In 2022, the AI boom started with the appearance of generative AI such as “Stable Diffusion”, “Midjourney” and “ChatGPT”.
In this article, to give an overview of what is said to be the “fourth artificial intelligence boom,” we will discuss what types of AI exist, what is expected, what are the concerns, and what AI will be like in 2023. Here is a summary of basic information.
What is “AI” in the first place?
Definition and a brief history of AI
“Strong AI” “Weak AI”
3 categories of AI
Will AI Surpass Humans?
Types of AI
Rule-based AI
Machine learning and deep learning
Other AIs
AI will realize a convenient and happy society
Concerns about AI
A most recent point of interest
Relationship with related fields
movement of the world
Summary: Beware of the “ELIZA Effect”
1. What is “AI” in the
Definition and Brief History of “AI”
What is “AI” in the first place?
AI (Artificial Intelligence) translates to “artificial intelligence/intelligence”. Professor John McCarthy proposed this term at a conference held at Dartmouth College in 1956, and AI was born as a new field when it was used as a term in the Dartmouth conference proposal. By the way, in 1958, Professor John McCarthy designed an easy-to-use programming language “LISP” for artificial intelligence research.
AI is a branch of computer science that studies intelligence using calculations and computers. Today, it is used as a term to indicate a technology that enables machines such as computers and robots to realize intelligent functions equal to or higher than those of humans at a level suitable for practical use. It can be rephrased as a technology or method that enables “intelligent” activities by computers and machines, but there is still debate about what constitutes “intelligent” and how. I’m here.
Currently, AI is partially being put to practical use as a computer system capable of performing advanced processing in fields such as natural language processing, image recognition, speech recognition, and predictive analysis, while research is also progressing. In particular, it is considered indispensable for the development of business, medicine, and agriculture, and is used in various fields such as manufacturing and transportation. Expectations for its importance and contribution to social development are increasing.
Short version: The history of AI
The concept of AI appeared in 1956. Initially, it seemed that it would be possible to realize it at an early stage, but that did not happen, and from the 1950s to the 1960s, research, and establishment of the theoretical foundation of artificial intelligence were promoted, and tools such as “LISP” were prepared. There were some impressive results. This period is sometimes referred to as the “first artificial intelligence boom.”
In the 1970s and 1980s, which is considered to be the “second artificial intelligence boom”, expert systems and other systems developed and are said to have reached a practical level. Difficulties and limitations are also clear. It was around this time that AI that could solve math problems and challenge chess and shogi appeared and became a hot topic.
From the 1990s to the 2000s, the application fields of AI technology expanded, and machine learning and neural networks developed. In 1997, IBM’s chess AI (supercomputer) “Deep Blue” made big news when it defeated world chess champion Garry Kasparov.
Deep learning, which is currently a hot topic, has developed significantly since the early 2010s and has triggered the attention of AI research. Especially since 2012 when “AlexNet” appeared, recognition spread rapidly, and it is said that the “third artificial intelligence boom” has arrived.
The “third artificial intelligence boom” has spread the recognition that model “depth” is essential for building high-performance AI. Deepening the depth increases the computational cost, but learning using a high-performance GPU results in a high-performance AI. The spread of this recognition has also accelerated investment in AI, and since then, we have seen major movements and surges.
In 2015, OpenAI Inc., which is attracting attention for its conversational AI “ChatGPT”, was established. From around 2020, generative AI (Generative AI) became a hot topic, and in 2022, “Stable Diffusion” and “Midjourney”, which generate images from the input text, were announced and became popular. It caught fire, and when “ChatGPT” was released in November 2022, it became a big boom all over the world. There are even articles describing the current situation as the “fourth artificial intelligence boom.”
With the emergence of “easy-to-understand” AI that anyone can use, it can be predicted that the generalization of AI will accelerate in the future, and AI will be used not only in business sites but also in all aspects of life. Attention is focused on how AI will permeate society and how it will be used.
“Strong AI” “Weak AI”
There are two types of AI: strong AI and weak AI. This idea was proposed by philosopher John Searle.
In fact, “strong AI” has not yet been realized and is a theoretical existence. For example, the computer “HAL” that appears in the sci-fi movie “2001: A Space Odyssey” (1968) and the villain “HAL” that also appears in the movie “Avengers: Age of Ultron”. I think Ultron is a “strong AI”.
“Weak AI” is capable of performing specific intellectual tasks. It can’t solve problems with self-awareness, but it can do things like language processing, image recognition, and voice recognition. The current AI is this “weak AI”, but it has functions and abilities that are sufficiently useful for humans, such as automatic driving of cars such as “Google Chauffeur” and the topic “ChatGPT”.
