The Tierra instruction set contains of 32 instructions. Any evolved Tierra ‘organism’ of length n will thus contain nlog(2)32 = 5n bits of information, so an ‘organism’ of length 100 will contain 500 bits of information. It’s not hard to adjust the Tierra settings* so that ‘organisms’ of that length evolve, including organisms with no redundant instructions (so they’re all functional) - especially if you start with a seed organism that uses a less compact reproduction algorithm* than the default one.
The ‘organisms’ generated by Dave Thomas’s Steiner tree GA consist of 4 3-digit numbers and 45 Boolean values, which works out at just under 85 bits of information - all of it functional. But that’s for a Steiner tree for ten nodes (six fixed, four variable) - it’s easy to modify the GA* for more nodes, and the 500-bit threshold would be exceeded by setting the GA to look for a Steiner tree connecting 32 nodes.
Similar to finding Steiner trees is the travelling salesman problem. This has been addressed by many different GAs, including at least one interactive web-based one that I can’t find at the moment. There are more than 10^130 possible traverses for a 100-city travelling salesman problem, so specifying any individual path takes more than 500 bits of information, and any path evolved by such a GA would likewise have more than 500 bits of non-redundant information.
Here is a report on successfully using a GA to create an algorithm to play checkers/draughts. It uses a neural network of 50 internal nodes and more than 1700 weighting values - so each possible network contains well in excess of 500 bits of information. Determining what proportion of that information is actually necessary (functional) is probably impossible, but even one bit per connection (e.g. on/off) would exceed 500 bits.
The GA I posted yesterday that finds routes from a starting point to a distant location represents routes by a series of angles representing directions to take. They’re specified in degrees, so have 6+ bits per angle. The version I posted generates paths that are typically 10 steps long, which is only about 60 bits per ‘organism’, but it can easily be updated* to generate paths that are 80+ steps long,** which require more than 500 bits - and they’re all functional, since without at least that number of steps the paths wouldn’t get close to the destination.
None of the ‘organisms’ produced by these GAs are designed, and none of them are biological. All of them contain more than 500 bits of information, and most wouldn’t work with less than that. In every case the search space exceeds 10^130 possibilities, and the target space is many orders of magnitude smaller.
I submit that these GA-generated ‘organisms’ are all non-biological, non-designed objects that exhibit high FI.
*I’ve done these things.
**It actually generated paths more than 200 steps long, which required considerably more than 500 bits to represent each path, but I’d hesitate to say it was all functional.