Saturday, September 27, 2014

Izabella Kaminska on Artificial Intelligence

Yeah, pretty much the poster child for the term autodidactic polymath.
From Dizzynomics:
Brute force, trial and error systems vs imagination
From the New Scientist article Daydream Believers (from last week):
“I would love to know if chimpanzees can entertain the notion of a unicorn, but we have no idea,” he says. “As far as I can tell, we don’t even know whether they can entertain two possible scenarios to solve a problem.” In Call’s view, it is impossible to say whether the animals that solve problems without trial and error are consciously imagining different solutions, or subconsciously integrating information to come up with the correct solution. “I’m not saying animals can’t imagine two different scenarios,” he says. “I just don’t see the evidence for it.”
As I have been learning this week from AI experts, the ability to learn depends mostly on trial and error processes which strengthen neural pathways that lead to positive results.

But as I am also learning, the field of artificial intelligence is split between three different types of AI learning processes.

First there are big data systems. These aren’t really intelligent because they depend on pre-programming by intelligent agents. They’re effective only because of brute force approaches to problems, processing huge amounts of data. You can imagine the problem. It’s a very energy intensive way to achieve the optimum action, and it remains dependent on instructions on what to look for. Like finding a needle in a haystack by processing every single bit of hay and checking that it isn’t by chance a needle. This is brute force trial and error.

Then there’s proper AI which achieves intelligence through self-learning algorithms that reward the AI for reaching objectives of their own accord. These are a different type of trial and error process. You don’t have to know what the task is, you’re simply learning to deal with the environment you are faced with and every time you discover something exploitable you develop a neural pathway that teaches you this is a good way to approach that sort of problem....MUCH MORE
See also:
In Search of the Universal Algorithm: Jeff Hawkins on Why His Approach to Artificial Intelligence will Become THE Approach to AI
Google's Plan To Make Your Brain Irrelevant (GOOG; EVIL;)
"Elon Musk: I'm Worried About A 'Terminator'-Like Scenario Erupting From Artificial Intelligence"
Robo-journalists: Beyond the Quakebot
"Why strange loops could be an argument for artificial intelligence"
Pew Research: "AI, Robotics, and the Future of Jobs"
Artificial Intelligence: The Painting Fool
Google Launches the Quantum Artificial Intelligence Lab (GOOG)
The First Conscious Machines will Probably Be on Wall Street
'Deep Learning' as Applied to Investing
"Stephen Hawking Joins Anti-Robot Apocalypse Think Tank"
"Why Is Machine Learning (CS 229) The Most Popular Course At Stanford?"
And many, many more.