以色列韦茨曼科学研究所的Michal Irani团队开发了一种人工智能工具，可以根据脑扫描数据准确重建人眼所见的图像，同时也能反向预测基于所观看内容产生的脑部活动1。该工具采用双分支脑解码器结合扩散模型的技术方案，利用高分辨率功能性磁共振成像（fMRI）扫描数据进行训练，每个体素覆盖约1立方毫米的神经元活动1。训练数据来自8个人各观看约9000张图像时采集的fMRI扫描记录，其中约70%的训练数据来自未与fMRI扫描直接配对的图像1。
相比以往需要40小时fMRI数据才能对新受试者进行预测，这一新工具仅需1小时的扫描数据就能实现类似功能，大幅提升了效率1。该研究在认知计算神经科学会议上展示，该会议在纽约举行1。fMRI扫描的成本约为每小时600至1000美元1。
这项技术突破在伦理层面引发了关注1。研究人员警告，此类技术可能被用于在未经同意的情况下提取个人思想1。研究团队成员Tommy Sprague表示："如果有办法秘密提取你正在思考的信息，150年的科幻小说可能随时成真"1。研究团队也指出该技术的潜在医疗应用，包括帮助完全瘫痪的患者进行沟通、解读创伤后应激障碍患者的闪回记忆等1。未来研究可能扩展到脑电图（EEG）等更便利的技术，以实现非侵入式的脑活动解码1。
Researchers at Israel's Weizmann Institute of Science have developed an artificial intelligence system capable of reconstructing images viewed by a person based on brain scan data 1. The tool operates bidirectionally, not only translating neural activity into visual reconstructions but also predicting brain responses based on visual input 1.
The system employs a dual-branch brain decoder combined with diffusion models and relies on high-resolution fMRI scanning data, where each voxel represents approximately one cubic millimeter of neural tissue 1. Training involved fMRI scans from eight individuals who each viewed roughly 9,000 images, with approximately 70 percent of training data derived from images not paired with fMRI scans 1. A significant efficiency improvement has been achieved: the new tool requires only one hour of fMRI data for predictions on new subjects, compared to the 40 hours required by previous approaches 1. Given that fMRI scanning costs between 600 and 1,000 dollars per hour, this represents a substantial practical advancement 1.
The research was presented at a cognitive computational neuroscience conference held in New York 1. While the technology offers potential therapeutic applications—such as enabling communication for completely paralyzed patients and interpreting post-traumatic stress flashbacks—it has raised significant ethical concerns 1. Neuroscientist Tommy Sprague warned that "if there's a way to secretly extract information about what you're thinking, 150 years of science fiction could become reality at any moment" 1. Future development may extend this capability to non-invasive EEG technology for decoding brain activity 1.
评论
还没有评论，欢迎留下第一条。