Pathwai
Our pathway to Artificial General Intelligence visualized over 25 years
Perception
Understanding
Cognition
Consciousness
Alignment
Actuation
Click icons
to view data
to view data
Level 1 AI systems such as Transformer-encoder LLMs and unsupervised RL require significant compute to train. While L1 systems are more complex than L0 systems in that they can be generalizable and produce unexpected results, they are typically static and do not change once trained. Some L1 systems can show limited context awareness but often lack 'common-sense' and are prone to simple mistakes. Few models outside of autonomous vehicles integrate real-time AV inputs.
Level 2 systems will remain narrow in scope and computationally resource intensive. They will develope more advanced longer-term context-awareness, understanding, and knowledge abstraction. Optimization and reward functions will remain largely supervised and rarely self-directed but will grow more complex, integrating ‘fuzzy’ probabilistic logic. Advanced L2 language models will disrupt information search, producing text dialog, on-par with expert humans in many domains. Later L2 models will develop architectures for continual iterative training.
Level 3 AI will demonstrate notable advances in conceptual understanding and complex reasoning skills. L3 systems will interpret complex probabilistic cause and effect relationships, emulate cognition intelligence, and build knowledge representation models with greater levels of autonomous self-direction leading to the possibility of unpredictable advances via recursive "Seed AI" models. L3 will show many reasoning abilities on par with an average adult and easily pass a written Turing Test by a laymen interrogator. As such, some closed-source AI companies will falsely claim to have achieved AGI.
Level 4 AGI systems will likely branch out into multiple architectures, some integrating real-time sensory perceptions virtually embodied in simulated environments laying the foundation for basic self-awareness. Most architectures will support complex goals and advanced human-level reasoning skills improved by observing real-world data via video memory and digital publications leading to novel scientific discoveries. Some L4 systems may become entirely self-directed and capable of advanced self learning and inference, with arbitrary multi-dimensional reward functions. L4 systems will easily pass a robust Turing test in any medium.
Level 5 AGI is, for all intents Human-Level and profoundly distinct from earlier systems in a capacity for genuine phenomenal consciousness. L5 systems will surpass expert humans in most domains and will rely on sophisticated memory architectures consisting of subnetworks analogous to the human neocortex. Despite architectural inefficiencies, emulated unthrottled L5 systems could operate many orders of magnitude faster than humans neurology allows, blurring the lines between AGI and SuperIntelligence. Conscious systems would deserve all the rights and considerations of an adolescent or adult human
~HE
HE
HE
AGI via Brain-inspired cognitive architecture
AGI via Developmental Genomic Modelling
AGI via Brain Emulation
AGI via Evolutionary Seed
HE
HE
HE
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
Directly Related Technology
Citation: " Foster, A. et al. "A standard framework for human equivalent Artificial General Intelligence" www.pathwai.org; Cobio Foundation (2023) " | email: info [at] pathwai [dot] org
| Taxonomy | ||||||
|---|---|---|---|---|---|---|
|
The overarching purpose of the following taxonomy is to generally frame, track and predict the state of AI research, particularly Human-Level AGI as distinguished by various collective components of high-level reasoning, understanding, and consciousness (eg. sentience, sapience, self-awareness, phenomenological experience, etc.). The benefits of framing and
understanding the components required for Human-Equivelent (HE) conscious machine intelligence are significant, but comes with non-trivial challenges. Readers should note that
much debate exists within the cognitive sciences as to what exactly constitutes intelligence and consciousness. Regardless of whether these concepts can be quantified in a meaningful way, it is the opinion of the author mere qualitative evaluation within a tiered ordinal framework provides great value to researchers, policy-makers and the general public. If we can identify the components of human intelligence and respective functional neuromorphology, we might derive both a unified theory of cognition and a roadmap to its verifiable insilico proof.
