What is artificial intelligence? Alan Turing posed the bold question, "Can machines think?" and proposed that, rather than defining "thinking," we should consider something to be "thinking" if it can satisfactorily simulate the act of thinking.
This is the 'Imitation Game', the name given by Alan Turing in his paper and also famously used as the title of a movie.
I, too, would like to answer the question, "What is artificial intelligence?" in the same way.
Instead of defining artificial intelligence like Turing, if we can satisfactorily prove the usefulness of artificial intelligence, then we can consider ourselves to have “understood artificial intelligence.”
So, I want to introduce you to the usefulness of artificial intelligence.
--- From "Entering"

In 1770, a novel chess machine appeared in Austria.
It was an automaton playing chess, with a carved doll wearing Ottoman costume and a turban on its head sitting behind a chessboard.
The surprising thing was that this machine played chess by itself.
Moreover, his skills were so outstanding that he could beat most people.
Naturally, many people suspected that there was a person hiding inside the machine.
Suspicious people looked into every nook and cranny of the machine, but found nothing but a mechanical device consisting of unknown levers and springs.
…what on earth was the identity of this chess-playing automaton?
--- From "Artificial Intelligence│Great Artificial Intelligence, Awakening"

Tesla recorded its first fatal accident in Florida in May 2016, just seven months after launching its self-driving feature.
It was an accident where a white trailer was hit from the side.
At the time, the Tesla Model S did not recognize the side of the white trailer as an obstacle at all and ended up ramming it without slowing down.
The vehicle was mangled and the driver died on the spot.
According to a later accident report, the driver was behind the wheel for only 25 seconds during the 37 minutes.
Not only that, the warning sound to keep your hands on the steering wheel sounded seven times.
He was overconfident in Tesla's self-driving capabilities.
--- From "Autonomous Driving│Tesla's Dream Machine"

How can search alone generate such enormous revenue? Randomly displaying ads doesn't increase clicks.
The key is to increase the likelihood of clicks.
We needed to display ads that were contextually relevant to the user so they would click on them.
For example, if a man in his 30s working in an office in a large city searches for 'shirts', famous brands such as 'Polo' and 'Beanpole' will be displayed, and further encouraging him to make an online purchase.
…This is the typical search advertising model and search engine revenue model.
It is also Google's representative revenue model.
Google generated $330 billion in advertising revenue this way in 2023 alone.

--- From "Search Engine│How Google Searches the World"

Siri is now under Apple's wing, just two months after its launch.
Plans to release Android and BlackBerry versions were of course canceled, and work is now underway to integrate Siri with the iPhone.
Even after the acquisition, Jobs showed great interest by attending Siri's weekly meetings.
A few months before the product's unveiling, the Siri development team and Jobs ran into each other in the cafeteria. Steve Jobs, who usually greeted everyone else perfunctory, stopped and greeted the Siri team warmly.
“Hey Siri friends! How are things going?”
--- From "Smart Speaker│Can Siri Become a Useful Secretary?"

Let's ask the speaker once, "How's the weather today?"
Will the speaker actually recognize "Fuck, shit" as "Fuck, shit"? Of course, the acoustic model can.
But language models never perceive it this way.
The language model is what we call prior knowledge.
The language model has never seen the sentence 'the weather sucks' before.
So, there is no way that you would have any relevant knowledge.
Clearly, the language model has seen the sentence 'what's the weather' countless times and has prior knowledge.
Therefore, we already know with a high probability that the phrase 'what is the weather like today' will be followed by 'how is it'.
The language model recalibrates the acoustic model based on probabilities even if it misrecognizes the voice.
It's a correction to 'The weather sucks today', which we often ask, 'How's the weather today?'
--- From "Smart Speaker│Can Siri Become a Useful Secretary?"

We no longer input rules into machine translation.
Learn the rules yourself from similar sentences.
It's about solving the complex problem of translation on your own using data.
All you need to improve performance is more data, more sentences.
The rest is all learned by the machine itself.
Neural network-based models are constantly evolving.
Research is becoming more active with the introduction of the concept of attention, which goes beyond simply learning entire sentences and focuses on important words.
Around this time, there was even talk that the research results of the past two years had surpassed those of the previous 20 years.

--- From "Machine Translation│If you don't know a foreign language, just have Papago"

As if to reward the long wait, GPT-4 boasts remarkably superior performance.
It outperforms existing models in many areas.
When I took the bar exam, the existing ChatGPT scored 213 out of 400, which was in the bottom 10%.
However, GPT-4 reached the top 10% with a score of 298.
This is enough to pass the New York State Bar Exam.
He also scored 700 out of 800 on the American SAT Math Test, which is equivalent to Korea's College Scholastic Ability Test.
Likewise, it is a score that is in the top 10%.
It's not just about being good at English.
We tested machine translation of some data in English into Italian, Spanish, German, French, Korean, and Japanese, and the results were significantly higher than those obtained with the existing ChatGPT for processing English sentences.
The new GPT-4's Korean answers are more accurate than the existing ChatGPT's English answers.
--- From "Chatbot│ChatGPT, Write a Report in 1 Minute"

The original recommendation system was actually Amazon.
Since the 1990s, Amazon has achieved significant success by recommending products related to what customers have purchased, products likely to be of interest to them, and ultimately products they are likely to purchase.
…Amazon's recommendation system initially only listed related products, but it has since become a successful service that influences click-through and purchase rates.
The key to recommender systems is that they become more sophisticated the more you use them.
As data grows and information becomes more abundant, more accurate recommendations become possible.
Recommendation systems account for a significant portion of total sales.
They say that 35% of Amazon product sales come from recommendations.

--- From "Recommendation Algorithm│The Unknown YouTube Algorithm Led Me Here"