Current status and future prospects of artificial intelligence technology
The current state of artificial intelligence technology
1) History of artificial intelligence
Although opinions vary about the history of artificial intelligence, artificial intelligence as a technology proposed the concept of a 'thinking machine' by Alan Turing in 1950 and the 'Turing Test', a method of evaluating it. It is accepted as the orthodox that the term 'Artificial Intelligence' began in earnest after the birth of the term 'Artificial Intelligence' by John McCarthy in 1955. Since then, the 'spring of artificial intelligence', which was followed by excessive expectations and investment, and the 'winter of artificial intelligence' due to technological limitations and market failures, recurred, and artificial intelligence became a reality from the concept close to mankind's goal of building a Mars colony or conquering cancer. The paradigm has been developed as an attainable technology. In 1997, IBM's Deep Blue beat chess champion Gary Kaasparov, and in 2011, IBM's Watson, also known as IBM's Watson, won the most streak champion in the popular quiz show, Jeopardy!, Ken Jennings. By defeating Ken Jennings and top prize winner Brad Rutter, expectations for artificial intelligence have once again risen, establishing itself as a leading company in the field under the term Cognitive Computing. Finally, in 2016, the artificial intelligence algorithm 'AlphaGo' of Google's subsidiary DeepMind in Baduk, where human intuition and reasoning skills are considered to be the most dramatic, is Se-dol Lee, one of the world's best Go articles The overwhelming defeat of people around the world is taking place in Seoul, the capital of Korea, and artificial intelligence is rapidly emerging as a social term beyond technical terms. What AlphaGo symbolizes is not that mankind has overcome the solution to a complex problem that has been developed for thousands of years through high-performance hardware-based computation, but is an area that has limited ability to judge and plan high-level situations that were thought to be unique to humans. Is that it was confirmed that it can be learned in a short time through learning. Now, beyond Go, where all information is shared, only part of the information can be observed, and although it is a limited environment in poker ([1]) or real-time strategy games that require a high level of strategy, it is approaching the level beyond the human champion ([2]).
2) Status of artificial intelligence technology
New York University professor Gary Marcus is currently artificial among intelligent individuals' abilities: Perception, Language, Reasoning, Analogy, Planning, and Common Sense. It is argued that the approach of intelligence technology to human abilities is still only an imitation of the ability to perceive, and cannot lead to further understanding [3]. However, awareness is the most basic and essential ability for intelligent entities to collect information. Even if machines are equipped with only these capabilities at the level of humans, the ripple effect can be beyond imagination. For example, in the case of image recognition, as Chinese facial recognition companies Sense Time and Megvii are growing explosively, the commercial value is very high in payment, security, and security, and real-time image recognition in autonomous vehicles. Technology is positioning itself as a core technology. In addition, the video-based search installed in Google Assistant and Google Photos is fundamentally innovating the existing keyword-based search and redefining the method of e-commerce or media consumption. In the case of voice recognition, which is another representative recognition problem, it completely innovates the method of using electronic devices by itself, and in connection with natural language processing technology, real-time translation, semantic analysis, question-and-answer system, or the recently introduced restaurant reservation service, Duplex. It leads to the development of the same dialogue and secretarial system, dramatically improving information accessibility and user convenience. Cloud service companies such as Google and Amazon provide this recognition technology based on machine learning as Algorithm-as-a-Service or AI-as-a-Service (AaaS). It is playing a role as a driving force behind the new sales pipeline by giving new value to the cloud storage and computing resource provision services that were provided as PasS).
1The state of artificial intelligence in specialized fields
a) Status of expert assisted artificial intelligence
Most of the examples of artificial intelligence technology described above were focused on productivity improvement that automates or streamlines simple and repetitive tasks by implementing the most basic human ‘recognition' intellectual ability with machines. The proliferation of artificial intelligence is rapidly progressing in the area of experts who obtain qualifications through experience. In the case of law, which is a representative area of expertise, artificial intelligence is used to automate specialized tasks such as analysis of legal data and precedents, such as Blackstone Discovery in the United States, and Automated Insight introduced by AP in the media area. ) Algorithm or an artificial intelligence reporter of Narrative Science introduced by Forbes, is sending an article of quality that cannot be distinguished from a professional reporter in real time [4].
