OpenAI researchers are making significant strides towards creating AI systems capable of performing human-level reasoning, according to recent reports from Bloomberg and Reuters. This development is crucial as it could reshape our understanding and application of artificial intelligence (AI) in various fields.
Why It Matters
Experts in AI are divided over whether current AI models, which are excellent at generating text and images, will ever reach a level of understanding and adaptability similar to humans. The debate centers on whether these models can truly comprehend the world and adjust to new information and situations as humans do.
Driving the News
OpenAI has outlined five levels of artificial general intelligence (AGI) internally, according to a report by Bloomberg. These levels are:
- Chatbots: AI systems that can engage in conversational language.
- Reasoners: AI capable of human-level problem-solving.
- Agents: AI systems that can take actions.
- Innovators: AI that can assist in invention.
- Organizations: AI that can perform the work of an entire organization.
During a recent company meeting, OpenAI leaders indicated that their systems are currently at level 1 but are “on the cusp” of reaching level 2, which involves human-level problem-solving.
Project “Strawberry”
Reuters disclosed information about a project at OpenAI named “Strawberry,” which aims to develop AI capable of human-level reasoning, planning, multi-step problem-solving, and independent deep research. This project is a continuation of an earlier effort known as “Q*.”
Recent Developments
Rumors of a significant breakthrough at OpenAI circulated just before the company’s board attempted to remove CEO Sam Altman in November 2023. A day before his firing, Altman mentioned at an Asia-Pacific Economic Cooperation event that recent weeks had seen the company pushing the “veil of ignorance back and the frontier of discovery forward.”
However, eight months later, OpenAI’s latest release is a new version of ChatGPT, GPT-4o, which combines text and visual modes in advanced ways. Human-level reasoning remains an objective for the future.
Understanding AGI
OpenAI’s mission is to ensure that AGI benefits all of humanity. However, the concept of AGI has often been described in vague terms. Optimists believe that AGI will evolve from today’s large language models (LLMs) as they grow bigger, train on more data, and improve through post-training techniques. Skeptics, on the other hand, think that the LLM approach might hit a ceiling, unable to achieve human-level reasoning before the resources required become unsustainable.
Comparing AI Progress
OpenAI’s five-level AI ladder is similar to other frameworks used to assess progress towards fully autonomous vehicles. For instance, researchers at Google’s DeepMind have also proposed a five-level breakdown for measuring AI advancements. Notably, neither OpenAI’s nor Google’s frameworks mention the possibility of AI systems becoming autonomous or self-aware.
The Bigger Picture
OpenAI, along with competitors like Google, Meta, and Anthropic, is investing billions to achieve AGI. It’s commendable that they are putting effort into defining what AGI means. However, there’s a risk that these definitions might just be shifting the ambiguity from one set of terms to another. Terms like “human-level problem solving” and “reasoning” could be as difficult to define as “artificial general intelligence.”
As OpenAI continues to push the boundaries of AI, the journey towards human-level reasoning remains an exciting and closely watched endeavor.