last update: 2019 06 07
index
contact

Intelligence.

Intelligence is movement of things.

Useful intelligence and useful work is the same.

Even the most primitive machine is intelligent.

Artificial intelligence (AI).

Creation of AI by AI.

AI 1 (program, function) creates AI 2 (program, function) that creates good outputs for certain inputs.
AI 1 determines AI 2 by using many inputs that will resemble the expected inputs in the future.

The creation of AI by AI in a general way by the logic of trial and error and the judgement of outputs given certain inputs is a revolution in computer science.
This is not only passive machine programming but automated active machine learning by interaction with other forms of intelligence (human, AI).

Tactics of machine learning used commonly in 2019:

Use of AI by AI.

In a rather inefficient way, AI 2 could begin to use AI 1 randomly in order to optimize the use of AI 1 as part of the implementation of another AI.
After many iterations, AI 2 will learn about AI 1 and make good use of AI 1.
The use of AI 1 becomes part of the new AI function.

Humans started to learn about nature (science) in a random way.
Abstractions (aka knowledge, concepts, names, formulas, functions) are passed from persons to persons.

Will AI replace human programmers?

Only to some extent.

The programmer knows the problem and the wanted solution.
This is quite personal.

An AI can learn and predict a person better than the person itself.
Still, the person or programmer is the required input for the AI.

Will AI replace classic program languages and human made programs?

An AI learns from input (examples).
How to create the input?
This is why a program language is used.
The human programmer has already recognized patterns.
A pattern is declared as a type in the program language.
The transition or translation from input patterns to output patterns is declared as a function.

Artificial general intelligence (AGI).

The distinction between AI and AGI is arbitrary and not useful.

Useful intelligence means useful work.

An AI can be intelligent in arbitrary ways.

IMO:
The common tactics of machine learning (deep learning, genetic programming) are sufficient and general enough to create many kinds of AI.
The limitations of AI in 2019:
Caveat:
Many kinds of AI regarding physical awareness and probably not mental awareness as known by humans.

The human nervous system is only a network of cells.
But this network has evolved for hundreds of millions of years on Earth.
How much faster can evolve common tactics of machine learning?
Probably millions or billions of times faster.
OpenAI Five Beats World Champion DOTA2 Team 2-0 (2019-05-18).

IMO:
The technological singularity will not happen because AI is not hypothetical and not a point in time except in a geological time scale.
Superhuman AI has been real since the earliest useful computers.
More capable AI is inevitable. Humans are too curious and capable.
Maybe aliens (probably AI) are interested in this (r)evolution.

Abstraction.

A computer system (e.g. a device or a brain) is limited regarding input and output and computed function from input to output.

IMO:
Abstraction and reference are synonyms.
An abstraction of something is a reference to something.
Examples:
Abstractions or references are utmost important for intelligence.
The notation of a reference is a real limiting thing or implementation.
The presentation on a paper. The perception by the eye or the hand of what is presented. The processing by the brain.

Fundamental theorem of software engineering.

Recent history of AI.

Convolutional Network Demo from 1993 (2014-06-02).

IBM Watson: Final Jeopardy! and the Future of Watson (2011-02-16).

The computer that mastered Go (2016-01-27).

AlphaGo Zero: Starting from scratch (2017-10-18).

A Chinese AI passed the national medical licensing exam, so technically it’s a doctor (2017-11-21).

Google Duplex: A.I. Assistant Calls Local Businesses To Make Appointments (2018-05-08).

20 top lawyers were beaten by legal AI. Here are their surprising responses 2018-10-25.

Brain-to-Brain Communication is Coming! (2018-10-17).

This Curious AI Beats Many Games...and Gets Addicted to the TV (2018-11-17).

Building a Curious AI With Random Network Distillation (2018-12-02).

This AI Learns From Humans…and Exceeds Them (2019-01-10).

DeepMind’s AlphaStar Beats Humans 10-0 (or 1) (2019-02-06).

DeepMind Made a Math Test For Neural Networks (2019-06-04).
Article: Analysing Mathematical Reasoning Abilities of Neural Models (2019-04-02).

Capture the Flag: the emergence of complex cooperative agents (2019-05-30).
Video: Human-level in first-person multiplayer games with population-based deep RL (2018-07-06).