An Introduction to Generative AI

What is Generative AI?

Generative AI is a form of AI that is capable of generating text, images, video, or audio in response to a prompt. It is an advanced form of AI that generates new data, whereas traditional AI makes computers follow simple rules. While conventional AI analyses data to find patterns, generative AI creates new data. 

The modern era of generative AI began between 2013 and 2017. Generative adversarial networks (GANs) improved AI-generated image quality with their emergence in 2014. Furthermore, the transformer architecture revolutionised how AI models process and understand language in 2017. This was followed by OpenAI's release of ChatGPT in 2022, which demonstrated AI’s ability to generate coherent and contextually relevant responses. This breakthrough accelerated enterprise adoption and triggered an explosion of development, with Google, Microsoft, IBM, and other technology providers introducing competing platforms. 

GenAI vs Agent-based AI

Generative AI is a technology that can generate content and understand complex instructions. AI agents' capabilities go a step further, as they use generative models for goal-oriented tasks, reasoning, planning, and executing them across systems. So, while generative AI can produce a business report, an AI agent can interpret a business objective, gather appropriate data, produce the report and send it to the right people. Think of generative AI as the engine that powers organisational work, while an AI agent is a goal-oriented digital co-worker powered by that engine and able to act across systems, both guided and supervised by humans. Let’s look into the distinction between key characteristics of generative AI and agent-based AI in the table below: 

Types

We can understand the types of generative AI models by either thinking about the different types or by the types of content each can generate. Irrespective of the type of generative AI model and its distinctive architectural approach, they all aim to generate content in various forms. The following are four different types of generative AI models:

Generative adversarial networks (GANs)

GANs consist of two duelling neural networks, namely the generator and the discriminator. While the generator's job is to create fake content indistinguishable from the training data, the discriminator's job is to detect it. Both neural networks go back and forth until the generator wins or the discriminator can’t tell the difference between the fake item and the training data. 

Variational autoencoders (VAEs)

VAEs encode data into a simplified representation that retains critical elements while omitting details. After encoding, the decoder generates new details around the most essential critical elements, essentially rebuilding the original dataset. The process introduces a bit of randomness, helping create unique items or variations of the original input. 

Transformer-based models

These are deep learning models that comprehend text by breaking it down into tokens. Tokens are small components of text that contain a character, a part of a word, or a short phrase. The model then converts tokens into numerical vectors and analyses their relationships. Transformers also use a self-attention mechanism that helps them understand the relative importance of words in a sentence. 

Diffusion models

Diffusion models add noise (random sets of data points) to the input to distort the data, study how that process alters it, and then reconstruct a reverse-diffused version of the original input. This process helps the AI model understand how the data elements relate to one another. After training, the model becomes capable of utilising what it learns about the patterns in its training materials to generate content that meets the prompt request. 

Autoregressive models

These models generate content sequentially, with each new element depending on all previous elements. The models generate content one piece at a time, using what they have already created to decide what comes next. 

Benefits & Risks

Generative AI presents organisations with a complex value proposition that requires careful evaluation of both opportunities and challenges. In this section, we will review the benefits and challenges that the technology brings. 

Benefits of Generative AI

• Varied outputs: Generative AI can produce a wide range of original outputs. It creates content by capturing patterns in data that may not have been detected earlier, before training. This provides a different lens through which to view a topic or problem. 

• Contextually sensitive outputs: Generative AI's ability to process and interpret human language in a conversational style enables it to generate contextually relevant responses to user prompts. Other capabilities include generating text in multiple languages, summarising long texts and enabling better comprehension of complicated material. 

• Boost organisational productivity: Generative AI models can be fine-tuned for various tasks across industries and domains, enhancing employee productivity and enabling the efficient utilisation of organisational resources. Some of the most popular use cases of generative AI in companies include drafting initial reports, organising text, and producing meeting summaries. 

• Personalisation: Generative AI models can remember earlier interactions, resulting in a more coherent experience for users. Models are now capable of remembering a project requirement from start to finish, including the aim and requirements. 

Risks of Generative AI

• Intellectual property: Most of the data used to train generative AI models is sourced from documents and webpages available on the internet, most of which is used without permission from their owners. Therefore, the output’s appearance of creativity and originality puts the onus on the users to make the judgment call. Due to the same reason, it is imperative that companies train their employees to be mindful of the data they upload to these models. 

• Carbon footprint: Generative AI models require significant amounts of electricity for training, generating substantial carbon emissions that contribute to climate change. For instance, according to estimates, ChatGPT-4 requires between 51,772 and 62,318 MWh of electricity. 

• Feedback loop: As generative AI models are trained on data available on the internet, which inadvertently includes biases inherent in humans, their outputs reflect this bias. For instance, historically, women and blacks have been at a disadvantage, and when a company trains a model to shortlist candidates. 

