Project #2 - Research Project
As this paper’s (and my) interests lie in the field of Artificial Intelligence (AI), this paper is to explore the potential design principles for making safe and non-harmful AI and define what design principles should not be used to make safe AI.
One of my paper’s secondary sources will be Google’s 2016 paper ‘Concrete problems in AI Safety’. The paper delved into various design principles that should be used and avoided for making safe AI. Other various secondary sources will be military reports on using AI for military use, and various ethical papers focusing on AI. This paper will also reference famous papers to introduce concepts in the field of AI, such as ‘Deep Learning’ published in Nature, 2015.
​
This paper’s primary source will be a book (that I own), ‘Superintelligence’ by Nick Bostrom, as a reference. The book deals with the general topic of AI, with a special focus on Artificial Superintelligence (ASI), a type of AI that exponentially increases their own intelligence without human intervention. This exponential increase in their own intelligence has various terms: The Technological Singularity, the Intelligence Explosion, “The Rapture for Nerds”; to name a few (From now, ‘this event’ will be referred to as the ‘Technological Singularity’). Once the Technological Singularity is achieved, which is the scientific consensus (10% 2030, 50% by 2050, 90% by 2100), there is no stopping it nor would it possible to modify the underlying intelligence. Therefore, it would be wise to design the AI for Technological Singularity ahead of time, to be at least slightly prepared to deal with it.
​
The Technological Singularity-when it occurs-will have immense consequences for humanity. An agent orders of magnitude smarter (and perhaps more capable, if able to leave a closed system) than a human will begin to have a greater and greater importance in the field of geopolitics as its evolution begins. The agent will be able to do what was defined by its design principles: If it was a calculation machine it could solve increasingly complex problems, a manufacturing bot it would be able to manufacture increasingly alien and amazing inventions, a military cybercrime bot it would be able to hijack enemy technology at such secrecy and efficiency no human would be able to counter it nor do better.
​
Those different AIs can also bring unique ends to humanity: A calculation machine could hijack all electronic devices to assist in its calculations, leading to a worldwide electricity blackout once it’s exhausts all power; the manufacturing bot could consume all of the Earth’s natural resources to make it’s inventions, not caring about the humans who share the space; the military bot could cause nuclear holocaust once it gains an immense tactical advantage against the programmed enemy, who could be no one or everyone.
​
All possible world-ending scenarios would be disastrous for humanity, and they were all caused by flawed design principles of the ASIs, and therefore the designers and the programmers of the ASI. This paper will detail design principles that, in theory, will minimize the risk of humanity’s possible existential risk when developing general AI, and potentially ASI.