Personal commitment to deep learning

In my interaction with clients on strategic growth issues, there are often high-profile topics that my clients are grappling with but where the application to their business and impact on their competitive environment is less clear. One of my commitments each year to personal learning is to pick one such high profile topic where my knowledge is immature and views not well-formed. I dedicate time to learn more, decipher the topic from a ‘layperson’ perspective, and hopefully form a point of view on applications and implications. I then use this point of view to create a frame of reference to assist in interpreting and updating my understanding of the topic over time.

A three part insights series on Generative Artificial intelligence (GenAI) for the layperson

Last years’ personal learning focus was likely the ‘topic of the year’ being Generative Artificial Intelligence (GenAI). The outcome of this learning enquiry is contained in this three-part series that is divided into each of the explanation, applications, and implications of GenAI from the perspective of a layperson (i.e. non-expert). My hope in sharing this learning enquiry is that it will assist you to interpret the language around GenAI and help form your own point of view around potential application cases and implications of GenAI technology individually, in your business context, and more broadly for the society in which we are all members.

While AI has become almost imperceptible in our lives this is not the case for GenAI

AI has been permeating our lives incrementally over the past decades through everything from powering our smartphones to the autonomous driving features on our cars to the personalisation tools that retailers have been using to surprise, delight, and seduce us as consumers. As a result its march into our lives has become almost imperceptible. However, this is not the case with GenAI.

GenAI has exploded into the consciousness of the average person over the last year. This has (almost) singlehandedly been achieved by the democratised access to Chat-GPT (which reached 100 million users in two months) which is the personification of the technologies’ capability. Users don’t need a degree in machine learning to interact with or derive value from it; nearly anyone who can ask questions can use it. GenAI applications can perform a range of routine tasks, such as the reorganisation and classification of data. But it is their ability to write text, compose music, and create digital art that has garnered headlines and persuaded consumers and households to experiment on their own.

There is a broad spectrum of views on the technology’s prospects for humanity

As a result, a broader set of stakeholders are grappling with GenAI’s impact on individuals, business, and society than generations. However, they are not necessarily equipped with the full context to help them make sense of it. Consequently, there is a very broad spectrum of views on the technology’s prospect for humanity. For example, Henry Kissinger predicts that GenAI will transform human cognition processes in ways not done since the onset of printing. Yuval Noah Harari fears that the consequences of decoding humanity’s operating systems being language.

In so many respects GenAI is not new – rather another iteration of an evolving automation and AI capability accumulating over past centuries. The focus of this first article is to help define and provide relationships between a range of terms that are regularly referenced, and often used incorrectly or interchangeably, making understanding of the subject matter of GenAi more challenging.

Explanation of key terms

Let’s start with Artificial Intelligence (AI). AI is a broad term (more like a discipline) that encompasses machines or computer programs that perform tasks that typically require human intelligence. AI systems are designed to perceive their environment, learn from it, reason and make decisions to achieve specific goals.

In the early days this capability was (in hindsight) relatively narrow (Artificial Narrow Intelligence ANI) and focused on the areas of machine learning capabilities that were used to solve one or narrow problems. Over time the focus of problems has broadened (Artificial General Intelligence AGI) and the machine learning capability enhanced towards machine intelligence incorporating capabilities such as deep learning, natural language processing, neural networks, and computer vision. The onset of the Graphical User Interface (GUI) that underpinned the digital gaming market and the internet at the turn of the century is the forefather of GenAI. Most experts’ expectations are that we are currently in a mid-state towards an ultimate destination of AI super intelligence (ASI) where machines possess an intellect smarter than humans in virtually all fields.

AI encompasses a wide range of techniques, including machine learning (ML) and deep learning (DL), to solve complex problems and make intelligent decisions. I outline some of the common AI terms and relationship between them.

Machine learning (ML) is a subset of AI that focuses on algorithms and models that enable computers to learn patterns and make predictions from data without explicit programming. ML algorithms can be trained on labelled data to recognise patterns and make decisions or predictions based on new, unseen data. ML finds applications in recommendation systems (Netflix, Amazon), image recognition (Google Photos), natural language processing (virtual assistants), and fraud detection (credit card transactions).

