If mathematical algorithms had been promoted as a research topic in the Victorian and first half of the 20th century. Problems like the Travelling Salesman, that could be explored in the context of decision making theory. While neural nets and other forms of genetic algorithm may need to wait until the computer age to allow for the computation to develop real 'self learning systems'. Certainty the classes of problems could have been arguably been explored in greater detail earlier, and certainty emergent behaviors in complex systems could have been achieved using analogue/mechanical computing that was around at the time - what was lacking was the conceptual framework as well as an actual use for the output of such computation in the world at the time.
I would somewhat argue that the modern development, has been close to the optimal path as it could reasonably have been without leaps of logic and funding pushed by visionaries for more than just research purposes.