2026年，人工智能与现实世界的关系，通过一系列具体事件变得可见：劳动者佩戴摄像机记录工作过程，身体动作成为机器人训练的数据；美军也公开表示，在战争中使用AI处理大量信息，辅助军事判断与决策。这件作品是艺术家对这些事件的影像回应。通过劳动者的第一人称记录、强化学习仿真与民用无人机影像的交织，作品关注人的经验如何被转化为机器的能力，以及这些能力如何重新介入现实。

理查德·布劳提根1967年的诗《All Watched Over by Machines of Loving Grace》中，人、计算机与自然共同构成一种“控制论生态”，人类免于劳作，由机器温柔照料。这份想象也与当时旧金山反文化对免费分享、共同生活的探索相连，寄托着一种关于如何生活的愿望。

半个多世纪后，这份愿望在2026年与劳动者教机器工作、AI参与现实判断的情境相遇。同一个动作既完成眼前的工作，也成为训练未来机器的样本。然而，已有报道指出，部分参与录制的劳动者并未因此获得额外报酬。当人的劳动持续转化为机器的能力，免于劳动的未来将由谁享有，仍然悬而未决。

作品以“现实↔数据集”的结构呈现这一循环：人类的日常行为被记录、解码与剥离，转化为训练材料；模型又通过行动与判断作用于现实，产生新的经验与数据。随着动作反复、网格扩张和视角转换，现实既是机器学习的来源，也逐渐成为其能力被部署和检验的场所。在这一过程中，谁提供经验、谁获得收益、谁承担判断的后果，构成了技术能力之外同样重要的问题。

片尾，机器的视角进入现实世界，呈现出摇摇欲坠的状态。这并非艺术家对机器人训练必然失败的预言，而是对“慈爱机器”得以形成的条件与正当性的担忧：当下的数据获取方式、劳动关系与AI应用，能否支撑机器将温柔照料人类的许诺？作品借诗中反复出现的“I like to think”——“我愿这样想”，让期待与疑问同时存在。人们依然可以向往那个免于劳作的世界，却也需要追问：我们正在通过怎样的方式走向它？

In 2026, a series of concrete events has brought the relationship between artificial intelligence and the physical world into view. Workers wear cameras to record their working processes, turning bodily movements into data for training robots. The US military has also publicly stated that it uses AI in warfare to process large volumes of information and support military judgments and decisions. This work responds to these developments through moving images. Interweaving workers’ first-person recordings, reinforcement learning simulations, and civilian drone footage, it examines how human experience becomes machine capability, and how those capabilities return to shape reality.

In Richard Brautigan’s 1967 poem All Watched Over by Machines of Loving Grace, humans, computers, and nature form a “cybernetic ecology”: people are freed from labor and gently cared for by machines. This vision resonates with the experiments in free sharing and communal life associated with San Francisco’s counterculture at the time. It carries a wish for another way of living.

More than half a century later, this wish encounters the realities of 2026: workers teaching machines how to work, and AI participating in judgments that affect the world. The same movement completes an immediate task while also becoming a sample for training future machines. Yet reports indicate that some workers receive no additional compensation for these recordings. As human labor continues to generate machine capability, the question of who will enjoy a future free from labor remains unresolved.

The work follows this process through a “reality ↔ dataset” loop. Everyday human behavior is recorded, decoded, and extracted as training data. Models, in turn, act upon reality through their actions and judgments, producing further experiences and data. Through repeated movements, expanding grids, and shifting perspectives, reality becomes both a source from which machines learn and a site where their capabilities are deployed and tested. Who supplies the experience, who receives the benefits, and who bears the consequences of these judgments become questions as significant as technical capability itself.

In the final sequence, the machine’s viewpoint enters the physical world, becoming unstable and precarious. This is not the artist’s prediction that robot training will inevitably fail. It questions whether the ways we acquire data, organize labor, and deploy AI can justify the promise of “machines of loving grace.” Can these conditions sustain the promise that machines will care for humanity? Drawing on the poem’s recurring phrase “I like to think,” the work holds hope and doubt together. We may still wish for a world free from labor, while asking how we are attempting to reach it.

片尾无人机画面（04:07.700–04:51.900）：节选自 TSF-Leo 于 2020 年 1 月 2 日在 Bilibili 发布的合集《【第一期】穿越机炸鸡集锦 每次5分钟 打杆慢又怂》。

Closing drone footage (04:07.700–04:51.900): excerpts from “【第一期】穿越机炸鸡集锦 每次5分钟 打杆慢又怂,” a compilation published by TSF-Leo on Bilibili, 2 January 2020.

https://www.bilibili.com/video/BV1sJ41187rZ/

在 CIFRA 观看 / View on CIFRA ↗