Frequently Asked Questions

What are the most important characteristics to look for in the ideal expert system for managing legal or philosophical knowledge?

Here are a list of principles that were taken into account during the design process of deCheem.

• Type-less knowledge entry
  • A word or phrase can mean different things in different contexts (e.g. club as in golf club or disco club), so any form of ‘type’ declaration (in the broadest sense of the word) will inevitably limit the ability of that word to take on different meanings (and hence roles) in future situations.
• Sequence-insensitive knowledge acquisition
  • Expert systems are only considered useful if it can produce accurate advice faster and cover knowledge areas larger than any human can.
  • To achieve this, acquiring and assimilating the knowledge from a large number of sources is essential. To enable large scale knowledge aggregation to take place efficiently, it is crucial that the sequence in which knowledge is entered into the system has no effect on it’s usefulness later on.
• Deterministic
  • Probabilistic methods (e.g. text-clustering, correlation-based) do not have a lot of use in philosophical and legal settings, as saying that something is ‘maybe’ or ‘likely’ true does not allow one to settle an argument completely.
  • It is therefore important that the framework deduces conclusions, rather than makes a guess of what it is (regardless of the likelihood that it is correct).
• Syntax-less knowledge entry
  • For expert systems to be truly useful in the fields of law and philosophy, it should not enforce specific syntax that the user needs to learn in order to document their knowledge. Not only does specific syntax decrease the usability of the system, it also can be seen as a form of ‘typing’, which has the downsides described above.
• Directionless reasoning
  • Forward and backward chaining are functionalities often looked at in expert systems, but any system that predefines types for reasoning/chaining will eventually be insufficiently generic for philosophical and legal use cases.
• Implicit deduction
  • A good expert system should be able to read what is explicitly mentioned in piece of documented knowledge, but also what is implicit. This allows a single statement to be more than just a declaration of a single relationship, but of all possible logically equivalent permutations of that relationship.

How does deCheem deal with forward and backward chaining?

deCheem intentionally does away with the technical distinction between these two forms of ‘chaining’. In the spirit of ‘typelessness’, deCheem instead has just one way of exploring situations, without taking into account if it’s the desired starting or ending situation.

The degree to which a particular explore is doing ‘forward’, ‘backward’ (or mixed??) chaining is determined by the human language used to describe those situations.

What is the best way to phrase a belief?

Refer to the style guide on this documentation website for extension guidelines.

How does deCheem represent beliefs about different subjects without using conditional/IF-ELSE statements?

IF-ELSE statements are needed when relating between two different subjects in terms of graphs/relations. In a set-based expert system like deCheem, this can be bypassed by phrasing everything as a property of a common subject. ‘Situations’ is the best choice of subject here, as it keeps everything in the same set-space by allowing anything under the sun to be phrased in terms of situations. This allows us to achieve conditionality without resorting to IF-ELSE clauses, and thus also achieves sequence-insensitivity at the same time.

Do relational databases fulfil the deCheem design principles?

Relation databases do not perform inference out of the box. If one were to (mis)use them for performing deduction, analyse 20 different possible situation types would need a table with 2^20 rows. This is unscalable and is not what relational databases should be used for.

Do graph databases fulfil the deCheem design principles?

Graph databases sees things as nodes with fixed relationships. deCheem forms relationships between different nodes based on certain conditions, and the inference engine layer is not native to graph databases.

Do decision-trees fulfil the deCheem design principles?

Decision trees are by nature hierachical and operates on branches. If an idea in a deep branch has links to another idea in an earlier branch, there is no efficient way to represent that relationship. Also, if the definition of a decision point at an earlier branch is changed, the validity of the decisions branches lower down will all be affected, which limits the maintainability of this solution.

Frameworks that share the same method and therefore the same pitfalls when used as philosophy and legal expert systems are:

• Decision Model and Notation (DMN)
• Argument-maps

Does Prolog fulfil the deCheem design principles?

Prolog is great for quantitative inferences and relationship deduction when properties share only inherit properties from a single parent. However, numerical methods are useless against analysis of beliefs, and the need for beliefs to take on different meanings (aka inherit properties) from any number of situations makes Prolog a bad choice to use for belief analysis. Prolog makes a distinction between ‘rules’ and ‘facts’, and that distinction takes away from the ‘type-less’ nature of a good general expert system.