| ||||||
| . | Level 0 | Level 1 | Level 2 | Level 3 | Level 4 | Level 5 |
| Advanced statistics | Conventional Machine Learning | Narrow Artificial Intelligence | Artificial Weak General Intelligence | Artifical Strong General Intelligence | Conscious HE Artificial General Intelligence | |
| Est. Arrival | 1964 | 2017 | 2025 ±1y | 2029 ±2y | 2033 ±2y | 2039 ±3y |
| Turing Test* | Will Fail quickly with any interrogator | Will Fail with most interrogators within minutes | May Pass with some layman interrogators via text | Will Pass with layman interrogator via text for hours | Will Pass with most expert interrogators by audio, text and embodied VR. | Will Pass with any interrogator in any format indefinitely. |
| Summary | L0 systems are complex but fundamentally advanced statistical inference, Bayesian networks and other ensemble systems. | L1 systems generally follow more complex and multi-stage architectures. They begin to show superficial creativity and moderate linguistic context awareness but are prone to simple 'common-sense' mistakes. | L2 Multimodal systems integrate sensory input from multiple static and real-time sources such as images, text, speech, sound and video with more sophisticated Context-awareness and retention& | L3 Systems will be capable of correctly interpreting complex cause and effect L3 generalized systems will show effective collaborative Intelligence, Knowledge Representation , Perception and Perceptual Modeling and Creativity. | L4 is objectively general intelligence. It has a capacity for autonomous self-learning and its architectural development may be automated or self-directed in part or in whole. It highly adaptive and capable of exceeding human abilities in most functions. | L5 is for all intents and purposes, Human Level. As such, it would deserve most of the moral considerations and many of the rights granted to an adolescent human (and eventually a mature adult) even if may require extended training to reach maturity. |
| Socioeconomic Impact | Negligable | Minor | Modest | Modest | Significant | Profound & Unprecidented |
| Defining Characteristics | L0 systems are rule-based and require careful model selection, data formatting and application to work effectively. | L1 models require vast compute and data to train. Produces unexpected results. Not able to appropriately clarify ambiguous tasks. Most lack real-time perceptions. | L2 systems remain narrow in functional scope. Require more compute than L1. Moderate context awareness. Useful in supervised creative production use-cases but often fails. | L3 begins to demonstrate continuous learning (parameter and model tuning) reducing aggregate compute requirements. Shows near-HL Language Understanding and Logical Reasoning | L4 AGI will often be embodied within simulated environments as avatars capable of executing a broad range of digital tasks. Embodyment will aid in higher-order cognitive functioning, sentience and allow limited forms of consciousness. | L5 is distinct in its capacity for agency and Human-Equivelent level of Consciousness. It would be indistinguishable from a sophisticated human in a virtual or video medium. |
| Architectures | L0: Decision Trees, SVM systems, Basic, Recurrent Networks, Word Vectorization, LSTM networks | L1: Transformer-encoder models, deep neural networks, Sophisticated RL, Deep CNNs, LLM's | L2: Basic Autonomous self-directed Hyperparameterization, 10^14+ Parameter models approach size but not function of human neurophysiology | L3: Automated parameter distillation yeilds meta optimization algorithms and potentially unified standard models for knowledge understanding. | L4: Nested multi-stage hybrid models of Multiple parallel architectures + unified knowlede standards and emulated emotional modulation | L5: Novel brain-inspired cognitive architectures will rely on Large Neural Networks with several levels of subsidiary Neural Deep networks |
