What is generative AI?
How is Generative AI different from regular AI?
Us humans often mix up generative AI with the generative capability behind it ie Large Language Models (LLM). Moreover, we confuse generative AI also with the tools that apply the technology such as ChatGPT or Bard. In our recent sub stacks, we have both with some seriousness and with some humour discussed ChatGPT, Prompt engineering and related topics.
So, what is Generative AI?
Lets take the assistance of one of the popular most talked about generative AI based tools called chatGPT to answer this question first and then give a human perspective on it. At the time of prompting, this what ChatGPT came up with as a crisp definition.
"Generative AI is a branch of artificial intelligence that enables computers to generate new content by learning from existing data. It uses complex algorithms and models to understand patterns and structures in the data, allowing it to create original and realistic outputs. For example, it can generate lifelike images, compose music, or even write stories. This technology harnesses the power of machine learning to mimic human creativity, opening up new possibilities for applications in various fields such as art, entertainment, and design. It's like having a digital assistant that can autonomously produce creative content."
So, I got greedier and prompted what is the difference between generative AI and normal AI to get the response below (I promise you this is not an article that is a prompt / response stream that we have gotten used to from obsessed as well as curious prompt engineers. The next response is the last use of chatGPT just to set the context, and after this human brains kick in :)
“The main difference between generative AI and normal AI lies in their objectives and capabilities. Normal AI, often referred to as traditional or conventional AI, focuses on specific tasks and is designed to provide specific outputs based on predefined rules or patterns. It operates within a set of given instructions and aims to solve problems or perform tasks efficiently. On the other hand, generative AI, also known as creative AI, aims to generate new content that goes beyond predefined rules or patterns. It leverages machine learning techniques, such as deep learning and neural networks, to understand and learn from existing data in order to produce original and creative outputs. Generative AI is more focused on creating something new and innovative, such as generating realistic images, composing music, or generating human-like text.”
There are three things worthy of note from the definition of generative AI:
Eventually the source data is still content created on the publicly available or sourced content from the internet and various sources. The ‘content’ is not lived experiences or lived emotions as a being, without getting drawn into the human vs AI anthropomorphic debates. Most firms are often guarded about this content as it opens the often talked about privacy, ethical and regulatory concerns
However, as the name says ‘generative’, what the AI is able to claim and perform at a very reasonable level of intelligence is fuse concepts, topics, images, algorithms and code to generate content which can again be text, images, videos, code, algorithms, opinions, summaries etc. How such input is converted into output is something other much better computer scientists, data scientists and machine learning engineers than me can answer. Example below:-
Finally, in the conversational form, the tools can hold context very well. Now, this part of contextualising responses is not necessarily completely a large language model trick per se. Its an evolution of conversational AI. Fascinatingly, what this means is that the days of frustratingly scripted chatbots are nearly over. My co-researcher and I published a paper just before the launch of chatGPT and amazingly, our theory behind conversational AI still remains relevant, whether applied with generative AI or normal AI
This leads us nicely into the response to the second prompt.
What is the difference between generative AI and normal AI?
The first difference is obvious in the data sets. Normal AI is trained on narrower data sets compared to the universal characteristic of generative AI’s data set. Secondly, applying the input / output analogy from the implications of the definition of generative AI, Normal AI with the same data and same prompts as input will ideally come up with the same response. However, generative AI won't respond the same and in a very human like fashion will potentially generate different responses. So, if the input prompt is deterministic, like what is 2+2, generative AI should respond 4, unless hallucinating. We will unpick hallucinating probably in a different post but what we mean is that when the tools are working just fine and not misbehaving.
House price forecasting as an example
However, when the inputs are not necessarily deterministic, generative AI can also take a creative license to generative different outputs to the same input. The implications of such a creative license are what makes generative AI different from normal AI as its free to analyse as well as apply patterns in its own ways compared to normal AI where the patterns are already baked in. For example, forecasting the price of a house using normal AI would typically model such a forecast on a set of finite dimensions such as location, historical prices, number of bedrooms, square footage etc. However, generative AI could be prompted to imagine and synthetically simulate how the area might change in 10 years time to then determine forecasts.
And yes, for all the data scientists out there, before you jump on this argument, I agree that if such dimensions can be included and trained in the model, its still possible to include in the forecasts. But, the larger argument is that with generative AI, we can pretend and imagine things like these and are only limited by our own creativity rather than availability of vast loads of historical data and a trained data model.
The fun has only just begun
As much as chatGPT claims about creativity in AI, the fun that us plain humans have been having as an evolving and curious species by asking questions, this is one more way for us to have the creative license to ponder, reflect and evolve. So, as we always have a techmindset nugget, we will continue to reflect on our podcast and our posts as the fun has only just begun. While we were chatting on the below podcast, we realised that keywords like input / output / data set etc means some technical knowledge is needed to understand generative AI, so in a next podcast we explored how to explain generative AI to a child and that will be the topic of the next post.
For our conversation that led to this substack, do checkout and the usual ‘like’, ‘share’ ‘subscribe’ of the podcast below:-