THE BEGINNINGS
The first use of the term ‘Artificial Intelligence’ was in 1955 by John McCarthy, a computer and cognitive scientist at Dartmouth College in the USA. Along with Marvin L. Minsky of MIT, Nathaniel Rochester from IBM, and Claude Shannon from Bell Laboratories, they were attempting to show that machines could be made to simulate intelligence if the principles of intelligence and learning were described to them very precisely.[1] In many ways, this laid the foundation stone and helped shape the field of Artificial Intelligence. The team held that “cognitive faculties could be abstracted from the supporting physical operations of the brain. Thus the former could, in principle, be reproduced in the different material substrate so long as the formal rules could be executed there[2]. Artificial Intelligence nowadays, however, has abandoned the approach from the Dartmouth Research Project, as intelligence is now viewed as a moving target in the field of AI.
WHY IS IT CALLED ARTIFICIAL INTELLIGENCE?
Artificial intelligence is built using specific mathematical functions and some very complex algorithms. These are the rules and instructions that have to be followed by the AI. That is to say, it is a system that has been built to address a need in such a way that it can take care of matters itself, without needing human intervention. So, one could say that AI is purpose-driven or outcome-oriented and it works with massive volumes of data by processing, analyzing, and learning from it.
HOW IS AI TRAINED?
MACHINE LEARNING – Machine Learning is the process by which machines ‘learn’ from the tasks they are performing, and eventually improve performance. Learning algorithms are used to analyze data, so that machines can learn by trial and error, to predict outcomes so that performance may be modified to get the desired goal.
DIGITAL NEURAL NETWORKS – Neural Networks or Neural Nets, are a collection of connected units or nodes, used to perform tasks and gather data so that the machine can modify future performance.
REINFORCEMENT TRAINING – This is akin to training a dog, where a task is given and the successful completion of said task results in a reward, whereas non-completion results in punishment. In this way, machines are nudged in the right direction and machine learning makes them improve their performance.
IMAGE PROCESSING – This is the analysis of images using computerized algorithms.
VARIOUS TYPES OF AI
There are various types of AI because not all AI is built for the same purpose. AI is built with specific targets. There are two broad categories used to differentiate AI. These are:
AI THAT IS BASED ON CAPABILITY
NARROW AI – This type of AI is built with a specific goal. A learning algorithm is designed to perform a task, but any information gathered cannot be applied to other tasks automatically, which is why it is also called WEAK AI.
ARTIFICIAL GENERAL INTELLIGENCE – Also known as STRONG AI. It can learn from the data and then use it in different tasks. OPEN AI’s DALL-E is an example of this.
SUPER AI – This is theorized as AI that can surpass humans so that it is superior to human intelligence.
Functionality based AI
REACTIVE MACHINES – As the name suggests, these machines work by reacting to their environment. They do not work with taught or recalled data and are therefore the most basic types of AI.
LIMITED MEMORY – These machines use data from previous experience to make decisions. A good example is self-driving cars that use proximity sensors.
THEORY OF MIND – This advanced type of AI will be able to infer emotions, designs, and intentions. A famous example is TOMNet (Theory of Mind Network).
SELF-AWARENESS – This has not been developed yet, due to hardware and appropriate algorithm constraints.
FUNCTIONS
AI is truly ubiquitous today. It can be used in nearly every sphere of existence. Not only is AI used to automate processes at production plants, but it is also used to keep levels of nutrients in check on big farms. It is used in hospitals to analyze MRI scans and to keep elderly patients company in a care home.
- Various chatbots make use of AI. For example, LaMDA (Language Model for Dialogue Applications), an AI developed by Google. It can mimic human speech and has been called ‘sentient’ because of its ability to think and reason like a human being, although this has been refuted by Google and it plans to embed this technology across its platforms, as laid out by CEO Sundar Pichai at Google’s Developer Conference in 2021.
- Another example is Replika, a chatbot that was developed by Eugenia Kuyda, CEO of Luka, a software company based in San Francisco. It is described as “a digital footprint of your personality”. It uses AI to provide companionship by having conversations with users. It not only listens, but it learns from users and replicates them.
- AI is also used as an image generator, where it can create images with only text-based instructions. Google Imagen and OpenAI’s DALL-E2 are prominent examples of the same.
- DeepBrain AI – a South Korean company that has utilized Deep Fake to make videos of, for example, a news anchor reading a news report. And a company called Metaphysics is helping those with tight finances, by recreating creative works and ideas.
THE FLIP SIDE
While AI has many advantages, as outlined by the creators of all these tools, AI has also led to many changes that people are not happy with. There is growing concern that since AI can be trained to do any tasks that humans can do, it will lead to unemployment for many whose jobs are susceptible to automation. Kai-Fu Lee, a Taiwanese pioneer in AI research and technology, said “…about 50% of jobs will be somewhat, or extremely threatened by AI in the next 15 years or so.” in the episode titled “In the Age of AI” on FRONTLINE. One of the biggest concerns about AI is that it will replace white and blue-collar jobs as AI is found to do jobs at a dramatically better level.
People have also used Deep Fake technologies to commit data breaches, financial fraud, phishing scams, and automated misinformation attacks. Since AI can be trained to perform tasks, its usage ultimately depends on the people who hold the reins, because whatever instruction is given to the AI, it will follow that to the best of its abilities.
CONCLUSION
It is clear, therefore, that while AI has many advantages and will lead to betterment in many spheres, it can also be used to make life difficult for many while it benefits some. Because only those who have access to the technology and the tools used to power it will be able to set the pace of inclusion. And of course, there is also the constant fear that AI will achieve singularity and turn on humanity. Neural networks are already said to be ‘striding towards consciousness’, according to Blaise Agüera Arcas, an engineer at Google.
With all the benefits that society gets due to AI, it can be easy to forget that at its core, AI is like a child that is learning about the world. It is a very intelligent child, and it can learn very quickly. It does not have an agenda or any vested interests, and therefore, humanity needs to be careful what it teaches this child so that it can continue coexisting with humans.
[1] McCarthy, Minksy, Rochester, & Shannon, 1955
[2] Kline, 2011, 2015
Author: Mr. Naman Sareen, Research Associate, CyberPeace Foundation
