by With AI, the goal of developing a particular program can often change to a goal of solving a problem in a specific way, to an exploration of what can be done. This is an appealing factor that has led to AI researchers often pushing the boundaries to see what AI is capable of. Throughout the many strides that AI has taken, it has at times fallen short of expectations and at times has been quite successful in the goal to provide intelligent behavior for a broad spectrum of issues. Always, the feasibility of what can be done varies, and thus the scope and success of any given intelligent behavior on an issue often depends on what is possible.
The changing world and the varying needs that result from the changing dynamics mean that the types of AI that are useful also change. AI has traditionally focused on areas such as game playing, expert systems, fuzzy logic, neural networks, and intelligent agents. During this time, our motivation has been twofold. In all of these areas, we are seeking to understand the nature of intelligence and trying to provide computers with some of the useful abilities that the intelligent behavior exhibits. Our focus, however, has changed from areas that are more theoretical with little practical application to areas with much opportunity for positive impact. We would like to see more research conducted on AI software based on available data that provides possible solutions to current problems in education, health, environmental sustainability, international development, national security, and human security. The changing world has also caused some to consider different types of intelligence and adaptive behavior for AI programs, which must be a consideration in the feasibility of various programs.
We begin with the issue of intelligence. We see AI as the study of the ideas that define intelligence. An AI system is one that acts in a way that, if a human were so acting, we would say that the human is intelligent. While trying to understand the behavior of intelligent humans, AI researchers are also interested in whether these behaviors can be produced by machines and in the endeavor to produce machines that are able to exhibit these behaviors. Often, intelligent behavior is taken for granted in AI research, and without concrete definition, there is no way to determine if a system is intelligent or not.
In this Article, we consider the nature of AI systems, the tasks for which they are designed, and their surrounding environment. We attempt to clarify a view of AI and indicate how its different aspects contribute to our goal. AI researchers are changing the goals and scope of their work. As we learn more, we continually expand the areas for which AI can be applied. As the field of AI has changed through the years, the feasibility of various AI programs has changed. Artificial intelligence can be viewed from a variety of perspectives. Some AI researchers think of AI as the construction of rational agents that can perform any intellectual task that a human can do. Others view AI as the development of computer programs for any task, while some others at times narrowly define AI as “the attempt to make computers do things which at present are done better by humans.”
2. Current Applications of AI
Most well-informed consumers realize that AI is embedded in their day-to-day world in the form of machine learning. Often, the concept of machine learning is associated with data mining, which consumers do not see but has shaped their experience. As consumers are well-versed with machine learning, they develop increased trust in the results being more accurate. A case in point is Google products. A study with the aim of comparing the accuracy of an automated hypertension risk prediction system to that of an expert concluded that the automated system was more successful at prediction.
Environments that are dangerous for humans are problematic for AI to explore. Whether it’s a nuclear power facility, space, underwater, or simply the battlefield, the ability to do something with high risk and no human casualties is very attractive. For instance, if a robot can disable a series of roadside bombs in Iraq with no risk to human life and minimal cost, the economic and strategically beneficial implications are evident.
Safety and Security: AI has been deployed by the US Department of Defense for predicting events that will lead to death and injuries, and the Australian Defense Force has used robotic planes to clear paths through minefields. AI is applied for homeland security and is high on the agenda for many governments with the proliferation of global terrorism. A recent and very specific example was the use of artificial intelligence (link is to a PDF file) to look through the vast amount of intercepted communication data by UK intelligence to determine if any of it was related to the kidnapping of peace activists in Iraq.
3. Limitations and Challenges of AI
At some point in the future, AI will surpass the best of human abilities across a range of skills requiring years of training. The prospect of such superhuman skill has raised fears of a sudden unemployment crisis, as a general purpose AI could automate any task that has a limited scope and a predictable solution. This in turn would moot the comparative advantage of human over AI labor in any skill areas. Measures of intelligence such as the IQ test are a prime example of such a task. It may be possible to use narrow AI to perform better than humans at an IQ test, without solving the task of general intelligence or building skills that would let the AI do anything else. This “race to the bottom” in the job market would also impede new human skill development. If human labor in a certain area could be replaced with AI, it would become harder for a human to enter that area and build up skill. This would compound the unemployment problem and could add to political pressures to provide some sort of guaranteed income for people whose skills have been replaced by AI.
