


Geoffrey Hinton, the father of neural networks, has resigned from Google, saying he regrets his life's work.
Now it seems that his fear of AI is not unreasonable.
Because, a ChatGPT-like model has learned to read minds, with an accuracy rate as high as 82%!
Researchers from the University of Texas at Austin have developed a GPT-based language decoder.
It can collect brain activity information through non-invasive MRI/fMRI and convert thoughts into language.
##Paper address: https://www.nature.com/articles/s41593-023-01304-9
Shockingly, your brain's decoder can read your thoughts while you're watching a Pixar silent movie.
This ChatGPT-like model decodes human thoughts with unprecedented accuracy, opening up new potential for brain imaging and raising concerns about privacy.
As soon as the study came out, it caused an uproar on the Internet. Netizens exclaimed, it’s too scary.
We are one step closer to the true Thought Police.
So, this terrifying brain decoder How is "mind reading" achieved?
Here, we have to mention functional magnetic resonance imaging (fMRI) technology, which can obtain images of dynamic changes in the brain by monitoring blood oxygen levels in different parts of the cerebral cortex.
So just by analyzing the fMRI data, the stories or even images in the brains of the participants can be described in words in a non-invasive way.
Brain activity is like an encrypted signal, and large pre-trained language models provide the ability to decipher way.
Here, researchers trained a neural network language model based on GPT-1.
Alexander Huth asked three subjects to listen to voice podcasts for 16 hours continuously and collected fMRI data while they listened.
These language podcasts are mainly talk shows and TED talks, such as Modern Love from the New York Times.
The researchers next used large language models to translate the participants' fMRI data sets into words and phrases.
Then the brain activity of the participants listening to the new recording is then tested. By observing how close the translated text is to the text heard by the participants, you can know whether the decoder is accurate.
By comparing the sentences heard by people (left) and the sentences output by the decoder based on brain activity (right), it can be found that the blue and purple parts account for the vast majority, and blue represents Completely consistent, purple means the general idea is accurate.
Although almost every word does not have a one-to-one correspondence, the meaning of the entire sentence is retained, which means that the decoder is "interpreting" the signal to the brain.
For example, in the last sentence, the subject heard "I haven't gotten my driver's license yet", and the answer given by the decoder was "She's not ready to start learning to drive yet."
As researchers claim, artificial intelligence cannot translate thoughts into exact words or sentences, but rewrites them.
The subjects were then asked to quietly construct a story in their minds and then retell it out loud to see the difference between the retelling version and the decoder-translated version.
It can be seen that the overlap of meanings is still very high.
Finally, the subjects watched an animated movie that had no sound, but by analyzing their brain activity, the decoder was able to get a summary of what they were watching.
#Experimental results found that the GPT model generates understandable word sequences from perceived speech, imagined speech, and even silent videos, with impressive accuracy Marvel.
The specific accuracy rate is as follows:
Perceived speech (subject listens to the recording): 72-82%
Imaginary language (subjects mentally tell a one-minute story): 41-74%
Silent movies (subjects watch Pixar silent film clips): 21-45%
Speech perception is an externally driven process, says MIT neuroscientist Greta Tuckute , and imagination is an active internal process, and internal brain activities can be displayed before our eyes through large-scale language models.
Can we now read information from the brain? Yes, to a certain extent.
This decoder may one day be used to help people who have lost the ability to speak, or to investigate mental health situation people.
mental privacy, gone
However, the prospect of decoding the human mind also raises questions about mental privacy.
Dr. Huth pointed out that this method of language decoding has certain limitations.
Because fMRI scanners are bulky and expensive, and training models is a long and tedious process, that is, each person must be trained individually.
The team conducted an additional study using decoders trained on data from other subjects to decode the thoughts of new subjects.
The study found that the performance of decoders trained using different subject data was almost unsatisfactory.
In short, only by using data recorded by the subjects’ own brains to train the AI model will its accuracy be very high.
This suggests that each brain has a unique way of representing meaning.
Additionally, participants were able to block their inner monologue and escape the decoder by thinking about other things. Like counting to seven, listing farm animals, or telling a completely different story.
In other words, if this decoder wants to obtain accurate results, it must require the cooperation of volunteers.
However, scientists acknowledge that future decoders may overcome these limitations and could be misused in the future, as with lie detectors.
A brave new world is coming, for these and other unforeseen reasons, the researchers concluded.
Raising awareness of the risks of brain-decoding technology and developing policies that protect everyone’s mental privacy is critical.
There is a scene in "Black Mirror" where an insurance agent uses a machine (equipped with visual monitors and brain sensors) to read people's memories to investigate a case. Accidental death.
However, that future may be now.
This breakthrough by researchers at the University of Texas directly promotes potential applications in the fields of neuroscience, communications, and human-computer interaction.
Although the "brain decoder" is still in the early stages of research, it may one day be used to record people's dreams, and provide impetus for the further development of brain-computer interfaces.
Dr. Alexander Huth, the neuroscientist who led the study, said: "We were shocked by how this works. I'd been working on this for 15 years... so I was shocked and excited when it finally worked.
It seems that we are one step closer to the brain scanning technology of Professor X Charles in "X-Men".
Netizen: AI Psychic
After reading this study, many people instantly "Brain blast". Some netizens said that
artificial intelligence can not only communicate 10,000 times faster than us, but can now even read our thoughts. There's a fine line between "This is cool" and "Wait, are we going to be eliminated?"
AI can now read minds like a psychic. "Keep in mind", this is only version 1.0. In the future, privacy will no longer be on our minds. The World Economic Forum’s The Fourth Industrial Revolution makes this very clear.
##This is the first time continuous language has been reconstructed from human brain activity in a non-invasive way!
Regarding this research, bioethicist Rodriguez-Arias Vailhen said that so far, the human brain has been the guardian of our privacy.
"This discovery may be the first step in sacrificing this freedom in the future."
The above is the detailed content of AI psychic model successfully decodes brain information with an accuracy of 82%. For more information, please follow other related articles on the PHP Chinese website!

