_Are instructions the same as knowledge?_ The question of whether artificial intelligence systems could ever achieve human-like consciousness and subjective experience remains open and debated. The philosophical perspective contends that true experiential consciousness requires some metaphysical component beyond physical computations in a machine. However, we currently lack a scientific theory of consciousness to prove or disprove such a claim. From a neuroscience lens, the sheer complexity and interconnectivity of the human brain with its billions of neurons and trillions of synapses suggests consciousness emerges from this intricate “wetware” in ways that simplified software and silicon may struggle to replicate. Nonetheless, we do not fully understand the mechanisms of human consciousness to definitively limit the potential for engineering machine-based consciousness. AI developers note that while today’s AI focuses on narrow task optimization, the long-term goal is to produce artificial general intelligence with the capacity for reasoning, creativity, planning and learning at a human level. Whether this could lead to subjective conscious experience in machines remains speculative. Achieving human-level artificial consciousness may require replicating biological learning processes and architectures we have yet to fully grasp. Examining language comprehension in non-human animals and AI systems reflects differences between behavioral capabilities, information processing, and meaning representation. Parrots can mimic human speech but lack the semantic and syntactic processing to truly comprehend language’s conceptual meanings and structure. AI chatbots can convincingly converse with scripted responses but do not acquire language in the developmental way humans innately do. While parrots and AI can behaviorally exhibit language use in social contexts, they do not possess innate mental grammars nor learn meanings in a causal, conceptual way. Their information processing is limited to statistical associations between words and outcomes. In contrast, human language acquisition involves creatively constructing conceptual representations and causal models of how words symbolically relate to objects, actions and abstract ideas in a generative, unbounded way. Therefore, replicating the functional conversational ability of language use does not equate to acquiring the richer conceptual knowledge that constitutes human linguistic comprehension. Human language mastery reflects cognitive capabilities for abstraction, reasoning and understanding not yet present in animals or machines. On instructions vs knowledge: The distinction between following instructions for task performance and developing knowledge reveals differences in cognitive requirements and mental representations. Instructions provide procedural rules that can generate behavioral outcomes without deeper understanding of the activity. But conceptual knowledge requires building explanatory causal models to make sense of experiences within a wider theoretical framework.s accumulate knowledge by meaningfully organizing memories, insights and beliefs into relational mental structures. This supports creative problem solving, deduction and inference generation beyond solely following prescribed instructions. AI systems trained on data at best capture statistical correlations between inputs and outputs, relying on programmed instruction sets rather than evolved cognitive architectures for contextual knowledge application. Therefore, while instructions can enable procedural capabilities, true knowledge requires holistic conceptual representation, semantic understanding and reasoning that evolves from experiences. This highlights a current limitation of AI systems, which lack the general learning mechanisms and knowledge frameworks characteristic of human cognition. Advancing artificial knowledge acquisition remains an open challenge. Analyzing these questions around consciousness, language, and knowledge highlights gaps between human and artificial systems while also mapping possible paths to resolving or narrowing these gaps in the future as AI capabilities progress. It poses important challenges for developing broader theories of mind, cognition and intelligence, both biologically and computationally. Bridging this knowledge gap to achieve human-level conceptual understanding in AI represents a significant challenge but also an important opportunity. Several approaches may help narrow or mitigate this gap: * Developing more advanced natural language processing and computer vision to allow AI systems to absorb information from text, speech, and visual data in a more human-like way. This could help expand the knowledge available to the AI. * Architecting knowledge representations and memory frameworks that mirror human conceptual models and allow learning abstract concepts in relation to one another. This goes beyond flat, statistical relationships in the data. * Adopting a developmental approach to knowledge acquisition where AI agents learn interactively in simulated or real worlds, allowing causal models to be inferred through exploration and social learning. * Integrating top-down cognitive architectures that leverage probabilistic reasoning, mental simulation and hierarchical planning to achieve understanding beyond bottom-up statistical learning. * Providing a pre-existing knowledge base of common sense facts and ontological concepts priming the AI models for general reasoning and problem solving. * Having greater transparency and explainability around the AI system’s confidence, knowledge gaps and causal inferences to identify areas needing further training. * Evaluating knowledge using standardized tests focused on reasoning skills, causal understanding and applying concepts in novel situations. While approaching human-level artificial knowledge remains extremely difficult, focusing on qualified knowledge acquisition rather than pure task performance may offer paths to narrow this gap. Combined progress in learning approaches, knowledge representation, reasoning architectures and evaluation methodology can incrementally impart AI systems with greater understanding of the world and language.