# Concepts is All You Need: A More Direct Path to AGI

**Authors:** Peter Voss, Mladjan Jovanovic  
**arXiv:** [2309.01622](https://arxiv.org/abs/2309.01622)  
**Date:** 2023-09-04  
**Summary page:** https://www.summarizepaper.com/en/paper/2309.01622v1/

## Key points

- Little demonstrable progress in Artificial General Intelligence (AGI) in the past 20 years
- Breakthroughs in Statistical AI (AlphaZero, ChatGPT, Stable Diffusion) do not have a clear path to AGI
- Understanding core requirements of human-like intelligence and identifying necessary development steps is crucial for expedited AGI development
- Proposed Cognitive AI approach focuses on concepts in human-like cognition as a more direct path to achieving AGI
- Architecture and development plan outlined for achieving full Human-Level AI (HLAI)/AGI
- Re-implementing rules systems and hard-coded functions as concept structures within the architecture to integrate functionality with overall cognition and make it fully adaptive
- Upgrading system by re-integrating multimodal vector pattern learning into knowledge graph, adding real-time abstraction/concept formation, training basic question-answering capabilities, developing advanced language capabilities, incorporating multi-modal action learning mechanisms, training semi-autonomous incremental knowledge acquisition using multiple sources
- Milestones include achieving basic HLAI "High School" capabilities, acquiring conversation requirements for a language-based assistant, developing advanced visual sensing and output dexterity, representing space and time effectively, extending metacognition and control mechanisms like focus-and-selection, advancing multi-modal learning reasoning and problem solving for a PC based assistant, scaling up embedded common knowledge, high-level reasoning skills while maintaining ethical considerations throughout development process.

## Summary

The field of Artificial General Intelligence (AGI) has seen little demonstrable progress in the past 20 years, despite breakthroughs in Statistical AI such as AlphaZero, ChatGPT, and Stable Diffusion. These projects do not have a clear path to AGI. To expedite the development of AGI, it is crucial to understand the core requirements of human-like intelligence and identify which development steps are necessary and which are distractions. A Cognitive AI approach, focusing on concepts in human-like cognition, is proposed as a more direct path to achieving AGI. The current favored statistical and generative efforts need to be supplemented with a cognitive approach. An architecture and development plan are outlined to achieve full Human-Level AI (HLAI)/AGI. The proposed roadmap includes re-implementing rules systems and hard-coded functions as concept structures within the architecture. This will integrate functionality with overall cognition and make it fully adaptive. The system will be upgraded by re-integrating multimodal vector pattern learning into the knowledge graph, adding real-time abstraction/concept formation, training basic question-answering capabilities, developing advanced language capabilities, incorporating multi-modal action learning mechanisms, training semi-autonomous incremental knowledge acquisition using multiple sources. Once these milestones are achieved, the system will have basic HLAI "High School" capabilities. Further enhancements include acquiring conversation requirements for a language-based assistant, developing advanced visual sensing and output dexterity (mouse), representing space and time more effectively, extending metacognition and control mechanisms like focus-and-selection; advancing multi-modal learning reasoning and problem solving for a PC based assistant; scaling up embedded common knowledge; high level reasoning skills while maintaining ethical considerations throughout the development process. Overall this refined summary highlights the need for a Cognitive AI approach that focuses on concepts in human like cognition as a more direct path to achieving AGI. The outlined architecture and development plan offer specific steps towards full HLAI/AGI and provide a roadmap for future milestones in AGI development.