Machine learning and data science are mainly made for processing large amounts of data nowadays, for example - a multitude of pictures.

But do these fields have some applications in the decision making?

I mean - do at least some of the companies make a decision making systems as a part of their products? And do they hire DMS specialists?

Is there any difference between DMS and "regular" DS?


The extraction of features from data and categorizing media are not business decision making. However, determining what patterns in real world information are distinctive and placing an item in one bin rather than another are of a simple kind of decision. That large amounts of data are required to perform these tasks does not discount these limited decision making abilities emerging either.

Consider that human brains can make sense of sparse data only because of a rather deep network the topology of which evolved for hundreds of thousands of years, long before higher cognitive capabilities emerged. Also consider that the volume of information processed during the development of an infant and the education of a small child is an essential factor in an older child's ability to begin making reasonable decisions with only sparse information.

In contrast, business decisions show capabilities we do not see in computer systems as of this writing. The ability to decide whether to shelve a product or invest into its further development to improve its market position draws on more than feature extraction and categorization. Decisions made by professional decision makers rely on much more than one dimensional thought.

A single decision may rely on a hundred mental models. These are just a few categories of models with many individual models contained in each.

  • market needs
  • market trends
  • general finance
  • specific budgets in the company
  • understanding of what the product is
  • imagining what it could be
  • comprehension of manufacturing
  • relationship between materials and product

In senior level decision making, the information that had been absorbed during the development of business acumen is critical to making choices. This development may resemble training in some ways, but there is one major distinction, which production system (rules based AI) researchers tried to address in the 1990s. The learning is self-directed.

This is distinct from unsupervised machine learning, where the decision to run the process on some particular data is still made by the people running the computer system. Business people go out and find challenges and address them through acts of the will autonomously. Nothing even close to that kind of self-directed, high level functioning is present in today's best computer systems. Some might argue that what the search engine vendors have accomplished is a step in that direction, however there are several arguments against that claim that are out of scope for this question.

In summary, the current machine learning capabilities fall quite short of handling any one of the above models well individually. The human ability to combine the models in a cognitive milieu, pull sensible options from a storehouse of many models and experiences that fall within them, and then analyze those options for the best choice is not something current technology even approaches.

Data science does, however fall well within the tool set used by many of the decision makers running multinational corporations and more locally successful mid-sized businesses. Manufacturing has used TQM and other approaches to statistical control for decades to reduce defects in product and lower production costs. 6-Sigma has been used in corporate decision making for a few decades too.

There is quite a bit of overlap between decision making and mechanics of data processing, employing elements learned in high school, such as probability, statistics, the scientific method, economics, and algebra. Add chaos theory, calculus, and network/graph theory, and we have a tool set for making computer systems that make decisions.

As the limitations of those computer systems are overcome, slowly, over time, the gaps between decision theory, data science, machine learning, cognition, natural language, and other aspects of artificial intelligence are likely to narrow.

Outside of corporate management and data center work, the trivial robotics of the early 20th century was outdone by the more sophisticated robotics of later 20th century for manufacturing and military applications. What is appearing in the 21st century is consumer robotics, mostly because of cost reduction. These robots that may be vacuuming our living room, manicuring our yards, and driving our cars will be making decisions.

This, as much as in any other sector of AI related product, can be hugely profitable, and that is where more capable decision making in computer systems is most likely developed, at least in an arena where the achievements are not either classified or company confidential.

| improve this answer | |

A possible application of Decision making systems are interactive computing, knowledge based tutorials and game-based learning. For example, it is possible to model the workflow in a hospital. The expert system has to support the decisions of the human operator. Before the software can explain “good decision”, the domain specific knowledge of the hospital example has to be modeled first. That means, Decision making systems are the result of ontology models created before.

The application are twofold. At first, it is possible to run the software in standalone mode and leave the human operator out of the loop (fully autonomous mode), or secondly, the software is used for training human personal to become familiar with their job. In any serious game such a submodule is implemented together with a multiagent architecture.

| improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.