why conceptnet won’t work

Conceptnet is a project I’ve been dreaming about on/off since at least 2013.

I wanted to list out why it won’t work so I can potentially talk myself out of dreaming about it.

  • Artificial Intelligence is too hard. People have been contemplating whether intelligence can be embedded into machines for thousands of years. And since we invented computers, its been on the thoughts of thousands of smart computer scientists with lots of work, but very little progress. Even with neural networks and deep learning, we are still in the stone age of artificial intelligence.

  • The human mind is thought to be the most complicated thing in the known universe. Neuroscience was only formed as a scientific topic of study around 1900. And we have barely scratched the surface of our understanding of the brain and mind. Many computer scientists think we don’t need to understand the human brain to build an AI, but the human brain is the only known example of human level intelligence, so what else would you base it off, a chicken’s brain?

  • There is no computational model for emotion. Emotion and homeostasis drive organisms to do certain behaviors and is deeply connected with intelligence. No computational models have been built that try to tie emotional systems with intelligence. The problem is thought to be too hard and not even attempted.
  • Computers are not good enough to build what is needed to simulate a mind. A CPU is faster than a brain to process a single bit of information, but a CPU only runs sequentially while the brain can run things concurrently. The architect of a CPU is thought to be not robust enough to emulate a brain. Some scientists believe analog computers or quantum computers are needed to be able to simulate brains.
  • A simulated body is needed to be able to implement a homeostasis model. This requires knowing what the right parts of a body to simulate are. We have no idea what the right areas we need to simulate are. Our knowledge of sensors in the body are very primitive. We know we have sensors for pressure, temperature, pain, orientation, hunger, pain, smell, touch, and more. Which of these is important to intelligence? All or just certain ones?

  • An environment is needed for intelligent agents to interact with. This means a physics engine is needed to simulate interactions between organisms and agents. Interactions such as gravity to give us feelings of weight, strain, “uphill battles” and pain. Liquid dynamics so we can feelings of “slippery situations”. Fine object collision and detections so we can simulate feelings of roadblocks and “rough situations”. Our physics engines are too primitive. They are also constrained by our CPUs being too slow.

Leave a comment

Your email address will not be published. Required fields are marked *