This is one of the cases where we don't need to push the limits. For one thing, the definition of intelligence is a really difficult moving target. For mid-range intelligences, we're pretty comfortable. However, when we start asking questions like "which is more intelligent: Stephen Hawking or Deep Mind, the computer that recently beat humanity at Go?" the definitions get tricky. If you're not careful, you'll custom tailor your ants into a corner trying to make them too intelligent, and they wont feel realistic.
So let's work with some knobs that let us play around with intelligence, without forcing ourselves into defining it too strongly. One great tool to consider is the Sentience Quotient, or SQ. SQ is the logarithm of the ratio of the information processing rate of a creature divided by the mass of its brain. If we measure processing in bits/s and mass in kg, we get a number between -70 and +50. -70 is the slowest you can get, processing 1 bit in the lifespan of the universe, using all the mass in the universe. +50 is a quantum limit based on how fast information can move. In practical settings, we find:
- Plants cluster around -2
- Carnivorous plants score around +1
- The aged Cray-1 is about a +9, while Watson, the new supercomputer, is around a +11 or +12
- Humans score around a +13
- Most animals which use neurons cluster around this (which makes sense: they're using the same basic hardware)
Its well accepted that SQ is not really a very good measure of Sentience nor Intelligence, but it has one great advantage: it is an easily measured quantity. It points out that, because your ants are using neurons, their general ability to process data will be about the same as any other animal. Brain size would be key. If your ant is the size of a Chihuahua, it is highly unlikely it will be any smarter than a Chihuahua.
However, that's not the end of the story. What makes ants interesting is not their body, but their society. They are social creatures which can solve remarkable problems. Next time you find a column of ants bringing back food, break it, put something in the way, and watch how they operate. In a remarkably short period of time, they will find a workaround and the column will operate as fast as ever. Now go remove that blockage. Frighteningly fast, they will realize the blockage is gone and resume the previously ideal path.
No one ant is smart enough to solve this kind of problem. However, ant colonies are so unbelievably good at these sorts of path optimizations that modern computer scientists literally have optimization algorithms written to emulate ants. You'll set a start, a goal, and you will have your simulated ants deposit pheromones as they path-find around your state space, just like real ants do. We, with all our intelligence, tell our computers to go act like ants because they do better at these problems than any algorithm we've thought up!
A hive mind solution for your ants is much more believable, and much more interesting. Suddenly you have much more brain mass to work with. The limit is no longer how much brain mass you have, but rather how well you can make all of those brains work together. If the ants develop a communal consciousness, it may be far more advanced than ours.
There is actually an measurable definition of consciousness out there which is applicable to these circumstances. The Integrated Information Theory of Consciousness is one model we have developed in an attempt to understand what consciousness is. In that model, consciousness is defined by the information stored in the system as a whole which is not immediately evident in any individual piece. This model has been used to explore the idea of groups of individuals having a collective consciousness. You could use such a model to explore the consciousness of your ant colonies.
I cannot actually answer the question "What would I have To change in order to make them sapient?" because we simply don't know enough about sapience to make such an answer. However, you may be able to use these numeric tools to explore your ants and see how smart they might be.