Evolutionary scientists are not very good at predicting the future. They are very good at reconstructing and explaining the past.
Reconstructing and explaining the past is not as easy as it seems. Evolution has resulted from natural selection—but natural selection has acted differently on every group of organisms, and in every part of the world, and at different rates. There have been countless examples of historical contingency, as Stephen Jay Gould liked to remind us. Genetic drift and the founder effect, among other things, produced unique and unusual sets of genes upon which natural selection acted. Natural selection can only act on whatever genes are available. The story of natural selection is therefore millions of stories.
This is why evolutionary scientists cannot predict the future very well. In the past, evolution has taken millions of directions—and it will presumably keep doing so. We cannot predict the conditions of the future, nor the contingencies (accidents) that may determine which populations will be at the right place at the right time to explode into a new and dominant evolutionary lineage. When scientists try to predict the future, they make it clear that they are using their imaginations, rather than the scientific process (for example, Future Evolution by Peter Ward and Alexis Rockman). Japanese film maker Akira Kurosawa (in his movie Dreams) imagined a post-nuclear-war world with giant dandelions and people who grew horns, which was a mixture of science fiction and traditional Japanese legends.
Just because there are a million directions evolution can take does not mean we cannot make reasonable guesses about some of them. I invite your comments about what you think some of the future directions of evolution might be.