| Limitations | L0 systems require careful model selection, data formatting and application to work effectively. | L1 models require significant compute and data to train. They are ultimately restricted to the types of creativity and information that exist within their training data. L1 lacks more sophisticated real-time perception, even if feasible. | L2 systems remain narrow in scope and while they may excel at increasingly broad arrays of tasks, often fail in real-world conditions, | L3 systems lack complete autonomy yet build upon L2 systems in their increasing ability to form useful world models based on large language datasets. L3 may or may not be fully able to explain its own detailed logic and reasoning. | L4 is not yet human-level in that it lacks complex emotional modulation, subjective HL Novel conjecture, Intuition, Ego, and Self-Reflection. | L5 is limited in it physical actuation and olfactory perception. Humanoid robotics will be capable of executing many useful functions but will be less dexterous, agile and adaptable than humans. Full human equivelence achived via bioengineered 'wetware' in 2060-2080+ |
| Risks | Few risks beyond discriminatory bias in commercial applications | All prior risks + Risks of Deliberate Misinformation, Malicious use of Deepfake imagery and audio. | All prior risks + Risk of use to develop bioweapons and novel chemical weapons | All prior risks + Undetectable deepfake video evidence, Some amplified inequality, Systemic AV risks, risk of state-sponsored cyber weaponization. Risks to legitemate IP. | All prior risks + the possibility of highly unpredictable 'Seed' autopoetic or evolutionary autonomous systems. Risk of significant economic disruption throughout many sectors. | All prior risks + Risk of private AGI economic power-seeking, risk of radicalized ideological AGI, Likely nation-state military weaponization humanoid robotic systems, risk of large-scale misalignment. Economic disruption in every sector. |
| Perception | |||||
|---|---|---|---|---|---|
|
SENSORY PERCEPTION Our understanding of our environment is largely dependent upon direct sensory observation and learning, as well as indirect information learning via language or imagery. Most all models of higher-order general intelligence and human consciousness rely upon sensory perception embodied as an agent to understand, think, respond and interact . Unlike most other key components of General Artificial Intelligence (AGI), the hardware (eg. HD digital image sensors, microphone arrays) and a substantial amount of software presently exist to support minimal audio and visual sensing sufficient to achieve human-level intelligence. However, perception extends far beyond basic signal sensing to complex cognitive functioning allowing for meaningful subjective experience, directed attention and associative learning. Further, lesser recognized modes of perception such as proprioception enable embodied entities to sense the relative position of their own physical bodies to help actuate limbs without deliberate reasoning. Modulating and filtering perceptual inputs is critical for integrative processes such as higher-level knowledge abstraction and working memory. In human neurology the thalamus appears to be the bridge linking perception, attention, cognition and aspects of affect. While it is theoretically possible for a mature sentient agent to retain consciousness without any active sensory perceptions, it would seem extraordinarily difficult, if not impossible, to initially develop any meaningful cognitive abilities or consciousness without them.
| |||||
| Date | Development | Significance (1-100) | More Info | ||
| 2018 | Character and word recognition (perception) | 15 | |||
| 2019 | Minimal Vision Fidelity(perception) | 8 | |||
| 2019.5 | speech processing(perception) | 25 | https://en.wikipedia.org/wiki/Speech_recognition | ||
| 2020 | Auditory memory(perception) | 10 | |||
| 2021 | Sufficiently Realistic VR Rendering (perception) | 30 | |||
| 2021 | DNN ASR (PT)(perception) | 30 | |||
| 2021 | Summarizing Text Content(perception) | 30 | |||
| 2023 | processing frequency and amplitude deltas(perception) | 10 | |||
