Thread by @Singularity_NET

1) NNs, LLMs, Diffusion - narrow #AI's are showing exciting capabilities right now. So, why do we need #AGI?

Simply, these algorithms are amazing pattern-recognizers...but they don't understand what they are doing.

A classic example of their unfortunate limitations:

2) A seemingly perfect-accuracy AI-driven tumor detector from tissue images... failed completely when given images from a new hospital.

On initial training data - 98% accurate; on data from this new source, it was no better than a coin flip

3) Turned out, all the training images with a tumor had a little white ruler in the picture to measure the size of the tumor. The new hospital did not include rulers in any images.

The model didn't understand tumors, and so has learned to be an amazing ruler detector.

4) This happens over & over in narrow-AI systems - an ever more sophisticated game of whack-a-mole.

This is where AGI frameworks, like #SingularityNET's @OpenCog, are needed - to bring reasoning, learning, and understanding capabilities together with these AI advances.

5) OpenCog Hyperon can utilize the strengths of advancing AI algorithms, by integrating their abilities into its framework directly.

- Increasing the capability of narrow-AI systems to genuinely solve problems
- Leveraging narrow-AI systems to augment OpenCog intelligence

6) Through the MeTTa AGI language and the Distributed Atomspace, narrow-AI modules, including NN's, are:

- given a collaborative 'space for thinking'
- and a 'thinking mechanism,'
- along with tools for pattern mining, probabilistic reasoning, and relevance...

7) Ready to understand more about the next generation of AI system being built by #SingularityNET, and the progress toward decentralized AGI?

Read our 2022 AGI Progress Overview: blog.singularitynet.io/singularitynet…