The strongest arguments against dynamic blocksizes which can be determined by miner actions, is that they do not necessarily represent the interests of other participants such as users, node operators, or even all miners.
If miners setting the blocksize have good equipment and connections, they may want blocks so large that others are forced off the network. This could include large miners (with the best equipment and network connectivity) forcing smaller miners off the network. Alternately, (some) miners may want a small blocksize while others would prefer a larger one, but be unable to exert any real influence.
Other issues may include concerns that a specific algorithm could grow too fast, too slow, down not allow for downward scaling if usage decreases, or could be attacked.
In summary there is significant:
- Difficulty in designing an adaptive blocksize formula that cannot be
"gamed"
- Difficulty in creating the best model to minimize cost of
node-option (CONOP)
- Difficulty in designing a system that can
efficiently scale both up and down
- Time and expense to compare
alternative formulas and thoroughly testing them
Finding the optimal CONOP model that can satisfy the above concerns will be difficult and time consuming. Until then blocksize limits previously set by humans (and possibly adjusted by humans to allow for more time to study/research codified adaptive blocksize solutions) is the best option available.
CONOP was introduced in this article by Paul Sztorc:
http://www.truthcoin.info/blog/measuring-decentralization/
The concept in the context of adaptive blocksizes was also recently discussed at the Bitcoin On Chain Scaling conference by Riccardo Spagni:
https://www.youtube.com/watch?v=mM2ra-LzMQk