# Systematic AI Approach for AGI: Addressing Alignment, Energy, and AGI Grand Challenges **Authors:** Eren Kurshan **arXiv:** [2310.15274](https://arxiv.org/abs/2310.15274) **Date:** 2023-10-23 **Summary page:** https://www.summarizepaper.com/en/paper/2310.15274v1/ ## Key points - The paper presents a new systematic AI approach for AGI that addresses three grand challenges: the Energy Wall, the Alignment Problem, and the Leap from Narrow AI to AGI. - The current state of AI solutions consumes unsustainable amounts of energy during model training and daily operations. - The leap from AI to AGI requires multiple functional subsystems operating in a balanced manner, which necessitates a system architecture. - The current approach to artificial intelligence lacks system design, despite its crucial role in how the human brain processes information and makes decisions. - Alignment and AI ethics approaches often overlook system design, even though studies show that the brain's system architecture plays a critical role in healthy moral decisions. - System design is critically important in overcoming all three grand challenges. - The proposed Systematic AI Approach for AGI utilizes system design principles for AGI while providing ways to overcome the energy wall and alignment challenges. - The paper is organized into several sections: describing the grand challenges, introducing the proposed systematic approach, discussing advancements in AI hardware systems, exploring system design highlights for alignment, presenting examples of system design principles derived from studying the human brain, and concluding remarks. ## Summary This paper presents a new systematic AI approach for AGI that aims to address the trifecta of grand challenges facing AI: the Energy Wall, the Alignment Problem, and the Leap from Narrow AI to AGI. The current state of AI solutions consumes unsustainable amounts of energy during model training and daily operations, with the amount of computation required to train each new AI model doubling every 2 months since 2020. This exponential increase in energy consumption is not sustainable. The leap from AI to AGI requires multiple functional subsystems operating in a balanced manner, which necessitates a system architecture. However, the current approach to artificial intelligence lacks system design, despite its crucial role in how the human brain processes information and makes decisions. Similarly, alignment and AI ethics approaches often overlook system design, even though studies show that the brain's system architecture plays a critical role in healthy moral decisions. In this paper, the authors argue that system design is critically important in overcoming all three grand challenges. They propose a Systematic AI Approach for AGI that utilizes system design principles for AGI while providing ways to overcome the energy wall and alignment challenges. The systematic approach emphasizes capturing the underlying complexity required for intelligence and highlights the importance of integrating complex cognitive capabilities into an overarching framework for AGI. The paper is organized into several sections. Section 2 describes the current grand challenges of energy consumption, alignment, and achieving AGI. Section 3 introduces the proposed systematic artificial general intelligence approach. Section 4 discusses recent advancements in AI-optimized hardware systems and their relevance to systematic AGI. Section 5 explores system design highlights for alignment, while Section 6 presents examples of system design principles derived from studying the human brain. Finally, Section 7 provides concluding remarks. The first grand challenge discussed is the Energy Wall, which refers to the significant energy consumption associated with contemporary AI solutions. Reports indicate that both run-time power dissipation and training energy consumption are major issues. For example, AlphaGo used 1920 CPUs and 280 GPUs consuming around 170 kW during championship games compared to only 20W consumed by human champions playing similar games at similar levels of difficulty or skill level . GPT-4 has an estimated daily energy consumption of around 260 MWh while its training period consumed between 51K - 62K MWh equivalent to monthly output of an average nuclear power plant operating at 876K MWh per year . If this trend continues , future generations may consume entire output of nuclear plants during training . The second grand challenge is Alignment Problem which refers ensuring that Artificial General Intelligence (AGI) systems align with human values & goals . Authors propose that systemic approach can help integrate alignment considerations into design process from building block level up till fully operational systems . The third grand challenge discussed is Leap from Narrow Artificial Intelligence (AI) towards Artificial General Intelligence (AGI). Achieving AGI requires multiple functional subsystems working together cohesively . Authors assert that paradigm shift towards focusing on system design is necessary for reaching AGI successfully . Overall , this paper highlights importance of system design in overcoming grand challenges facing Artificial Intelligence & proposes Systematic Artificial Intelligence Approach as solution . By incorporating system design principles into development & considering factors such as performance , efficiency , reliability & robustness from both perspectives , it aims at addressing Energy Wall & Alignment Problem while facilitating progress towards Artificial General Intelligence (AGI).