無意識物體辨識： 無意識與無覺察情境中無現象辨識下的神經辨識

Abstract

清晰化或是一次性學習在演化上扮演一個重要的角色，幫助人類可以透過少量過去經驗快速學習辨識模糊潛在的危險，提高存活的機率。在本研究希望探討過去經驗或先備知識 （prior）對於人類辨識模糊視覺圖像的影響。過去研究利用慕尼影像研究派典 （Mooney Images Paradigm）在有意識觀看模糊刺激時發現清晰化效果的行為和神經上的證據支持，包含描述的行為清晰化效果(behavioral disambiguation effect)以及神經清晰化效果 （neural disambiguation effect）。其中，行為清晰化效果為描述圖片經清晰化後，受試者主觀辨識率和辨認正確率上顯著提升。而神經清晰化效果為描述經清晰化後第二次觀看慕尼影像時的神經反應相較於一開始清晰化前觀看模糊刺激時的神經反應，變得和觀看灰階圖的神經反應更為相似。過去研究發現有不論是有意識或是無意識觀看的灰階圖，做為先備經驗 （prior） 確實對後續在有意識觀看模糊刺激時產生了行為和神經表徵改變。建立於此基礎上，本研究希望進一步了解此做為先備經驗的灰階圖，在受試者無法意識到刺激存在的無意識觀看的情境下，先備經驗是否可以被自動地應用到模糊影像辨識中。 為回答此研究問題，本研究設計透過慕尼影像研究派典控制各組圖片的清晰化前、清晰化和清晰化後觀看階段；並透過不連續閃光抑制技術（discontinuous flash suppression, dCFS）以遮蔽圖片進入到意識中，控制有意識和無意識的刺激呈現情境。在實驗過程中，透過功能性磁振造影（fMRI）測量受試者的腦部神經表徵計算是否發現神經清晰化效果 （neural disambiguation effect），回答在無意識觀看模糊影像的情境中先前在灰階圖中所學習到的圖片經驗，是否可以被自動的應用在無意識觀看模糊影像的情境中？實驗結果顯示在無意識觀看慕尼影像時，在現象/行為上，受試者應無法辨認慕尼影像，而實際上受試者受試者的神經反應在所觀測腦區中皆出現神經清晰化效果，也就是說出現了現象辨識（phenomenal disambiguation）和神經辨識 （neural disambiguation）不一致的結果。支持大腦在無意識觀看模糊刺激時，會自動將經驗應用於辨識模糊刺激；就算受試者並沒有感覺到，也無法回報看到刺激，但是神經反應上確實出現了自動化應用先備經驗的神經辨識支持。 而此神經辨識和現象辨識不一致的現象也在刺激以有意識呈現但受試者現象上回報不能辨識的無覺察辨識情境中被發現。且在有意識觀看、無覺察觀看和無意識觀看的情境下，在 V1、V2、FG、IT和 MT 等觀測腦區中都可以觀察到神經清晰化的效果或是明顯的趨勢，說明了在有意識、無覺察和無意識觀看中，先前學習的經驗都可能對模糊刺激產生神經表徵上的影響。本研究的結果提供了有關大腦如何利用過去經驗來處理模糊視覺刺激的新見解，除了發現了在無意識觀看模糊影像時先前在灰階圖中所學習到的圖片經驗可以被自動應用在無意識觀看模糊影像的情境中；還發現了在無覺察和無意識辨識中，神經辨識 （neural disambiguation）和現象辨識 （phenomenal disambiguation） 皆存在不一致的現象。希望此研究發現與討論，可以做為未來進一步討論意識與無意識資訊處理所涉及的機制異同提供證據與基礎。

Disambiguation, or one-time learning, enabling humans to rapidly recognize potential threats based on limited past experiences, thus enhancing survival chances. Our study aims to delves into the impact of prior experience on human recognition of ambiguous visual images. Previous research has unveiled behavioral and neural evidence supporting the disambiguation effect during conscious viewing of ambiguous stimuli. The behavioral disambiguation effect manifests as a significant increase in subjective recognition rates and accuracy of verbal identification of post-disambiguated images, while the neural disambiguation effect describes changes in post-disambiguation neural representations, aligning more with grayscale images compare to pre-disambiguation. While previous studies demonstrated that consciously or unconsciously viewing grayscale/prior images consistently leads to behavioral and neural changes when consciously perceiving ambiguous stimuli; Our study aims to answer whether prior experience with grayscale images can be automatically applied to post-disambiguated images during unconscious viewing. To address this, we designed an experiment using the Mooney Images Paradigm, controlling recognition stages (pre-disambiguation, disambiguation, and post-disambiguation) for each image. The discontinuous flash suppression (dCFS) technique masked images, creating conscious and unconscious stimulus presentation conditions. Functional magnetic resonance imaging (fMRI) measured neural representations to determine if a neural disambiguation effect could be observed in regions of interest (ROIs), V1, V2, FG, IT, and MT. Results indicated that during unconscious viewing of Mooney images, participants subjectively reported an inability to consciously recognize the images. However, neural responses across ROIs exhibited a disambiguation effect, highlighting an inconsistency between phenomenal/behavioral and neural recognition effects. This supports the notion that the brain automatically applies prior experience to recognize ambiguous stimuli, even during unconscious viewing, challenging the idea that such automatic application is exclusive to conscious perception. Furthermore, the inconsistency of neural and phenomenal recognition was observed in scenarios where stimuli were unrecognized in conscious presentation, known as unaware viewing. Concluding across conscious, unaware, and unconscious viewing, neural disambiguation effects or similar trends were observed in brain regions like V1, V2, FG, IT, and MT, suggesting that prior learning influences neural representations of ambiguous stimuli in various viewing contexts. In summary, our study provides fresh insights into how past experiences influence the neural representation of ambiguous visual stimuli. It supports the evidence that prior learning from grayscale images can be automatically applied during unconscious viewing of ambiguous images and reveals inconsistencies between neural and phenomenal recognition in unaware and unconscious recognition scenarios. These findings contribute evidence and a foundation for future investigations into the mechanisms involved in conscious and unconscious information processing.