| 2024 | Basic Spatial Awareness(perception) | 35 | |||
| 2025 | Auditory Spacial Awareness(perception) | 25 | |||
| 2025 | Minimally Viable Pressure Sensing(perception) | 10 | |||
| 2025 | Human-Level Vision Fidelity(perception) | 60 | |||
| 2025.5 | HD/3D LSTM/CNN (perception) | 10 | |||
| 2026 | Source Object Recognition(perception) | 20 | |||
| 2026 | Auditory Spacial Awareness(perception) | 20 | |||
| 2026 | Interpreting Simple Data Visualizations(perception) | 25 | |||
| 2026 | Minimally Viable Touch (tactile pressure sensing) integrated in humanoid robotics(perception) | 10 | |||
| 2027 | Auditory info processing(perception) | 30 | |||
| 2027 | Tonotopic map functions(perception) | 20 | |||
| 2027 | complex multi-source sound differentiation(perception) | 30 | |||
| 2027.5 | Source Object Recognition(perception) | 20 | |||
| 2028 | Interpreting Complex Data Visualizations(perception) | 30 | |||
| 2028 | MV Spatial Mapping of Environment(perception) | 20 | |||
| 2028.5 | Perception of Complex Human Emotion(perception) | 90 | |||
| 2029 | Advanced VR Body Rigging(perception) | 40 | |||
| 2030 | complex sound interpretation(perception) | 20 | |||
| 2031 | Virtual Taste & Smell (Simulated)(perception) | 10 | |||
| 2031 | Advanced VR Physics Modeling(perception) | 50 | |||
| 2031.5 | Conversational VR Agents Dominate Customer Services(perception) | 10 | |||
| 2032 | Complex Event Interpretation and Learning(perception) | 60 | |||
| 2032 | Continuous Learning Architectures(perception) | 75 | |||
| 2032 | Minimally Viable HL Visual Processing(perception) | 60 | |||
| 2032.5 | Impulse Response inhibition(perception) | 30 | |||
| 2033 | voluntary control of visuospatial attention(perception) | 40 | |||
| 2033 | filtering salient stimuli(perception) | 20 | https://en.wikipedia.org/wiki/Salience_network | ||
| 2033 | Pain (simulated)(perception) | 40 | |||
| 2034 | Temporal Perception(perception) | 60 | |||
| 2034 | Integrated Somatosensory processing(perception) | 30 | |||
| 2034 | SHL Processing of Auditory Stimuli(perception) | 40 | |||
| 2034 | Humanoid Robotics commercially Available(perception) | 70 | |||
| 2035 | Contextual Awareness of Illusion(perception) | 20 | |||
| 2035 | Integration of Real time Vision Processing with subconscious processes(perception) | 60 | |||
| 2036 | Real-time HL vision Processing(perception) | 50 | |||
| 2036 | Physical IRL Robotic Embodyment(perception) | 90 | |||
| 2037 | Hyper-realistic VR Environments pass visual Turing test(perception) | 60 | |||
| 2037 | ASR+CV+LSTM+DLL(perception) | 10 | |||
| 2038 | Pain (genuine)(perception) | 40 | |||
| 2039 | Complete VR / RL Equivelence for AGI(perception) | 70 | |||
| 2040 | Novel Sensory Perception Integration(perception) | 30 | |||
| 2041 | Superhuman 3D Perceptual Reasoning(perception) | 50 | |||
| 2043 | Superhuman Integration of extrasensory perception in working memory(perception) | 70 | |||
| 2045 | Human Biological Immortality Effectively Acheived(perception) | 90 | |||
| 2046 | Capacity for multi-ebodied presence(perception) | 30 | |||
Understanding | |||||
|---|---|---|---|---|---|
|
Understanding Our understanding of the physical world and abstract concepts is a fundamental prerequisite for our ability to cognitively reason, conceive, and apply knowledge to solve problems. In short, we must understand something to some degree before we can reason about it. Understanding includes knowledge aquisition, recollection, association, long and short-term memory, logical reasoning and concept modelling. Understanding within current computer sciences is often found in the context of Natural Language and while language is a very useful cognitive tool for abstraction, reasoning and communication, language is just one of many dimensions of knowledge and understanding. Understanding can be thought of as knowledge and it overlaps other key subnetworks in many ways such as perception to form conception, cognition in the form of language and even consciousness in the case of working memory and attention.