Despite the tremendous advancements in AI research and technology in the past few decades, the “reality check” of strong AI is that R&D in this area is still at a very early stage. Some critics have found it more useful to pursue weak AI applications, such as the Desert Tortoise robot Gaui, which seeks out and removes land mines. Some have questioned whether strong AI can be achieved, at least in the near future, and have suggested that the resources allocated toward this end might be more effectively applied to other projects. Finally, the prospect of superintelligent AI has raised a different set of concerns. A super intelligent AI, while still only a narrow AI, could exceed the ability of humans in every skill area, leading to a replacement of all human workers, and a general intelligence that could solve many of the world’s problems to a degree that is beyond human capacity. This has the potential to make current day humans obsolete, as a new species or a timeless human-machine hybrid could emerge. This possibility is seen as something that could eventually lead to a human extinction event, unless AI’s goals align with humanity’s, in which case our impact to the universe could potentially be leveraged by the AI (a theme obliquely addressed in Bostrom’s book).
4. Future Possibilities of AI
The next question to arise is, if an AI were to be created, could it surpass all intellectual activities of man? This is the question of whether or not AI can have what is called super intelligence. Could it create a machine that could create another machine more intelligent than itself? This idea is logical in the sense that if the smarter machine were able to solve problems faster and with more accuracy than itself, then technically it is surpassing its own capabilities. Logically this step is very far away and the possibilities are just speculation. However, it does bring about the question of what is the future of AI. It is rational to say that we need AI to automate work that is currently done by humans, and then advance further to needing AI to do the more difficult intellectual tasks. However, what happens if we get to the point where an AI can do any and all tasks better than a human? At this point, the AI would be the most cost-effective way to further development as there are no salaries, pensions, and vacations to pay for. This is essentially where one could claim that we have created a replacement for humans. This leads us into the topic of job displacement due to AI. Currently, there are many jobs being replaced by automation. This is a move to more narrow AI, but it is still effective. Many repetitive tasks can be more efficiently and cost-effectively done by a machine. In the short term, displacing labor with AI can have bad effects on economies and societies. If many low-skilled jobs begin to be replaced by AI, there is potential for higher unemployment rates as those who would have done those jobs do not have the skills for the new positions that emerge. This would create a division of unemployed lower class and a more educated higher class, as the new positions for developing AI and for maintaining and monitoring the work of AI would require higher education. The potential long-term effects are unknown, as the displacement of the lower class could drive them to seek education and fill the newer positions, but it is not known if there would be enough new jobs emerging to replace the lost ones. If there comes a time where AI is doing all tasks better than humans and it is cost-effective, it could result in a massive unemployment of humans. However, by this time, there is speculation whether or not we would have reached a state of singularity and whether the AI would be able to improve itself and make discoveries beyond the capability of humans.
5. Ethical Considerations
As AI becomes more advanced, we begin to consider the possible ethical considerations that come into question in relation to its capabilities. By creating a machine with the ability to imitate intelligent behavior, such as language and emotions, some of the responsibilities and duties of human beings are bound to be within the capacity of the artificial entity. This in turn poses the question of the extent in which an artificially intelligent machine should be considered as a person, morally and legally. Should a separate set of laws be implemented for an intelligent machine? If so, when will the point be reached when these laws are necessary? This could be difficult to determine if the advance in AI is steady, as what might not seem necessary one day could be apparent the next. On the other hand, there is the possibility that machines will exceed the intelligence of human beings. An advanced intelligent machine may be able to make valuable contributions to knowledge and culture. If a machine creates a piece of music, writes a book or a poem or any other form of original work, is the work to be considered as that of the machine’s or its human designer? This is just a few of the issues which will become more and more relevant as AI advances.