3 categories of AI
AI is said to fall into three broad categories. There are three types of “specialized artificial intelligence (ANI)”, “general artificial intelligence (AGI)”, and “artificial superintelligence (ASI)”. I haven’t been able to.
Artificial Narrow Intelligence (ANI)
Specialized artificial intelligence (ANI) is artificial intelligence that is specialized for a specific environment or task. is. Examples include image recognition and voice recognition. All existing AI can be considered to fall into this category. It has already been put to practical use and is being used to improve the convenience of business and daily life, such as voice recognition and autonomous driving, and is expected to continue to be used in the future.
Artificial General Intelligence (AGI)
Artificial general intelligence (AGI) is artificial intelligence “strong AI” that can understand, learn, and perform all intellectual tasks performed by humans. There is no unified opinion on how to define the target “intelligence” and use it as a criterion for judgment, and some predict that it will take time to realize it, but on the other hand, research is progressing all over the world. Yes, and some predict that the realization is near.
A famous method for measuring the intelligence of artificial intelligence is the “ Turing test ” devised by mathematician Alan Turing. Alan Turing thought it was futile to debate the question “Can machines think?” and devised the “Turing test”. In 2014, it was reported that Russian AI “Eugene Goostman” became the first “pass” in history.
In 2010, Steve Wozniak, one of Apple’s co-founders, devised a judgment test for general artificial intelligence called the “Wozniak test”, also known as the “Mr. Coffee test”.
Artificial Super Intelligence (ASI)
Artificial Superintelligence (ASI) is artificial intelligence with intelligence that exceeds human intelligence. It can be said that it is a hypothetical existence at this time that it has intelligence far exceeding that of humans. “HAL” and “Ultron” mentioned above are probably classified here.
Will AI Surpass Humans?
As mentioned above, artificial superintelligence (ASI) and artificial general intelligence (AGI) will not be completed by 2023, but we may reach a technological singularity (or singularity) at some stage. It has been pointed out that there is
This is the idea that once a “strong AI” that operates autonomously is completed, it will be recursively upgraded, and an intelligence that is superior to the human imagination will be born. advocating. There is an ongoing debate about whether or not the Singularity will come, and when it will.
2. Types of AI
There are various types of AI technology, such as rule-based AI, machine learning, and deep learning, depending on the technology used.
Rule-based AI
Rule-based AI solves problems and judges situations based on the knowledge and rules of experts, etc., programmed in advance. The advantage is that advanced knowledge can be used by non-experts. Recently, it is sometimes used as a “gateway” for user support in some web services.
It is used for medical diagnosis support and financial risk assessment and is also called an expert system (expert system). Manual configuration is required to set rules and change content. They are not good at answering vague questions.
Machine learning and deep learning
Machine learning and deep learning are major areas of AI in recent years and play a wide-ranging and important role.
machine learning
Machine learning is a term coined by scientist Arthur Samuel in 1959. AI that statistically learns and analyzes large amounts of data to automatically generate rules. Learn using “training data” or “learning data”. Create prediction models by automatically learning from large amounts of data using statistical algorithms. Machine learning can be further divided into three broad categories. They are supervised learning, unsupervised learning, and reinforcement learning.
deep learning
Deep learning is a type of machine learning, and the mainstream in AI research and development in recent years is deep learning. It is characterized by learning by associating concepts in a hierarchical structure using a deep neural network that mimics human brain functions.
“Generative AI” such as “ChatGPT”, which has become a hot topic recently, is a large language model (LLM) using deep learning. LLM is a language model consisting of a neural network, trained on large amounts of unlabeled text using “supervised learning”. Since it appeared around 2018, it is said that there is no official definition yet. It has become popular because it can handle a wide range of tasks and interact with humans in natural language.
Difference between machine learning and deep learning
Machine learning and deep learning are considered to differ mainly in algorithms, but the process of feature extraction is also different. Machine learning uses statistical methods to learn patterns from given data, but feature extraction is usually done by humans. Deep learning, on the other hand, uses multilayer neural networks to automatically learn features and perform advanced pattern recognition.
Machine learning can learn a wide range of patterns from relatively simple to complex patterns, but when processing complex patterns, more feature engineering and pre-processing are required. On the other hand, deep learning can automatically extract features, so the point is that it can recognize complex patterns efficiently. Therefore, deep learning can handle advanced tasks such as image recognition, speech recognition, and natural language processing. However, it requires huge datasets and a high degree of computational power (usually using GPUs) and computing power. This is because deep learning multilayer neural networks have a large number of parameters, and processing these parameters requires a large amount of data and computational power.
