Facebook’s parent company Meta has announced that they are launching a long-term research project to build the next generation of artificial intelligence that can learn and process speech and text like the human brain. Meta describes efforts to create human-level AI.
Meta is working with NeuroSpin, a neuroimaging company that images the human brain, and Inria, a software company, to study how the human brain processes speech and text, and then compare that to how AI language models do.
NeuroSpin is a research center dedicated to brain imaging. The researchers consist of physicists, mathematicians, neuroscientists and doctors working together to build tools to understand the human brain differently.
NeuroSpin explains what it does:
“Focusing on neuroimaging, the research conducted ranges from technology and method development (data acquisition and processing) to preclinical and clinical neuroscience, including cognitive neuroscience.”
Meta release:
“Today, we announced a long-term artificial intelligence research initiative to better understand how the human brain processes speech and text. In collaboration with the Center for Neuroimaging Neurospin (CEA) and Inria, we are comparing how AI language models and the brain Respond to the same spoken or written sentence.
We will use the insights from this work to guide the development of artificial intelligence that can process speech and text as efficiently as humans. “
The problem with AI language models is that they require a large number of examples to learn. The human brain only needs a few examples to learn.
Current research on brain-like AI language models has found that:
“Language models that most closely resemble brain activity are those that best predict the next word from context (e.g., before…time).
While the brain predicts words and ideas long in advance, most language models are trained to only predict the next word. Unlocking this long-term predictive power could help improve modern AI language models. “
The announcement cites current research into AI for modeling human brain activity, which uses MRI and other imaging tools to look at human brain activity as they perform a variety of language-related tasks.
The cited research paper is from 2021 and is titled, Language Processing in the Brain and Deep Neural Networks: Computational Convergence and Its Limitations (PDF).
The opening paragraphs of the research paper discuss a summary of the findings:
“It turns out that (1) the position of the layers in the network and (2) the ability of the network to accurately predict words from context are the main factors that lead to the emergence of brain-like representations in artificial neural networks.
Taken together, these results show how perceptual, lexical, and combinatorial representations unfold precisely within each cortical region and help shed light on the governing principles of language processing in the brain and algorithms. “
The importance of the above studies is to show that studying how the brain processes data can yield insights that create similar processes in algorithms.
The Meta research team is using thousands of scans of human brain activity to see which areas of the brain are activated during tasks.
The research is said to demonstrate the “computing organization of the human brain” and it yields insights useful to Meta’s goal of developing “human-level artificial intelligence.”
The benefits are not limited to producing human-level AI, the research also helps neuroscientists better understand the human brain.
Citation
Read the official meta announcement
Building artificial intelligence that processes language like a human
Read a more in-depth description of Meta’s human AI research
Studying the brain to build artificial intelligence that processes language like humans
