Howard Lovy has an interesting post on how misperceptions among the public have led to an unnecessary fear of molecular manufacturing. If Ralph Merkle is right (and he's a pretty smart guy, based on my own interactions with him), we can turn off the gray goo.

You can only turn it off if it's made that way. And you don't have to make it that way.

It's even worse than that Mitch. Even if it is made that way it is likely to have some onboard software (really tiny software :-) ) to interpret the broadcast along with some canned routines for doing it's job. All it takes at that point is a single bit error in the right place that is then replicated. So when the broadcast says stop! one of your millions of nanomachines makes a copy of himself, those two makes copies, those four make copies... goo.

The real saving grace is not the instruction source but the energy source, IMHO.

As I understood it, the broadcast was not a yes/no command, but the actual blueprints/software needed for a nanomachine to replicate itself. So if the onboard memory isn't sufficient to store the entire set of blueprints and the broadcast stops, so does replication. A single 'point mutation' should not be able to get around that. Which isn't to say it's foolproof, but not too bad.

While it's true that there are dangers, it's even more important not to listen to the Luddites who say let's call the whole thing off before we even get started.

It might be a good law to say nanites can't be manufactured to operate "too independantly", but that doesn't stop anyone from doing it on the sly. We're going to need fast-response teams to handle malicious (or just out-of-control) nanites in every city. But we already have police and fire departments, so this is just one more. The good news is the defender nanites generally have the edge as "the art of destruction is infinitely easier than that of creation." You can zap/slash/stomp a nanite faster than it can complete the longer and more complex task of building a twin to itself and uploading it's programming as well. And there will generally be MORE of them on hand than the attacker nanites, unless we're dealing with a whole ARMY of terrorists all in the same place (unlikely).

All going precisely according to the prognostications of Neal Stephenson, I see. The Toner Wars, they are a-comin'.

AndrewS, yes that is one design that would likely be safe, but Mitch's point is still valid. Broadcasting simple instructions makes sense because we already have the processing power and don't have to worry about how to fit it or parts of it into a nanomachine. However, when you think about the tasks involved with self replication these may not be conducive with a broadcast to uniform machines working in lockstep.

While conceptually simple to have a single type of machine that replicates itself, it is more likely to have different types of machines that perform different task. Even with a single type of machine, each may be performing different tasks at the same time. Regardless of whether a single type or multiple types, if we assume simultaneous different tasks it requires addressing the machines so they listen only to the the instructions they are to carry out and ignore other instructions meant for other machines. This in itself is a distinction implying various machines are different even if of the same design.

The first type of machine might only have a few instructions it is capable of following... simple things like move, turn, grab, release; for whatever action that implies.

But then you've got more complicated, higher level instructions... identify, position relative to, etc. for which on-board canned instructions start to make much more sense. The model here is more biological than mechanical.

So, if we develop a machine that mimics a biological process... will we also have to create an antibiotic (antimechanic?) or risk gray goo?

I'm no Luddite myself, I suspect we are not smart enough yet to produce gray goo... but we are certainly stupid enough to let it get out of hand.