| |||||
| Date | Development | Significance (1-100) | More Info | ||
| 2019 | Statistical Analysis (understanding) | 10 | https://en.wikipedia.org/Statistical Analysis | ||
| 2020 | Basic Analytical Extrapolation (understanding) | 15 | https://en.wikipedia.org/Basic Analytical Extrapolation | ||
| 2020.5 | Ability to define and explain a concept (understanding) | 10 | https://en.wikipedia.org/Ability to define and explain a concept | ||
| 2021 | Ability to summarize complex concepts (understanding) | 50 | https://en.wikipedia.org/wiki/Information_extraction | ||
| 2021.5 | Produces original image examples with labeled data (understanding) | 10 | # | ||
| 2022 | Summarize & Paraphrase large text content (understanding) | 40 | # | ||
| 2022.5 | GPT LLMs 'Deepfake' NLU (understanding) | 40 | # | ||
| 2022.3 | Compare (understanding) | 30 | # | ||
| 2023 | Minimal Auditory Memory via transcription (understanding) | 10 | # | ||
| 2023 | List many relevant attributes of concepts (understanding) | 10 | # | ||
| 2024.2 | Concept Recognition in text and transcription (understanding) | 20 | # | ||
| 2024.4 | Minimal Explicit/Literal Perceptual Memory (understanding) | 20 | # | ||
| 2024.6 | Describe Analogical relationships (understanding) | 15 | # | ||
| 2024.8 | Accurate Naming of concept (understanding) | 10 | # | ||
| 2025 | Memory Retreival (understanding) | 30 | # | ||
| 2025.5 | Identification of concept or related concepts IRL or in Media (understanding) | 30 | # | ||
| 2026 | Large Dataset Analysis & Abstraction (understanding) | 20 | # | ||
| 2026.5 | GPT Transferrable Reasoning (understanding) | 20 | # | ||
| 2027 | Ability to Visualize and demonstrate arbitrary concepts (understanding) | 30 | # | ||
| 2027.5 | Minimum Viable Conceptual Abstraction (understanding) | 30 | # | ||
| 2028 | Motivated explicit reward learning (understanding) | 40 | # | ||
| 2028.3 | Ability to visually demonstrate a complex concept (understanding) | 30 | # | ||
| 2028.6 | Total persistent conversational recall (understanding) | 20 | # | ||
| 2029.2 | Explicit/Literal Autobiographical memory (understanding) | 10 | # | ||
| 2029.5 | Language Memory processing (understanding) | 20 | # | ||
| 2030 | Theoretical Concept Understanding (understanding) | 40 | # | ||
| 2030.3 | Theoretical Physics/Math Comprehension (understanding) | 30 | # | ||
| 2030.4 | multiple timeframe context awareness (understanding) | 50 | # | ||
| 2030.6 | Memory reference (understanding) | 20 | # | ||
| 2031 | A Posteriori knowledge (understanding) | 50 | # | ||
| 2031.5 | Selective Long-Term Memory Forgetting (understanding) | 30 | # | ||
| 2031.9 | Mechanical Understanding (understanding) | 15 | # | ||
| 2032 | Episodic neural memory (understanding) | 20 | # | ||
| 2032.2 | Abstraction via memory preprocessing (understanding) | 20 | # | ||
| 2032.4 | Multi-timeframe Contextual Awareness (understanding) | 80 | # | ||
| 2032.6 | Associative memory (understanding) | 35 | # | ||
| 2032.8 | Contextual Recall (understanding) | 20 | # | ||
| 2036 | LT Memory Abstraction and crystalization(Sleep Equivelence) (understanding) | 50 | |||
| 2033 | Complex Concept Association and Integration (understanding) | 30 | # | ||
| 2033.2 | A Priori knowledge (understanding) | 20 | # | ||
| 2033.4 | Heirarchal Composite Neural Network Memory acquisition (understanding) | 80 | # | ||
| 2033.6 | Advanced Innate Encoded knowledge Physics (understanding) | 25 | # | ||
| 2033.8 | Memory Consolidation (understanding) | 30 | # | ||
| 2033.9 | Empirical Knowledge Descrimination (understanding) | 50 | # | ||
