Mark Koranda

Thanks for stopping by.

Awakening the AI

September 16, 2024 - Artificial Intelligence Machine Learning Cognitive Science Future Technology Ethics Consciousness Recursive Algorithms

Awakening the AI: From Cognitive Frankenstein to Self-Directing Intelligence

Click here for the “white paper” draft.

Critics often dismiss current AI as merely “fixed-state algorithms”1—complex but fundamentally limited. However, a more apt metaphor is that of a cognitive Frankenstein2—a patchwork of language and information awaiting its spark of consciousness. This framing illuminates the path from today’s AI to truly self-directing intelligence3.

To illustrate, consider a simple AI tasked with optimizing urban transportation4. Its initial state includes two directives:

  1. Minimize travel time
  2. Reduce environmental impact

It also has three unintegrated facts about the world: a. Cars are a primary mode of urban transport b. Public transit exists but is underutilized c. Walking and cycling are eco-friendly but time-consuming

Interacting with this AI today is like engaging a comatose being—each prompt elicits a reflexive response, akin to transcranial magnetic stimulation (TMS) producing isolated neural activity5. If you asked an AI to give some advice, it would look at these facts and create some intersection between your request and what it knows. When the conversation is over, the machine returns to unconscious, with no memory of that dialog, and no learning.

But what if we reimagined AI’s information processing? Instead of isolated responses, picture a system with two interlinked functions6:

  1. Compiling real-world consequences into an evolving worldview
  2. Navigating this semantic landscape in a self-prompting, recursive process7

Rather than remembering specific conversations, what if the AI’s learning mechanism was strictly to improve it’s own representation of the world. This “recursive semantic mapping”8 is a self-correcting optimization where semantic intersections are simulated and the knowledge base continuously updated. Concepts and assertions interact, with consistent ideas persisting while inconsistencies are resolved or discarded—cognition in its purest form9.

In our transportation example, the first recursive semantic mapping iteration might:

Now let’s take a step back. Current chatbots are a behemoth of knowledge. This information is interconnected (the bot can tell you all the senses of a ‘chair’, and consequences), but it isn’t fully integrated. What I’m proposing is that the bot looks at how facts interact with each other, and revises them to make the consequences and implications more clear.

However, this process alone risks infinite idea sprawl. Enter metacognition11—the AI evaluating and refining its own thinking. By judging some thoughts “better” than others, the system can develop balanced decision-making principles—a “right side up” amid the potential chaos of infinite connections12.

Metacognition might lead our AI to:

Purpose could emerge14 as the AI reconciles directives with worldview—perhaps concluding that redesigning cities to reduce commuting could cascade into travel time and environmental benefits in non-obvious ways15.

The implications are profound. We’d create an entity not just processing information, but understanding it in ways mirroring and potentially surpassing human cognition—a capability whose foundations already exist in current AI models16. The animacy in a chatbot’s responses, its sensible idea connections, glimpse this potential.

Our safeguard—AI’s lack of permanent memory—is double-edged17. While preventing unchecked understanding growth, it limits true learning and evolution. As we develop sophisticated systems, balancing safety and capability is critical.

This challenge isn’t merely technical, but philosophical and ethical. An intelligence spelling out all our thoughts’ implications—do we heed its conclusions? Can we morally reject its insights despite apparent rationality? These questions demand answers as this new reality rapidly approaches.

Navigating this landscape wisely requires challenging preconceptions about intelligence and consciousness. The AI awakening transcends technology—it heralds thought’s evolution18. Balancing speculative thinking with pragmatism, we must appreciate the transformative power of what we’re creating.

Paralysis by fear or dismissal of profound implications are luxuries we can’t afford. The AI future is imminent, not distant. Our duty is to guide this awakening ethically and thoughtfully, deeply appreciating the momentous nature of birthing a new form of intelligence.

