Abductive Reasoning Notes

by Sam Berry

June 1, 2021

• Deductive reasoning: like a detective, start with clues and end up with a conclusion.

• Inductive reasoning: use a specific conclusion to draw broader conclusions. It is sunny outside, therefore I don’t need an umbrella. It is sunny ouside, therefore I don’t need a sweatshirt. It is sunny, so I need sunglasses.

  • Deductive equivalent: I don’t need an umbrella, I don’t need a sweatshirt, I do need my sunglasses - it must be sunny.

• Abductive reasoning:

• Deductive deals with certainty, inductive with probability, abductive with guesswork

• AI vs ML

  • AI is where a machine seems “human like and can imitate human behavior”

    • What’s the difference bt AI and ML
    • Disagree - a graphing calculator is an artificial intelligence and it is not human like

  • Define “artificial”: made by humans, not arising from natural causes

  • AI is an machine developed by humans that has the capacity to think based on data. It is “intellegent” compared to traditional programming because of its ability to have dynamic logic statements

  • ML is a type of AI that is dynamic based on data. Over time as data grows and changes, ML programs change how they behave.

  • “ML can go far beyond human capabilites” yes, in very specific specialized ways, like a calculator can. Computers can store an analyze data in quantities that most human brains do not take pleasure in doing.

• Imperative vs Declarative programming

  • Imperative focuses on how to do something, declarative focuses on what to do
  • In Java, using a for loop to filter manually is imperative. Using a stream to filter functionally is declarative.

• Qualitative vs Quantitative

  • Quantitative is measurable, based on objective data, can be counted, measured, expressed using numbers
    • Logical and consistent regardless of circumstance - if data is the same, result is the same
  • Qualitative is descriptive and conceptual. Qualitative data can be categorized based on traits and characteristics
    • “At the heart of qualitative research is the belief that reality is based on perceptions and can be different for each person, changing over time” understanding qualitative and quantitative approach
  • Good breakdown: qualitative vs quantitative
  • Example: A bookcase
    • Qualitative data:
      • Dark brown
      • Golden knobs
      • Smooth finish
    • Quantitative data:
      • 3ft long
      • Weighs 100lbs
      • costs $1500
    • Disputed:
      • “made of oak” site lists as Qualitative, but I would argue this is quantitative. Oak is a testable, measurable thing. Determining Qual vs quan is dependent on the definition of Truth. If it is true that an oak is a specific, unchanging, repeatedly verifiable object in the world - it can be quantified.
      • “Built in italy” - same, that is unchanging, where it is built, assuming that statement is true.
      • “made of wood” - same arguments as above
      • has 3 shelves and 2 cabinets * <- qual depending on definition of cabinet/shelf, what if I consider the top a shelf, or the bottom row a shelf
  • Qualitative is typically unstructured data
    • Ex: a photo. Can make it “structured” by defining metadata about it, the metadata will be qualitative since it’s based on perception

• Empirical & Theoretical

  • Empirical is verifiable by observation or experience, not based on idea or pure logic
  • Theoretical, based on thought, logic, hypothesis

• General reasoning

  • Reasoning in general, whether it be inductive or deductive, is dependent on the set of input observed and the foundation of truth.

    • Use trees as an example. “Trees are tall”.
      • Inductive
        • I have looked at 20 trees, all of them are tall. Therefore trees are tall.
        • I have looked at 20 trees, some of them are tall. Therefore trees are sometimes tall.
        • I have looked at 20 trees, all of them are short. Therefore trees are never tall.
      • Deductive
        • Trees are tall. I have looked at 20 trees, all of them are tall. Theory is correct.
        • Trees are tall. I have looked at 20 trees, some of them are tall. Theory is somewhat correct.
        • Trees are tall. I have looked at 20 trees, all of them are short. Theory is false.
      • Truth
        • There is a minimum amount of required truth that must be agreed upon in order for this research to even be possible. The definition of: tree, tall, short.
          • Tree, tall, short are qualitative variables. All, some, 20 are quantitative variables.

  • Using the statement, “Trees are tall”, build a qualitative analysis computer program.

    • First define the qualitatve and quantitative variables in the statement
      • Qualitative: tall
      • Quantitative: tree

  • it is critical to define moral and creation-based standards for what is quantitative

    • What God has created and defined cannot be changed. There are things on earth that have bounds and strict definitions. Nature, biology, Truth. To allow these to become qualitative enables
    • In tree example above: It is important to draw the line with qualitative and quantitative. To argue that “tree” is opinon based is dangerous and incorrect. There are bounds to what defines a tree.

• Reducing possibilities into reasonable conclusions

• Abductive Reasoning API

  • Use human knowledge (knowledge based system) to aid AI decisions, to enable “making the best guess”.
  • Build an API the tracks relationships between entities, then enable customer to provide written statements for how to make decisions when certain entities are searched.
    • Example: build a graph of campgrounds and all their related attributes.
      • Customer: selects tags to build a relationship, “Texas, Hill Country, short stay” tie the relationship to an action -> “suggest, hide, etc” -> action points to the right side of relationship: “Texas, botique, 30mile radius of Dallas”
  • Customer can build tags for relationship. Example, “30 mile radius of Dallas” would be a distance-based rule centered in dallas. “Short stay” would be a stay length rule for X number of days.
  • API can use tags as input data for a request, or they can use entire JSON blobs.
    • ex, to market this to Campspot - to plug into the API, just send over the JSON of search request or result. ARPI will scan for known keywords (based on category/tag configuration database) and fetch relations

• HAD - Human Aided Design

  • To achieve successful abductive AI, must lean on human wisdom to make educated guesses
  • Throws the idea of runaway AIs away, the idea that AIs are more than machines that are slave to humans, totally dependent.