A bot has generated this blog entry based on the linked YouTube video. There are no intentions to steal the information or research from the original owner. This blog post is exclusively for "Educational Purposes Only," and I have no intention of earning, selling, or monetizing it. The main objective of this site is to share knowledge to as many people as possible. I strongly advise people to watch the original video for a better understanding.
How to mitigate problems like algorithm bias in artificial intelligence? This is a very good question, so we believe that artificial intelligence is a god-like phenomenon which knows everything etc. well. That is not true. Artificial intelligence is essentially something that emerges mostly well. There are a number of ways to dimmer to construct or create artificial intelligence. One of the ways is machine learning in which you have in which you have an algorithm which is fed data it, and it learns on its own, based on pattern, recognition, etc., and that is how it develops its own so-called intelligence.
It learns on its own. It is sometimes supervised learning, sometimes it's unsupervised, learning etc., but you find that these algorithms they learn on their own. For example, you have these chess playing algorithms that basically learn how to play chess by playing lots of chess games against themselves or by going through a great amount of data of hundreds of thousands of chess games that are played in the past. So it learns on its own the rules of chess, and then it can start defeating human players. Now you have these other algorithms, like language, recognition, algorithms, automatic translation, algorithms and text generator algorithms, which are able to generate text based upon certain keywords that you input and it gives you very human-readable text.
It looks like it's been written by human beings, so this is basically the thing we're talking about, and this is where algorithm bias comes in so algorithm bias is. It occurs in a variety of ways. First uh, you may have a faulty algorithm, an algorithm that produces results that are systematically biased or prejudiced because of erroneous assumptions in the machine learning process which could arise from faulty or poor or inaccurate data yeah, because if garbage goes in garbage comes out, you know. That'S the old adage in computer science, GIGO garbage out. So if you have data that is inaccurate, it is poor. It is biased in some manner.
It is not representative of the entire population etc. for example. That'S what happens or you can have problems introduced by people who design and train the machine learning system who create algorithms that reflect either unintended cognitive, biases of theirs or actually real life prejudices right. For example, I have seen this myself. You have this GPT 3 text generation, artificial intelligence. Basically, and if you input keywords related to India into it is outputs text that looks like it has been written by Marxists Hindu, phobic texts that reflect certain kinds of biases against India.
So it is because these people, who have been training the algorithm when it comes to India related subject, they have been basically feeding it, texts that have been written by leftist and Marxist academics and left oriented journalistic publications. It'S very clear. So whenever you input something related to India, keywords related to India into gpd3, it gives you it outputs you, paragraphs of text that look like they have been written by some professor in Jun, jawaharlal Nehru university. So that is a clear demonstration of algorithm bias. So either these biases are created by poor data or data that is prejudiced or it is slanted in a certain ideological or cultural direction. It either reflects real life prejudices of certain people of the people who are introducing the data or unintended cognitive, biases or biases.
That are introduced by incomplete or faulty or prejudicial or biased data sets to train and validate these machine learning systems, so that is algorithm bias. So you have to address the real reason why this bias has occurred either it is a faulty algorithm or it is data that is not representative of the entire t of knowledge about that particular thing. For example, in the case of India, if you feed only data that comes from left-leaning journalistic organizations or leftist Marxist professors, then the machine learning algorithm will get trained on that and every time you give it keywords related to India. It'S going to give you output. That looks like it's been produced by Jun professors, so that is what algorithm bias is so the mitigation is, it depends on how it has been introduced.