Moshe Vardi's CACM editor letter The Moral Hazard of Complexity-Theoretic Assumptions practically begs a response from this blog. I also encourage you to read the discussion between Moshe and Piotr Indyk and a twitter discussion between Moshe and Ryan Williams.
I take issue mostly with the title of Vardi's letter. Unfortunately we don't have the tools to prove strong unconditional lower bounds on solving problems. In computational complexity we rely on hardness assumptions, like P ≠ NP, to show that various problems are difficult to solve. Some of these assumptions, like the strong exponential-time hypothesis, the Unique Games Conjecture, the circuits lower bounds needed for full derandomization, are quite strong and we can't be completely confident that these assumptions are true. Nevertheless computer scientists will not likely disprove these assumptions in the near future so they do point to the extreme hardness of solving problems like getting a better than quadratic upper bound for edit distance or a better than 2-ε approximation for vertex cover.
If you read Vardi's letter, he doesn't disagree with the above paragraph. His piece focuses instead on the press that oversells theory results, claiming the efficiency of Babai's new graph isomorphism algorithm or the impossibility of improving edit distance. A science writer friend once told me that scientists always want an article to be fully and technically correct, but he doesn't write for scientists, he writes for the readers who want to be excited by science. Scientists rarely mislead the press, and we shouldn't, but do we really want to force science writers to downplay the story? These stories might not be completely accurate but if we can get the public interested in theoretical computer science we all win. To paraphrase Oscar Wilde, as I tweeted, the only thing worse than the press talking about theoretical computer science results is the press not talking about theoretical computer science results.