I guess it depends on your definition. I can think of many examples that are self replicating in some sense, but probably not in the same way as a bacterium.

A 19th century machine shop is probably capable of self-replicating, with metal as an input and a human 'catalyst'.

A crystal is self replicating in a way.

A computer virus and a meme.

This was published in 2004 and was highly speculative at the time. We are still a long, long way from the Diamond Age fantasy of self assembling nanotech.

Recent work in this area seems to mostly involve modding 3D printers to try and integrate conductive materials (basic circuits) into plastic structures.

Example:

https://3dprintingindustry.com/news/botfactory-to-develop-u-...

What do you define as an "artificial physical self-replicating machine"?

A genetically engineered bacterium is a self-replicating machine created via artificial means that exists to effect some purpose in the physical world.

No, it's never been done.* While factories are built all the time, they have never been built entirely from their own products and without human labor as an essential ingredient (a "closed" manufacturing process).

There's an economic reason for this: the constraints that make a factory "self-replicating" mostly make things more difficult/expensive; they're only beneficial when operating in a human-hostile area disconnected from existing supply chains, so it's always been better to use open-loop factories + human labor to make products quicker, cheaper, and of higher quality than a self-replicator would. Space industrialization is where self-replicators will be exceptionally useful.

These are the key functions of a truly closed-loop self-replicating factory:

1. collect energy

2. use energy to collect matter.

3. use energy and matter to create parts.

4. use energy to assemble parts.

For example, consider a water-powered saw mill, made of wood with some metal "vitamins":

1. Wooden water wheels collect energy from flowing rivers.

2. Humans use wood/metal saws to cut trees, and wooden rollers/sleds and transport them to the mill.

3. The mill's saws, drills, and lathes turn the tree trunks into useful wooden parts.

4. Humans move wood parts from station to station in the mill and can assemble another mill from the boards, dowels, and a few metal saw blade/drill bit "vitamins."

Could you replace the humans with logging robots, and use robot arms to move things inside the mill? Sure, but humans are so much better suited for these tasks, and now your humble saw mill has to be able to create, power, and control these robots in order to be closed-loop.

*Yes clever HN, you can get all pedantic about biological self-replicators, self-assemblers, and nano-bots, but you're just muddying the waters. We haven't ever managed a von Neumann-style "clanking automaton" self-replicator.

P.S. don't worry about rampancy—there are physical limits that make sci-fi's Grey Goo floods impossible.

Well, there are degrees of "artificial".

Years ago, we spliced an insulin producing gene into a bacteria's DNA:

https://www.nlm.nih.gov/exhibition/fromdnatobeer/exhibition-...

This is taking a naturally occurring bacteria, and modifying it slightly. We are using the bacteria's existing protein production infrastructure to produce a peptide hormone.

Check out how a ribosome works:

https://www.youtube.com/watch?v=morl5e-jBNk

It is, in some sense, a programmable robot arm that assembles structures from parts.

As we have gotten better at understanding how these mechanisms work, we can tweak things further. We can imagine many directions for this to go in. Not just producing proteins with existing amino acids, but having the bacteria produce new ones with interesting properties.

We could eventually be able to produce a wide array of chemicals, and not just organic compounds. Though the range of things that can be produced is still limited.

The main thrust of molecular nanotechnology is to leave behind the protein synthesis that biological life uses, and instead be able to produce arbitrary structures from arbitrary atoms.

This is a difficult and ambitious goal, but one that is imminently achievable.

I didn’t see any references in this thread to Von Neumann replicator cellular automata, so in this particular sense the idea has been implemented.

https://en.wikipedia.org/wiki/Von_Neumann_universal_construc...

I have three of them at home. They're easy to make but relatively high maintenance. Because of this, I try to make sure that they learn self-maintenance before self-replication. Let's see how that plays out.

Not until we've automated every single step of the manufacturing chain.