This question seeks to identify who will first publicly reveal human-level AGI.

Timeline: Voting closes at the end of 2029. The question may be resolved earlier if a clear winner emerges. If human-level AGI is not achieved by 2030, the question will resolve based on who has made the most progress towards the criteria.

Definitions:

Human-level AGI: Artificial Intelligence capable of doing almost all human cognitive tasks to a high level (equivalent or better than a median professional in the field). For this question, we will not require embodiment or physical labour. If there is widespread consensus among AI researchers that AGI has been achieved, this determination takes precedence.

Evaluation Criteria:
If the AI can perform at or above the level of human professionals in at least 95% of remote jobs that exist in 2024, with salaries up to the median level (for that job), it should be considered human-level AGI.

Software Developer

For a software developer, human-level AGI would be expected to understand and implement complex software solutions equivalent to or better than a median professional. This includes the ability to:

• Analyse software requirements, ask appropriate clarifying questions and design comprehensive software architecture.

• Write clean, efficient, and well-documented code in appropriate programming languages. Demonstrate ability to integrate with existing large codebases, not just new projects.

• Assess how well solutions satisfy requirements (taking into account technical correctness, user experience, security, maintainability etc).

• Debug and resolve software issues with minimal guidance.

• Adapt to new programming languages or frameworks as needed. Adapt to changes/clarifications in requirements.

• Collaborate with team members on software projects, understanding and integrating feedback.

• Communicate well e.g. give estimates for work along with levels of uncertainty, make suggestions for beneficial changes etc.

The AGI would demonstrate not just technical proficiency, but also the problem-solving skills and creativity typically exhibited by experienced developers in creating or improving software products.

Scientific Discipline (Remote)

For a scientific discipline that can be conducted remotely, such as computational biology, human-level AGI would be expected to:

• Conduct complex data analysis and interpretation with proficiency equivalent to or exceeding that of a median professional in the field.

• Develop and test hypotheses using established scientific methods.

• Use and understand a wide range of computational tools and software specific to the discipline.

• Integrate new research findings into ongoing work.

• Design and execute experiments (simulations, in this case) to validate theoretical models or explore new scientific questions.

• Write research papers and reports that meet peer-reviewed publication standards.

The AGI's performance would reflect a deep understanding of the field's theoretical foundations, as well as the practical application of its methods and technologies.

Educational Content Creator

For an educational content creator, human-level AGI would be expected to:

• Design and develop engaging and pedagogically sound educational materials (e.g., video lectures, interactive exercises, written content) across various subjects, tailored to different learning styles and educational levels.

• Assess and incorporate feedback from learners to improve content quality and effectiveness continually.

• Stay informed about the latest educational trends, technologies, and pedagogical research, applying this knowledge to create innovative learning experiences.

• Adapt content for diverse educational platforms and formats, from online courses to mobile learning apps.

• Evaluate the effectiveness of educational materials through data analysis and learner outcomes, making data-driven decisions to enhance learning engagement and achievement.

This AGI would embody the creativity, pedagogical knowledge, and adaptability required to produce educational content that is both informative and captivating for a wide range of learners.

Counter Examples

An AI will not be deemed competent if it consistently makes critical errors that a competent professional would not make, even if exceptionally skilled in other areas.

The AI should be able to adapt to the communication styles and tools commonly used in a professional environment. It should not require instructions to be formatted in a specific or unusual manner.

General Considerations
It is inherently difficult to define comprehensive criteria due to the general nature of the intelligence required. Existing benchmarks may not be sufficient. Focus on the AI's demonstrated capabilities across a variety of professional tasks rather than theoretical arguments, recognising that a degree of judgment may be needed when announcements are made.

I have attempted to reflect a balanced view of human-level AGI. It is strict in some ways e.g. the breadth of knowledge and ability required. Being as capable as prescribed in almost every field could be considered superhuman. However, it is lenient in other regards e.g. no embodiment requirements - which covers a broad spectrum of human skill that the AGI would not be evaluated under.

Option Clarifications

• Collaborations and mergers: If there is a clear lead partner in terms of conducting the core AI research and development, choose this option. Otherwise, use the order of preference from top to bottom.
  Example 1: If Meta were to acquire Anthropic after the question is posted Anthropic would be considered the winner if they lead the research efforts even if funding, infrastructure etc are provided by Meta.
  Example 2: If Stanford University were to develop AGI with funding from Google, the "University..." option would win. However, if Google were to collaborate with Stanford University and either lead the research or if there is no clear research leader, Google would win due to the specified order of precedence.

• Who counts as a 'winner'?: The entity that achieves human-level AGI and is the first to publicly reveal it will be considered the winner. Public revelation includes demonstrations, peer-reviewed papers, or other verifiable evidence of capability.

• Government initiatives: Includes government-led research institutions/labs, but not government-funded work primarily carried out by a private company.

• Other considerations: Fame in the AI field is not a requirement for the "Another specific company" category.