9 hrs ago
OpenAI Targets Fully Automated AI Researchers by 2028
OpenAI says its computer helpers can already do some research jobs with people watching them.
These helpers can write code, run experiments, and help plan research.
Researchers sometimes use several helpers at the same time.
The helpers are doing more complicated jobs than they did before.
However, people still need to guide them and fix problems.
OpenAI says it wants to build an automated AI researcher by March 2028.
Such a system could suggest ideas, test them, and study the results.
OpenAI says these systems might also help make AI safer.
The company warns that it may slow down or stop development if safety cannot be guaranteed.
OpenAI says it has created a research assistant that can perform well-defined tasks under human supervision.
The company reports that its researchers used the equivalent of 3.1 agent workdays for every human workday by mid-August.
AI agents are handling coding, experiments, research planning, technical specifications, training oversight, and infrastructure troubleshooting.
More than half of successfully completed tasks requiring four to eight hours of human work involved at least one human intervention over the past six months.
OpenAI aims to make significant progress toward an automated AI researcher by March 2028 while warning that safety and oversight challenges may increase.
- Who
- OpenAI and its research teams, using AI agents under human supervision.
- What
- OpenAI announced progress toward automated research assistants and a goal of developing an automated AI researcher by March 2028.
- Where
- OpenAI’s research operations; no specific physical location is stated.
- When
- The announcement describes current use, data from mid-August and the past six months, and a target of March 2028.
- Why
- The goal is to accelerate AI research and potentially improve AI safety and alignment.
Potential Research Benefits
Safety and Oversight Concerns
Research productivity
Potential Research Benefits
AI agents can write code, run experiments, plan research, and handle increasingly complex tasks, potentially accelerating scientific work.
Safety and Oversight Concerns
The agents still require significant human guidance, particularly as tasks become more complex.
Automated AI research
Potential Research Benefits
An automated researcher could generate ideas, develop them, conduct experiments, and analyze results, while contributing to AI safety and alignment.
Safety and Oversight Concerns
More advanced systems may be harder to oversee, requiring safety and alignment efforts to evolve alongside their capabilities.
Development pace
Potential Research Benefits
OpenAI is pursuing significant progress toward an automated AI researcher by March 2028.
Safety and Oversight Concerns
OpenAI says it is prepared to slow down or halt development if safety cannot be sufficiently guaranteed.
Key facts
- Current capability
- AI assistants can handle well-defined research tasks that might take a skilled researcher several days.
- Agent usage
- By mid-August, OpenAI’s researchers used the equivalent of 3.1 agent workdays for every human workday.
- Concurrent use
- An increasing number of researchers were running four or more agents simultaneously.
- Task expansion
- Agents were being used for research planning, technical specifications, experiment analysis, training oversight, and infrastructure troubleshooting.
- Human intervention
- Over the past six months, more than half of successfully completed tasks estimated at four to eight hours of human work involved at least one human intervention.
- OpenAI target
- The company says it is making significant progress toward an automated AI researcher by March 2028.
- Safety position
- OpenAI says it may slow or halt development if a system’s safety cannot be sufficiently guaranteed.
Sources
Related news
OpenAI’s AGI Timeline Raises Hopes and Serious Safety Questions
OpenAI Says AI Agents Now Supply More Research Work-Hours
OpenAI launches GPT-6 Astra for autonomous computer tasks
Anthropic Study Finds AI Models Improving Other Models
Nvidia CEO Says OpenAI GPT-6 Astra Has Reached AGI
OpenAI Scientist Warns Smarter AI Requires Global Safety Standards
Matt Clifford Steps Down as ARIA Chair Over Anthropic Conflict
OpenAI Chief Scientist Warns of Rapid AI Progress Risks
ChatGPT May Soon Learn and Mimic Your Writing Style
OpenAI Chief Scientist Warns AI Risks Require Tighter Critical-System Security
