Artificial Intelligence Anonymous part Deux Ex Machina

This is my attempt at providing a brief introduction to another subject I know nothing about. Much to my dismay, I never took AI. Ignorance hasn't stopped me from blogging on topics before (see my Lojban, income inequality, lottery and capitalist ideology pages) and it surely isn't going to stop me now.

1. Artificial intelligence anonymous, part one.


2. Artificial intelligence anonymous, part two.

How do you make artificial intelligence?

To make intelligence is easy. You look at what Homer Simpson does and do the exact opposite. To make a machine realise that being intelligent means being the opposite of stupid all by itself is a different story, however.


Homer Simpson. Artificial? Yes. Intelligent? Well..

There are three main approaches to creating artificial intelligence. These are symbolic, sub-symbolic and statistical.

Symbolic AI is sometimes called Good Old Fashioned AI or GOFAI. This approach is based on the premise that human thinking is largely based on manipulating symbols. All our thoughts require symbols and the relationships between them. The first wave of the rise of the machines was heavily based on lambda calculus. Lambda calculus involves a kind of bootstrapping thinking where a function can call itself. It is meant to simulate the human ability to think about thinking. Symbolic AI is still intimately tied with programming languages that can support this mode of thinking such as LISP.

Sub-symbolic AI is concerned with describing human thinking processes algorithmically in order to simulate these processes on a machine. Algorithmically means that the process is describe in a step-by-step manner like a cooking recipe. Sub-symbolic AI is thus based on simulation of specific thinking processes instead of on the underlying thinking process of symbolic processing. Neural networks fall under this category.

Statistical AI is an extension of sub-symbolic AI in the sense that it is heavily steeped in mathematics and algorithmic thinking. Certain tasks are solved mathematically and then implemented in a machine so the machine may exhibit the same intelligent behaviour. While GOFAI researchers are often contempt with having some fuzzy logic, statistical AI researchers require that their solutions are determinable and optimal. This means they require that the machines try to do something constructive by attempting problems that do have solutions and that they do so in the best way possible.

Computational intelligence is the new blanket term for AI. While it incorporates all the waves of AI so far, it relies heavily on mathematical and statistical techniques without much regard for replicating the fuzziness of human logic.

The three main approaches are still far from creating general intelligence, even though they've made progress in creating very smart artificial thinkers for specific domains.

The lunatic fringe of AI

By lunatic fringe, I don't mean emo hair. Of course you always get those people who aren't satisfied with the norm. I am one of those people so I sympathise with the lunatic fringe of AI. Two fields that I admire are artificial life and technological singularity.

Artificial life researchers are not satisfied with merely simulating one aspect of humanity. Nay, nay, they want machines that smoke, drink and fuck. Provided of course that they evolve to this amazing level of civilisation by themselves.

Singularity researchers reason that we need not be content with creating human-level AI. We can reach further and create superintelligent beings so we can worship them. This is much the same as any other religion except you can actually reach out and touch Far-fetched Artificial Intelligence Theology or F.A.I.T.H. No, not really, I just made up that acronym.