This sounds like a bunch of boring truisms when stated this way, but perhaps a little extra insight can be wrung out by binding this to some more specific claims.

There's a corner of the cognitive science literature about iterated learning in a Bayesian framing. A question it asked was roughly, "If we partition a dataset over a sequence of agents who each see only a slice, and we allow each agent to communicate something to the next agent down the line, what must be true of the communication between those agents for them to learn the same thing as a single agent who saw all of the data?" And roughly, for the sequence to converge to the same posterior as a single agent, each agent had to adopt an unbiased estimate of the previous agent's posterior as their prior. See esp work from Griffiths and colleagues.

In a somewhat looser analogy, there's a family of techniques for distributed convex optimization (like ADMM), where a group of agents are collectively trying to optimize a function over data, and each agent sees only a slice of that data. Note especially that the way the data is partitioned can be entirely arbitrary, including e.g. training an SVM classifier where some agents see only positive data points. And for the group of agents to collectively converge to the global optimum, each local agent basically adds a large term to the function which penalizes disagreement with the other agents.

In slightly hand-wavey terms, in either case, for the agents to reach a "correct" conclusion as a group, each needs to be able to give potentially an arbitrarily high weight to the information distilled from others. In the iterated learning case, the posterior distribution received from a previous agent may have more information than the likelihood distribution of the data that an individual agent sees. In the optimization case, the consensus term may contribute more to the local loss than the data that an individual agent sees.

So in contexts where one is actually trying to solve the _same problem_ as a group, placing very high weight on the honest information received from previous experts can be critical.

But in cases when you're _not_ all working on the same problem, it can be disastrous.