+1000 to Markus K's main arguments above. Yes noise will be introduced as different people come in to make edits. Administrators locking them out isn't the best way to solve that problem in an open system. There are other options, as he raised - both technical and social.
Our group maintains hundreds of thousands of items on Wikidata with bots that import data about genes, drugs, and diseases from a large variety of 'trusted authorities'. We have also experienced some frustrations when users make changes that don't fit our views (e.g. there was a thread here about people merging gene and protein items that caused some major headaches for us). But, we resolved it by engaging with the people doing the problematic edits via the talk pages and by adapting our bot code in various ways - both by automatically fixing things that we judged to be clearly broken (producing edits that could be tracked and argued with), by educating community members about why things were modeled the way they were, and by fixing processes that led to the confusion. We could have argued that because of our PhDs or whatever, we should 'own' these claims that are "clearly facts...", but that would have been a huge mistake.
Looking ahead, we see that there will be an increasing challenge to keep things in order as more and more people get involved. Figuring out this process will entail a lot of work for us - much more than if the 3 _total_ people that are writing the gene, disease, drug bots currently were left to their own devices to create our own little data warehouse. We are here because we want to be a small part of something much more profound than that - that will take not 3 people in a closed room, but 30,000 people collaborating and fighting it out together. We are a long way from 30,000 here (at least in our part of this). Lets keep the gates as open as possible.
-Ben
p.s. In terms of 'making our data more reliable for re-use'. The formula for me here is (1) get references on statements (2) develop code that automates the processes of checking them, (3) provide the re-user with straightforward ways to filter the data based on the combination of 1 and 2. This can even be accomplished directly inside of Lua code that builds Wikipedia templates from Wikidata - we have already started prototyping that.