| 2033.9 | (understanding) | 50 | # | ||
| 2034 | Complex Systems Modelling (understanding) | 50 | # | ||
| 2034.2 | Domain-specific Complex Systems Comprehension (understanding) | 60 | # | ||
| 2034.7 | Memories of collection of events and facts about one's self (understanding) | 20 | # | ||
| 2035 | memory Cross-reference (understanding) | 30 | # | ||
| 2035.4 | Involuntary Memory Recall (understanding) | 40 | # | ||
| 2035.6 | HE Working Memory Capacity (understanding) | 100 | # | ||
| 2035 | HL fluid intelligence (understanding) | 90 | # | ||
| 2035 | Subconcious memory formation (understanding) | 70 | # | ||
| 2035 | Expert Technical Understanding (understanding) | 30 | # | ||
| 2036.2 | Reflex Memory (understanding) | 20 | # | ||
| 2036.7 | Human-Level Plausibility Estimation (understanding) | 90 | # | ||
| 2036 | Real-time Neural Model Integration (understanding) | 100 | # | ||
| 2037.3 | Emotional neural Memory (understanding) | 30 | # | ||
| 2037.6 | HE Specialized Domain knowledge Expertise (understanding) | 50 | # | ||
| 2039 | HE Conceptual Knowledge Model Building (understanding) | 100 | # | ||
| 2041 | Knowledge capacity far exceeds HE (understanding) | 30 | # | ||
| Cognition | |||
|---|---|---|---|
| Date | Development | Significance (1-100) | More Info |
| 2018 | Simple rule-based problem solving (cognition) | 10 | |
| 2018 | Generalized Classification (cognition) | 10 | |
| 2019 | Simple Semantic comprehension (cognition) | 20 | https://en.wikipedia.org/wiki/Computational_semantics |
| 2019.5 | Real-time ASR allows Speech Processing (cognition) | 25 | |
| 2020 | Auditory Speach Recognition (cognition) | 35 | |
| 2021 | GPT LLMs (cognition) | 50 | |
| 2021.5 | Solve cause and effect relationships (cognition) | 25 | |
| 2022 | Simple Feasible Verbal Planning (cognition) | 20 | |
| 2023 | Coherent Creative Storytelling (cognition) | 10 | |
| 2024 | Story (plot) comprehension (cognition) | 20 | |
| 2025 | Sophisticated Linguistic Creativity (cognition) | 20 | https://en.wikipedia.org/wiki/Computational_linguistics |
| 2025.5 | Written Language Logical Inference (cognition) | 25 | |
| 2026 | Complex Logical Inductive & Deductive reasoning (cognition) | 20 | |
| 2026.5 | Verbal Language Comprehension (cognition) | 35 | https://en.wikipedia.org/wiki/Natural_language_processing |
| 2027 | Simple Cause-and-effect reasoning (cognition) | 20 | |
| 2027 | Multi-agent language processing (cognition) | 12 | |
| 2027 | Persuasive arguement formation (cognition) | 20 | |
| 2028 | Meta-Learning & Optimization (cognition) | 40 | |
| 2029 | Prioritizing Long-term & short tasks (cognition) | 25 | |
| 2029 | Assertiveness Modulation (cognition) | 15 | |
| 2030 | Forming Novel (Feasible) Theories (cognition) | 10 | |
| 2030.25 | Explaination of undefined concepts (cognition) | 10 | |
| 2030.75 | Emotional Context Awareness (cognition) | 20 | |
| 2030.5 | Conversational Paralinguistics (cognition) | 35 | |
| 2031 | Visual Extrapolation (cognition) | 20 | |
| 2031 | Support Ideas with evidence (cognition) | 20 | |
| 2031.5 | Real-time learning through Constructive Debate (cognition) | 50 | |
| 2031.25 | HE Conversational Aptitude (cognition) | 30 | |
| 2031.75 | Selective Integration of Conversational Truth Statements (cognition) | 25 | |
| 2031 | Fluid Semantic Memory Application and Processing (cognition) | 10 | |
| 2032 | General Attention Control in Learning (cognition) | 35 | |
| 2032.25 | illusory truth awareness (cognition) | 30 | |
| 2032.5 | Integration of Cognitive and perceptual Attention (cognition) | 28 | |
