Progress on AI since Turing
- It has been frequently claimed that AI will be solved in 10 years,
but it hasn't nor will it be.
- There has been progress in lots of areas.
- Expert Systems (rule based) have been a success.
- Robotics and machine vision are commonly used.
- Machine Translation works relatively well.
- Datamining works relatively well.
- However, anything that works is domain specific.
- The Turing test is a domain general one.
- People are domain general and robust.
- So my big research question is
How do we build a machine that can pass the Turing test?
- Assuming humans are doing computations,
- Turing completeness shows that it can be done in any programming
- So the question is not what mechanism can build such a machine,
but which is the best way.
- My hypothesis is that The best way is to build a machine
that follows human behaviour at neural and psychological levels