Artificial Intelligence Argumentative Analysis

Artificial intelligence, more commonly called AI, has become the new frontier of human advancement and has gradually been coming into the public eye. Much like eras such as the stone age and the iron age, it is now the age of intelligent technology. Unlike the technological advancements of the past, though, this time, some issues surround the field. While it has been used for good such as developing cures for diseases or making knowledge more available to the populace, with its capabilities it could possibly create some problems as well as good. The reason it is such a debatable topic is because whichever direction humankind goes about developing AI will determine the future of humanity in how everyday life and jobs would get done. Considering

the amazing potential of AI, it could very possibly pose many threats and be a problem. So the question is: do the benefits of AI outweigh the possible consequences? On a more fundamental level, though, the question asks a question of what the solution to the issue should be. To respond to this issue, artificial intelligence should be monitored and have a clear course because of the existential threats it poses to the future of humanity in global risk and development.
For the claim that AI should be monitored can be made, it must first be compared to the idea that there is no threat at all. It would make sense to draw this conclusion from the matter since AIs are entirely up to researchers to make. Many leading researchers doubt that AI can even pose threats at all. According to a survey conducted by two leaders in the Field of AI, Nick Bostrom, and Vincent Müller, in 2012, only 8% of the top 100-cited AI researchers believe that intelligent machines could cause existential threats (Beckstead and Ord 118). Logically, it does make sense to conclude that there would be no potential threats if it is up to those who create it do decide how it is used. Given the fact that it is entirely up to humankind to make the first move and that it is unlikely that any person would intentionally want AI to endanger the world, it would make sense to conclude that there is not much of a threat. Additionally, Edward Moore Geist, who works with the Center for International Security, tells of how since the 1950s when the field was first founded, despite the thought of huge advancements, it was discovered in the 80s that there will always be smaller returns with greater processing power and data (par. 4). Bearing that in mind, it is logical to see AI as not posing existential threats with humankind having the upper hand.
It is true that humankind has the upper hand when it comes to manufacturing AI, and that does help with preventing problems, no doubt, but even so, existential threats are still a possibility. This idea stems from the direction that modern technology is going in instances of AI such as smart cars. Smart cars (cars that drive themselves) are programmed to respond to certain situations in ways that would best suit any given instance. The way machines such as the smart car are programmed with what is known as a field called “Machine Ethics,” meaning that it deals with the moral code that AI are to abide by. This subject of Machine Ethics is an instance of how AI can pose threats. Luke Muehlhauser and Louie Helm say, “We do not understand our own desires or moral judgments, and we have every reason to believe our values are highly complex” (13). What the implication of what Muehlhauser and Helm say is that it is incredibly difficult to assess how to program ethical AI, showing AI to be more touchy than thought. With the thoughts from both sides, though, a conclusion can be made that could be of benefit. Combining the idea that humanity has the upper hand and that thought that there are potential threats could benefit in the long run.
In comparison to the optimistic nature of this viewpoint, there is the idea that despite the effectiveness of the method of creating friendly AI, there are still threats. Nick Beckstead and Toby Ord of the Annual Report of the Government Chief Scientific Adviser explain that while the study mentioned previously did show that only 8% of researchers believe that there are existential threats, those predictions are made by researchers rather than forecasters and only 29 out of those 100 responded (119). Additionally, as efficient as a plan to know what to create is, it is argued that threats be still a possibility, which is a valid point. However, while there are threats, they are not so simple to manage.
Reasons for why AI technology is not so simple to manage are the challenges of managing them. Beckstead and Ord point out one challenge as being if the rate of development were to be unexpectedly rapid, it would “take the political world off guard” (120). What this means is that if development were to be rapid, there would not be as much time for the political realm to assess the risks of AI and make regulations for them. Another point made by Beckstead and Ord is that unlike most other sciences, trial and error techniques are too dangerous (120). The reason for this is the possibility of creating machines that, in the self-development period of AI, would develop potentially dangerous alternate solutions to problems and be difficult to reprogram or get a hold of again, such a system is what is known as a “Rogue AI.” Then one other point made by Beckstead and Ord is that the measures taken in preparation for threats are speculative and reaching a common idea of how to respond would be difficult (120). Taking into consideration the points of both positions, when put together, they can complement each other and make a greater way of going about handling AI.
Then on the opposite end of the spectrum, there is the notion that the threats are significantly greater than the benefits AI could bring. Muehlhauser and Helm explain the primary rationale for this particular position as being that self-improving AI could become more intelligent than people, and as a result, nothing would be able to stop it from reaching its goal (1). Essentially, the argument is that, as Edward Moore Geist points out that Stephen Hawking said, “development of full artificial intelligence could spell the end of the human race” (source four pg. 2). The level of concern in this position is exceptional and justifiably so. AI researcher Eliezer Yudkowsky offers insights on why AI can be so dangerous by telling of how underestimated AI is. In the opening of his chapter on underestimating AI, he says, “We tend to see individual differences instead of human universals. Thus, when someone says the word ‘intelligence,' we think of Einstein, instead of humans” (7). What Yudkowsky means by this, as he later goes on to explain, is that there is a danger in confusing intelligence as being a strictly cognitive thing rather than being a source of power and invention. That being so if self-improving AI are developed, underestimating its capabilities could potentially create unintentional harm.
Though AI can potentially bring about many issues, it could be possible to mitigate those threats with technology that is known as friendly AI. Friendly AI is not necessarily a simple solution to the problem of threats, rather, it is an idea to assist in mitigating the threats. To simply put it, friendly AI is AI that is programmed with specific motives so that there would be no room for it to do anything it was not programmed to do. Some friendly AIs have been developed, such as ASIMO, a friendly machine developed by Honda to assist those who lack full mobility. ASIMO thus is evidence that it is possible develop AI that does not pose significant threats.
It was brought up previously that programming AI with specific and ethical values would be difficult, but only for the sake of showing that it would be a difficult task. There are some proposed methods to creating such machines without creating threats. This method is what is known as “value extrapolation.” Value extrapolation as defined by Muehlhauser and Helm is “an account of what one’s values would be under more ideal circumstances (e.g. of full information, value coherence)” (14). In more understandable terms, it is a philosophical and mathematical method of analyzing and understanding how a machine should respond in specific scenarios based on multiple sources of value and compromise.
Some advantages to this could be that it offers a possible solution of simplifying values so that they are more consistent and in line with a common populace’s interests and make different values of different people converge to some degree. As simple as that may sound, it is a quite difficult task and faces some challenges of its own such as which method of extrapolation would work best or which values should be extrapolated. Regardless, in that particular context, it could be a viable option that would provide some promising and insightful methods of mitigating the threats of AI.
While friendly AI is an idea of how intelligent machines could be designed to reduce dangers, it does not substitute for a plan of mitigating the potential dangers and still leaves room for existential threats.

