In typical situations incremental innovations of a given technology or organizational practice will lead to exponential performance increase. That also means that a local optimum will be reached at exponentially fast times. In artificial intelligence applications those optima can be found with relative ease e.g. by traditional neural net algorithms. In the context of genetic algorithms the system can improve beyond local optima by using crossover methods: If two methods are independently successful that some form of a (non-linear) combination of both methods is even more successful. Examples of innovations that are based on combining separate ideas are plentiful. For instance since medieval Europe it was a custom that craftsmen learned with a master until they knew basically all he could teach them (exponential phase). Then they had to go on travel and explore the best practices in other regions (crossover phase). Very rarely some fundamentally new innovation (such as the printing press) occurred (mutation phase). In programming genetic algorithms one important parameter is the ratio between crossover and mutation rates. In some areas with a high rate of innovative changes (such as the area of modern computers and related areas) it might be profitable for a company to hire a group of creative individuals that just spend their time playing with off-the-wall ideas.
Another innovation enhancing strategy involved the interaction of experts from different fields. These interdisciplinary approaches were systematically encouraged in research centers of non-linear dynamics and complex systems such as the Santa Fe Institute. Someone who naively looks at a problem in an area far from her expertise often can suggest solutions that have been given up by the experts in the field long time ago as being completely unfeasible. What experts sometimes overlook is the fact that the changing technological and other conditions often make it worth while to reconsider a solution approach in the new situation. Environmental conditions and their rate of change will determine in a more practical environment how many resources can be invested in that kind of crossover and mutation activities.