What is the difference between Artificial Intelligence and Computational Intelligence?
The short answer is that they are two parallel research efforts working on similar problems, but with different methodologies and histories. Essentially, they study similar things, but with different tools. In the modern context, computational intelligence tends to use bio-inspired computing, like evolutionary and genetic algorithms. AI tends to prefer techniques with stronger theoretical guarantees, and still has a significant community focused on purely deductive reasoning. The main area of overlap is in machine learning, especially neural networks.
The longer answer is that your source from 1948 says they are synonyms in part because it predates the split in the research community, which took place later.
The two communities have always some overlap in topics, but in my experience, mostly are skeptical of each other's methodologies, and mostly publish in separate journals. Some authors consider CI to be a subset of AI however, particularly those writing in the 1990s.
Example topics that are solidly in AI but definitely not in CI are logical and expert systems, and statistical approaches to machine learning like regression.
Example topics that are solidly in CI but perhaps not in AI (depending on whether one views CI as a subset of AI or not) are genetic programming, fuzzy logic, and ant colony optimization.
As a rule, AI-rooted techniques have better theoretical guarantees, and better developed theory in general (there are exceptions though). For example, Fuzzy Logic has been strongly criticized for the lack of a solid theoretical foundation (good modern summary here), as have genetic and evolutionary approaches (most famously, both lack a proof of convergence within finite time to a global optimum on a smooth surface, even though they do quite well in practice).
CI-rooted techniques nonetheless often see major performance advantages in specific problems (see, for instance, deep learning results), and tend to have a strong experimental and engineering tradition. The No Free Lunch theorems are often used to justify their use when theoretical certainty is missing. Basically, the theorems say that, in learning and optimization problems, a technique can only perform well on a problem by performing poorly on some other problem. CI authors argue that there are some problem domains in which their techniques work well (which must be true, because simpler algorithms like hill-climbing outperform them on simple problems).
Check out this paper for lots more references on CI, or this book for a list of core topics in the field.
