-23
Posit: In the future, generative A.I. will be thought of as the unconscious part of a general A.I.'s mind.
(lemmy.world)
| Attribute | Unconscious Mind | Generative AI |
|---|---|---|
| Information Processing | Processes vast amounts of information rapidly and automatically, often without conscious awareness (From the first studies of the unconscious mind to consumer neuroscience: A systematic literature review, 2023) | Processes large datasets quickly, extracting patterns and generating outputs without explicit programming for each task (Deep Learning, 2015) |
| Pattern Recognition | Recognizes complex patterns in sensory input and past experiences, influencing behavior and decision-making (Analysis of Sources about the Unconscious Hypothesis of Freud, 2017) | Excels at identifying patterns in training data, forming the basis for generating new content or making predictions (A Survey on Deep Learning in Medical Image Analysis, 2017) |
| Creativity | Contributes to creative insights and problem-solving through unconscious incubation and associative processes (The Study of Cognitive Psychology in Conjunction with Artificial Intelligence, 2023) | Generates novel combinations and ideas by recombining elements from training data in unexpected ways (e.g., GANs in art generation) (Generative Adversarial Networks, 2014) |
| Emotional Processing | Processes emotional information rapidly, influencing mood and behavior before conscious awareness (Unconscious Branding: How Neuroscience Can Empower (and Inspire) Marketing, 2012) | Can generate text or images with emotional content based on patterns in training data, but lacks genuine emotions (Language Models are Few-Shot Learners, 2020) |
| Memory Consolidation | Plays a crucial role in memory consolidation during sleep, strengthening neural connections (The Role of Sleep in Memory Consolidation, 2001) | Analogous processes in some AI systems involve memory consolidation and performance improvement (In search of dispersed memories: Generative diffusion models are associative memory networks, 2024) |
| Implicit Learning | Acquires complex information without conscious awareness, as in procedural learning (Implicit Learning and Tacit Knowledge, 1994) | Learns complex patterns and rules from data without explicit programming, similar to implicit learning in humans (Deep Learning for Natural Language Processing, 2018) |
| Bias and Heuristics | Employs cognitive shortcuts and biases that can lead to systematic errors in judgment (Thinking, Fast and Slow, 2011) | Can amplify biases present in training data, leading to skewed outputs or decision-making (Mind vs. Mouth: On Measuring Re-judge Inconsistency of Social Bias in Large Language Models, 2023) |
| Associative Networks | Forms complex networks of associations between concepts, influencing thought and behavior (The associative basis of the creative process, 2010) | Creates dense networks of associations between elements in training data, enabling complex pattern completion and generation tasks (Attention Is All You Need, 2017) |
| Parallel Processing | Processes multiple streams of information simultaneously (Parallel Distributed Processing: Explorations in the Microstructure of Cognition, 1986)) | Utilizes parallel processing architecture (e.g., neural networks) to handle multiple inputs and generate outputs (Next Generation of Neural Networks, 2021) |
| Intuition | Generates rapid, automatic judgments based on unconscious processing of past experiences (Blink: The Power of Thinking Without Thinking, 2005) | Produces quick outputs based on learned patterns, which can appear intuitive but lack genuine understanding (BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding, 2019) |
| Priming Effects | Unconscious exposure to stimuli influences subsequent behavior and cognition (Attention and Implicit Memory: Priming-Induced Benefits and Costs, 2016) | Training on specific datasets can "prime" generative AI to produce biased or contextually influenced outputs (AI Fairness 360: An Extensible Toolkit for Detecting, Understanding, and Mitigating Unwanted Algorithmic Bias, 2018) |
| Symbol Grounding | Grounds abstract symbols in sensorimotor experiences and emotions (The Symbol Grounding Problem, 1990) | Struggles with true symbol grounding, relying instead on statistical correlations in text or other data (Symbol Grounding Through Cumulative Learning, 2006) |
| Metaphorical Thinking | Uses embodied metaphors to understand and reason about abstract concepts (Metaphors We Live By, 1980) | Can generate and use metaphors based on learned patterns but lacks deep understanding of their embodied nature (Deep Learning-Based Knowledge Injection for Metaphor Detection, 2023) |
| Dream Generation | Produces vivid, often bizarre narratives and imagery during REM sleep (The Interpretation of Dreams, 1900) | Some generative models can produce dream-like, surreal content (Video generation models as world simulators, 2024) |
| Cognitive Dissonance | Automatically attempts to reduce inconsistencies between beliefs and behaviors (A Theory of Cognitive Dissonance, 1957) | MoE architectures can handle a wider range of inputs without ballooning model size, suggesting potential for resolving conflicts between different AI components by synthesizing expert opinions into a coherent whole (Optimizing Generative AI Networking, 2024). |
The appearance of similarities between Generative AI and the unconscious mind do not mean there is any actual equivilance to be had. We gotta stop using the same terminology to describe generative AI as we do humans because they are not the same thing in the slightest. This only leads to further unintentional bias, and an increased likelihood of seeing connections with the unconscious mind that don't actually exist.
Typically speaking, the appearance of similarities should be explored for functional similarities. Even in the world of prompting, a user gets demonstrably better results (in some contexts) by acting as-if they're talking to a person. We can explore the similarities between interpersonal interaction and AI-interaction for overlap in effective modalities without claiming they're equivalent.
Who is "we"? The reality is that these grifters are constantly claiming equivalence. The term "AI" is itself a misleading equivalence. I would much rather not be part of that "we".
"We" being anyone interested in scientific exploration, hence the papers linked in the post and comments.