The other day, I was on my couch watching TV, stuck in front of the countless commercials that delayed the evening movie, when an ad spot caught my attention. A female voice extolled the virtues—if there are any—of a car that was supposed to allow me to pursue my dreams and become a better person. Then the voice said, “Make your life easier with our brand new, onboard, generative AI!”
In my professional activity, I’m accustomed to software salesmen selling me, rightly or wrongly, the merits of AI. Given the frequency and demand, I decided to train myself on this new artificial intelligence thing. But I certainly didn’t expect a commercial to be used to sell AI to the general public.
We know that digital change generates erroneous beliefs. But does AI deviate from this great rule? Perhaps it is time to break the myth.
“Intelligence is, of all things among men, the most equally distributed; for every one thinks himself so abundantly provided with it, that those even who are the most difficult to satisfy in everything else, do not usually desire a larger measure of this quality than they already possess.”
René Descartes
Different types of artificial intelligence
There are two primary types of AI: symbolic AI and connectionist AI, and some would argue a third type—generative AI.
Fabio Caversan explains in a Forbes article that to understand why the “how” behind AI functionality is so important, we first have to appreciate that there have historically been two very different approaches to AI. The first is symbolism, which deals with semantics and symbols.
Symbolic AI
Symbolic AI simulates human reasoning by exploring a set of rules and facts predefined in advance, reflecting the solutions to a given problem. It is based on the interpretable nature of the models and the explainable nature of the decisions. It uses two kinds of inference:
- Abduction: reasoning inferred according to a logical formula based on an established knowledge base. For example, when a rate calculated on the fly exceeds a fixed threshold.
- Causal trees: infer according to a “fuzzy” formalization of explanations that lead to established and known facts. D develops an abstract argumentation system that links observations to facts. This system aims to identify a cause—such as fraud or an operating incident–from the accumulation of abnormal facts, for example.
Connectionist AI
More recent AI-based systems use a bottom-up approach called connectionism. Caversen notes that connectionism is known for its most successful techniques, deep learning or deep neural networks. It is the architecture behind the vast majority of machine-learning systems.
In this model, AI simulates mental phenomena through formal neural networks. These artificial neural networks are made up of several layers, each made up of neurons. There are at least three types of layers: an input layer, a hidden layer, and an output layer.
An input layer that will capture the raw data, including shapes, colors of an image, letters, sentences of a text, sounds, etc. The neuron layer then transmits this information according to determined parameters to the second layer.
The second layer processes the information received according to a set of parameters. Then, the sorted and processed information will be transmitted to the output layer, which processes the information in the same way as the second layer and formulates an output response.
Generative AI
Although we talk a lot about Generative AI, or GenAI, lauding it as the most advanced form of AI, this term is a misnomer. It describes an AI capable of generating images, videos, and even music, but it requires a global and versatile cognitive capacity. In essence, GenAI is another term for the latest generation of Connectionist AI.
Sorting out intelligence
There are three types of AI: artificial narrow, artificial general, and artificial super.
Artificial Narrow Intelligence (ANI): Weak or narrow abilities — Weak AI refers to AI systems specialized in a specific task that does not have the ability to learn or general understanding. In the business world, narrow AI is used in the financial sector for financial data analysis, fraud detection, risk management, and algorithmic trading. Chatbots are another example of ANI in use. For now, all existing AI is ANI.
Artificial General Intelligence (AGI): Strong or general abilities — AGI, or general AI, represents a higher level of artificial intelligence. It can learn independently, understand context, and adapt to new situations. This level of AI is still largely theoretical and has not yet been fully achieved.
Although strong AI is an exciting prospect for the future, its practical application in the business world is still limited. However, some companies are beginning to explore the possibilities of strong AI for complex and evolving tasks, such as strategic decision-making or the automation of very sophisticated processes.
Artificial Super Intelligence (ASI): Super abilities — ASI refers to a form of AI that would significantly surpass human intellectual capabilities in everything from artistic creativity to scientific problem-solving. Unlike AGI, which is equivalent to human intelligence, ASI could be superior to the best human intelligence in virtually all disciplines.
Although still in the realm of science fiction, the advent of ASI could open up almost limitless possibilities. ASI could solve humanity’s currently insurmountable problems, such as climate crises, incurable diseases, or even the fundamental mysteries of the universe. It’s unclear if or when ASI might emerge. However, we do know that with incredible potential comes a series of ethical and existential challenges that we can start tackling now.
AI learning models
There are four major learning models for AI: supervised, unsupervised, reinforcement, and deep. Let’s explore each.
Supervised learning
Supervised learning is a machine learning technique in which an AI model is trained on a set of labeled data. The model learns to associate data features with corresponding labels to predict labels for new unlabeled data.
In the business context, it is commonly used for classification and prediction. For example, banks use this technique to assess customer creditworthiness based on historical data.
Unsupervised learning
In contrast, unsupervised learning does not require labeled data for training. The AI model seeks to discover hidden structures or patterns in the data without receiving prior answers. It groups similar data, identifies clusters, and finds correlations.
In the business context, it is often used for customer segmentation, customer sentiment analysis on social media, or supply chain optimization.
Reinforcement learning
Reinforcement learning is an approach where AI learns to make decisions by interacting with an environment. It takes a series of actions and receives rewards or penalties depending on the quality of its decisions. The objective is to maximize the cumulative rewards over time.
This form of AI is particularly useful for companies that need to optimize processes in dynamic and uncertain environments.
Deep learning
Finally, deep learning is a subcategory of machine learning that uses neural networks to model and solve complex problems. These networks are inspired by the functioning of the human brain and are capable of learning hierarchical representations of data. Deep learning has revolutionized many areas of business, including speech recognition, machine translation, and more.
Businesses leverage these technologies to automate data-intensive tasks, improve operational efficiency, and create more personalized customer experiences.
Written with the kind participation of Ana Salvador and all the CACEIS experts.