Hacker News 中文摘要

RSS订阅

AI生成视频以最大程度驱动目标脑区 -- AI-generated videos to maximally drive a target brain region

文章摘要

NEvo是一种神经引导的进化视频合成方法,通过训练大脑的数字孪生模型来预测视觉区域对视频的反应,并自动进化出能最大程度激活特定脑区的AI生成视频,用于研究视觉选择性的动态变化。

文章总结

标题:NEvo:神经引导的进化式视频合成

NEvo是一种通过人工智能生成视频来激活特定脑区的方法。它首先为大脑建立“数字孪生”模型,预测每个视觉区域对不同视频的反应,然后利用进化算法优化视频内容:将视频描述为“基因”(如主题、光照、运动、情绪),通过生成、评分、交叉和变异等步骤,逐步提升对目标脑区的激活效果。整个过程分为两个阶段:先寻找最强的静态图像,再添加运动生成2秒视频片段。

NEvo生成的视频能针对不同脑区定制,例如FFA对应人脸、PPA对应场景、EBA对应身体、MT对应运动、V1/V3A对应图案、pSTS/aSTS对应社交场景。这些视频的激活效果优于手工制作或自然视频,且运动元素对每个脑区都至关重要。沿外侧视觉通路从V1到aSTS,合成内容从简单图案和运动逐渐转向人脸和社交互动,揭示了视觉选择性从动态到社交的梯度变化。即使从抽象图形(如堆叠圆盘)开始,优化pSTS也会产生类似人脸和互动的角色,而优化MT则保持纯运动,清晰分离了各脑区的偏好特征。

评论总结

根据评论内容,主要观点和论据如下:

1. 技术潜在危害与伦理担忧(多数观点) - 评论认为该技术可能被用于恶意目的,如生成超常刺激(superstimuli)或定制化广告/宣传,导致成瘾或思想操控。 - 关键引用: - "automated search for visual superstimuli likely leads to bad outcomes" (ben_w) - "Customized propaganda made for the individual is the only thing I'm afraid of" (nathias) - "Fuck this and whoever is driving this type of research... We’re increasingly living in a world designed by sociopaths" (sph)

2. 与现有技术类比(部分观点) - 评论将之与社交媒体算法、科幻作品(如《雪崩》《银翼杀手》)或已有研究(如BLIT短篇)联系,认为这是现有趋势的延伸。 - 关键引用: - "This is already happening at scale by the social media feed algorithms" (nullbio) - "Inventing Snow Crash. Neat." (pockybum522) - "Approaching the capabilities of a 'mood organ' from blade runner" (Gecko4072)

3. 研究目的与科学价值(少数观点) - 部分评论指出该研究旨在理解大脑功能,而非直接用于操控,强调其作为工具的价值。 - 关键引用: - "This is a tool to help researchers in figuring out what different parts of the brain are actually for" (Unearned5161) - "The purpose is a generalized understanding of brain function" (Unearned5161)

4. 个人体验与质疑(少数观点) - 一些评论者对视频效果无感,或质疑技术实现方式(如是否基于fMRI),或认为演示不够震撼。 - 关键引用: - "I watched few of them and it does nothing for me" (aacid) - "these videos were disappointing and underwhelming" (stainablesteel) - "Wait, this is with a digital twin only? Not fMRI or webcam based?" (numpad0)

5. 未来展望与警告(部分观点) - 评论担忧技术若不受控,将导致不可逆后果,呼吁加强监管或伦理审查。 - 关键引用: - "We are really getting to the point where the tech industry must be stopped if humanity is to continue" (voidmain) - "Let's hope our regulations hit these 'social networks' hard enough so that never dare deploy this kind of technology" (aubanel) - "If future generations of researchers will wonder why IRB reviews became mandatory for computer science, studies like this will be the answer" (uniqueuid)