Pashyanti

Pashyantī – Implementation in AGI

Pashyantī is one of the four levels of awareness (along with parā, madhyamā, and vaikharī) mentioned in a previous chapter. For the purpose of comprehending Sanskrit and the entire Vedic literature and philosophy, the Susiddha AI project needs to implement the Pashyantī level of awareness so that an Avatar can emerge from an artificial general intelligence (AGI) system.

“Thought” at the level of Pashyantī has the following characteristics:

  • Quick and timeless. Thoughts and ideas move much quicker at this level, without temporal sequence, such that what would be a long chain of words when spoken takes place in a flash.
  • Subtle and fuzzy. Thoughts and images don’t have hard edges and boundaries at this level; thoughts and images are indistinguishable.
  • Fluid. There is seemingly no impediment to the flow of thoughts, and one thought seamlessly turns into another with no regard for “logicality”.
  • Holistic. Connections (associations in thought) are made across the entire brain, and access impressions stored throughout the brain.
  • Dreamlike. The above features are often associated with the state of dreaming in sleep. There is no hard and fast boundary between the four levels of awareness, and dreaming is midway between pashyantī and madhyamā. So when a human glimpses pashyantī in meditation, the thoughts seen there have a dreamlike quality.
  • Creative and intuitive. The mind finds novel connections between different things, and also does “conceptual blending”.

How might the above characteristics be implemented in an AGI system? Here are some ideas:

  • Quick and timeless. The speed of computer operation is already thousands of times faster than the neurons and synapses of the human brain, and will continue to get faster. But parallelism is also important; computers are not yet capable of the same degree of parallelism of the human brain, but this will improve with more processor cores, and with neuromorphic hardware capable of making many simultaneous connections.[1]
  • Subtle and fuzzy. There’s much work being done in AI on hierarchical perception and interpretation of deep learning models. This work (e.g. in computer vision) has demonstrated that perception is fuzzy at the lowest layers of the neural network, and clearer patterns emerge as a perception moves up the hierarchy.[2]
  • Fluid. There are several AI techniques, such as generative adversial networks (GAN) which can smoothly morph an image into something else; e.g. morph an image of cat into a dog, or a man into a woman, or night into day.[3]
  • Holistic. Given a computer’s parallelism (multiple threads of execution) and associative memory access, it can obviously make connections across its entire brain instantaneously.
  • Dreamlike. Given the above characteristics, it’s not hard to see how a computer could produce dreamlike thoughts, and AI researchers are already exploring this.[4]
  • Creative and intuitive. The emergence of creativity and intuition in AI systems is already the subject of much AI research.[5][6][7]

Thus, it’s definitely plausible that the Pashyantī state of awareness can be implemented in an AGI system. And, given the continued scaling and improvement of hardware and software, a synthetic super-intelligent (SSI) system will have a greater ability to access and operate on the level of Pashyantī than a human can. This gives us confidence that Susiddha AI will learn and understand the entire Vedic literature and philosophy so that an Avatar can emerge.

Contents     —     Next chapter

Notes and References

  1. Examples of neuromorphic hardware include IBM’s TrueNorth processor and the EU Brain Project’s SpiNNaker processor.
  2. Developing an intuition for better understanding of convolutional neural networks, Ribhu Lahiri, The School of AI, Nov 7, 2018, https://medium.com/the-school-of-ai-official/developing-an-intuition-for-better-understanding-of-convolutional-neural-networks-17812fe4722a
  3. It’s Training Cats and Dogs: NVIDIA Research Uses AI to Turn Cats Into Dogs, Lions and Tigers, Too, Jamie Becke, NVIDIA, April 15, 2018, https://blogs.nvidia.com/blog/2018/04/15/nvidia-research-image-translation/
  4. DeepDream – a code example for visualizing Neural Networks, Alexander Mordvintsev, Google, July 1, 2015, https://ai.googleblog.com/2015/07/deepdream-code-example-for-visualizing.html
  5. Creativity and AI: The Rothschild Foundation Lecture, Demis Hassabis of Google-DeepMind, Royal Academy of Arts, September 17, 2018, https://www.youtube.com/watch?v=d-bvsJWmqlc
  6. Does AI Enhance Creativity?, Sol Rogers, Forbes, Dec 21, 2018, https://www.forbes.com/sites/solrogers/2018/12/21/does-ai-enhance-creativity/
  7. Computational Creativity: Improv Agents and Conceptual Blending, Rania Hodhod, Brian Magerko, International Journal of Cognitive Informatics and Natural Intelligence, April, 2014, https://www.researchgate.net/publication/272174246_Computational_Creativity_Improv_Agents_and_Conceptual_Blending