Autocomplete on smartphones is convenient, but sometimes it leads to unintended errors. Let’s take a look at how AI autocomplete works and its limitations.
If you’ve ever used a smartphone, you’ve probably been frustrated by autocomplete. Autocomplete is a branch of artificial intelligence (AI) that is designed to make people’s lives easier, so why do we find it frustrating? Autocomplete’s inconvenience reminds us of the reality of artificial intelligence. Artificial intelligence was first proposed by John McCarthy in 1956.
It refers to the ability of computer programs to learn and reason like humans. Artificial intelligence is most valuable when it is applied to other scientific and social fields. AlphaGo, which gained worldwide attention last year, is an example of artificial intelligence applied to Go. Beyond these cutting-edge games, AI is also transforming industries such as healthcare, finance, transportation, and more. For example, in healthcare, AI systems are being developed to help diagnose and plan treatment for patients, and in finance, algorithms are being used to analyze market data and suggest investment strategies.
In reality, AI exists in many different fields and has varying degrees of intelligence. Therefore, AI can be categorized into two types: strong AI and weak AI. Strong AI is AI that can actually think and solve problems. Weak artificial intelligence is when the system does not have sentience, but it mimics sentience. You can think of AI that can talk to humans naturally as strong AI, and AI that can only analyze data as weak AI. In the movie “Iron Man,” Tony Stark’s assistant Jarvis is a strong AI, while AI such as autocomplete on smartphones is a weak AI.
So what is it that makes A.I. a unique technology? It is machine learning, or ‘machine learning’. Machine learning is “the field of developing algorithms and techniques that enable computers to learn.” In 1959, Arthur Samuel first defined machine learning as “the field of study that develops algorithms that enable machines to learn from data and execute behaviors that are not specified in code.” Machine learning focuses on making predictions based on attributes learned from training data.
In current commercialized AI, machine learning is primarily accomplished through a method called pattern matching. Pattern matching is a way to figure out what you already know about a problem and then try a known solution. Your smartphone’s autocomplete feature is an example of pattern matching. It learns about the characters being typed and matches them to the corresponding text. However, this technology has its limitations. Many people use pattern matching on their smartphones to correct grammar mistakes, such as omitting an upper strophe, or autocomplete unintentional words.
Another example of pattern matching is Facebook’s content recommendations. If you’ve ever been on Facebook, you’ve probably noticed that after watching a video in your newsfeed, other videos are recommended below the video. However, this example also illustrates the limitations of pattern matching. When recommending content, it keeps doing the same thing to users. You don’t know how users react to your recommendations, and you don’t get much feedback.
This is where pattern recognition comes in to overcome the limitations of pattern matching. Pattern recognition means that software detects new patterns by monitoring user behavior. Whereas pattern matching is applied to all users at once, pattern recognition identifies and corrects patterns on a user-by-user basis. Current applications include automatic postal mail recognition and automatic fingerprint identification. Some smartphones already have a user-specific autocomplete feature. It collects information about the characters you type and analyzes what comes next. Unlike traditional pattern matching, autocomplete is user-specific, which makes it a little more convenient for us.
Just like pattern matching gave way to pattern recognition, the system keeps improving, but it takes quite a bit of time to cross the gap. It still struggles to understand complex phrases, and in the case of speech recognition, it fails to recognize different accents and pronunciations. These issues will only get better as more data is collected. For now, most of our real-life situations are weak A.I., and research on strong A.I. is ongoing. AI is suddenly all the rage, but its progress may be slower than we think.
This technological advancement is more than just a technological breakthrough; it also brings with it social and ethical issues. AI is likely to disrupt the labor market as it gradually replaces human jobs, and we need to consider and prepare for the social implications of these technological advances. It’s important to prepare for the future and how AI will change our lives.