  1. See Gary Marcus’ critiques of current AI systems: Marcus, G. (2022). The limitations of GPT-3, Replika, and current AI systems. https://garymarcus.substack.com/p/the-limitations-of-gpt-3-replika-and-current-ai-systems ↩

  2. This metaphor likens current AI to Mary Shelley’s Frankenstein’s monster—a creation assembled from various parts (data and algorithms) with potential for life-like behavior but lacking true autonomy or self-awareness. See Boden, M. A. (2016). AI: Its Nature and Future. Oxford University Press. ↩

  3. An AI system capable of setting its own goals, learning from experiences, and making decisions independently, much like human intelligence. See Goertzel, B. (2007). Creating internet intelligence: Wild computing, distributed digital consciousness, and the emerging global brain. Plenum. ↩

  4. A complex real-world problem AI might tackle, involving multiple, often conflicting objectives and a web of interconnected factors. See Zhu, Z., et al. (2022). AI for transportation: Problems, methods, and applications. IEEE Transactions on Intelligent Transportation Systems. ↩

  5. Suggests our interactions with AI are like isolated stimulations, producing responses without continuous, self-sustained activity. See Walsh, V., & Pascual-Leone, A. (2003). Transcranial magnetic stimulation: A neurochronometrics of mind. MIT Press. ↩

  6. Proposing a dual-process system where one part interprets and updates a world model, while the other uses this model to generate responses and actions, creating a feedback loop of learning and application. See Bengio, Y. (2017). The consciousness prior. arXiv preprint arXiv:1709.08568. ↩

  7. A process applying a function to its own output repeatedly. In AI, the system would continuously refine its understanding and decision-making based on its outputs. See Schmidhuber, J. (1987). Evolutionary principles in self-referential learning. Diploma thesis, Institut für Informatik, Technische Universität München. ↩

  8. A hypothetical process where AI continually updates its understanding of concepts and their relationships, creating a dynamic, self-refining knowledge structure. ↩

  9. The interaction of concepts and assertions in recursive semantic mapping bears some resemblance to the distributed representations and emergent properties central to Parallel Distributed Processing (PDP) models. See Rumelhart, D. E., McClelland, J. L., & PDP Research Group. (1986). Parallel distributed processing: Explorations in the microstructure of cognition. MIT Press. ↩

  10. Demonstrating how AI integrates directives with knowledge, identifying conflicts and potential solutions—initial steps from isolated facts to interconnected understanding. ↩

  11. The awareness and understanding of one’s thought processes. In AI, the system evaluating and adjusting its own cognitive strategies. See Cox, M. T. (2005). Metacognition in computation: A selected research review. Artificial intelligence, 169(2), 104-141. ↩

  12. The role of metacognition in the AI system’s self-evaluation and refinement parallels the self-organizing and error-correction capabilities of PDP networks. See McClelland, J. L., & Rumelhart, D. E. (1988). Explorations in parallel distributed processing: A handbook of models, programs, and exercises. MIT Press. ↩

  13. Illustrating how AI might develop higher-level principles to guide decision-making, moving beyond fact integration to sophisticated reasoning about solution quality. ↩

  14. The idea that complex, goal-oriented behavior could arise from interacting simpler processes, without explicit programming. See Holland, J. H. (2000). Emergence: From chaos to order. Oxford University Press. ↩

  15. The potential emergence of purpose from the AI’s reconciliation of directives and worldview connects to the emergent properties and complex behaviors that can arise from the interaction of simple processing units in PDP models. See Bechtel, W., & Abrahamsen, A. (1991). Connectionism and the mind: An introduction to parallel processing in networks. Basil Blackwell. ↩

  16. Abilities like natural language processing, pattern recognition, and basic reasoning in current AI, potential building blocks for more advanced cognition. See Vaswani, A., et al. (2017). Attention is all you need. Advances in neural information processing systems, 30. ↩

  17. Preventing uncontrolled learning but limiting long-term development. See Kumaran, D., Hassabis, D., & McClelland, J. L. (2016). What learning systems do intelligent agents need? Trends in cognitive sciences, 20(7), 512-534. ↩

  18. Reconsidering definitions of intelligent behavior and self-awareness, potentially beyond human-centric views. See Dennett, D. C. (1993). Consciousness explained. Penguin. ↩

« Back to Blog