There is a lot of buzz around Generative AI (Gen AI). But what does it mean? McKinsey Explainer recodes it and says: It is a subset of machine learning that grants computers the remarkable ability to create a wide range of content, from music and art to entire virtual worlds. Yet, its applications extend beyond mere entertainment. Generative AI plays a pivotal role in practical domains such as product design and business process optimisation.
Let’s delve into the creative potential of generative AI and explore the remarkable innovations it can usher into existence.
One system that has recently captured considerable attention is ChatGPT, short for Generative Pretrained Transformer. Developed by OpenAI and made publicly available in November 2022, ChatGPT has swiftly gained recognition as a cutting-edge AI chatbot. In just five days, it attracted over a million users who eagerly embraced its capabilities. These range from generating code and essays to crafting poetry and delivering humor with surprising precision. This has left content creators, from advertising copywriters to esteemed professors, both impressed and, in some cases, apprehensive.
Despite the concerns, it is crucial to acknowledge that machine learning, encompassing generative AI, has demonstrated its potential across diverse industries. Its achievements span from advancing medical image analysis to offering high-resolution weather forecasts.
A 2022 McKinsey survey underscores a significant upsurge in AI adoption over the past five years, accompanied by a substantial increase in AI investments. Generative AI tools like ChatGPT and DALL-E, designed for AI-generated art, hold the potential to transform various job sectors. Nonetheless, the full extent of their impact and the associated risks remain uncertain.
Distinguishing between artificial intelligence (AI) and machine learning
It reveals that AI aims to replicate human intelligence for task execution. You’ve likely interacted with AI through voice assistants like Siri and Alexa or customer service chatbots aiding you during website navigation. Machine learning, on the other hand, is a subset of AI that empowers models to autonomously “learn” from patterns within data, a capability increasingly essential in today’s complex data landscape.
The foundations of machine learning can be traced back to classical statistical techniques developed between the 18th and 20th centuries. The 1930s and 1940s witnessed pioneering efforts by computing luminaries like Alan Turing, who laid the groundwork for basic machine-learning techniques. However, it wasn’t until the late 1970s, with the advent of powerful computers, that these techniques became practical. Initially, machine learning primarily focused on predictive models, which aimed to recognise and classify patterns in data.
For example, a classic machine learning task involved analysing adorable cat images, identifying patterns, and then recognizing other images that matched the “adorable cat” pattern. The emergence of generative AI marked a significant breakthrough. Instead of merely recognising and categorising images, machine learning models gained the ability to create images or generate textual descriptions on demand.
How do text-based machine learning models function, and how are they trained?
Initial machine learning models for text processing were manually trained by humans to classify various inputs based on predetermined labels, a process known as supervised learning. For instance, a model could be trained to classify social media posts as either positive or negative, with a human instructor guiding the learning process. The next generation of text-based machine learning models, however, relies on self-supervised learning. This method entails feeding an extensive dataset of text to the model, enabling it to predict outcomes.
Building a generative AI model has typically been a substantial undertaking, one that only a handful of well-funded tech giants have pursued. OpenAI, the company behind ChatGPT and previous GPT models, has received billions in funding from high-profile donors. DeepMind, a subsidiary of Alphabet (the parent company of Google), and Meta have also entered the generative AI domain. These companies boast some of the world’s most skilled computer scientists and engineers. However, it’s not solely about talent.
Training a model using nearly the entire Internet as a data source incurs significant costs. While OpenAI has not disclosed exact figures, estimates suggest that training GPT-3 involved around 45 terabytes of text data — a quantity equivalent to approximately one million feet of bookshelf space or a quarter of the entire Library of Congress. Such resources are beyond the reach of most typical startups.
What kinds of outputs can generative AI models produce?
Outputs can range from human-like to uncanny, contingent on the model’s quality and suitability for the task at hand. Generative AI has applications in various industries, from IT and software benefiting from AI-generated code, to marketing and even medical imaging, saving time and resources. Businesses can utilise off-the-shelf generative AI models or fine-tune them for specific tasks.
Despite its potential, generative AI presents risks. Outputs may be inaccurate or biased due to internet and societal biases. Organisations must meticulously select training data, opt for smaller models, incorporate human oversight, and refrain from making critical decisions solely reliant on generative AI. As generative AI continues to evolve, businesses and leaders must adapt to evolving regulations and risks. Stay vigilant as this dynamic field integrates into our lives and industries.
