Designing a rudimentary general AI

I like to think of the human brain as a heterogeneous collection of neural nets dedicated to different tasks. Each net is specialized to do one thing well, and they can be improved and repurposed over time. They are capable of influencing each others’ behavior to lesser or greater extent by “piping” data to each other, and they are also influenced by chemicals, moods, rhythm, heartbeat, nerve signals from the rest of the body, it’s all jumbled together in a chaotic mess because evolution. An AI designer should be aware of how the human mind works and try to reproduce the most important traits with a much cleaner and simpler design.

To design a conscious AI we need to solve the various sub-problems. Classifying visual information and pattern recognition is something neural nets are getting good at. Storing and retrieving data is something computers can be programmed to do very well. All our sensory input is unstructured data. Our brains structure it in a way it understands. An AI also needs to structure input in a way it understands. Then that processed, structured data must be fed into a higher level of cognition that looks for connections, assesses significance and attaches emotions to the data. Discard or use? Interpret data. What does it signify? Does it change the AI’s world view in a meaningful way? Do such changes to the world view trigger changes in behavior? What are the logical implications of the data in relation to connected data points? How does it relate to the AI’s goal pool? How does the AI determine what its goals are? It needs the ability to run simulations on real and hypothetical data to predict consequences. The Tesla autopilot seems to be pretty good at some of this stuff. Other AIs have solved other tasks.

Another angle to consider is how neural nets can help the human mind function better. We already offload many cognitive tasks to our electronic devices. As AI becomes more powerful and brain-computer interfaces become better, we should also gain better insight into our own minds.