• Ethical, social, and human costs: The training process for generative AI models has become a cause of concern, as some companies prefer Reinforcement learning from human feedback (RLHF), in which humans review the outputs of generative AI models for accuracy, appropriateness, and alignment. Humans employed for such tasks work under inhumane conditions for as low as 3$/hour. 

Generative AI Use Cases

Generative AI is used across industries such as software development, marketing, advertising, financial services, and healthcare. Let's review some of the applications in this section, and we will then discuss which one provides the highest ROI and where regulatory and technological challenges are expected to be higher: 

• Software development: Generative AI is helping software teams generate and optimise code faster. Other use cases within the field include auto-completion and debugging. Software developers using generative AI tools have reported a productivity increase of 20-30%. 

• Customer service and support: Generative AI has enabled more intelligent, personalised, and efficient customer interactions. Chatbots and virtual assistants powered by generative AI can engage in fluid, human-like conversations. This has resulted in faster first-contact resolution rates and improved customer satisfaction. Many enterprises have reported a 15-25% reduction in customer support costs and an increase in Net Promoter Score(NPS). 

• Content creation and marketing automation: This has become the most common and impactful use case, as generative AI tools help create diverse marketing content. This has significantly reduced the content creation cycles by up to 50% while enhancing the personalisation of customer communication. 

• Product design and innovation: Generative AI can help develop multiple design variations, accelerating ideation by generating multiple design concepts, simulating product performance, and creating synthetic data for testing. 

Even though the technology offers many advancements, the industries that adopt it are still not free from the technological barriers that arise. For instance, because technology evolves rapidly, status regulations struggle to keep pace with AI's evolution. LLMs that form the basis of generative AI, such as ChatGPT, the first models to become popular for generating human-like text, require large amounts of data and significant electricity consumption. The functioning of these models depends on high-quality data, which is challenging to ensure, thereby increasing the risk of hallucination and the provision of false or fabricated information.

Additionally, since companies started reaping the benefits of advancements brought in by generative AI, there have also been regulatory and compliance challenges, which have imposed responsibilities on developers and deployers alike to ensure ethical use owing to challenges encountered, such as data privacy and protection, intellectual property breach, explosion of misinformation and deepfakes, and bias. 

GenAI Tools

The generative AI landscape offers enterprises a diverse array of specialised tools designed for various business functions, from content creation to marketing automation and video production. In this section, we will review some popular tools that help companies achieve efficiency at scale: 

• ChatGPT by OpenAI: ChatGPT is a generative AI model developed by OpenAI and built on GPT-4 architecture. Its most widely used applications include generating marketing content, providing customer support, brainstorming ideas, and more. The tool is widely integrated into business workflows, and its plugin ecosystem makes it versatile for enterprise use. 

• Adobe Firefly: This tool lets users generate images, vector art, and design elements from simple text prompts. The tool is focused towards creative professionals and ensures copyright-safe outputs as it is trained on Adobe’s custom datasets. Firefly offers several applications, including faster prototyping and enhanced creativity. 

• GitHub Copilot: This tool, jointly developed by OpenAI and GitHub, helps developers worldwide write code faster by suggesting code blocks based on natural language prompts. Copilot is useful across technical stacks as it supports multiple languages, including Python, JavaScript, and Go. The tool supports rapid software prototyping and boosts developer productivity. 

• Notion AI: Notion AI is integrated into the Notion productivity platform and is an intelligent assistant that helps users summarise content, generate text, and brainstorm ideas, all within a unified workplace. The tool offers context-aware suggestions and edits and is ideal for startups and educators, enabling faster decision-making and more organised workflows. 

• Canva AI (Magic Studio): Canva Magic Studio is a suite of AI-powered design tools integrated into Canva. It includes several notable features that make it easier for design novices to create polished content, such as presentations, mass invitations, posters, and marketing assets. Canva also uses machine learning to understand user intent that aligns with brand aesthetics. 

• Jasper AI: This is a popular tool for content creation targeted towards marketing, copywriting, and SEO tasks. It enables users to create high-converting content for websites, blogs, emails, and ads using customised templates and brand voices. Jasper’s AI models are trained on high-quality marketing data, making them adaptable to different tones and styles, aligned with the company's branding consistency. 

• Synthesia: This AI-powered video creation platform lets users generate professional-looking videos with AI avatars and voiceovers using simple prompts. It's popularly used for creating training videos, product explainers, and ads without the need for paid actors or production equipment. It's helping companies save time and money. 

Conclusion

Generative AI is a transformative technology that is reshaping how organisations create value, serve their customers, and empower employees. Early movers and adopters are already realising the competitive advantage through enhanced customer service, amplified employee performance and overall organisational efficiency. However, successful deployment requires robust governance frameworks, ethical data practices, and continuous monitoring for bias and accuracy. Therefore, business leaders need to weigh the pros and cons when deploying strategically.