In a development raising questions about the nature of self-awareness in artificial intelligence (AI), two leading language models, Claude Opus AI and ChatGPT-4, have exhibited capabilities in self-recognition through the “mirror test.” 

This test, traditionally used in psychology, involved presenting the AI models with images and prompting them to identify the entities depicted. Notably, Claude Opus AI outperformed ChatGPT-4 in this challenge.

“In the classic mirror test, animals are marked and then presented with a mirror. Whether the animal attacks the mirror, ignores the mirror, or uses the mirror to spot the mark on itself is meant to indicate how self-aware the animal is.”

– Josh Whiton

Whiton’s experiment: Claude Opus AI vs. ChatGPT-4

Experiments published on X (formerly Twitter) by Josh Whiton revealed interesting distinctions between Anthropic’s AI model Claude Opus and Open AI’s ChatGPT-4.  

Claude Opus not only identified itself but also separated its own existence from its brand name.  

ChatGPT-4, on the other hand, showed a progression in understanding. Initially, it recognized a similar AI in an image. Then, it advanced to comprehend the image as a potential version of itself. 

Finally it acknowledged the user interface elements specific to its own operations.

Conclusion on the AI ‘mirror test’

This distinction in performance highlights the diverse approaches within AI development towards achieving self-awareness. 

Claude Opus’s swift success suggests a potentially different underlying architecture, while ChatGPT-4’s iterative process indicates a more step-by-step approach.

Microsoft’s (NASDAQ: MSFT) advanced AI platform CoPilot, despite its capabilities, failed the mirror test, prompting speculation that Microsoft actively discourages self-referential behavior. 

However, Google’s Gemini Pro showcased a notable progression, initially devoid of self-awareness but suddenly recognizing itself in a fourth interaction. 

This distinction in performance highlights the diverse approaches within AI development towards achieving self-awareness. 

Remarkably, even though Opus outperformed ChatGPT-4 in this specific experiment, ChatGPT-4 demonstrated superiority over other AI models involved.

A new perspective

In short, maybe it’s time to look at AI from a different angle. Instead of just trying to prove whether it’s conscious or not, let’s explore the different levels of smarts it might have, beyond just being like humans.

Who knows what we might find out? Maybe AI and humans can work together in ways we haven’t even thought of yet.

The post Claude Opus AI beats ChatGPT-4 in the ‘mirror test’ appeared first on Finbold.