b) Status of medical artificial intelligence
Among these specialized areas, the field in which the introduction of artificial intelligence is the most active is the medical and healthcare areas, among which the development and introduction of artificial intelligence technology is accelerating for the diagnosis of diseases based on medical images or biological signals. While the amount of medical images and bio-signals for diagnosis is explosively increasing, the current nurturing system for medical staff who need long-term training and clinical experience cannot meet all these demands, and diagnosis based on training and clinical experience among medical staff. The high inconsistency of results contrasts with the consistency, diffusion, and accuracy of artificial intelligence, increasing the necessity of introducing medical fields. Accordingly, the US FDA approved the world's first cloud-based artificial intelligence medical imaging diagnostic assistance solution in January 2017 ([5]), and in April 2018, artificial intelligence independently developed diabetes based on retinal images. The artificial intelligence solution for determining whether to refer to an ophthalmologist for sexual complications has been approved ([6]). In Korea, Buno was the first in Korea to receive approval from the Ministry of Food and Drug Safety for artificial intelligence-based pediatric bone age reading software through clinical trials ([7]), followed by Lunit and JLK Inspection, respectively, with chest X-ray-based pulmonary nodule detection assistant software. It was approved by the Ministry of Food and Drug Safety as a brain MRI-based cerebral infarction diagnostic aid software.
2 Challenges for clinical application of medical artificial intelligence
a) Regulatory taskThere are still problems to be solved in order for medical artificial intelligence to be introduced and used in the clinical field in earnest. Technically, issues related to the explanatory power of the deep learning model used as the core of the current artificial intelligence technology, quantification of uncertainty about the result, response to hostile data, and security remain, and the diversity of artificial intelligence-based medical solutions There is a limitation that this is low and the level of integration and linkage with the existing hospital system is low. There is still a need for regulatory improvement, but Korea has to go through a multi-stage process such as medical device approval, new medical technology evaluation, and insurance benefit decision, and the cost and time required for this are very large. In addition, since the introduction of artificial intelligence medical devices and business outcomes are determined by whether a fee can be recognized by passing a new medical technology evaluation in a single insurance system, there is no choice but to face great uncertainty even after technology development. Recently, the Ministry of Health and Welfare has stated that if the role of the medical artificial intelligence solution is only to improve the convenience of medical staff or improve hospital productivity, it is a principle not to apply health insurance. As evidence should be prepared, long-term research and follow-up observations are required, which makes it expected that active clinical introduction and commercialization in a short period of time will not be easy.
b) Ethics and legal issues
The daily use of artificial intelligence in the professional field raises various ethical and legal controversies. In particular, in the case of a medical diagnosis and prescription, if the patient's health or quality of life is seriously affected by a misdiagnosis or incorrect prescription of artificial intelligence, the responsibility is currently with the medical staff who made the final judgment using the artificial intelligence. However, if the AI is applied routinely and used without doctor's intervention, the developer who developed the AI cannot be completely free from responsibility. In parallel, various discussions and social consensus are still needed on who and how much of the AI developer, the medical staff who are users, and public and private insurance will share the liability for damages in case of medical accidents caused by artificial intelligence.
On the one hand, voices of concern that the proficiency of medical staff will be hampered by the increased dependence due to the introduction of medical artificial intelligence are also rising. In a situation in which ECG reading is replaced by a machine or diagnosis using a stethoscope is disappearing due to the introduction of new technology, medical staff in the future will be able to perform diagnosis or procedures that depended on artificial intelligence in situations where artificial intelligence does not work or malfunction. Whether it can be done should be considered in terms of education and ethics of medicine.
3 Future prospects of artificial intelligence
a
) Artificial intelligence and human coexistence
As mentioned earlier, artificial intelligence technology has already reached deep into our lives through various services and products, and in the future, almost all data-driven industries will be automated and optimized by artificial intelligence technology. In addition, not only the innovation of the traditional industry, but also the field of experts that require a high level of intellectual ability and field experience is not unfamiliar with artificial intelligence replacing or collaborating with experts
Come on. However, this rapid proliferation does not mean technological completion of artificial intelligence. Rather, it is possible that most of the existing intellectual activities performed by humans can be implemented using a large amount of data, sufficient computational amount, and a machine learning model that can be trained efficiently. It is appropriate to see that it has been confirmed. The artificial general intelligence, which is pursued by Elon Musk and Stephen Hawking, such as Super AI, which will threaten the survival of humanity, or Strong AI or Deep Mind, is still It is difficult to achieve ([8]), and the current weak AI (Weak AI) or narrow AI (Narrow AI) has a lot of room for improvement in terms of computational efficiency and ease of learning. Currently, the most efficient way to overcome the limitations of artificial intelligence is through the harmonious division of roles and collaboration between experts in the field and artificial intelligence, and building a user environment that enables such collaboration efficiently is an artificial intelligence product or It will be the core competitiveness of the service.
0 Comments