Editing Self improving programs (section)

= The dispute process =

To start with: the bounty contract already has crypto-asset locked inside of it to pay for solutions, but the same idea also needs to be applied for solutions so there is something to prevent spam.

The idea is that every solution will require a bond that is returned on success to its owner or given to someone else based on a dispute process. A solution has a period where it may be disputed by anyone else in the network where there are two main ways to dispute a solution:

# Prove that there is a known input and output for the original algorithm that doesn’t occur with the new solution (i.e. the new solution doesn’t implement the full functionality of the original algorithm.) OR
# Prove that the new solution is insecure.

As per point 2 - every single computable solution should also be a [http://roberts.pm/exploit_markets bonded exploit bounty] against itself, including solutions for other exploit bounties, such that disputes are recursive and settlement will eventually elapse with no one to claim a dispute. This process ripples up the chain, causing funds to be transferred based on a final success or failure.

For more complex cases it should be possible for a human intermediary to verify that a solution is invalid by committing to a larger deposit and asking for their intervention. If the humans convene that the solution is flawed the bond offered for the flawed solution will be distributed between the humans and the person claiming that its flawed. This is the last line of defence. The primary lines of defence can all be automated.

Now what about readability? Suppose we use this process to prove that the new algorithm is equivalent and secure. We can always use the original algorithm as documentation for the subsumed functionality of the new algorithm as if the improvement were its complication.

Improvements won’t necessarily come from humans in the future so the need for readability is reduced. If the routines need to be replaced in the future by a human the same process can be repeated to improve any new changes, thus the original code is still readable.

<span id="consequences"></span>