# Ten Hard Problems in Artificial Intelligence We Must Get Right

**Authors:** Gavin Leech, Simson Garfinkel, Misha Yagudin, Alexander Briand, Aleksandr Zhuravlev  
**arXiv:** [2402.04464](https://arxiv.org/abs/2402.04464)  
**Date:** 2024-02-06  
**Summary page:** https://www.summarizepaper.com/en/paper/2402.04464v1/

## Key points

- Address hard problems hindering AI's full potential and posing risks to society
- Developing general capabilities of AI systems
- Ensuring performance and training processes
- Aligning system goals with human goals
- Enabling real-life applications
- Addressing economic disruptions
- Ensuring participation for all
- Socially responsible deployment
- Geopolitical disruptions caused by AI
- Promoting sound governance of the technology
- Managing philosophical disruptions for humans living in the age of AI
- Emphasize starting work on these issues now to achieve positive outcomes by 2050
- Review recent literature on hard problems, outline research agenda for each one, explore history of AI advancements, and highlight significant work published between 2017 and 2022
- Consider hard problems as "wicked problems" with potential for partial solutions through focused efforts
- Stress collaboration across various fields concerned with AI to ensure positive outcomes by 2050
- Provide additional resources in electronic appendix for each hard problem discussed in the paper

## Summary

The  aims to address the hard problems that currently hinder the full potential of  and pose risks to society. These problems include developing general capabilities of AI systems, ensuring their performance and training processes, aligning system goals with human goals, enabling real-life applications, addressing economic disruptions, ensuring participation for all, socially responsible deployment, geopolitical disruptions caused by AI, promoting sound governance of the technology, and managing philosophical disruptions for humans living in the age of AI. The program emphasizes the importance of starting work on these issues now to achieve positive outcomes by 2050. The paper reviews recent literature on these hard problems and outlines a research agenda for each one. It explores the history of AI advancements and presents significant work published between 2017 and 2022. The authors highlight the need for continued research in these areas over the coming decades. The paper also addresses the perception that these problems are unsolvable by considering them as "wicked problems" with potential for partial solutions through focused efforts. The outlook for the future is optimistic, envisioning a scenario where AI has greatly benefited society due to successful resolution of key challenges. Acknowledgments are made to funding sources and individuals who contributed to the paper's development. The authors stress collaboration across various fields concerned with AI to ensure positive outcomes by 2050. The electronic appendix provides additional resources for each hard problem discussed in the paper. Overall, the  seeks to harness the potential of  while mitigating risks through proactive research and collaboration among experts in the field. By addressing these hard problems now, society can reap widespread benefits from  in the future.