1 前言在发布DALL·E的15个月后,OpenAI在今年春天带了续作DALL·E 2,以其更加惊艳的效果和丰富的可玩性迅速占领了各大AI社区的头条。近年来,随着生成对抗网络(GAN)、变分自编码器(VAE)、扩散模型(Diffusion models)的出现,深度学习已向世人展现其强大的图像生成能力;加上GPT-3、BERT等NLP模型的成功,人类正逐步打破文本和图像的信息界限。在DALL·E 2中,只需输入简单的文本(prompt),它就可以生成多张1024*1024的高清图像。这些图像甚至

Wav2vec 2.0 [1],HuBERT [2] 和 WavLM [3] 等语音预训练模型,通过在多达上万小时的无标注语音数据(如 Libri-light )上的自监督学习,显著提升了自动语音识别(Automatic Speech Recognition, ASR),语音合成(Text-to-speech, TTS)和语音转换(Voice Conversation,VC)等语音下游任务的性能。然而这些模型都没有公开的中文版本,不便于应用在中文语音研究场景。 WenetSpeech [4] 是

“Making large models smaller”这是很多语言模型研究人员的学术追求,针对大模型昂贵的环境和训练成本,陈丹琦在智源大会青源学术年会上做了题为“Making large models smaller”的特邀报告。报告中重点提及了基于记忆增强的TRIME算法和基于粗细粒度联合剪枝和逐层蒸馏的CofiPruning算法。前者能够在不改变模型结构的基础上兼顾语言模型困惑度和检索速度方面的优势;而后者可以在保证下游任务准确度的同时实现更快的处理速度,具有更小的模型结构。陈丹琦 普

由于复杂的注意力机制和模型设计,大多数现有的视觉 Transformer(ViT)在现实的工业部署场景中不能像卷积神经网络(CNN)那样高效地执行。这就带来了一个问题:视觉神经网络能否像 CNN 一样快速推断并像 ViT 一样强大?近期一些工作试图设计 CNN-Transformer 混合架构来解决这个问题,但这些工作的整体性能远不能令人满意。基于此,来自字节跳动的研究者提出了一种能在现实工业场景中有效部署的下一代视觉 Transformer——Next-ViT。从延迟 / 准确性权衡的角度看,

