The difference is that when there is an issue with a legal contract you can defer to an arbiter (a judge) and discuss whether that is a bug or a feature as soon as a divergence of interpretation is detected.
Hell, even just having a sentient empowered human in the loop is sufficient, with fully automated response systems we'd never have survived the cold war as a civilisation, the first false-positive detection (and there have been several) would have ended it.
The problem (and the thing you propose as a solution) is deferring to an arbiter. This is not human judgement as much as it is human authority. We conflate the two in meatspace because of the social rank conferred upon such positions (this is the same drive that makes humans bow before gods and dictators)
The very idea of decentralization is a different authority model than what is typically in human institutions. In theory, institutions managed in a decentralized and anonymous way have the potential to achieve a form of democratic governance that is far more resistant to corruption than any form previously invented.
In meatspace, a judge must be chosen, elected/appointed, confirmed, etc., and when that judge needs to be replaced we take an entirely different (largely unknown) judge and do a full-scale migration to that new judge's "firmware".
With smart contracts, we can divide the execution into many smaller smart-contracts, each with a specific domain of expertise. This makes versioning, incremental improvement, and extreme transparency possible.
If you listen to us SCOTUS argument, particularly when a case involves discussion of the intent of specific words in a law, it becomes clear how utterly mutable all of the constructs are. While the SCOTUS is a world-class institution in terms of its overall quality, it is reinforced using a highly centralized model of human authority, with all its problems.
Perhaps, but nobody can write a bug-free legal contract either, and no legal system is without bugs.
Sadly, for the legal system, many of the bugs are due to corruption, so they are actually more akin to systematic exploits being done again and again by malicious actors (who often happen to be wealthy or powerful).
Also, in terms of how drastic this bug is, suppose $5M gets stolen over a few days. If the community can strengthen itself and become resilient to a whole class of attacks, that is likely far superior to a meatspace improvement in contract law, which would likely take years to become law (and would be selectively enforced once it did).