Other AIs
genetic algorithm
Another known AI is a genetic algorithm (GA). A genetic algorithm is one of the optimization methods that mimic the mechanism of biological evolution. It is said that it has excellent searchability and can find the optimal solution to complex problems. It was proposed in 1975 by Professor John H. Holland of the University of Michigan.
3. AI will realize a convenient and happy society
Advances in AI are expected to lead to greater efficiency and accuracy, especially in fields such as medicine, agriculture, and science and technology.
Improving the performance of AI will improve the speed of drug development, make the treatment and prevention of diseases more efficient than ever before, and improve the efficiency of growing crops. Being (happiness) is predicted to lead to improvement.
AI is expected to bring the following social benefits:
- It can solve labor shortages and reduce labor costs.
- Improving productivity and efficiency in the industrial and agricultural fields
- Improving social safety through advances in automated driving, danger detection technology, etc.
- Deepening and expanding communication through advances in automatic translation, etc.
The “weak AI” that exists at the moment cannot completely replace human judgment and thinking, so it is necessary to discuss and prepare for the appropriate use of AI.
Four. Concerns about AI
Concerns have also been expressed about advances in AI. Physicist Stephen Hawking has expressed concern that if AI evolves out of control, there is a possibility of human extinction. Many celebrities such as Bill Gates and Elon Musk share similar views.
Even if the crisis is not that serious, it has been pointed out that AI may generate a large amount of fake news, and deep fakes may spread biased opinions. In particular, it is said that the risk is high when used for military purposes.
In addition, the development of surveillance technology using AI may limit privacy and freedom, and may even violate human rights. There are also discussions about copyrights when AI’s “learning data” is obtained from the Internet, and how copyrights are handled when pictures and texts generated by AI are similar to existing ones.
It has been pointed out that it is necessary to emphasize ethical and social perspectives in AI technology development, considering the risk of misuse, including cases where the user does not intend.
Five. A most recent point of interest
There are two aspects to watch out for when looking at the future movement of AI-related technology: the relationship with related fields and the movement against risks.
Relationship with related fields
Recent AI-related highlights include:
Development of quantum computers = evolution of AI
It is expected that quantum computers, which can perform a large amount of calculations faster than conventional computers, will enable AI to solve more complex problems and perform more advanced learning. We cannot take our eyes off the development of quantum computers.
Collaboration with manufacturers, etc. = resolution of labor shortage
Advances in automation, efficiency, and robot control using AI will make it possible to respond to labor shortages and improve work efficiency, and improve the productivity of factories and other facilities. The key point is whether it will lead to a solution to the labor shortage problem.
Evolution of biotechnology = sophistication of AI
With the development of technology that handles ecological information, it will become possible to use AI in conjunction, and there is the possibility that the medical and bioindustries will evolve further.
movement of the world
As AI advances, some moves indicate an increasing focus on AI safety and ethics. It is believed that this will have an impact on future AI development.
Impact of “Asilomar AI 23 Principles” (2017)
“Asilomar AI Principles” is a gathering of AI researchers and legal, ethical, and philosophical experts from around the world in Asilomar, California, USA to discuss what AI is beneficial to humankind. Ethical principles and guidelines for AI utilization. It summarizes ethical issues, measures for safety management, transparency of research, etc. so that AI can benefit humankind.
Movements related to the EU’s draft AI regulation (2021)
The EU is putting together a draft AI regulation bill that sets standards for the ethical design, development, and deployment of AI and regulates market access. As for high-risk AI systems here, they need their assessment before entering the EU market.
Future of Life Institute Requests (2023) Ripples
In March 2023, the non-profit Future of Life Institute presented ethical guidelines for AI researchers and engineers. The Institute for the Future of Life, which translates the name of the organization, announced the principles of AI safety and ethics. The request has been signed by many celebrities such as Elon Musk and Steve Wozniak.
Summary: Beware of the “ELIZA Effect”
The “ELIZA effect” is a phenomenon in which humans overreact to AI’s reaction when AI reacts in the same way as humans do during a conversation. Developed in the 1960s, ELIZA was an early chatbot that could converse with humans through simple wordplay. Compared to the current system, it was a simple system, but there were still people who misunderstood “ELIZA” as a real person.
Today’s AI can respond surprisingly human-like and can do incredible things in a short amount of time. Therefore, while using it, you may mistake it for a real person or expect too much. Now that the speed of evolution is fast, it may be the point for humans not to get caught up in the flow.
