With the advent of ChatGPT, AI has become a focus not only for researchers and developers, but also for the general public. ChatGPT is a hot topic because it is a chat-type and easy-to-use service, but various AI has been developed and put into practical use, and companies and researchers around the world are still competing to develop AI for all purposes.
Is AI, including ChatGPT, a "dream tool" that surpasses human intelligence?We will explain through familiar examples and the characteristics of AI.
Supervision
FRONTEO Inc.
Behavioral Information Science Institute
DEPUTY DIRECTOR GENERAL
Keisuke Tomiyasu
Obtained a Ph.D. (Science) through physics research.He is engaged in the fusion of condensed matter physics research and experiments and data analysis at universities and UK science and technology facilities.He researched and developed the algorithm of artificial intelligence KIBIT, and succeeded in major innovation.Currently, he is leading the improvement and creation of further algorithms, the design of KIBIT-equipped products, and business reforms.
What is AI (artificial intelligence)?
AI is an abbreviation for Artificial Intelligence."AI" can be said to be a general noun, but in fact there is still no clearly unified definition.It may be defined as a perfect imitation of human intelligence, intellect, and emotion in a wide range of fields.However, in general, AI often refers to systems programmed to perform specific tasks previously performed by humans by analyzing and learning from data*.For example, there are various AIs around us.Services that use AI include ChatGPT, which allows you to exchange text, self-driving cars, voice assistants (smart speakers), and facial recognition systems.
* The former AI that perfectly mimics human intelligence is sometimes classified as "strong AI", and the latter AI that performs specific tasks is sometimes classified as "weak AI".All the AIs currently in practical use are weak AIs.
Are you misunderstanding or misunderstanding about AI?
For example, ChatGPT is a very useful AI service that answers text questions naturally like a human being, and if you ask a good question, it will organize and tell you the necessary information.Therefore, you may feel that AI is much more intelligent than humans, that it knows everything, and that it is omnipotent.But is it really so?
Is AI smarter than humans? Will AI steal your job?
AI is still not as intelligent as humans. AI is developed to be highly competent in specific tasks, but it is different from having human-like general intelligence.
So, will AI take away jobs?Some people say that AI is actually helping humans in their work. In many cases, it is appropriate to regard the development of AI as enabling intellectual labor on a scale that was not possible before, and saving labor in the work of humans so far, rather than taking away human jobs.
Is AI perfect and never wrong?
You may have an image that AI is perfect and does not make mistakes, but in reality AI also has limitations and errors.There are two reasons.As can be said for machine learning in general, first of all, AI's answer is "probability".You can't always get 0% or 100% chance.Second, AI makes predictions and judgments based on the data it has learned, so accuracy varies depending on the quality and quantity of the learning data. AI does not know everything, and its ability to deal with new situations and unknown data is inevitably limited.This is because AI models show very high accuracy for tasks within the range of training data (interpolation), but cannot demonstrate sufficient performance outside the range where training data does not exist.
Therefore, it may be difficult to make appropriate judgments or make complex ethical judgments in the event of a sudden accident.Highly capable AI is convenient and seems to be able to do anything, but it is important for humans to have a good understanding of the problems they want to solve and the characteristics of AI, and to place and utilize AI in the right place.Only by making a comprehensive judgment can mankind achieve breakthrough development with AI.
What is natural language processing AI, one of many types of AI?
AI is used in a wide range of fields, including AI that can play Go, which is famous for AlphaGo, AI-equipped shogi software, image recognition AI used in face recognition and automatic driving, speech recognition AI used in voice assistants such as Alexa and Siri, and natural language processing AI used in ChatGPT.
Natural Language Processing (NLP) is a technology that allows computers to process the natural language that people use on a daily basis.With the advent of deep learning, advanced natural language processing that learns patterns from large amounts of data has become possible, and among them, the advent of Transformer has brought about a major evolution.
What is deep learning?
Deep learning is a type of machine learning, a learning method that uses a multi-layered neural network that resembles the structure and connections of nerves in the human brain.It provides great flexibility in creating approximation functions (models) between inputs and outputs.The resulting model outputs a list of optimal solutions for the input with probabilities and scores.
What is Transformer
Transformer is a model of natural language processing announced in 2017.Learning is done by deep learning.The model's ability to better capture context and meaning was a major turning point in natural language processing. With the appearance of Transformer, the performance of machine translation, question answering, text generation, sentence summarization, etc. has greatly improved.
What ChatGPT is good at, generation AI represented by ChatGPT
ChatGPT, which has been a hot topic since the end of 2022, is a conversational and chat-type AI service developed by an organization called OpenAI.It is characterized by the ability to generate and respond to natural and human-like sentence expressions in response to user questions and input.
ChatGPT is a conversational AI service, and GPT is a language processing mechanism
To say that ChatGPT is an AI is actually not accurate, ChatGPT is the name of an AI service. GPT is an abbreviation of "Generative Pre-trained Transformer", which is a language processing mechanism, and it means a pre-trained generative transformer.
