Generative AI

Background

Now that artificial intelligence can automate complex tasks, extract useful insights from large data sets, and help workers boost their productivity, AI has crossed over from science fiction to the mainstream.

Generative AI, meanwhile, is an emergent form of AI that can actually mimic human imagination and creativity. The term “generative” refers to models’ ability to generate original content, including text, video, audio, and more.

How it works

Prediction is at the heart of how generative AI systems work. Most users interact with these systems through different applications—for example, by typing a prompt into OpenAI’s ChatGPT.

In the case of ChatGPT, the chatbot relies on a large language model that has been “trained” on huge data sets. This underlying model has analyzed much of the public internet, giving it a general understanding of which words are most likely to go together in a sequence. See a more technical breakdown of how ChatGPT works here.

Other AI tools, particularly those creating images and videos, rely on a different approach known as generative adversarial networks. The machine-learning framework involves two systems working against each other—one, the “generator,” produces something like an image to try and trick the other model, the “discriminator.”

In the process (deep dive here), the generator’s output gets incrementally more realistic. This approach is particularly useful for applications that create images and videos, like Midjourney and OpenAI’s popular DALL-E.

How are people using generative AI?

More than 100 million people are now using ChatGPT each week for tasks that range from the mundane to the creative. ChatGPT can do everything from writing an essay to drafting an email, debugging computer code, summarizing articles, and explaining complex topics in a human-like style (try ChatGPT here).

The use of text-to-image and text-to-video applications has also become increasingly popular. These tools allow the rapid creation of content ranging from abstract art to hyper-realistic footage based on simple inputs (Midjourney, DALL-E, and Sora).

Beyond creating text-, image-, and video-based content, generative AI underpins new products for searching the web (Perplexity), digital assistants (Google’s Gemini, Microsoft’s Bard), and more.

What are the risks?

Estimates vary, but due to generative AI, it’s believed that jobs involving largely rote, repetitive tasks will eventually shift from humans to machines, potentially displacing a large number of jobs across a number of industries.

From a technical perspective, generative AI programs sometimes “hallucinate” erroneous or made-up information (for more on AI hallucinations, what they are, and why they happen, Google offers this guide).

Meanwhile, hackers are also already using generative AI to augment their efforts. Generative AI makes some things easier—like lowering the barrier to content creation, for example, in terms of both cost and speed. But lowering that barrier for everybody, by definition, also means lowering it for bad actors.

Generative AI has also made it easier to create content like fake news articles as well as disinformation and propaganda.