Does the Carneades system fulfil the deCheem design principles?

When it comes to how knowledge is represented, the Carneades argumentation system is one of the closest to the deCheem belief language. Subjects and predicates are represented together in ‘statements’ (belief properties in deCheem’s terms), which is one step closer to true ‘typelessness’. Carneades also represents only relations between statements in a single direction, while deCheem does that but also allows statements to have true modality (e.g. represent assertions that are true in all cases/directions).

However, when it comes to how conclusions are generated (aka the inference engine), Carneades takes a graph-based approached (e.g. linking nodes to each other through edges) while deCheem goes for a set-based approach. Graphs are meant to show (cor)relation, and it can at best only deal with forward-chaining use-cases, and only for the situations that have been explictly documented either in part or full. deCheem does away with directionality altogether thanks to it’s set-based approach, and also allows for deduction of all possible implicit conclusions.

Why are OWL or RDF-based formats not used for representing beliefs or statements in deCheem?

OWL Web Ontology Language and RDF make heavy use of object properties and relationship declarations (e.g. subClassOf, oneOf, childOf) to represent information, which takes away from the typelessness that deCheem trieds to strive for.

Other formats or frameworks that share the same pitfalls are :

• LegalRuleML
• Protége.

Does AceRules fulfil the deCheem design principles?

AceRules attempts to make rule entry very similar to typing regular English, which is admirable. However, it’s strength but also its pitfall lies in its use of Attempto Controlled English as its foundation. As Attempto Controlled English restricts the set of standard English that can be used, it already introduces a limit to the kinds of ideas or relationships that we can express with it. Furthermore, AceRules requires relation types to come from a predefined list, which also takes away from the goal of typelessness.

What can deCheem not deal with (natively)?

Working with beliefs with a temporal or numerical nature is possible, but cumbersome without tools to help generate the arbitrary number of beliefs needed to cover temporal or numeric graduations.

deCheem does allow for plugins to be added on top as pre-processors, allowing data input sources that accepts graduated input (e.g. time, amounts, decimals) to be translated to input congruent with the set-based system of deCheem.

How do you compartmentalise belief-systems in deCheem?

If you want to categorise belief-systems based on their provenance, simply add that as an additional description of the situation.

Is deCheem a NLP project?

Nope. deCheem neither is nor aims to do Natural Language Processing in any way at this stage. NLP plugins are however possible on top of the deCheem framework to allow for beliefs to be generated in a much quicker way than human input.

Why can’t deCheem automatically solve all confusion in conversations?

deCheem shifts the complexity of reasoning away from code and mathematics and into the realm of language. While this grants it enables the user to utilise any corner of his/her vocabulary, it cannot help the user extend or correct his/her vocabulary.

Why doesn’t deCheem use any form of weighting?

Using weightage/votes to determine the correctness of a belief is fundamentally against the idea of deCheem, which is to use logical deduction to arrive at facts about our world. If you find yourself struggling with the correctness of a certain belief, think about a specific subset of situations with this belief that you for sure is correct, and document that instead.

Since ‘not good’ is not necessarily ‘bad’, how can things be binary?

Indeed, ‘not good’ is not the same as ‘bad’, just like ‘not hot’ is not necessarily ‘cold’. deCheem leaves it to the user to determine what the opposite of each situation is, be it as a new situation or simply the negation of the former.

How do you deal with ‘scales of things’ or ‘rankings’ or ‘priorities’.

‘Scales’ have similar pitfalls to weightage - something has to be more important than everything else, and when it’s not, then something else is. This means that any arbitrary ranking-list can be expressed as a series of beliefs about the utmost importance of a certain thing under certain circumstances.

How efficient is deCheem in dealing with large numbers of beliefs and arguments?

deCheem uses set-theory to perform deduction, meaning it is able to perform implicit deduction without the computational overhead of generating 2^n scenarios or relationships in order to achieve it.

The limitations on the number of beliefs that can be in a belief base is only limited by the RAM available on your computer. As for the computability of the inference, deCheem can generate arguments with up to 500000 chained implications easily on any modern laptop.