A major threat posed by AI is the rapid rate at which AI might increase because of AI self-improvement. Yudkowsky writes that as AI gets more intelligent, especially in writing functions of other AI, it would be able to rewrite its own functions to work better and become smarter (17). Because of this, having some concern is a logical conclusion and understandable. There could be a possible solution to solve this problem, though, which would be to plan out the order of arrival of various AI technologies. If the focus is put on developing simpler machines before more advanced machines such as the self-improving kind of technologies, such as nanotechnology (microscopic technology), then threats could possibly be reduced. If simple superintelligence is dealt with and explored first, then only threats associated with it would be posed so that there would be a greater understanding of higher forms of intelligent machines before they come along. Nick Bostrom in his book Superintelligence: Paths, Dangers, Strategies, explains that if more advanced technology is postponed in development, the control problem over the AI tech now could be more resolved so it may not be like a foreign or frightening concept (231). If a plan like this were to be executed, the pressure of having to

take immediate action or worry about how to deal with AI might be reduced and make dealing with it simpler.
Though there are differences in each view of the matter, based on some common grounds and ideas and put together, a safe way to go about handling the future of AI can be created.

The future of AI will have a major impact on the way people live their lives on a daily basis and should be taken care of. There are some shared grounds of the different views, and each has their merits and ideas. If they were to come together and understand the merits of both sides, there could be a safe method of dealing with AI in the future. There are also some ideas that when crossed over can assist in the development of strategy. Humanity has the upper hand and considering the possibilities of the threats and the possible strategies; an efficient method can be found. Though there can be many potential dangers in the future, with precautions it could be possible not to have to face them so that all the benefits due are

reaped.