View source for Model-based reasoning

{{Short description|Inference method used in AI expert systems}}
In [[artificial intelligence]], '''model-based reasoning''' refers to an [[inference]] method used in [[expert systems]] based on a [[model (abstract)|model]] of the physical world.  With this approach, the main focus of application development is developing the model.   Then at run time, an "engine" combines this model knowledge with observed data to derive conclusions such as a diagnosis or a prediction.

==Reasoning with declarative models==
[[File:Model based goal based agent.png|thumb|Model based goal based agent]]
A robot and [[dynamical system]]s as well are controlled by software. The software is implemented as a normal computer program which consists of if-then-statements, for-loops and subroutines. The task for the programmer is to find an algorithm which is able to control the robot, so that it can do a task. In the history of robotics and [[optimal control]]<ref>{{cite journal |title=Autonomous systems |author=Watson, David P and Scheidt, David H |journal=Johns Hopkins APL Technical Digest |volume=26 |number=4 |pages=368–376 |year=2005 }}</ref> there were many paradigm developed. One of them are [[expert system]]s, which is focused on restricted domains.<ref>{{cite journal |doi=10.1002/int.4550050202 |year=1990 |publisher=Wiley |volume=5 |number=2 |pages=135–151 |author=Newton S. Lee |title=A computational paradigm that integrates rule-based and model-based reasoning in expert systems |journal=International Journal of Intelligent Systems |s2cid=26907392 |doi-access=free }}</ref> Expert systems are the precursor to model based systems. 

The main reason why model-based reasoning is researched since the 1990s is to create different layers for modeling and control of a system.<ref>{{cite conference |title=Model-based programming using golog and the situation calculus |author=McIlraith, Sheila A |conference=Proceedings of the Tenth International Workshop on Principles of Diagnosis (DX’99) |pages=184–192 |year=1999 }}</ref> This allows to solve more complex tasks and existing programs can be reused for different problems. The model layer is used to monitor a system and to evaluate if the actions are correct, while the control layer determines the actions and brings the system into a goal state.<ref>{{cite conference |title=Model-Based Monitoring of Dynamic Systems. |author=Dvorak, Daniel and Kuipers, Benjamin |conference=IJCAI |pages=1238–1243 |year=1989 }}</ref>

Typical techniques to implement a model are [[declarative programming language]]s like Prolog<ref>{{cite journal |title=An expert system shell for uncertain rule-and model based reasoning |author=Niederlinski, A |journal=Methods of Artificial Intelligence in Mechanics and Mechanical Engineering AIMech |year=2001 }}</ref> and Golog. From a mathematical point of view, a declarative model has much in common with the [[situation calculus]] as a logical formalization for describing a system.<ref>{{cite conference |title=Representing actions and state constraints in model-based diagnosis |author=McIlraith, Sheila A |conference=AAAI/IAAI |pages=43–49 |year=1997 }}</ref> From a more practical perspective, a declarative model means, that the system is simulated with a [[game engine]]. A game engine takes a [[Feature (machine learning)|feature]] as input value and determines the output signal. Sometimes, a game engine is described as a prediction engine for simulating the world.

In 1990, criticism was formulated on model-based reasoning. Pioneers of [[Nouvelle AI]] have argued, that symbolic models are separated from underlying physical systems and they fail to control robots.<ref>{{cite journal |doi=10.1016/s0921-8890(05)80025-9 |title=Elephants don't play chess |author=Brooks, Rodney A |journal=Robotics and Autonomous Systems |volume=6 |number=1–2 |pages=3–15 |year=1990 |publisher=Elsevier |citeseerx=10.1.1.588.7539 }}</ref> According to [[behavior-based robotics]] representative a reactive architecture can overcome the issue. Such a system doesn't need a symbolic model but the actions are connected direct to sensor signals which are grounded in reality.

== Knowledge representation ==

In a model-based reasoning system [[knowledge]] can be [[knowledge representation|represented]] using '''[[causal rules]]'''. For example, in a [[medical expert system|medical diagnosis system]] the [[knowledge base]] may contain the following rule:
: <math>\forall</math> patients : Stroke(patient) <math>\rightarrow</math> Confused(patient) <math>\land</math> Unequal(Pupils(patient))
In contrast in a [[diagnostic reasoning]] system knowledge would be represented through [[diagnostic rules]] such as:
: <math>\forall</math> patients : Confused(patient) <math>\rightarrow</math> Stroke(patient)
: <math>\forall</math> patients : Unequal(Pupils(patient)) <math>\rightarrow</math> Stroke(patient)

There are many other forms of models that may be used.   Models might be quantitative (for instance, based on mathematical equations) or qualitative (for instance, based on cause/effect models.)  They may include representation of uncertainty.  They might represent behavior over time.  They might represent "normal" behavior, or might only represent abnormal behavior, as in the case  of the examples above. Model types and usage for model-based reasoning are discussed in.<ref>[http://gregstanleyandassociates.com/whitepapers/FaultDiagnosis/Model-Based-Reasoning/model-based-reasoning.htm Model Based Reasoning for Fault Detection and Diagnosis]</ref>

== See also ==
* [[Diagnosis (artificial intelligence)]], determining if a system's behavior is correct
* [[Behavior selection algorithm]]
* [[Case-based reasoning]], solving new problems based on solutions of past problems

== References ==
{{reflist}}
* {{Russell Norvig 2003|pages=260}}

== External links ==
* [https://web.archive.org/web/20120716213632/http://www.cs.uu.nl/docs/vakken/mbr/ Model-based reasoning at Utrecht University]
* [http://ti.arc.nasa.gov/ NASA Intelligent Systems Division]
{{Automated reasoning}}

[[Category:Expert systems]]
[[Category:Automated reasoning]]


{{compu-AI-stub}}