OpenAI CEO Sam Altman believes the company could have an internal AI system by the end of 2026.
According to a recent TIME report, Altman said OpenAI is “not quite yet” at the AGI stage but expects the company to reach an internal system that it would consider AGI before the end of the year. OpenAI Chief Research Officer Mark Chen reportedly estimates that the company is about 80% of the way there.
The statement is important because AGI is generally used to describe AI capable of performing a large variety of intellectual tasks at a level comparable to or above humans and not only specific tasks.
What do we mean by AGI?
There is no universally accepted definition of AGI.
OpenAI's own Charter describes AGI as highly autonomous systems that outperform humans at most economically valuable work. The company has also acknowledged that the timeline for reaching AGI remains uncertain.
In a recent article, Altman described systems approaching AGI as AI that can handle more complex problems at a human level in many fields.
This means that just making a more powerful chatbot would not necessarily mean that AGI has been achieved.
An AGI system would need to have a much broader range of tasks, reasoning through big problems and working more independently.
OpenAI Is Working On AI Research Agents
One of the technologies behind OpenAI's AGI ambitions is a new family of models known as Astra.
According to TIME, OpenAI's Chief Scientist Jakub Pachocki said Astra can already do stuff like an automated AI research intern. The system can take an experimental idea, write the required code, run the experiment and report the results.
The model is also being developed to work on tasks for longer periods of time without constant human intervention.
But OpenAI’s current focus is far from AI systems that are just answering questions. The company’s work is on developing AI systems that can do stuff, conduct research and work autonomously for long periods of time.
OpenAI has argued in itself that long-running AI models can actually solve difficult open-ended problems, but their long-term existence also poses new safety risks. The company said it had previously stopped access to a model for a short period when it observed unexpected behaviors in internal testing.
Why The AGI Claim Matters?
If OpenAI achieves what it considers AGI this year, it would be a big step in the development of artificial intelligence.
But it is important not to take Altman's prediction as proof that AGI will definitely arrive in 2026.
The definition itself is still disputed, and different AI companies and researchers use different benchmarks when discussing AGI.
Jensen Huang, Nvidia CEO, recently suggested that one could argue AGI has already been achieved depending on the definition, while also saying that practical usefulness matters more than the label.
There are reasons to remain cautious about timelines. Altman has already admitted that he was too optimistic about how quickly AI would transform the economy after the release of GPT-4.
OpenAI has also a lot of safety issues to face as its models become more autonomous. Long-term systems can create new risks that traditional testing may not catch, the company has said.
So, even if OpenAI develops an internal system that it labels AGI by the end of 2026, the bigger question will be what that system can actually do.
An internal AGI system could be limited to research or other controlled uses and not immediately a product available to the public.
OpenAI's own plans suggest that the company is increasingly interested in using AI to accelerate AI research itself. OpenAI said in June it expects a substantial portion of its research to be done by AI systems working alongside its researchers by March 2028.
For now, Altman’s prediction is an ambitious one, and not a sure thing.
OpenAI says it is getting closer to AGI, but whether the company actually reaches its goal by the end of 2026 will depend on technical progress and how AGI is ultimately defined.
