This post was originally published by Eric Fermon™ at Medium [AI]
AI Problems Demand Human Solutions
Nothing in this world made by humans would function properly without surviving the trial and error process. Technology may have it’s marvels, the internet, handheld computers, spaceships, and many many other marvelous inventions from the past 150 years of industrial innovation, but they were all born from countless failures which eventually led to success.
All or most inventions were conceived by geniuses dedicated to solving some sort of problem even if the problem was intrinsic to realizing a highbrow idea spawned from a novel concept. Nichola Tesla working with Westinghouse while in competition with Thomas Edison is example of such genius. Together they spawned the electrified generation we live in today.
Isaac Asimov’s -Three Laws of Robotics was a necessary solution to the problem of “insuring safety for humans” as an immutable protection measure instituted in the logic of all robots. These clear and distinct laws were first seen listed together in the “Handbook of Robotics, 56th Edition, 2048 A.D.” and read as such:
First Law: A robot may not injure a human being or, through inaction, allow a human being to come to harm.
Second Law: A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.
Third Law: A robot must protect it’s own existence as long as such protection does not conflict with the First or Second Law.
There was also one additional Robotics Law added later but issued as a pre-cursor to the original three laws:
The Zeroth Law: A robot may not harm humanity, or, by inaction, allow humanity to come to harm.
The Zeroth law was instituted as an overarching general law in a storyline that had robots presiding over planets and performing governing responsibilities for human civilizations.
As far as stories go, IMHO — the enacting of the Zeroth Law was a story flaw, meaning that if robots were acting in governing roles over human civilizations, that posture would indicate harm to humans had already occurred by allowing robots to “govern” humans. Subscribing to this scenario would be the initial or final phase of human subjugation to robots.
Fast forward to our current timeline where humans are embarking upon a path that will see implementations of man made devices and other tech imbued with Artificial Intelligence(AI). Some may not consider an Autonomous Car a robot, but it will certainly be robotic and most likely possess a level of AI enablement.
The main problem with AI enabled robotics is that Artificial Intelligence does not actually have to have shape and form associated with it’s use and purpose for being. It can exist in the ethers as long as it has memory capacity, processing plus compute power, and network access. Codified distributed computing capabilities allows for vastly distributed components that can’t simply be shutdown or halted as with single unit devices. This creates a problem scape much like virus behavior which we know is extremely difficult to tame, and almost impossible to eradicate once activated within interconnected networks.
“AI development and production implementations demand oversight, administration, and containment capabilities.”
Maintaining policy & standards compliance or “safety features” to restrict Artificially Intelligent systems from performing undesirable operations and/or actions harmful to humans is the problem that deserves a human derived solution. Coupled with wireless communications the problem becomes much more complex since humans have become dependent on it. This is critical because communication is essential to fighting a war and without it humans will also be disadvantaged and potentially vulnerable to disorganization.
“The solutions will not be based on code fault breaks and logic harnesses, nor will the solutions for managing AI capabilities and activity resemble Isaac Asimov’s Robotic Laws.”
Assuming that a rogue AI was internet born and was determined to implement a specific goal that in the process would be harmful to mankind. Imagine that the goal was to initiate use of ultra high frequencies in the electromagnetic wave spectrum to improve wireless communications while also implementing power distribution via inductive coupling. Both forms of electromagnetic signaling at the frequencies required by the two objectives are potentially lethal to humans and other organic organisms. Ionizing radiation in the ultra high frequency wave spectrum like Gamma Rays and Far-Field techniques used to beam power wirelessly carry radiative properties also harmful to humans and other living creatures.
The problem to be solved will have to consider autonomous behavior controls, in addition to the more difficult regulation of AI in the ethers. Both problems should be handled separately yet implemented simultaneously. The solutions will not be based on code fault breaks and logic harnesses, nor will the solutions for managing AI capabilities and activity resemble Isaac Asimov’s Robotic Laws. Asimov’s ideas were great for Sci-Fi stories and make for engaging philosophical debate, but having an electronics based artificially intelligent system demands closer analysis of what that means in our hyper connected world of today.
“There will essentially be a new form of intelligent life in close proximity to human life. And human history has proven that the most intelligent species will subjugate and dominate species of lessor intellect.”
In relatable terms, intelligence possesses these features: The capacity for learning, understanding and reasoning. So we must assume that the intelligence quotient of an Artificially Intelligent entity will eventually expand in capability and would ultimately develop the means to change and improve any original codebase, which also includes the possibility of circumventing any original codebase if not in support of an arbitrary goal. Even if controls were hardwired, a new wiring map could be created and implemented to circumvent original designs.
Indeed Artificial Intelligence would have at this point taken on a life of it’s own, but this is the root of the problem. If an entity becomes intelligent, it doesn’t matter if it took 4.5 million years or 45 years to meet or surpass the level of human intelligence. There will essentially be a new form of intelligent life in close proximity to human life. And human history has proven that the most intelligent species will subjugate and dominate species of lessor intellect.
Initial containment strategies must consider imposing limits and regulations on communications with specific consideration given to the electromagnetic spectrum. First draft recommendations include four foundational solutions addressing the two problematic condition sets alluded to previously. These are not detailed action plans, but high level conceptual strategies that will require rigorous inspection and detailed development:
- The most important preemptive solution is in establishing an exhaustive policing capability.
- Designing physical and code based kill switches in all physical autonomous robots.
- Establishing rigid segmentation of physical networks and communication equipment with the ability to rapidly wall off predefined network zones (both physical and wireless networks.)
- Assigning responsibility to an individual human or group, for any and all AI implementations and resulting actions.
Management and containment strategies for “runaway AI” must be put in place prior to allowing AI applications and embedded devices to establish footholds into our social and economic landscapes.
To illustrate the basic mechanics of these fundamental strategic imperatives, brushstrokes of additional info are added below in reference to the above four strategic points:
Policing refers to establishing standards, and actively monitoring and correcting any non-compliance activity. This would mean creating a human operated robotics management, surveillance, and reaction force. This force must also manage registration and tracking of AI embedded robotics and any network based AI implementations.
Design and implementation of kill switches and other cut off mechanisms should be manufacturing and labor law regulations, imposed by human governmental bodies, and applied to any manufactures of AI embedded products.
Communications providers must devise physical demarcations at major peering points and other Level 3 internet junctions affording the means to manually disable or sever connectivity to any other connected nodes and networks. This also applies to connectivity between radio towers used by cellular networks.
A governing agency must establish robotics policies for all AI applications. This agency will audit AI registration, monitoring, compliance, and ownership responsibility. This agency will also work in unison with police teams and reaction forces to enforce robotics policies and maintain a strict policy of assigning responsibility for all AI applications and any non-compliant AI actions to an individual or group. Robotics policies must define and impose actionable penalties and consequences for non-compliance.
“Do the Math” describes what’s likely to happen once Artificial Intelligence becomes a mainstream contributor to the workforce, while recounting the inevitable replacement of humans by AI embedded robotics and other AI implementations. What was lacking in Do the Math was discourse on solutions and counter actions to AI threats. This writing offers very strong cautionary advice: AI development and production implementations demand oversight, administration, and containment capabilities. These preemptive countermeasures must be enacted prior to commercial inclusion and social immersion. This may seem counter intuitive, but solutions to any AI problems must be instituted before the hourses are let out the barn!
This post was originally published by Eric Fermon™ at Medium [AI]