How fast can directed evolution make the leap from unicellular to fishlike ability level?

Question

Premise target:

A few thousands of intelligent machine-scientists have been tasked on the creation of an organism with the abilities of an animal. The exact requirements that they are given are:

• Can move itself away from harm and towards what it considers good. It should be able to move away from being repeatedly bothered by something before 10 seconds, for example.
• Can reproduce if not under too harsh conditions. Good conditions is earth-like, either water, land or mixed, temperate climate, up to the answer requirements.
• Can autonomously/gregariously do what it need to do to feed. Whether it's hunting, absorbing nutrients, etc.
• Can adapt to environment changes on geological timescales.
• An individual should have the potential to learn based on some sort of memory and adaptable interpretation of stimuli/action/consequence loop. If should be able to remember at short and medium term that there is something bothering it in a general location.
• Its macroscopic: Could be clearly visualized if taken a photo of it with a mundane camera from a fair distance. About ten centimeters is okay.
• It does not explicitly need to be multicellular, an animal, or even reproduce sexually! If you reach the previous described characteristics, it's as good as any.

I figured out the simplest organism would be would be some fish-like animal, but you can end up with something else.

Premise scientists:

• They know way more than us in the fields of inorganic chemistry, physics, math and computation.
• They posses no previous knowledge about organic chemistry, biochemistry or biology. If the answer includes those sciences, it will need to include the time necessary to discover them to the desired depth (the necessary depth can vary depending on the method chosen to reach the goal).
• They can work on it until the project is done, however long it takes.
• Other than that, treat them as a few thousands of scientists with the ability to perfectly share information between the members and without other academic duties.
• The scientists have only access to a single type of creature, an DNA based waterborne photosynthetic unicellular organism (ABWPUO). Don't limit yourself to a Microalgae if it better suits your answer, but is must start as unicellular, waterborne and photosynthetizer.
• They have never seen any other form of organic life. They just know that an animal can be "evolved" from it. This means that they do not know how internal organs work, how creatures metabolize or how tissues mature from embryonic cells. This is a huge factor in the time it takes if the answer focuses on designing a creature insted of evolving. Maybe alternating evolution and design will speed things up a lot.
• They can study, breed and toy to their hearts contempt with the ABWPUO both in laboratory setting and in its natural habitat.
• They have at their disposition the influence, natural resources and manpower of a country the size of Japan, which they can use at will, for example to set up laboratories or alter the natural habitat of the ABWPUO. (So they could build a dam and enclose whole sections of the ocean under their sovereignty to experiment, but could not, for example, alter the atmosphere of the whole planet)

Question:

Approximately how long will it take the scientists the creation of the creature described? What magnitude of time are we talking about? A hundred years? A million?
I know it's an estimate, but I'd like to have a well founded estimate, and they think they can do better than just waiting 3500 million years.

I'll accept the shortest believable time-frame that does not cheat the whole thing. Outside influences to the described situation are not allowed.

Answer 1

100 years by brute force. They would need to learn about biochemistry and genetics to about our level, then turn their minds to achieving a creature that fits your requirements. Note that as in any requirements-bound challenge, the solution will tend to 'game' the system. With your current setup (learning, but no requirement for fast neuronal-like learning), and my current knowledge, i'd go for a hollow-sphere-type creature composed of copies of one celltype, with flagella for movement (very slow), relying on cell -death and -spawn for adaptable behaviour (slow).

Guided Evolution, aka breeding, would take much, much longer. The known examples of breeding show that variations of a species can be easily achieved within a few generations - this takes advantage of the already present variations in the genome, though, and cannot be easily used to break ground towards completely new features. The added speed of higher mutation rates (see Atomic Gardens, where radiactivity was used to increase mutation rate, and the resulting individuals where then evaluated for value would not come into play as much, because a lot of the modifications needed for multicellular life (as in tissue-based creatures, not just the 'cheat' i described above) need other mods to be effective, so on their own would not seem like a step towards the stated goal but rather towards something else, and thus would get weeded out by the scientists (that mod1 plus mod2 then work into the direction of the stated goal cannot be foreseen without very detailed knowledge of genetics).