Do simulated annealing and ant colony optimization count as machine learning?

yes, see my blog https://charlesmartin14.wordpress.com/2013/11/14/metric-learning-some-quantum-statistical-mechanics/ Simulated Annealing is a fundamental approach  to solving the kinds of non-convex optimization problems based on the energy models that arise in advanced machine learning methods such as Logistic Metric Embedding and Deep Learning It is particularly useful in discrete optimization problems, and/or in problems where the derivatives are hard to calculate.  For example, prior to the development of subgradient methods, Simulated Annealing  was a goto method for optimization of energy models with non-differentiable optimization functions (i.e. L1 regularization), although it has been  more commonly  used in statistical physics for problems such as simulations of hard spheres http://prb.aps.org/abstract/PRB/v44/i2/p477_1

I like both Charles Martin and Justin Rising answers and they say different things! I think this is about semantics and sometimes things are rather gray. I would say SA and ACO are optimization algorithms. Now you have to decide if optimization is inside machine learning or if its something different. Semantics...