During my younger days, I spent much time playing computer strategy games, and the AIs that powered them were simplistic and rules-driven. A rules-based AI follows a large flowchart where the programmer has to think ahead, consider all possible scenarios, and predict each response. In computer science, this kind of system is considered non-scalable, meaning that as the system becomes more complex, it cannot efficiently handle larger volumes of data or requests. Obviously, a rules-based system cannot produce real intelligence, and no matter how many times the game is played, the machine will not become more intelligent.
AI has made tremendous progress in recent years. Nowadays, modern AI employs machine learning, which eliminates the need for humans to write complicated flowcharts. Deep learning is a subset of machine learning that utilizes artificial neural networks to imitate the human brain’s operation. Although AI and the human brain are two very different systems, machine learning can help us to understand how our brain works.
So, what is machine learning exactly?
“Machine learning is the study of computer algorithms that can improve automatically through experience and the use of data. Machine learning algorithms build a model based on sample data, known as training data, to make predictions or decisions without being explicitly programmed to do so.” (Wikipedia)
Much like machine learning algorithms, our brains also construct numerous models of the world around us. While our brains are not as quick as modern computers, we make up for this with a vast number of brain cells that allow us to process information in parallel. It is impossible for us to analyze every situation we encounter from scratch, so we must rely on mental models we have constructed and compare them to similar situations we have experienced before. This is particularly crucial in primal situations where humans must make quick decisions to avoid danger and ensure their survival.
Throughout my Ph.D. research, I frequently presented my supervisor with data that was not yet fully developed. Rather than outrightly highlighting my errors, my supervisor would caution me about the fallibility of pattern recognition in our brains. He explained that while our brains excel at identifying patterns, sometimes these patterns are not grounded in reality. As a result, our innate pattern recognition abilities can mislead us and lead us astray.
My brothers and sisters, believers in our glorious Lord Jesus Christ must not show favoritism. Suppose a man comes into your meeting wearing a gold ring and fine clothes, and a poor man in filthy old clothes also comes in. If you show special attention to the man wearing fine clothes and say, “Here’s a good seat for you,” but say to the poor man, “You stand there” or “Sit on the floor by my feet,” have you not discriminated among yourselves and become judges with evil thoughts? James 2:1-4
As opposed to discrimination, favoritism is showing a preference for one person over another even though they have equal claims. Favoritism is also due to our data model thinking that those people will give us better survival value. Unlike discrimination, which is easy to spot, some behaviors that look like favoritism are not really favoritism. For example, if you were invited to have dinner with the US president, you would wear your best suit, whereas if your friend invited you for dinner, you would only wear regular clothes. Fairness is not treating everyone equally; fairness is treating people according to their needs or respecting the position that they hold. There is nothing wrong with spending more time with my wife than with other people around me. Giving more to the needy is not favoritism either. We need to constantly watch out and examine our behavior to identify favoritism and try our best to get rid of it.
Although we long for a world where justice is truly impartial and equitable, no society is entirely just, and there have been recent scandals in the US Supreme Court. An AI judge, in principle, has the potential to deliver impartial justice without any biases or prejudices. However, the effectiveness of an AI judge depends on the quality of the training data it receives. Since humans create the data, it may inadvertently incorporate societal biases and prejudices. As a result, AI judges may fail us, just like human judges, if they are not trained on unbiased and representative data and continuously evaluated for potential biases.
Like machine learning, our models can be retrained. Brain researchers tell us our mind is very plastic, and we can reconfigure neurons to conduct different tasks. For example, if you became blind, you could train your brain to enhance the sense of sound. To eliminate our racial stereotyping, we can always understand cultures other than ours and learn more positive things about them. So, when our brain extracts information from our model, we have more positive things to say. No matter how hard we try, our data might still inadvertently incorporate societal biases and prejudices. We need to train our model with the perfect law of God and meditate on His word.