JackRose

JackRose is a new entry in the spaced repetition software category.

It is not intended to be especially simple, or easy for non-technical people to use. If you want a simple system, you are invited to consider Anki. This advice is not flippant or sarcastic; JackRose requires some computer expertise to set up and is not suitable for newcomers.

In contrast to the small personal systems that currently dominate the SRS arena, JackRose is designed to handle large data sets efficiently. It is borne of the author's assessment that the existing SRSs don't scale well to 100s of users with ten of thousands of learnt items.

The name has no particular significance; Jack and Rose are two English words that have multiple meanings; that is all. It is pronounced with the stress on the first syllable.

• It is a web application. This avoids having to implement a user interface, but it also satisfies the design aim of a large-scale system, in that it can be centrally administered and access granted or denied securely. Unlike other web-based SRSs however, it is entirely open; anyone with the knowledge and means can install and run their own JackRose server.
• Open data: in contrast to systems that require learning data to be imported into their own database, JackRose connects to external data sources, which (currently) can be SQLite3 tables, Postgres tables, or tab-separated flat files. The only requirement is that each data record have a unique and unchanging key. This allows a business or other organisation to maintain, for example, a lexicon (``foreign-language dictionary'') and JackRose will use the `live' data from it and won't require a periodic refresh or re-import. Similarly, media (sound and video files) can be organised according to the wishes of the user, and do not have to be imported into the application.

• It implements the Anki concept of separate data (notes, in Anki parlance) and learnt items (cards in Anki parlance; really, views on the data).
• Written in Haskell; one clear research finding is that dynamically-typed languages suffer from exponential maintenance complexity and that strong static type checking is an important tool of software engineering to achieve higher reliability. This of course precludes the Anki model of add-ons (monkey patching) and JackRose must provide user-customisable behaviour either by configuration options or by source modification. Modern DVCSs allow the preservation of user customisations when source is updated or pulled.
• Importing data is achieved simply by specifying a new data source, having previously used bulk load etc. to populate the source.
• Configurable spacing algorithms; this is to determine, for a particular learner, which is the most `efficient' algorithm; be it SM-2, SM-x, Leitner or ANN.
• The business model is that the software is open-source (free) but that you can choose to pay for hosting (which means that you only need a browser to perform daily repetitions) and consultancy.

Research

Artificial Intelligence and Human Thinking by Robert Kowalski

The Voidness Way

Anki Text Playback

Ja-Dark Blog Archive Jan. 2014 to June 2017

Need to remember something? Try forgetting it!

"Any item you encode into your memory can be described by two characteristics: storage strength and retrieval strength. Storage strength is a general measure of how well learned that item is, while retrieval strength measures how accessible the item is at that time. Storage strength increases monotonically— the more you are exposed to an item, the stronger the storage strength gets. One important thing to note is that storage strength does not have a direct effect on memory performance; the probability that you will be able to reproduce something from memory (e.g., a name or phone number) depends almost entirely on its retrieval strength.

Retrieval strength, unlike storage strength, varies with context and intervening information. This type of strength is what we need to increase if we want to be able to remember an item later.

(...) The real kicker—and where forgetting comes in—is that the more difficult the retrieval is, the more beneficial it will be when you finally do come up with the answer!"

Now, read this excerpt from the attached paper "Physical Nature of Intelligence" written by a polish cybernetician - M. Mazur - in 1974):

"This is the most general concept of memory. (...) There are two factors for storing information: a) The conductance as the passive record, i.e. which can last even without any energy flow but alone cannot provoke any process; b) the control power as the active record, i.e. which can provoke other processes but does not last if there is no energy flow.

Consequently, there are two kinds of forgetting and remembering: Disappearance of the passive record owing to the decrease of the conductance with time, till the forgotten stimulus reappears and thus restores the record; disappearance of the active record owing to the cease of the energy flow or to its displacement into another area of the correlator ("the brain"), till energy returns to the former path constituted by the still existing passive record. This explains, for example, the difference between having forgotten the number of a bus ticket bought ten years ago and having forgotten the topic of the conversation suddenly disturbed. The first cannot be remembered unless the forgotten number will be seen anew, whereas the latter can be remembered after some thinking."

And now, read this excerpt from the attached paper "The Logical Way to Be Artificially Intelligent" written by a logician R. Kowalski:

"Both Thagard and Baron fail to recognize that, to be used in practice, logic needs to be controlled. This could be put in the form of a pseudo-equation:

Thinking = Logic + Control.

Here the term “Logic” refers to goals and beliefs expressed in logical form and “Control” refers to the manner in which the inference rules of logic are applied. Control includes the use of forward and backward reasoning. In the case of backwards reasoning, it includes strategies for selecting sub-goals, as well as strategies for searching for alternative ways of solving goals and sub-goals. It also includes the application of inference rules in sequence or in parallel."

Three different perspectives, and yet very similar conclusions:

1. 1. storage strength,
   2. the conductance as the passive record,
   3. logic (knowledge or "the web of beliefs": goals and beliefs),
2. 1. retrieval strength,
   2. the control power as the active record,
   3. control (search) includes the use of forward and backward reasoning.

By paraphrasing, we can paradoxically say "The real kicker—and where forgetting comes in—is that the more difficult the SEARCH is, the more beneficial it will be when you finally do come up with the answer!".

And what about that? Is human memory like a river? A B.

More about thinking, knowledge, search and the web of belief here.