What is generative AI?

Imagine if a parrot could not only repeat words but also learn the patterns in language to create new sentences. Then imagine that parrot consumes petabytes of content from the internet and remembers it all. It doesn’t understand human language, but it can believably respond to anything you say. You’d have something similar to generative artificial intelligence (gen AI)—at least on the surface.

Gen AI uses machine learning models to understand and recreate patterns. Not simply to mimic what they’ve learned, but to produce novel recombinations for original (and believable) text, images, videos, and more.

But, gen AI is far more than a parrot on steroids. Buckle up to learn how it’s going to change the world as we know it.

Generative AI uses algorithms to create new content based on existing data. This process is driven by advanced machine learning techniques, primarily deep learning and neural networks.

Neural networks

Neural networks mimic the human brain's structure and function. These networks consist of layers of nodes, or neurons, that process data and learn patterns.

When trained on large datasets, neural networks can generate new content by predicting and assembling elements based on learned patterns.

1. Train the model. The first step in generative AI is training the model using a huge amount of data. For example, a generative AI model designed to create text (such as GPT-3) is trained on diverse text datasets, including books, articles, and websites. The model learns grammar, context, and nuances of language during this phase.
2. Pattern recognition. As the model processes data, it recognizes patterns and relationships within the input data. For text generation, this might include understanding sentence structure, word associations, and contextual relevance. For image generation, it could involve recognizing shapes, colors, and textures.
3. Generate new content. Once trained, the model uses its learned patterns to generate new content. For text, it can write essays, stories, or articles by predicting what comes next based on the initial input. For images, it can create new visuals by blending learned elements in novel ways.

Deep learning

Deep learning, a subset of machine learning, drives the capabilities of neural networks through multiple layers of processing. These layers allow the model to learn complex patterns and representations from large data sets.

• Layered learning. Deep learning models consist of many layers, each extracting higher-level features from the raw input. For example, in image generation, initial layers might detect edges and simple shapes, while deeper layers identify complex structures such as objects and scenes.
• Backpropagation. This algorithmic approach trains deep learning models. It involves adjusting the weights of connections between neurons based on the error rate of the output and expected result.
• High computational power. Deep learning requires substantial computational resources. Specialized hardware such as graphics processing units and tensor processing units handle the intense computations involved in training deep learning models.

Types of generative AI

Most generative models fall into the following three categories:

• Large language models (LLMs) predict the next word in a sequence based on extensive training data. LLMs are the best for creating textual content and they power many chatbots, writing assistants, and text completion tools.
• Generative adversarial networks (GANs) employ two competing neural networks to produce new content. One network generates content, while the other evaluates it. This back-and-forth process results in increasingly realistic outputs. They work well for creating visual and audio content, such as artificial images or synthetic voices.
• Variational autoencoders (VAEs) compress input into a coded form and then decode it to create a new output. VAEs often produce visual content or code and excel at tasks such as image generation and style transfer.

Generative AI will revolutionize almost every aspect of business in the coming years, but it’s not perfect, nor is it free of negative consequences. Here at Talbot West, given our focus on AI implementation, we spend more time on the use cases of gen AI than on the downsides. However, we still keep our eye on the limitations and adverse effects of artificial intelligence technologies.

As they say, you can’t put the cat back into the bag. Generative AI is here to stay, so we may as well leverage its upsides while managing the ethical issues and other considerations it brings.

Uses for gen AI

Creators have adopted generative AI to create new content, from articles to images to video. Content creation is the most obvious use case for the tech, but it’s far from the only one. Here is a small sampling of some of the uses for gen AI in enterprise:

• Product design: gen AI can automate many subprocesses in the design of new products, prototypes, and simulations. It can also ideate and iterate rapidly to augment human creativity.
• Customer service: enhance chatbots and virtual assistants for more natural and efficient customer interactions.
• Data augmentation: generate synthetic data to improve model training and performance, especially in data-scarce environments.
• Marketing: gen AI can help with audience segmentation, predictive analytics, and automated A/B testing to create targeted, effective marketing campaigns and optimize budget allocation.
• Fraud detection: identify patterns and anomalies in financial transactions to prevent fraudulent activities.
• Drug discovery: accelerating the discovery and development of new drugs by simulating molecular structures and interactions.
• Supply chain optimization: predict demand, optimize inventory management, and improve logistics through advanced simulations.
• Training simulations: create realistic training environments for employees in healthcare, aviation, the military, and other sectors.
• Code generation: automate code writing for software development, speed up the development process, and reduce human error.
• RAG systems: retrieval-augmented generation (RAG) systems give enterprises an internal AI expert for instant and accurate information on SOPs, policies, inventory, pricing, or anything else you want to include.
• Market research: analyze vast datasets to uncover market trends, customer preferences, and competitive insights for more informed decision-making.
• Agentic AI: utilizing autonomous AI agents for such tasks as managing IT systems, optimizing workflows, automating routine processes, and even negotiating contracts.
• Scientific research: generate hypotheses, design experiments, and analyze complex data sets to accelerate discoveries in fields such as biology, chemistry, and physics.
• Engineering: optimize designs, simulate performance, and predict failures in engineering projects for more efficient and innovative solutions.
• Advancing innovation: assist in patent analysis, discovering new materials, and innovating new technologies by synthesizing vast amounts of scientific literature and data.

As with the internet in the 1990s, it’s impossible to predict all the ways generative AI will impact our lives and our organizations. Here at Talbot West, we’re discovering new uses for it all the time.

But make no mistake: the impacts of gen AI will be on a similar scale to the internet. These impacts are already showing up, though in the early stages.

