OpenAI: the 5 levels of AI

AI has grown exponentially in recent years and is having a major impact on daily life. To make sense of these developments, OpenAI has laid out five levels of AI capability. In this article we walk through what each level means, which GPT models match each one, and where we currently stand.

The first level of AI focuses on basic patterns and simple responses to stimuli, with no form of memory. These systems respond to direct input and cannot remember context.

What this involves:

• Limited functionality: reactive AI responds to specific inputs without memory or the ability to learn.
• Examples: simple chatbots and voice assistants like Siri answering standard questions.
• Comparable GPT model: GPT-1. The first step in using large language models for NLP, with limited context understanding and no long-term memory.

At this level, AI can remember and use contextual information within a single interaction to improve its responses. The model can learn from historical data, but its memory is limited to short-term information.

What this involves:

• Contextual memory: AI can remember and process information within one session, but loses it after the session ends.
• Examples: chatbots that remember previously shared information within the same session to give more relevant answers.
• Comparable GPT model: GPT-2. Improved text generation and better in-session context understanding, but still no long-term memory.

AI at this level can better understand human emotion, intent, and mental state, and respond accordingly. It has more advanced context understanding and can engage in more complex interactions.

What this involves:

• Human interaction: the AI picks up on subtle nuances of language and context, and uses them to hold conversations closer to natural human communication.
• Examples: hypothetical AI that infers human emotions and reacts to them, offering personalised experiences.
• Comparable GPT model: GPT-3 and GPT-4. Both have very advanced language understanding and can hold complex conversations with deeper contextual knowledge than their predecessors. They are not yet able to fully understand human emotion or be self-aware.

Self-aware AI would not only understand human emotion, but also be aware of its own existence. This level involves self-reflection and the ability to take actions and decisions based on self-awareness.

What this involves:

• Awareness: the AI has a sense of self and can reflect on its own decisions and 'thoughts'.
• Examples: there are no current AI systems operating at this level. It remains a theoretical concept.
• Comparable GPT model: no existing GPT model meets the criteria for self-awareness. Even GPT-4 lacks self-awareness and cannot reflect on its own 'thoughts' or 'feelings'.

Superintelligence represents the end-point of AI development, where AI surpasses every human capability. This would mean AI can solve any conceivable problem better than humans, with capabilities far beyond our understanding.

What this involves:

• Unprecedented capability: AI that exceeds human intelligence on all fronts, from creativity to problem solving.
• Examples: not yet reached, and remains a theoretical possibility.
• Comparable GPT model: no existing GPT model comes close to superintelligence. This would require a future, completely new category of AI.

Currently we sit somewhere between Level 2 and Level 3, thanks to models like GPT-3 and GPT-4. These models have very deep context understanding and can hold complex conversations, but they lack long-term memory and emotional intelligence. Although GPT-4 represents a significant leap over its predecessors, it is still limited in its ability to fully understand human emotion or to reflect on itself.

It is worth emphasising that while GPT-4 delivers impressive results across language-related tasks, it does not yet have the capabilities of a true 'theory of mind' AI in the fullest sense. It cannot infer intent or emotion at the level required for genuine human understanding, and it is certainly not self-aware.

OpenAI's five levels of AI give a clear picture of where the technology is now and where it might go. With GPT-3 and GPT-4, we are in the transitional phase between Level 2 and Level 3. While these models deliver striking performance in language processing, there is a long road ahead before we reach self-aware or superintelligent AI. The future of AI is promising, and it will be exciting to see how the technology develops over the coming years.