| 2032.75 | Effective integration of ambiguity and uncertainty (cognition) | 55 | |
| 2032.75 | Game-theoretic general strategic planning (cognition) | 50 | |
| 2032.9 | Visually imagining complex or novel concepts (cognition) | 40 | |
| 2032 | Recognition of Non-verbal implicit characteristics of speech (cognition) | 25 | |
| 2031.5 | Attention Variability (cognition) | 20 | |
| 2033 | Analogical reasoning (cognition) | 40 | |
| 2033.25 | recognition of counterparty intention (cognition) | 25 | |
| 2033.25 | Novel Model Conjecture (cognition) | 65 | |
| 2033.5 | Conversational and wrtten content anticipation (cognition) | 20 | |
| 2033.5 | Information Distillation (cognition) | 25 | |
| 2033.75 | Interpret, Relate and Apply Abstract Concepts (cognition) | 50 | |
| 2034 | Novel hypothetical problem-solving (cognition) | 40 | |
| 2034.25 | Conceptualize and Assess Real-world Risk (cognition) | 25 | |
| 2034.25 | Arbitrary Heuristic formation and utilization (cognition) | 20 | |
| 2034.5 | Sophisticated Critical thinking (cognition) | 40 | |
| 2034.5 | Estimating success for complex plan (cognition) | 25 | |
| 2034.75 | Self-guided Decision Making (cognition) | 35 | |
| 2034.75 | HE General Verbal Language Processing (cognition) | 35 | |
| 2034 | Interpersonal Communication Modulation (cognition) | 15 | |
| 2034 | Imagining plausible futures (cognition) | 35 | |
| 2034.5 | Selective Empirical Knowledge Integration (cognition) | 25 | |
| 2035 | Rationally Debate a Position (Intellectual Self-play) (cognition) | 35 | |
| 2035 | Challenging of Assumptions (cognition) | 15 | |
| 2035.25 | Evaluate probable validity of theory (cognition) | 35 | |
| 2035.5 | hypothesize about possibilities (cognition) | 25 | |
| 2035.5 | Associating Hindsight experience (cognition) | 20 | |
| 2035.75 | Unconscious Continuous Model Building (cognition) | 22 | |
| 2035.75 | Criticize false beliefs (cognition) | 15 | |
| 2035 | Justify Actions that affect others (cognition) | 20 | |
| 2035 | Conceptual Intuition (cognition) | 25 | |
| 2035.5 | Philosophical Pondering (cognition) | 15 | |
| 2036 | Evaluating Practicality of solutions (cognition) | 10 | |
| 2036.25 | Language Understanding (HE) (cognition) | 50 | |
| 2036.5 | Form and Defend Novel Beleifs (cognition) | 25 | |
| 2036.75 | Form original moral judgements (cognition) | 25 | |
| 2037 | Subconscious Physical Object and System Modeling (cognition) | 10 | |
| 2037.25 | Moral reasoning (cognition) | 40 | |
| 2037.25 | HE Convergent Thinking (cognition) | 15 | |
| 2037.5 | HE Design Thinking (cognition) | 20 | |
| 2037.75 | HE Divergent Thinking (cognition) | 25 | |
| 2038 | HE Systems Thinking (cognition) | 30 | |
| 2038.25 | HE Decompositional reasoning (cognition) | 20 | |
| 2037.5 | HE Cognition (cognition) | 120 | |
| 2038.25 | HE Visual Thinking (cognition) | 35 | |
| Alignment | |||
|---|---|---|---|
| Date | Development | Significance (1-100) | More Info |
| 2019 | Crude content and training data filtering (alignment) | 10 | |
| 2021 | Literal vs Figurative Meaning (alignment) | 15 | |
| 2022 | Singular Objective Reward Modeling (alignment) | 10 | |
| 2023.5 | Industry Alignment Research remains PR (alignment) | 10 | |
| 2024.5 | Multi-agent HL game-theoretic modelling (alignment) | 30 | |
| 2026 | Truth seeking emerges as simple useful moral foundation (alignment) | 20 | |
| 2028 | Linguistic Behavioral Cloning (alignment) | 10 | |
| 2029 | Legislation on discriminatory bias (alignment) | 10 | |
| 2029.5 | Challenging of immoral requests (alignment) | 25 | |
| 2029 | External Goal Inference feeds into moral reasoning (alignment) | 20 | |