The reason why ChatGPT became such a hot topic is because it was a "service that allows conversations" that allows chats to be returned smoothly.One of my strengths is that I am now able to respond with natural sentences and words.
Generative AI that responds to textual questions and inputs and generates sentences
ChatGPT learns a lot of data by deep learning, finds patterns in it and generates responses.It can be said that natural language processing is extremely superior in that it can generate an appropriate and natural response that matches the question rather than outputting the learned data as it is.
It can answer general conversations and questions with fairly high accuracy and the sentences are natural, but it tends to be inaccurate when it comes to specialized knowledge and complex questions.That said, ChatGPT is good enough for information and basic support, and can be useful as a companion or assistant, so to speak.
What ChatGPT does
The content of ChatGPT, which has made such an impact, is surprisingly simple. "Keep predicting the next word (token)", that is, keep predicting the next word of the text that has been input so far.Words other than the most probable are sometimes used to introduce naturalness.
On the other hand, let's put this in the context of the situation we're talking about.You speak one word, choose the natural word that comes next, think of the natural word that comes next, and speak, and so on, repeating this activity at high speed.What do you think about your own behavior like that?It may be closer to the reality that we specialize in speaking naturally rather than speaking with an image in mind, or that we remember patterns, syntax, or something close to grammar that connects many words rather than thinking logically.To put it the other way around, what I thought was a complicated and mysterious language might turn out to be such a relatively simple expression tool.
What is Generative AI
Generative AI is AI that generates content based on learned data. ChatGPT is an AI that can generate sentences from a large amount of pre-trained data, and there are also other AIs that can generate images, designs, music, program codes, etc.
What FRONTEO's AI "KIBIT" is good at, features as a natural language processing AI
FRONTEO's in-house developed AI engine "KIBIT" is also a natural language processing AI, but it is not a "generation type" AI, but a "discovery type" AI that searches for and finds necessary information from a huge amount of text data. It's AI.
In today's world where information continues to increase every second, the selection of information is a major issue for any company in any field.For example, a tax accountant who investigates accounting fraud finds documents and interactions related to fraud from a large amount of financial statements and related documents, and a lawyer in charge of litigation needs to find documents that can serve as evidence from a huge amount of related materials.Call center representatives search for opinions that can be used to improve products and develop new products on a daily basis from various customer feedback such as inquiry histories and questionnaires.
Instead of sending dozens of people to check all such data and all documents, the AI engine "KIBIT" reproduces the judgments and thoughts of experts such as lawyers, tax accountants and people in charge of each business by learning and selects the necessary information.
AI that learns human-like contexts and quickly finds information
KIBIT is a type of machine learning that uses an algorithm different from deep learning to analyze and learn text from a very small amount of training data on a normal notebook PC level computer at high speed and with low power consumption.
For example, documents related to fraud and documents not related to fraud are given to KIBIT as training data (original data). A document can be a large unit, such as several pages of A4 paper or an email. At this time, KIBIT captures the atmosphere (subtleties) of the entire document and the surrounding text of the text in question, not only from the words but also from the structure, number, and combination of characters used in the text, and features that quantitatively express the atmosphere. We will build an AI model to extract and “discover” trends.
This is how KIBIT is resistant to ambiguity and fluctuation of words, and can analyze as if it were reading the context.As a result, even if the words used in the document to be analyzed are different from the first given document, KIBIT will judge it as a "related document" if the overall document has a similar atmosphere.
Each AI has its own strengths and uses
There are various types of AI that have been put into practical use in the world, such as automatic driving, voice/face recognition systems, and chatbots.Even for natural language processing AI that targets text and language, the type is different between ChatGPT and KIBIT.
Features include ChatGPT, a ``generative AI'' that is good at creating and summarizing sentences, and KIBIT, a ``discovery AI'' that reproduces the thinking of experts and searches for necessary information from vast amounts of text data. it's different. The humans who use AI should use the right people in the right places, or use them together. In other words, KIBIT and ChatGPT are not rivals competing for market share, but rather "AI friends" with different features.
There is text data that can be used in KIBIT for any company and any business
KIBIT, which features high-precision analysis of text data, was developed to support experts.
For drug discovery researchers, for example, we propose “targets” that serve as the basis for drug discovery from tens of thousands of papers.Attorneys in charge of litigation are presented with high accuracy documents that can serve as evidence from a huge amount of related materials.In general companies and organizations, emails and SNS are analyzed, and useful information for business decisions and strategy formulation is provided.We will provide necessary and appropriate information to experts in various fields such as medicine, law, intellectual property, finance, and human resources, and support the resolution of issues.
Text data that can be used for KIBIT exists in any industry and any business.Any data that can be converted to text or text, such as daily emails and reports, telephone conversations, interview records, and postings on SNS, are targets that can be analyzed by KIBIT.
At that time, a large amount of training data is not required.In general, AI tends to require thousands to tens of thousands of training data, but KIBIT can learn and start using even dozens of training data. With KIBIT, it is possible to utilize all text data and support experts in various fields.