The primary goals of AI research include the application of expert knowledge, environmental perception, manipulating physical objects and natural language processing.
The approaches used to accomplish these goals include computational intelligence, machine learning, statistical methods and traditional symbolism. The tools that computer scientists use to implement these approaches include mathematical logic, search optimization and other methods based in a particular field of study. In addition to computer science, AI draws on other fields such as linguistics, mathematics, neuroscience, philosophy and psychology.
AI research is based on the concept that human intelligence can be described sufficiently precisely that a machine can simulate it. The question “What is artificial intelligence?” raises strong debate on the nature of the human mind and the ethics of endowing machines with human-like intelligence, which have been explored since antiquity.
The algorithms for difficult problems require great computing resources, especially processing speed and memory. These requirements typically experience exponential growth, meaning that many AI problems require more computing resources than are currently available. Increasing the efficiency of algorithms is therefore a high priority in AI research.
Many problems that AI machines are expected to solve require extensive knowledge of the physical world, making knowledge representation an essential element of AI research. For example, researchers must be able to represent objects, their properties and their relationships with other objects.
AI experts generally consider a 1956 conference at Dartmouth College to be the beginning of AI research. Attendees of this conference included Marvin Minsky, John McCarthy, Allen Newell, Herbert Simon and Arthur Samuel, who became leaders in the field. These researchers wrote programs that appeared to simulate human intelligence, with feats such as playing checkers at a competitive level, solving algebraic word problems and speaking English.
The Department of Defense (DOD) was heavily funding AI research by the mid-1960s and had established laboratories throughout the world. Herbert Simon even predicted the problem of creating AI would be substantially solved within 20 years, allowing machines to do anything humans could do.
However, this early optimism failed to appreciate the difficulty of accomplishing some of the remaining tasks of AI. Progress in AI began to slowdown, and the U.S. government began canceling the funding for research in the mid-1970s. Well-funded AI projects were virtually non-existent from this point until the early 1980s, a period now known as the first AI winter. The commercial success of expert systems simulating the analytical skills and knowledge of humans revived interest in AI research in the early 1980s, resulting in an AI market worth over $1 billion by 1985.
AI was used in new areas from the late 1990s to the early 2000s, including data mining, logistics and medical diagnosis. These successes were primarily due to the rapid increase in computing power and a greater commitment to developing standards and methods for AI systems. A chess-playing system beat a current world chess champion for the first time on May 11, 1997 when Deep Blue defeated Garry Kasparov. Watson, IBM’s question-answering system, easily defeated world Jeopardy champions Ken Jennings and Brad Rutter in 2011. AlphaGo was the first AI program to defeat a Go champion without a handicap, when it won four games out of five against Lee Sedol.
Google rarely used AI projects in 2012, but this usage increased to over 2,700 projects by 2015. Its error rates for processing images also fell significantly during this period, primarily due to the decreasing costs of developing neural networks. The greater availability of affordable neural networks is generally attributed to the increase in cloud computing and other AI research tools. Other recent examples of AI’s use in mainstream computing include Facebook’s system for describing images to blind people and Microsoft’s Skype system for automatically translating languages.
Major artificial intelligence companies currently include Amazon, Facebook, Google, IBM and Microsoft, who created a non-profit partnership in October 2016 for establishing best practices in AI technology. Apple and other tech companies also joined this partnership in January 2017. This partnership will conduct research, provide thought leadership and create material to educate the public on AI technologies. Partners will also make financial contributions and recruit members of the scientific community to the board.
Software must run on some type of hardware or software framework, which is commonly known as a computing platform. A platform affects the software’s AI capabilities just as much as the software defines the platform’s AI features, meaning that AI problems must be solved on real-world platforms rather than an isolated environment. Platforms specifically designed for AI software include the expert system Cyc and the robot platform Roomba. AI Software libraries such as Deeplearning4j, Torch, Theano and TensorFlow have become possible due to recent advances in distributed computing and neural networks.
Artificial intelligence applications often no longer considered to be AI once they enter mainstream use, a phenomenon commonly known as the AI effect. Well-known examples of this effect include search engines such as Google and online assistants such as Siri. Autonomous vehicles such as self-driving cars and aerial drones are also common applications for AI. Other examples include image recognition, spam filtering and targeted online advertisements.
The healthcare industry has provided some of the most recent uses for AI. More than 800 drugs for treating cancer currently exist, which makes treatment selection more difficult for doctors. Microsoft has developed a system that assists doctors in selecting the best treatment for a particular cancer patient. It’s also developing an expert system called Hanover, which memorize all of the research papers related to cancer treatment. This will help predict the most effective combination of drugs for cancers such as myeloid leukemia, a cancer whose treatment hasn’t improved significantly in decades.
The automotive industry has also seen recent advancements due to AI systems. More than 30 companies have developed driverless cars as of 2016, including Apple, Google and Tesla. These vehicles use AI in a variety of functions including braking, collision prevention, lane changing and navigation. A computer integrates these functions into a single driving system, which can be pre-programmed with a map of the local area. This map provides data that gives the vehicles an awareness of their surroundings, including street light locations and curb heights. Google is also developing a system that can adjust to a new surrounding without the need for a pre-programmed map.
Researchers are developing short-term and long-term goals that can help to predict the course of AI technology. For example, researchers at the Future of Life Institute have developed some short-term goals to predict the ways in which AI will affect the world’s economy, legal systems and ethics. Their long-term predictions generally focus on optimizing AI’s functions while minimizing the security risks posed by AI.
The IT research firm Gartner, Inc. predicts that the trend towards practical applications for artificial intelligence projects will drive lasting business changes for both consumers and enterprises. For example, Gartner predicts that machines will write 20 percent of business content by 2018. Automated composition engines will foster the transition to machine-generated content by using data and analytics to perform natural language writing. The best candidates for this use of AI include legal documents, market reports, press releases and shareholder reports.
Gartner also predicts six billion devices will be connected through the Internet by 2018. The increasingly blurred distinction between the digital and physical world means that all connected devices will be treated as customers, resulting in a greater need to use AI for fulfilling service requests. The strategies for responding to these requests will be distinctly different than those used in traditional human communication.