MSGID: 1:317/3 6451e366   
   PID: hpt/lnx 1.9.0-cur 2019-01-08   
   TID: hpt/lnx 1.9.0-cur 2019-01-08   
    University of California, Berkeley, researchers have measured brain waves   
   in participants and artificial intelligence systems -- a comparison they say   
   provides a window into what is considered a black box of AI.    
      
     Date:   
         May 2, 2023   
     Source:   
         University of California - Berkeley   
     Summary:   
         New research shows that artificial intelligence (AI) systems   
         can process signals in a way that is remarkably similar to how   
         the brain interprets speech, a finding scientists say might help   
         explain the black box of how AI systems operate.   
      
      
         Facebook Twitter Pinterest LinkedIN Email   
      
   ==========================================================================   
   FULL STORY   
   ==========================================================================   
   New research from the University of California, Berkeley, shows that   
   artificial intelligence (AI) systems can process signals in a way that   
   is remarkably similar to how the brain interprets speech, a finding   
   scientists say might help explain the black box of how AI systems operate.   
      
   Using a system of electrodes placed on participants' heads, scientists   
   with the Berkeley Speech and Computation Lab measured brain waves as   
   participants listened to a single syllable -- "bah." They then compared   
   that brain activity to the signals produced by an AI system trained to   
   learn English.   
      
   "The shapes are remarkably similar," said Gasper Begus, assistant   
   professor of linguistics at UC Berkeley and lead author on the study   
   published recently in the journal Scientific Reports. "That tells you   
   similar things get encoded, that processing is similar. " A side-by-side   
   comparison graph of the two signals shows that similarity strikingly.   
      
   "There are no tweaks to the data," Begus added. "This is raw."   
   AI systems have recently advanced by leaps and bounds. Since ChatGPT   
   ricocheted around the world last year, these tools have been forecast   
   to upend sectors of society and revolutionize how millions of people   
   work. But despite these impressive advances, scientists have had a   
   limited understanding of how exactly the tools they created operate   
   between input and output.   
      
   A question and answer in ChatGPT has been the benchmark to measure an AI   
   system's intelligence and biases. But what happens between those steps   
   has been something of a black box. Knowing how and why these systems   
   provide the information they do -- how they learn -- becomes essential   
   as they become ingrained in daily life in fields spanning health care   
   to education.   
      
   Begus and his co-authors, Alan Zhou of Johns Hopkins University and T.   
      
   Christina Zhao of the University of Washington, are among a cadre of   
   scientists working to crack open that box.   
      
   To do so, Begus turned to his training in linguistics.   
      
   When we listen to spoken words, Begus said, the sound enters our ears   
   and is converted into electrical signals. Those signals then travel   
   through the brainstem and to the outer parts of our brain. With the   
   electrode experiment, researchers traced that path in response to 3,000   
   repetitions of a single sound and found that the brain waves for speech   
   closely followed the actual sounds of language.   
      
   The researchers transmitted the same recording of the "bah" sound   
   through an unsupervised neural network -- an AI system -- that could   
   interpret sound.   
      
   Using a technique developed in the Berkeley Speech and Computation Lab,   
   they measured the coinciding waves and documented them as they occurred.   
      
   Previous research required extra steps to compare waves from the brain   
   and machines. Studying the waves in their raw form will help researchers   
   understand and improve how these systems learn and increasingly come to   
   mirror human cognition, Begus said.   
      
   "I'm really interested as a scientist in the interpretability of these   
   models," Begus said. "They are so powerful. Everyone is talking about   
   them. And everyone is using them. But much less is being done to try   
   to understand them."  Begus believes that what happens between input   
   and output doesn't have to remain a black box. Understanding how those   
   signals compare to the brain activity of human beings is an important   
   benchmark in the race to build increasingly powerful systems. So is   
   knowing what's going on under the hood.   
      
   For example, having that understanding could help put guardrails on   
   increasingly powerful AI models. It could also improve our understanding   
   of how errors and bias are baked into the learning processes.   
      
   Begus said he and his colleagues are collaborating with other researchers   
   using brain imaging techniques to measure how these signals might   
   compare. They're also studying how other languages, like Mandarin,   
   are decoded in the brain differently and what that might indicate about   
   knowledge.   
      
   Many models are trained on visual cues, like colors or written   
   text -- both of which have thousands of variations at the granular   
   level. Language, however, opens the door for a more solid understanding,   
   Begus said.   
      
   The English language, for example, has just a few dozen sounds.   
      
   "If you want to understand these models, you have to start with simple   
   things.   
      
   And speech is way easier to understand," Begus said. "I am very hopeful   
   that speech is the thing that will help us understand how these models   
   are learning."  In cognitive science, one of the primary goals is to build   
   mathematical models that resemble humans as closely as possible. The newly   
   documented similarities in brain waves and AI waves are a benchmark on   
   how close researchers are to meeting that goal.   
      
   "I'm not saying that we need to build things like humans," Begus   
   said. "I'm not saying that we don't. But understanding how different   
   architectures are similar or different from humans is important."   
       * RELATED_TOPICS   
             o Matter_&_Energy   
                   # Albert_Einstein # Engineering # Optics #   
                   Materials_Science   
             o Computers_&_Math   
                   # Neural_Interfaces # Communications #   
                   Artificial_Intelligence # Computer_Modeling   
       * RELATED_TERMS   
             o Mathematical_model o Artificial_intelligence o Bioinformatics   
             o Earth_science o Computer_vision o Security_engineering o   
             National_Security_Agency o Computer_simulation   
      
   ==========================================================================   
   Story Source: Materials provided by   
   University_of_California_-_Berkeley. Original written by Jason Pohl. Note:   
   Content may be edited for style and length.   
      
      
   ==========================================================================   
   Journal Reference:   
      1. Gasper Begus, Alan Zhou, T. Christina Zhao. Encoding of speech in   
         convolutional layers and the brain stem based on language   
         experience.   
      
         Scientific Reports, 2023; 13 (1) DOI: 10.1038/s41598-023-33384-9   
   ==========================================================================   
      
   Link to news story:   
   https://www.sciencedaily.com/releases/2023/05/230502201343.htm   
      
   --- up 1 year, 9 weeks, 1 day, 10 hours, 50 minutes   
    * Origin: -=> Castle Rock BBS <=- Now Husky HPT Powered! (1:317/3)   
   SEEN-BY: 15/0 106/201 114/705 123/120 153/7715 218/700 226/30 227/114   
   SEEN-BY: 229/110 112 113 307 317 400 426 428 470 664 700 292/854 298/25   
   SEEN-BY: 305/3 317/3 320/219 396/45   
   PATH: 317/3 229/426