Wednesday, November 5, 2014

Superintelligence

One hundred and twenty thousand years ago, Neanderthals were the dominate humans on the planet. How much time do you think they spent worrying about Homo Sapiens?

Where are the Neanderthals today?

Today, Homo Sapiens aren't spending a lot of time worrying about it, but soon there may be an intelligence explosion of either biologically enhanced humans or smarter machines. Perhaps both.

Machines that could mimic human common sense along with the ability to learn, reason, and solve complex challenges has been expected since the dawn of computers in the 1940s. But human level intelligence is not the end result—superhuman machine intelligence is!

The capacity to radically outperform the best human minds is called superintelligence. Creating superintelligence may be humanity's legacy since once achieved, superintelligence will self-improve and surpass itself.

If we start with the 21st century human and add significant cognitive and physical enhancements the post-human may achieve biological superintelligence. The real question is—will machine superintelligence arrive first.

Trans-humanism may be science fiction right now, but the approach toward superintelligence is being pursued. Machine artificial intelligence is the study of many projects around the world.

Ways to enhance machine intelligence include: statistical patterns, neural networks, pattern recognition, and problem classification are old tools in a multilevel approach toward enhanced intelligence. A large class of native decision trees and statistical analysis are being used. Machine learning using probabilistic based Bayesian networks and advanced Monte Carlo approximations are also in the running. They are used to create learning rules. To create an AI we need to add decision rules to learning rules. Computational structure resulting from the analysis is used to reproduce intellect memory and personality.

In planning for the future, safeguards must be employed to prevent letting the genie had of the bottle by creating a super artificial intelligence bent on replacing humans.

Technological developments may radically change humans brain through enhancements and genetic improvements or by redesigning the human condition around parameters of aging and limitations physical and mental. Mind scans or capturing the 'mental individual' and moving it to a new physical environment is another issue. An alternative to greater than human intelligence is to enhance functioning of biological brains through selective breeding biochemical enhancements or drugs. All of these methods require a considerable number of years to effect change.

The emergence of superintelligence may be sudden and dramatic. Paths to superintelligence include intellect that exceeds the cognitive performance of humans in all domains of interest.

The state-of-the-art of artificial intelligence already has cases where the human intelligence is being outperformed by specifically designed machines checkers backgammon, chess, and scrabble. All are performed by computer at a super human level.

However, in sense and natural language understanding have turned out to be difficult for computers to solve. We use Markov models with statistical techniques to accomplish speech recognition. And while machine translation is imperfect, it is improving. Face recognition and their improving our other areas of development.

In whole brain emulation, intelligence software scans and models computational structures of the logical brain requires detailed scanning of human brain including dissecting the tissues of the brain into thin slices studying the structural and chemical properties.

Brain computer interfaces are hybrid systems to radically outperform the un-augmented brain requiring significant medical complications including and fictions hemorrhages and brain damage cyborg is a period. Networks and organizations can be enhanced on a collective basis to improve collective intelligence examples of these are writing and printing in languages gradual accumulation of institutional capital systems collective intelligence from the member minds and overhead of communication integration.

Enhancement of biological organization intelligence might increase ability to design machines super intelligence that is safe and beneficial. Brain computer interface is unlikely as a source of super intelligence improvements. Networks and organizations may yield narrow increases to capability.

Collective superintelligence is a system composed of a large number of small intellects such that the system's overall performance vastly outstrips current systems. Superintelligence at least as fast as the human mind and vastly qualitatively smarter.

The timing and speed of the takeoff to reach superintelligence is critical. A slow takeoff offers opportunity to intervene on the human level. A fast takeoff is so quick it is unnoticed till the game is over and humanity has lost. A moderate takeover is one that occurs over media global time of months or years giving humans some chance to respond but not much time to analyze the situation.

Building a seed AI might be quick and only require an insightful algorithm that can recursively improve itself so rapidly as to outstrip all competition.

Superpowers of strategic intelligent advantages include intelligence amplification wherein a system can bootstrap its own intelligence. A system can achieve distant goals and overcome intelligent opposition, and even conduct social manipulation to leverage external resources.

A boxed AI may persuade its gatekeeper to let it out. A boxed AI may exploit security holes steal financial resources and hijack.

An AI takeover scenario would include a first superintelligence gaining a distant decisive strategic advantage in the pre-critical phase. The creation of a seed AI would enable it to improve its own intelligence secretly, recursive self-improvement phase would be rapid, covert preparation phase where it understands its strategy to escape and dominate, and in the overt implementation phase where the AI gains power and self-preservation could be a motivation of the AI itself or it could be just merely cognitive enhancements or seeking technological perfection or research acquisition in order to do calculations in any case it would be a direct threat to humanity. Controlling an AI starts with boxing physical input and output or designing the AI with inherent limitations or controls.

Machine intelligence can take several forms including: oracle, genies, sovereigns, and tools.

An Oracle, is a question answer system working with super intelligence in an extremely limited domain. Such as the Internet search engine.

Genies and sovereigns use a command executing system. A Genie receives high-level commands and carries them out then pauses and waits for the next command. A sovereign is a system that has an open-ended mandate to operate in the world to pursue the broadest possible objectives.

Reference: Bostrom, N., Superintellience, Oxford University Press, Oxford, UK, 2014

No comments:

Post a Comment