To those who haven't researched what's behind modern machine learning and artificial intelligence work, all of this effort and research often looks like one big amorphous jumble. However, when you scratch the surface and look at what scientific leaders are doing in these fields, you see that in a way, there are really five different major approaches to the issue of pushing artificial intelligence forward.
These five "schools" or "tribes" have been popularized by the work of Pedro Domingos in his "Master Algorithm" book on AI development, but they are also under consideration elsewhere in various parts of the scientific world.
|Free Download: Machine Learning and Why It Matters|
The first school of artificial intelligence is called connectionism. This school focuses on the actual neural connections and the physics of the human brain. It relies on the idea of backpropagation, which traces these connections to form results. Some people call the connectionist school an "effort to reverse engineer the human brain."
The next school of artificial intelligence is symbolism. Symbolists use logic and pre-existing knowledge to build models that work intelligently. In some ways, the symbolist approach is similar to what emerged early on in the artificial intelligence world before neural networks were developed. If you compile a big enough knowledge base and deal with it in particular ways, it starts to create a form of artificial intelligence, and that's what's behind the symbolist approach which has now been combined with some of the other modern approaches.
The third school is the school of evolutionism. Here, there's a focus on not only evolution theory, but also on genetics and biophysics as well as bioinformatics. You could see this arm of artificial intelligence as the category that works with the human genome and applies modern technologies to the field of genetics. In that sense, evolutionist artificial intelligence is unique. It's a somewhat different kind of project than the other four schools.
The Bayesian school is the fourth school of artificial intelligence. This is, again, one of the older schools and was applied early on, for example, in the elimination of spam from email folders.
The Bayesian model and approach is a heuristic model. It works on the idea of probability to evolve models that will cut out undesirable results, or pursue other objectives, based on where events are most likely to happen, or on other metrics. Another popular application of Bayesian logic is in network security – over the past few years, security engineers have widely used Bayesian logic to spot threats to a network by modeling where those are likely to occur, and how.
The fifth and last school of machine learning is called analogizing. This is also a school that's perhaps more easy for the average consumer to understand. Recommendation engines from companies like Facebook and Google are based on an analogizing approach. They take algorithms like "nearest neighbor" and combine them with various types of signaling to try to match ideas to other ideas, or alternately, to people. A computer that claims to know what kind of music you like is a good example of this approach.
All of these schools of thought combine to form the body of research on modern artificial intelligence. Scientists are working to push each of these forward in conjunction with one another, and generally advance the field – and they're trying to do so in a very interesting context. Some of the top leaders of technology in the past few decades have warned that in addition to pushing AI forward, there must be a focus on ethics and responsible use of technology in order to prevent serious social problems. That has to be applied to each of these five schools of machine learning.