Machine reasoning is the concept of giving machines the power to make connections between facts, observations, and the different things machines can be trained to do with machine learning. This is done through the machine applying logical techniques like deduction and induction. Early in the development of artificial intelligence, researchers realized that for machines to navigate the real world, they would have to gain an understanding of how the world works and how various different things are related to each other. Machine reasoning is the enhanced form of AI to perform the logical reasoning that leads the machine through the steps of a complex process as it adapts to real-time change and execute functions.
Computer scientist and futurist Jerry Kaplan describes a reasoning system as a "concept that deconstructs tasks requiring expertise." The deconstruction is described as reducing a task into two components:
- A knowledge base—which acts as a collection of facts, rules, and relationships about a specific domain of interest, represented in a symbolic form a computer can understand, and
- A general purpose inference engine that describes how to manipulate and combine these symbols.
One of the expected advantages of machine reasoning is that the systems will reason based on facts and rules, and the operation and performance of this kind of system can be modified more easily since new facts and knowledge can be incorporated over time. In this kind of process, reasoning systems can be taught by individuals who can, in turn, incorporate expertise into computer programs. Through this, a common language of concepts and inter-relationships can be constructed and contained in an ontology and the machine has a language it can work with to solve problems. These machine reasoning systems feature four components:
Four components of machine reasoning systems
These machine reasoning systems generate conclusions from available knowledge and build the foundation for knowledge-based environments. Reasoning systems come in different approaches that vary in expressive power, in predictive abilities as well as computation requirements. Reasoning expert Leon Bottou classifies seven types of approaches:
- First order logic reasoning
- Probabilistic reasoning
- Causal reasoning
- Newtonian mechanics
- Spatial reasoning
- Social reasoning
- Non-falsifiable reasoning
Current machine reasoning systems have different abilities and different requirements than those described by what machine reasoning will, or is expected to, become. The necessary characteristics to empower a machine reasoning system include the abilities to:
- learn on its own,
- find solutions on its own,
- discover the world on its own, and
- understand the world based on concepts (ontology).
Machine learning is a widely used form of artificial intelligence that relies on using collected datasets that can be analyzed for patterns. While machine learning has success in areas involving big data and patterns, it faces obstacles on having to overcome the reliance on tribal knowledge. IT and security practices using machine learning tend to put a great deal of emphasis on innate knowledge possessed by the individual, while also relying extensively on data-driven analysis. Machine learning is best applied in scenarios where the outcome is probabilistic, like determining a risk level.
Machine reason is one order or more of complexity beyond machine learning. It accomplishes the tasks of reasoning out of the complicated relationships between things. Overall, machine reasoning is considered a more human-like approach within the AI spectrum that allows for more flexible adaptation than machine learning. However, machine reasoning requires heuristics and curation, which are usually done by knowledge domain experts. This process is where machine reasoning is difficult to scale as it requires a great deal of expert human effort for the curation to take place.
Machine reasoning is best applied in deterministic scenarios, such as determining whether something is true or whether something will happen. As well, machine learning and machine reasoning should not be seen as competing approaches to understanding data, but complementary ones. And the use of either system should be considered based on the specific use case.
An example of how machine reasoning can be used is network automation. This can be applied in a network that is organized in geographical regions or sectors where a customer needs to define business intent, such as improving network quality in a specific region. This customer need is then broken into service level goals, such as reducing time to content delivery. In turn, the service level goals are further broken into network level goals at individual nodes. These goals can be to improve throughput or defined under the scope of core network. This in turn can be defined to the machine reasoning machine as a desired state.
Machine learning and machine reasoning then work together to devise a strategy upon which transitions need to be followed. Even if the goal is not reachable, the process may uncover problems inside or outside the network. The reasoner, in turn, looks at the predictions and builds a path to transition from the network's current state to the desired state and even offers different paths to achieve the desired state with probabilities for the success of each path.
