Madhav Malhotra
To understand Generative Artificial Intelligence (Generative AI) and why it matters as a skill, it helps to first take a step back and look at the broader field of Artificial Intelligence (AI) from first principles.
What is Artificial Intelligence?
Artificial Intelligence (AI) is all about making machines smarter. It involves giving the machines the ability to perform tasks that usually need human intelligence. For example recognizing objects, reasoning, conversation, etc. Essentially, we are trying to make the machines behave close to human intelligence. Hence the word Artificial Intelligence
How does AI learn?
Just as humans learn from a variety of sources such as school, books, travel, interactions with friends, family etc. In the same light, AI systems learn from data.
For instance, to train an AI to recognize cats in photos, you would provide thousands of cat images. Over time, the AI learns what makes a cat—a process that improves as it gets more data. There are a number of different learning methodologies that can be used to make AI learn from data.
Once an AI has been trained with sufficient amounts of data and an appropriate learning algorithm, we get what is known as a model. Typically, each AI model is built for a specific purpose/task.
What is Generative AI?
In the “cat recognizer” example, the AI model identifies cats in images. This is called a classifier model, because it classifies or categorizes inputs
Generative AI is a special type of AI model designed to create new content. Generative AI can actually generate things that didn’t exist before, a job previously reserved for humans. If we train an AI system on thousands of cat images and it can generate completely new images of cats, that is a Generative AI model.
Some examples, but are not limited to:
- Writing essays and articles
- Creating artwork
- Solving complex math problems
- Generating code
As we are getting more advanced generative AI models, we are still learning about their various capabilities.
Some examples of Generative AI models include:
- ChatGPT, which is a language generative AI model (more on what that is lateR)
- Dall-E: An AI model that creates images.
- Midjourney: Another text to image tool.
But why is it important?
The simple answer to this question, as with many things in life, is it depends. Whether Generative AI is a skill that you need depends on what kind of use case you want to use it for. Generative AI models are especially powerful in generating new things because of their learned patterns from massive amounts of data.
A simple mental model for deciding if you need Generative AI models is -
- Do you need to create something new (an email, a design, a campaign, a prototype, a script, a model)?
- Is there a generative AI model that exists out there which can used for me to get started.
The get started aspect is really important. We at QurioSkill believe that generative ai models, depending on your use case, can be an augment in creating something new. But, the human in the loop is very important because, as we are going to see in the next section, these models come with downsides that are important to be considered. A skilled human continues to be an important quality checker for the output of these models.
Having said that, with the proliferation of hardware capabilities, we are seeing more and more useful generative ai models. And being able to use them for improving productivity and quality can be an important skill of the future.
What are some Generative AI model gotchas?
Generative AI models can have some gotchas, some of them being:
- Hallucinations: They can produce outputs that are incorrect, illogical, or completely fabricated.
- Reasoning Skills: Their reasoning abilities are improving but still limited; outputs must be checked.
- Knowledge Limitations: They are trained on fixed datasets, so they may not reflect the most current information.
- Plagiarism and Ethics: There are ethical concerns regarding how the data is sourced and used in model training with pending court cases.