Ethically Used AI

In recent years, colleges have seen a rapid increase in the use of AI and its expansion of capabilities. It has been used to explain topics, create images and videos, but is also used to cheat. How can students use AI as a tool for learning while maintaining RPI’s mission to foster a culture of knowledge, thoroughness, and above all, ethics?

What is AI?

General Definition: “At its core, artificial intelligence refers to machines or computer systems designed to perform tasks that traditionally required human intelligence. These tasks could include things like recognizing speech, making decisions, solving complex problems, or even understanding emotions. AI strives to mimic how humans think and make decisions, but it does so using algorithms, data, and statistical models rather than biological processes”

From sciencenewstoday.org

Capabilities:

General/Strong AI – Artificial intelligence that can think, act and make decisions almost equal to a human’s level of intelligence. Theoretical advances include AI that can understand, learn, and apply. The main goal of this application is replicating human intelligence. Common examples are robots, language processing, and image recognition. There is very limited human interference. We are in the concept and design stages.

Narrow/Weak AI – Weak AI is much like your Alexa. It is AI that is created to perform specific tasks. Another name for Weak AI is narrow because it performs only a limited number of tasks. They excel at performing these functions because they are built with parameters. Another key detail of Weak AI is lack of consciousness. It has no understanding and is programmed. Common examples are self-driving cars, voice-activated assistants and virtual assistants. Chatbots on websites are also a very common use.

Super/Super Intelligent Ai – This form of AI has an intelligence level higher than that of a human. This type of AI can solve problems, think critically and interpret emotions. It is hypothetical with building blocks being built as of 2025. This step can only occur once Strong AI is developed, with researchers currently working to design the beginning stages. This level of Artificial Intelligence would be a cutting-edge, significant invention.

Functionalities:

Limited Memory AI – Extremely common. Limited Memory AI is used in modern applications to retain data it has found and improve after making mistakes. Using data it finds, it improves its predictions and decisions. ChatGPT uses limited memory AI, forgetting conversations once you leave the tab, unless you have an account in which your conversations are stored. “Ephemeral” memory. Self-driving cars also use this technology.

Reactive Machines/Basic AI – The most basic form and function of AI. They do not have memory; inputs always deliver similar outputs. It cannot predict information and must be fed appropriate information for its task. It cannot learn. The most common use of this type of AI is amazon recommendations or simple tasks in a self-driving car. The reason you get recommendations from sites like Amazon is because a reactive machine reads customer data and picks similar items to show you.

Self-Aware AI (Concept) – Can be the most capable AI. This type of AI is aware that it exists and is also aware of its thoughts and decisions. It experiences the world around it. It uses neural networks as its form. This is mostly hypothetical so far. It would require itself to have motivations, this is a concern for many people. How far is too far.

Theory of Mind AI (Concept) – Theory of mind AI is a predictive AI. At its core it must have emotional intelligence and be able to understand emotions. This type processes signs from humans and animals to predict emotions and mental states. It must understand social interaction and dynamics. Empathy drives it. It is still in conceptual stages. Like self-aware AI, people have concerns about the ethics behind such powerful and human-like AI.

Technologies:

Computer Vision – Interprets real world data through visual input. It uses deep and machine learning. Machine learning helps it derive meaningful information from the data it collects. It has the capabilities to recognize or reconstruct patterns and visuals. It can see in 3 dimensions using multiple camera angles for depth. It has the capacity to understand thermal vision as well. Used in traffic, airports and in military technology. New iPhones come with a form of computer vision.

Deep Learning – Subset of machine learning. Models involve layers of neural networks that process mass data. The backbone of most new state-of-the-art AI technology. Machine learning helps adjust connections in neural networks giving weight to certain models. Powerful and versatile the more layers used. Was enabled by significant development of GPUs. Strong GPUs made a difference in the processing power and speed of deep learning.

Generative AI – AI that can create original images, videos or text based on prompts given to it. ChatGPT, Sora, Copilot. Uses machine learning and, by proxy, deep learning. It uses mass amounts of data to understand requests and respond the best it can. Landslide of AI platforms was caused by the introduction of generative AI. You train, tune, and evaluate generative AI to improve its understanding and output.

Large Language Models (LLM) – Combination of deep learning and natural language processing. They are built on deep neural networks and can understand and generate text as if they were humans. Natural communication between machines and humans.  They are given massive datasets to use and many parameters. They can generate code, translate foreign languages and understand paragraphs.

Machine Learning – These systems are created to learn from data, improve outputs, and analyze carefully. They are not programmed for every task and instead are asked to complete a variety of tasks by learning. They can find patterns and make predictions.