, does it qualify as artistic production, then the question is slightly different. When we arrive at an age where content is sustainably produced, instantly available and fully automated, it may be worth looking at how art is defined in other ways – if this is still necessary. To a certain extent, generative AI may not be a challenge to art, but the historical peak of "creativity dispositif", such as the myth of "the god of agriculture devouring his son" (note: "dispositif" from The philosophical thought of French philosopher Foucault refers to the system, mechanism or knowledge structure that constructs power relations).

Discussing authorship outside the paradigm of legal intellectual property rights is also complex. As mentioned above, the question is whether we should focus on "Who is the author? (ontological level)" or "Who counts as an author? (Legal level)". We often face this dilemma when working across disciplines in academia. The concept of the individual author has long since been deconstructed, with many believing that all creation is dialogical.

Q: What do you think are the limitations of using generative AI to create music? Is there anything missing from the music generated this way?

Answer: To provide a historical framework for my answer, it is enough to look at the artistic changes represented by photography in the mid-to-late 19th century. Photography was initially seen as a mechanism of reproduction, with some commentators stating that it could never capture "the exquisite feeling and emotion that animates the work of genius" or "the necessary qualities of creation and feeling that constitute a work of art... Photography will never be better than Sculpture has a higher status" (These were published in a British art journal in 1855).

So, what are the limitations when AI generates music, and will anything be missing? We can look at the current mainstream commercial AI interfaces. Most of them were created since the introduction of the transformer model (note: a deep learning model) in 2017 and later evolved into the large language model (LLM). They are extremely limited to a descriptive approach, that is, a blinking cursor or chatbot model (such as Suno and Udio), from the user (the idea) to the computer (the executor), from top to bottom, through many A linear dynamic process of the semiotic translation stage. However, we found that in the field of musical instruments and music performance, there are a large number of non-linear, co-creative relationships between humans, materials and sounds. How to integrate them and their interface design are the keys to breakthroughs.

Period28 Jun 2024

Media coverage

1

Media coverage

  • Title學者:AI音樂突破關鍵 在「非線性關係
    Degree of recognitionInternational
    Media name/outletMing Pao Newspaper
    Media typePrint
    Country/TerritoryChina
    Date28/06/24
    Description明報專訊】獨立樂隊南洋派對N.Y.P.D. 在音樂前線,身先士卒利用AI創作音樂,學術界又如何看待?香港浸會大學音樂學院助理教授Roberto Alonso Trillo,研究興趣之一是「音樂與AI」,他接受本報專訪,討論AI跟音樂創作的看法(問答經翻譯及節錄)。
    答:Roberto Alonso Trillo 問:嚴嘉栢
    問:AI按道理應該做一些單調重複的工作,以釋放人類的創造力。然而今天趨勢走向相反,AI反而正在創造新的音樂,你怎麼看這種趨勢?
    答:人類智慧已經是一個已簡化的模型、地圖和反思性的自我敘述,AI只能是模型的模型(model-of-a-model)。然而AI作為記憶機器,它數位化濃縮人類經驗,如果它能夠走向更強大的AI範式,超越單純的統計推斷和沒有智慧的模仿,就有可能偏離我們試圖表達和揭示的人類智慧,成為一種源自我們、但超越我們自身模擬和理解世界能力的異形力量。


    問:有些音樂人使用生成式AI來創作音樂,你認為生成式AI是一種創造力嗎?你認為使用AI創作的音樂仍能稱是他們的作品嗎?
    答:這個術語(創造力)保留許多含義——它是英國認知科學家Margaret A. Boden的評估標準,如新穎、價值、驚喜等;或者是影響資本主義自由經濟的驅動力,這對我們至關重要。如果問題是,在藝術領域內生成內容,它是否符合藝術生產的條件,那麼這個問題就略有不同。當來到一個內容可持續生產、即時可用且完全自動化的時代,可能需要看看在其他方面如何定義藝術——如果這仍然有必要的話。某種程度上,生成式AI可能不是對藝術的挑戰,而是「創造力裝置(creativity dispositif)」的歷史頂峰,如神話中「農神吞噬其子」(按:「裝置(dispositif)」出自法國哲學家傅柯的哲學思想,指建構權力關係的制度、機制或知識結構)。
    在法律知識財產權的範式外討論作者身分也很複雜。如上所述,問題是我們是否應該關注「誰是作者?(本體論層面)」或「誰算一個作者?(法律層面)」。在學術界跨學科研究時,我們經常面臨這種困境。個人作者的概念早被解構,許多人認為所有創作都是對話性(dialogical)的。
    問:你認為使用生成式AI創作音樂的限制是什麼?使用這種方式生成的音樂是否會有所缺失?
    答:若給我的回答提供一個歷史框架,只要看看19世紀中後期,以攝影為代表的藝術變革就夠了。攝影最初被視為一種複製機制,有些評論指出,它永遠無法獲得「天才的作品中充滿活力的精緻感覺和情感」,或者「創造和感覺構成藝術品的必要品質……攝影永遠不會比雕刻的地位更高」(這些內容發表在1855年英國的一份藝術期刊上)。
    那麼,AI生成音樂時有哪些限制,以及是否會有所缺失?我們可以看看現在主流的商業AI界面,大多數是自2017年引入transformer模型(按:一種深度學習模型),後來演變大型語言模型(large language model,LLM)而產生。它們極其限於以描述的方式(transcriptive approach)來達成,即閃爍的游標或聊天機器人模型(例如Suno和Udio),是從使用者(想法)到電腦(執行者),由上至下、歷經多種符號轉譯階段的線性動態過程。然而我們發現在樂器和音樂表演領域,人類、素材和聲音之間存在大量非線性、共同創造的關係,如何能夠整合它們,以及其界面設計,都是突破的關鍵。
    Producer/Author嚴嘉栢
    PersonsRoberto ALONSO TRILLO