OK, hang on. If we're going to go here then let's get this right.
<gets on soapbox>
"Software Defined" anything as it relates to networking is the concept of separating the control plane and the data plane functions. This almost always takes the form of there being a "controller" that oversees all of the network, and then a number of "forwarders" that take their programming from the controller and do the actual forwarding of network traffic. Technically speaking, there's an API that runs between the controller and forwarders that's referred to as the Southbound API.
Now, if you apply this concept to WANs you open up some interesting possibilities. By having a centralized controller that maintains a holistic view of the network, while being fed real-time statistical data about the state and performance of multiple WAN connections you create an environment where forwarding decisions can be made dynamically based on the current conditions of the network. Dynamic policy based routing if you will.
Add to this a little bit of probability theory applied to the performance of multiple Internet services compared to a single dedicated private circuit and now you end up with a system that can statistically provide very high levels of service over cheaper links, comparable, and even exceeding in some cases, what a dedicated private link can provide.
But to be clear, SD-WAN isn't necessarily an MPLS killer, quite the opposite in fact. MPLS and Internet can co-exist in an SD-WAN deployment very well, each providing a different level or type of service that compliments the other.
So, to reign all this back in, SD-WAN is the idea of applying SDN concepts to the WAN. It has nothing to do with the number of Internet services connected, however, the more links you have the higher the chances are you have a link that is performing well enough to meet even the most stringent SLAs, and as those links change so too will the routing decisions being made by the forwarder.
</gets on soapbox>