This guide introduces key terminology, mechanics, capabilities and considerations relevant to business leaders seeking to deploy AI at work.
Image credit: DALL-E
The concept of Artificial Intelligence (AI) has long captivated the human imagination. Before it became a field of scientific inquiry in the late 1950s, authors and filmmakers were painting vivid pictures of intelligent machines capable of independent thought and action.
For almost 70 years, people paid little attention to AI. But with the release of ChatGPT, the hype surrounding the technology has exploded. AI now dominates the news, stock markets and political discourse with raging debates about its potential to change the world for the better, render humans obsolete, or simply fail to deliver on the hype.
To non-technical business professionals (i.e., those of us without a PhD in mathematics), these discussions often seem abstract or irrelevant. Even if you believe that AI has the potential to revolutionize the world of work, sifting through the endless barrage of information and actually capitalizing on the opportunity can feel like an overwhelming task.
This guide filters out the noise, distilling information and insights that are critical for understanding AI and successfully integrating the technology within a business.
Artificial Intelligence (AI) is a branch of computer science. It deals with the creation of machine-based systems that can perform tasks normally requiring humans. Given a set of human-defined objectives, these systems can simulate intelligent behavior.
Data is the most vital ingredient of effective AI systems. Data is any information that can be processed or transferred. This includes numerical, textual, visual, and aural information. The quality, quantity, and diversity of data directly impacts the effectiveness of AI systems.
Algorithms are the sets of instructions followed by AI systems. An algorithm is the programming that tells a computer how to operate on its own.
Machine Learning (ML) is an application of AI in which a system is designed to automatically learn and improve from experience. ML models are trained on existing datasets and can make useful predictions or generate content when new data points are introduced to the system.
Deep Learning is a powerful subset of ML methods inspired by the human brain. It uses artificial neural networks – many layers of interconnected systems – to enable processing of more complex tasks, such as:
Image credit: Simon J.D. Prince ('Understanding Deep Learning')
Large Language Models (LLMs) are deep learning models that are pre-trained on vast amounts of data. They can understand the relationships between words and phrases and extract meaning from a sequence of text. LLMs are incredibly flexible, with a single model capable of multiple tasks, such as answering questions, summarizing documents, translating languages or completing sentences.
Generative AI (GenAI) models are capable of generating text, images and a variety of other data that has similar characteristics to their input training data. GenAI models are often guided by user prompts, the quality of which greatly affects the quality and accuracy of output content.
A high-level awareness of AI training methodologies is useful for understanding what AI can and can’t do. The machine learning methods used to train AI systems can be divided into three main areas: supervised, unsupervised and reinforcement learning, each of which has unique applications and mechanisms.
Supervised learning relies on labeled datasets to teach models to predict outcomes.
By analyzing examples with known answers, the model trains itself by iteratively making predictions and adjusting to minimize error. In this way, the model learns how to infer the "correct" response from new, unseen data.
This approach is key for applications like fraud detection, where the system discerns patterns by comparing transactions against a historical database of fraudulent and legitimate activities.
Image credit: Google ('Introduction to Generative AI')
In contrast, unsupervised learning analyzes unlabeled data to uncover hidden patterns or characteristics, without having been given any prior instructions.
This method excels in tasks that require segmenting data into distinct groups, such as for targeted marketing campaigns where there is a need to identify similarities and differences among customers without prior categorization.
Image credit: Google ('Introduction to Generative AI')
Reinforcement learning is used to train models to make sequences of decisions in highly complex environments.
Similar to training a pet with treats, algorithms use reward-and-punishment as they process data and work towards a desired objective; actions that move the system closer towards your goal are reinforced, while actions that detract from the goal are ignored. Systems learn from the feedback of each action and self-discover the best processing paths to achieve optimal outcomes, backtracking and incorporating delayed gratification when necessary.
This technique shines in dynamic environments requiring adaptation and optimization over time, such as in autonomous vehicle navigation or optimizing trading strategies in financial markets.
Image credit: MathWorks ('What Is Reinforcement Learning?')
AI has the capability to fundamentally change how work is done. The University of Pennsylvania predicts that "approximately 19% of workers may see at least 50% of their tasks impacted" by the introduction of LLMs.
With such a dramatic shift taking place, businesses risk getting left behind unless they pivot to reimagine internal processes that will foster innovation, drive efficiency and enhance customer experiences.
Josh Kaufman's '5 Parts of Every Business' is a useful framework for analyzing a company and assessing opportunities for improvement. Although every organization differs in its specific operations, the following examples should help to spark ideas about how AI could be applied within your business.
Discovering what people need or want, then creating it.
Attracting attention and building demand for what you've created.
Turning prospective customers into paying customers.
Giving your customers what you've promised and ensuring that they're satisfied.
Bringing in enough money to keep going and make your effort worthwhile.
The number of tools vying to address these opportunities is enormous. At the time of writing, There’s an AI for That (an AI tool aggregator) lists "12,236 AIs for 16,604 tasks and 4,847 jobs". These numbers do not include the millions of existing software applications that are currently integrating AI into their feature sets.
Picking tools at random and hoping for results is a recipe for failure. To ensure the successful implementation of AI within an organization, business leaders must consider the following points.
There are abundant resources available to those who wish to continue learning about AI. These are just a few that I have found valuable – and most are entirely free!