Certainly and let me at the same time give you a lesson in the breakdown of definition of terms.
Novel situation - a situation previously not experienced e.g. tomorrow, nuclear war, getting knifed.
Deterministic model - Our choices and responses are based on instinct and previous experience, in fact any thing in our past or inherited past. If something is determining our response then it suggests memory or knowledge or indeed preprogramming
Feel free to square the circle of how past experience handles novelty, in view of saying novelty is rare being an attempt at shuffling.
Not sure why you think there is a problem here. Past performance is not a cast iron guide to the future, but learning systems use retained memory to help guide future decision making. The present moment always brings with it some degree of novelty but stored memory can be used to guide responses.
Right now you are having a novel experience; web browsers are capable of displaying 16,777,216 colours and I picked one at random for this text so in all likelihood you have never in your entire lifetime encountered this exact colour before. However it is similar to previous colours you have encountered and it does not present an insurmountable challenge to your visual system to process itAnd absolutely, yes, memory retention, or preprogramming, as you put it, is key to the success of any learning system. We make decisions in response to change and we retain trace memory of the outcome, be it good or bad, and these memories form the basis of emotions that guide similar decisions next time around. Memories, in a fundamental sense, are also intergenerational; when a baby duckling follows its mother into the water for the first time, it has never encountered wetness, or buoyancy, and yet it immediately knows what to do on hitting the water and this is because a form of memory is passed through inheritance in that the duckling's neurological pathways for processing the experience of paddling the water are already built from birth so it will seem to the duckling as if it already knew what to do with water all along.
We also often make the sloppy analogy of the brain being like a digital computer whereas it bears significant similarities to the principles of analogue computing. A synapse is not equivalent to a digital bit, either on or off, as synapses have a gradient, or variable strength, and this allows for fuzzy comparisons and intuitive solving of differential equations - when you race to return a serve in tennis for instance you are solving the equations of motion on the run by analogue methods which are far faster than digital methods. Nature has already solved many of the computational problems that we are now trying to replicate in machine learning systems and I don't see that a deterministic substrate is a problem to this in principle, in fact, any indeterminacy, or randomness in the workings of logic would weaken the effectiveness of learning systems.