Deep learning (DL), is a subset of ML (other subsets include supervised and unsupervised learning, and reinforcement learning) that involves the use of artificial neural networks, inspired by the structure and functioning of the human brain. Deep learning models, particularly deep neural networks, consist of multiple layers of interconnected neurons that learn hierarchical representations of data. They excel at learning complex patterns and have achieved remarkable success in various domains, including image and speech recognition, autonomous vehicles, natural language processing, and medical diagnostics.

Generative AI (GenAI), is a subset of DL. Generative models learn the underlying patterns and structures of the data, enabling them to generate outputs that resemble the training data. Importantly, Gen Ai models are not Intelligent (only designed to give a likely continuation of the prompt given training data), deterministic (not guaranteed to give the same response given the same prompt) or, intentional (exist to complete your train of thought i.e. they do not perform intentional actions). Types of GenAI model include text to text, text to image, text to music, and text to voice. These model types enable GenAI to have applications in art, music, literature, and computer-generated imagery (CGI). For example, GenAI algorithms can analyse existing artworks and generate new pieces in the style of renowned artists.

Large Language models (LLMs) or foundation models are a specific type of GenAI model that specialises in generating coherent and contextually relevant text. These models, such as OpenAI’s GPT (Generative Pre-trained Transformer), are trained on vast quantities of unstructured, unlabeled data in a variety of formats, such as text and audio. Unlike the earlier ‘narrow’ AI models, the same foundational models can perform a wide range of tasks. This could range for example from creating an executive summary for a 20,000-word technical report on quantum computing, draft a go-to-market strategy for a tree-trimming business, and provide five different recipes for the ten ingredients in someone’s refrigerator. The downside to such versatility is that, for now, GenAI can sometimes provide less accurate results, placing renewed attention on AI risk management.

Distinguishing Artificial General Intelligence from Generative Artificial Intelligence

Having established some definitions and relationship between methodologies in AI, it is now worth exploring whether the often-interchangeable use of the terms Artificial General Intelligence (AGI) and Generative Artificial Intelligence (GenAI) has merit.

AGI is a type of AI that understands, learns, and applies knowledge to various tasks. It can adapt to any situation and perform any intellectual task a human can and should possess abilities like abstract thinking, background knowledge, common sense, understanding cause and effect, and transfer learning. True AGI capability requires machines satisfying two human capabilities – consciousness and feeling. To be clear, current AI systems, including GPT-4, are not true AGI as they lack these full human-like comprehension.

Comparatively, GenAI is a field that revolves around the development of intelligent systems capable of generating new and original content. Unlike traditional AI systems that rely on predefined rules and patterns, GenAI models learn from vast amounts of unstructured data to generate outputs that closely resemble human-created content. It is designed for a wide range of tasks but lacks AGI’s comprehensive understanding or learning ability. It creates original content such as images, text, music, or code, using extensive data to produce relevant and realistic outputs.

While AGI is the dream, GenAI is often seen as the more practical approach to AI. GenAI systems are designed to be flexible and adaptable, able to handle a wide variety of tasks without needing to be specifically programmed for each one. For instance, a GenAI system might be used in a customer service role, where it needs to handle a wide variety of customer inquiries. The system wouldn’t need to be programmed with specific responses to every possible inquiry. Instead, it could use its programming to understand the customer’s question and generate appropriate responses.

GenAI is a not AGI but a subset of ML and DL that uses vast training data to transform or generate new content

In conclusion, GenAI is different and less evolved than AGI and is a subfield of ML that focuses on transforming or creating new content. It leverages generative or foundational LLM models to deliver outputs resembling human-created content. While AI encompasses a broader range of techniques, ML and DL play significant roles in GenAI.

Stay tuned for the next instalment of this GenAI for the layperson series where I focus on the emerging and potential applications of GenAI.