If you’d like to explore how generative AI can drive efficiencies in your workplace, request a free consultation with Talbot West. We can discuss specific tools, implementations, and risk management strategies.

Gen AI limitations

Generative AI technologies are not a golden panacea. They have the following limitations and must be thoughtfully combined with human ingenuity for the best results. We view gen AI as a force multiplier for humans, not a replacement for them.

• Biases: generative AI systems may contain built-in biases from their training data. These biases can be difficult to detect. AI tools are blind to their own biases. The solution: careful attention by human experts to spot and account for biases.
• Closed-loop reinforcement: generative AI tools sometimes participate in a closed feedback loop of reinforcement that’s divorced from reality. One example: generative AI creates internet content which is used to train generative AI models. The solution: original human input into the system, as well as subject matter expertise to spot erroneous closed-loop cycles.
• Hallucination: when a generative AI model doesn’t know something, it doesn’t admit its ignorance. Instead, it will fabricate an answer. Solution: human experts should vet information generated by AI.
• Can't think independently or generate truly new ideas. While generative AI creates original content, it can’t generate truly novel paradigms (at least as of August 2024). This is because it’s limited by the scope of its training dataset.
• Lack of first-hand experience and personal opinions. Generative AI tools can't think for themselves, can’t experience anything, and can’t have an opinion. Human reviews, experiences, and opinions should always be used to enrich generative AI outputs.

Despite these limitations, generative AI’s abilities are truly mindboggling and are increasing by the day. We believe the ultimate gamechanger is the harnessing of human and AI abilities.

While much of the public discourse on AI is fanciful and disconnected from reality, there are some legitimate critiques of the technology, which can broadly be grouped into ethical, ecological, and doomsday categories of concern.

Ethical concerns

Depending on the specific model, generative AI can process, understand, and output a wide range of data types:

• Text. AI can write stories, articles, social media posts, and computer code.
• Images. AI can create new images or edit existing images.
• Audio. AI can create fake voices that sound real, make music, and create sound effects.
• Video. AI can generate video, including realistic-seeming “live-action” footage, animations, and more.
• Multimodal. Advanced generative AI systems can work with different types of data at the same time. For example, creating a video with sound, or a website with text, images, and graphics.
• Synthetic data. Generative AI can create synthetic datasets that mimic real-world data. This is useful for training other AI systems, especially when real data is hard to get or needs to be kept private.

Foundation models are large-scale machine learning systems that form the core technology for generative AI applications. 

Think of a foundation model as a car engine, and generative AI applications as the vehicle that the engine powers. The engine on its own can’t do much, but an engine-powered car can be driven to the grandparent’s house, take one’s teenager to soccer practice, and do many other tasks. 

Similarly, a foundation model is the “engine” that powers a generative AI tool such as ChatGPT. 

Parameters are the building blocks of foundation models, functioning like the model's "brain cells." They store the patterns and relationships the model learns from its training data. More parameters allow a model to capture more complex language nuances and store broader knowledge. This increased capacity translates to better performance, but more parameters also equate to higher computational demands and increased resource usage.

Here are some of the leading foundation models on the market:

• 1956: the term "artificial intelligence" was coined by John McCarthy at the Dartmouth Conference and represents the birth of AI as a field of study.
• 1960s: early AI systems such as ELIZA, developed by Joseph Weizenbaum, simulate human language.
• 1972: the Prolog programming language is developed, which becomes essential for AI research and the development of logic-based AI systems.
• 1980s: the rise of expert systems, which use rule-based models to simulate human expertise in specific domains, advances the understanding of AI's potential.
• 1997: IBM's Deep Blue defeats world chess champion Garry Kasparov and showcases the power of machine learning and AI in complex problem-solving.
• 2006: Geoffrey Hinton and his team introduce the concept of deep learning, which involves training artificial neural networks with multiple layers. This breakthrough leads to significant advancements in AI's ability to process and generate complex data.
• 2014: Ian Goodfellow and his colleagues introduce Generative Adversarial Networks (GANs), a revolutionary approach that pits two neural networks against each other to improve the quality of generated data. One network, the generator, creates fake data, while the other, the discriminator, evaluates its authenticity. This competition drives both to improve.
• 2015: OpenAI is founded with the mission to ensure that artificial general intelligence (AGI) benefits all of humanity. This organization is one of the most important in advancing generative AI research.
• 2018: OpenAI releases GPT-2, a generative pre-trained transformer model capable of generating coherent and contextually relevant text based on a given prompt.
• 2020: OpenAI releases GPT-3, the successor to GPT-2, with 175 billion parameters. GPT-3 showcases unprecedented language generation capabilities and leads to widespread adoption in different applications, from chatbots to creative content writing.
• 2023: the release of GPT-4 by OpenAI brings further advancements in natural language processing.
• 2024: Anthropic’s Claude beats GPT-4 in several accepted benchmarks of generative AI performance.

By 2025, generative AI is expected to produce 30% of outbound messaging, and by 2026, it could create up to 90% of online content. This rapid adoption signals a transformative shift in how businesses operate.

AI will be an integral part of creative processes, data analysis, and decision-making across industries. As the technology evolves, we can expect more sophisticated, context-aware AI systems that can handle a range of tasks and collaborate seamlessly with humans.

This growth will also bring challenges, including the need for robust AI governance, ethical considerations, and the evolution of human roles in an AI-augmented workplace. The future of generative AI promises increased efficiency and innovation, but it will require careful navigation to maximize benefits while addressing risks.