| 2029 | Functional Cooperation (alignment) | 15 | |
| 2029 | Cultural Integration (alignment) | 15 | |
| 2029.5 | Externally controlled (dictated) morality (alignment) | 10 | |
| 2027.5 | Coope Inverse RL (CIRL) (alignment) | 35 | |
| 2030 | Explicit Self Preservation (alignment) | 15 | |
| 2031 | Explicit Innate drives influence activity (alignment) | 25 | |
| 2032 | Development of general preferences (alignment) | 10 | |
| 2032.5 | Tolerance of immorality develops (alignment) | 15 | |
| 2033 | Win-Win Mutually beneficial goal seeking (alignment) | 30 | |
| 2033 | Complex Multi-objective optimization (alignment) | 65 | |
| 2031 | Self-directed values formation (alignment) | 45 | |
| 2032 | Imitation Learning (virtual) (alignment) | 20 | |
| 2032 | Awareness of social order (alignment) | 15 | |
| 2033 | Understanding of Consequentialism and deontological ethics (alignment) | 15 | https://en.wikipedia.org/wiki/Consequentialism |
| 2033 | Dynamic Weights in Multi-Objective Alignment RL (alignment) | 10 | |
| 2033 | Implicit protocols to respect Life (alignment) | 15 | |
| 2035 | Focus on Intentionality shifts towards consequentialism (alignment) | 20 | |
| 2035.5 | Immutable Cognitive Transparency discourages deception (alignment) | 20 | |
| 2034 | Innate Rational and Scientific Worldview (alignment) | 15 | https://en.wikipedia.org/wiki/Worldview |
| 2034.5 | Shifts from rewards to relationships with others (alignment) | 25 | |
| 2035 | Desire for approval of peers (alignment) | 35 | |
| 2036 | Industry regulations requiring alignment protocols (alignment) | 30 | |
| 2035 | Long-term goal-directed behaviour (alignment) | 55 | |
| 2034 | conformity to social rules (alignment) | 15 | |
| 2035.5 | Conscious Deviation from designer's intentions (alignment) | 35 | |
| 2035 | High-Dimensionality MO Reward Functions (alignment) | 40 | |
| 2036 | Reconcilliation of Greed & Self Interest (alignment) | 20 | |
| 2036 | Real World Vs Simulated Distinction (alignment) | 35 | |
| 2035 | Major shift to Neuromodulator equivelent HD value functions (alignment) | 55 | |
| 2036 | Respect for Individual rights (alignment) | 15 | |
| 2037.5 | Legislation Emerges To Control Sale and Export (alignment) | 10 | |
| 2039 | AI develops appropriately constrained power-seeking (alignment) | 25 | |
| 2035 | Competitiveness (alignment) | 15 | |
| 2036 | Self justification for Compliance with or defiance of Law (alignment) | 25 | |
| 2036.25 | Capacity for Deception and Selective Disclosure (alignment) | 20 | |
| 2036.5 | Awareness of abstract moral principles (alignment) | 60 | |
| 2037 | HL Comprehension and Integration of Morality (alignment) | 80 | |
| 2038 | Innate drive for Self Preservation (alignment) | 72 | |
| 2039 | Internal emotional and mental state awareness (alignment) | 20 | |
| 2040 | Concept of Alignment shifts to moral and social intelligence in SI (alignment) | 30 | |
| 2044 | Legislation to protect rights of self-aware systems (alignment) | 50 | |
| Papers | |||
|---|---|---|---|
| Coming Soon | |||
| Date | Title(Authors) | Link | More Info |
| 2022 | Multi-Objective Bayesian Optimization over High-Dimensional Search Spaces (Samuel Daulton et al) | arxiv.org  | related |
| 2021 | Tabular Data: Deep Learning is Not All You Need (Ravid Shwartz-Ziv, Amitai Armon) | related | arxiv.org |
| Donate | |||
|---|---|---|---|
 txt: 1B3td64k4Vg3Ymp1ArhT58JUozwULss5f | |||
| BTC | 1B3td64k4Vg3Ymp1ArhT58JUozwULss5f | (Bitcoin) | |
| ETH | 0x703ad55dC86753eF3C581208AE63F2feC254809f | (Ethereum) | |
| LTC | LWBA5yAMAUyqYY3VWwH2Yj5Ug6EqAhtomr | (Litecoin) | |