Some things to try with brain farms...

1. Connect the brains to virtual worlds and try to get them to learn, grow, and become intelligent. They may need participation of grown people in the virtual worlds to teach them.

2. Speed up the physics in the virtual world to see how this affects intelligence.

3. Try to make a brain intelligent enough in a virtual world to educate other new brains that are connected to the virtual world.

4. Put a new brain that is to be educated in a virtual world in a virtual world that is embedded in the virtual world of a brain that was educated in that virtual world. Increase the simulation speed (and/or vary other parameters) in each level of the hierarchy of virtual worlds. Oversight by brains whose access spans multiple levels of virtual worlds might be needed.

5. Connect new brains in different ways at different times in their development. For example, connect the output of one brain's visual cortex past the basic image processing functions to the input of the visual cortex of a new brain so that higher levels of abstaction may be formed. This could also be done in the code of a virtual world. Or, try to connect multiple brains to one agent in a virtual world.

6. Vary the physics of the virtual worlds. For example, make random things less random, make everything squishier, make physics non-local, or omit conservation of mass and energy.

7. Vary the basis of "reward". Make the reward mechanism be based on arrangement of symbols that are valid syntax of a programming language with increasing rewards for code that calculates elementary mathematical functions, generates random numbers, or has an output of high Kolmogorov complexity.