Introduction to Argumentative Learning

Let’s study the rationale for introducing a new approach in machine learning called “Argumentative Learning”. (It’s still unpolished for now- sorry for the inconveniance…)

Argumentative theory

argumentative learning : Mercier and Sperber : Why do humans reason ?

Humans do not use perceptrons : they never learn by themselves (collaborative)
human learns when they are corrected (this is why MOOCs is a hoax; books are better, yet global population isn’t smarter)

argumentative rates the source of the idea, (it can be a merit function)
Merit functions can be automatically crafted, see Cornell’s Eureqa (now Nutonian)

argumentation has the deep beauty of being able to keep information in a somehow analytic level : there is no need to quantify

A first attempt in argumentative learning can be fuzzy logic; the problem is that logic is not argumentation, but the spirit is there

The lack of mathematical foundations in machine learning

Kolmogorov complexity : the idea is the creation of the data, what ultimately matters is the retrieval (this is where perceptrons can be efficients)

xkcd travelling salesman//kolmogorov complexity has disappeared with networks

mathematics is not a language : it is based on equalities, while a language is itself based on approximations

Turing test can be seen as the proof a the lack of an argumentative power
(Autotelicity ?)

Epistemology and argumentation

Poincare and russell (perception)

am a physicist myself, and I can help (most computer scientist have no idea how to gather data, which is why they do’t try these ideas)
Wright brothers invented the plane not because it looks like a bird, but because it has the same functions (this is why neuron networks are deeply flawed)

not  big proponent Feyerabend or lakatos

this is what lies behind the idea of falsablilty : it must be contradicted

There’s no equivalent for double negation : right-right does not add more information, while… it should

Science itself is based on argumentation, not discovery : who ever found something he was looking for ? By definition of research, that can’t be + quantum physics
photons vs electrons? fermi-dirac vs. bose-einstein : interaction ? polarization = mixed state

Philosophy is not logical, contrary to what we may think, but argumentative : theories do not disappear
Bayes theorem : a start in argumentation; but the hypothesis must be formulated beforehand (can it be opposed to frequentists ?)

Google is ad hominem (reputation)

The Art of Being Right  Schopenhauer

The problem with neuroscience

ability to anticipateAI hasn’t evolved in years (hoffstadter->Lecun), only big data and conputing
the probleme with big data : it tries to use large number laws to cover its intrinsic inefficiencies

http://boingboing.net/2013/11/28/brainwashed-neuroscience-vs-n.html

Example of the brain on a chip

deep blue and jeopardy somehow manage to be argumentative : they have to predict beforehand

Statistics and stability

The signal and the Noise, by Nate Silver

Can an argumentative theory be stable ? Can it be modelized ?
Financial markets are argumentative.

Democracy is stable, and argumentative (all technocratic societies have failed at some point, example of war as a proof of stability)

fallacies and falsification

Decision-making, psychology and behavioral sciences  – The power of bias

Kahneman : WYSIATI//cognitive bias

Gigerenzer : bias is good

List of Cognitive Bias

State of the art

A Comprehensive Workflow for General-Purpose Neural Modeling with Highly Configurable Neuromorphic Hardware Systems

Argumentation-Based Computer Supported Collaborative Learning (ABCSCL): A Synthesis of 15 Years of Research

free online lessons about machine learning.

Pattern recognition and Macine Learning, by C. Bishop

Argumentation in Artificial Intelligence by Iyad Rahwan

You think i’m wrong, i think i’m right : this is the beginning of argumentation
Eh, you know what? I gave you my arguments !

Now it’s your turn !

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>