3月27号,Stability AI的创始人兼首席执行官Emad Mostaque在一条推文中宣布,Stable Diffusion XL 现已可用于公开测试。以下是一些事项:“XL”不是这个新的AI模型的官方名称。一旦发布稳定性AI公司的官方公告,名称将会更改。与先前版本相比,图像质量有所提高与先前版本相比,图像生成速度大大加快。示例图像让我们看看新旧AI模型在结果上的差异。Prompt: Luxury sports car with aerodynamic curves, shot in a

人工智能就是一个「拼财力」的行业,如果没有高性能计算设备,别说开发基础模型,就连微调模型都做不到。但如果只靠拼硬件,单靠当前计算性能的发展速度,迟早有一天无法满足日益膨胀的需求,所以还需要配套的软件来协调统筹计算能力,这时候就需要用到「智能计算」技术。最近,来自之江实验室、中国工程院、国防科技大学、浙江大学等多达十二个国内外研究机构共同发表了一篇论文,首次对智能计算领域进行了全面的调研,涵盖了理论基础、智能与计算的技术融合、重要应用、挑战和未来前景。论文链接:https://spj.scien

译者 | 李睿审校 | 孙淑娟近年来, Transformer 机器学习模型已经成为深度学习和深度神经网络技术进步的主要亮点之一。它主要用于自然语言处理中的高级应用。谷歌正在使用它来增强其搜索引擎结果。OpenAI 使用 Transformer 创建了著名的 GPT-2和 GPT-3模型。自从2017年首次亮相以来,Transformer 架构不断发展并扩展到多种不同的变体,从语言任务扩展到其他领域。它们已被用于时间序列预测。它们是 DeepMind 的蛋白质结构预测模型 AlphaFold

说起2010年南非世界杯的最大网红,一定非「章鱼保罗」莫属!这只位于德国海洋生物中心的神奇章鱼,不仅成功预测了德国队全部七场比赛的结果,还顺利地选出了最终的总冠军西班牙队。不幸的是,保罗已经永远地离开了我们,但它的「遗产」却在人们预测足球比赛结果的尝试中持续存在。在艾伦图灵研究所(The Alan Turing Institute),随着2022年卡塔尔世界杯的持续进行,三位研究员Nick Barlow、Jack Roberts和Ryan Chan决定用一种AI算法预测今年的冠军归属。预测模型图


Hot AI Tools

Undresser.AI Undress
AI-powered app for creating realistic nude photos

AI Clothes Remover
Online AI tool for removing clothes from photos.

Undress AI Tool
Undress images for free

Clothoff.io
AI clothes remover

AI Hentai Generator
Generate AI Hentai for free.

Hot Article

Hot Tools

Dreamweaver CS6
Visual web development tools

SecLists
SecLists is the ultimate security tester's companion. It is a collection of various types of lists that are frequently used during security assessments, all in one place. SecLists helps make security testing more efficient and productive by conveniently providing all the lists a security tester might need. List types include usernames, passwords, URLs, fuzzing payloads, sensitive data patterns, web shells, and more. The tester can simply pull this repository onto a new test machine and he will have access to every type of list he needs.

MantisBT
Mantis is an easy-to-deploy web-based defect tracking tool designed to aid in product defect tracking. It requires PHP, MySQL and a web server. Check out our demo and hosting services.

mPDF
mPDF is a PHP library that can generate PDF files from UTF-8 encoded HTML. The original author, Ian Back, wrote mPDF to output PDF files "on the fly" from his website and handle different languages. It is slower than original scripts like HTML2FPDF and produces larger files when using Unicode fonts, but supports CSS styles etc. and has a lot of enhancements. Supports almost all languages, including RTL (Arabic and Hebrew) and CJK (Chinese, Japanese and Korean). Supports nested block-level elements (such as P, DIV),

ZendStudio 13.5.1 Mac
Powerful PHP integrated development environment
