Last Updated on February 18, 2024
Imagine human intelligence as a combination of senses – our eyes, ears, mouth – along with our inherent ability for common sense reasoning. Now, take this intricate tapestry of human cognition and reimagine it within the realm of computers. This is the essence of Artificial Intelligence (AI).
The Chessboard Conundrum: Learning from Every Move
To better understand AI and Machine Learning, think of AI as a discipline, like Electrical Engineering. Machine Learning is a subfield of AI, just like Electronic Engineering is a subfield of Electrical Engineering.
To understand how AI works, let’s delve into the world of chess. Picture a board, the pieces, and every potential move. Traditional programming would approach this as a problem of possibilities. Using the Minimax Algorithm, a program would consider the best move for itself and the worst move for its opponent.
Yet, the complexity deepens quickly; after only 5 moves, there are approximately 70 million possible outcomes. And considering 5 x 10^44 moves overall, it’s infeasible to hard-code them all. Hence, modern AI needs software that can learn the optimal moves rather than being explicitly programmed for each one.
This software AI writes itself – we call this kind of Machine Learning, Deep Learning. In this instance, computers learning by playing endless solitary games of chess.
The Nuances of Intelligence: Weak vs. Strong AI
The domain of AI is broadly divided into weak (or narrow), and strong AI. The former automates tasks without true comprehension, like determining what constitutes a chess move. The latter, however, attempts to replicate human thought, discerning not just ‘what’ but ‘why’ – what is a good chess move? (more: https://www.ibm.com/topics/strong-ai)
This delineation further extends into logical conclusions drawn from direct information (symbolic programs) versus associative results based on patterns and inference (subsymbolic programs). (more: https://ceur-ws.org/Vol-2699/paper06.pdf)
The Limitations of Generative AI
The power of AI in our work places and homes is being talked about by everybody. While many laud the power of AI, its current capabilities, especially in the generative domain, remain bounded. Generative AI can produce countless examples based on statistical data, context, and abstractions, especially in language training. Yet, its primary limitation is the inability to truly explain its output. It can generate, but not rationalize.
Neural Networks: The Pervasive Threads of Modern Life
Today, the term ‘neural networks’ resonates far beyond academic circles. AI models are everywhere in our lives, from smartphones to social media networks and even production centers. AI traces its lineage traces back to 1956, marking decades of evolution. This is only the beginning.
As AI continues to progress and intertwine with our daily lives, it’s essential to recognize its capabilities, distinctions, and limitations. The future beckons with promise, but like any tool, its utility rests in our hands.
Further Reading
Artificial Intelligence and Online Marketing
Artificial Intelligence is crucial for online marketing in small businesses, nonprofits, and schools. Why? AI personalizes engagement, streamlines operations, and provides data insights. Using AI in your marketing mix has benefits such as personalized content and automated tasks. AI can help you create a website. AI can support your SEO strategy too. However, there are challenges, such as additional costs and potential privacy issues.
Using AI tools in a balanced way helps create effective and ethical online marketing strategies. Find out more about how you can use AI in your online marketing!
AI Terms and Definitions
A foundation AI model refers to a substantial AI model that undergoes pretraining on an extensive dataset, with the intention of being “adapted” or fine-tuned for various downstream tasks, such as sentiment analysis, image captioning, and object recognition.
A generative AI model might be trained on a dataset of cat images and subsequently employed to create novel cat images. A discriminative AI model could be trained on a dataset featuring both cat and dog images, then utilized to categorize new images as either cats or dogs.
A prompt, in the context of a large language model, is a brief piece of text provided as input to control and guide the model’s output in various ways.
Sometimes you don’t get what you expect, do you? An AI hallucination occurs when an AI model generates incorrect information but presents it as if it were actual facts. Hallucinations in Generative AI pertain to words or phrases generated by the model that are often nonsensical or grammatically incorrect. Several factors can contribute to such hallucinations:
- + The model is trained on data that is noisy or contains errors.
- + The model lacks sufficient training data.
- + The model lacks the necessary contextual information.
Image Content credentials: Post Image Generated with Bing Image Creator AI ∙ October 10, 2023 at 12:03 PM
Article notes 11.10.2023 from Digital Workshop VHS Wildeshausen by Warren Laine-Naida. Article by ChaptGPT4, Editing by Grammarly and Hemmingway AI.