We rely on experimentation to be non-lethal. As kids, we crawl around touching and tasting everything; if we find something sweet and poisonous, we end up like cats on antifreeze.

Life at the edge, experimenting with new things has a lot of appeal to people, in addition to playing with new things that nobody else has, you also generally get paid more because coming up with better technology has historically proved decisive between evenly matched opponents. The Economist recently harped on the negative side of this by pointing out that:

They [regulators] are paid less than those they oversee. They know less, they may be less able, they think like the financial herd, and they are shackled by politics.
-- Economist "Fixing Finance" 2008-04-05
Financial innovators make big money, and we as a society rely on their innovations to be non-lethal when adopted en masse.

Responding to Jeff Sachs on biodiversity, Tyler Cowen assumes that experimental changes over time have more benefit than the occasional mishap that can occur. Sachs implicitly argues that we have incorrectly or improperly specified ecological models (unexpected things happen), so we run the risk of creating ecological recessions (toads and rabbits in australia). However, here we run into an experimenter's dilemma: how do you safely test your model when this combination hasn't ever happened before in recorded history?

As technology develops, we also increase our ability to accidentally (or purposefully) destroy life as we know it. We need better insulation between our experiments and the global world, just as much as we need better understanding of how things work. In the worst case, we fail to pass Hanson's Great Filter. In the best case, we settle multiple planets before we lose one.