Generative AI is all set to transform lives

This article was first published by the author in Sampada magazine

Before we begin, let’s start with a piece of historical trivia. In 1736, when the Shanivar Wada in Pune was constructed, the famous artist Leonardo Da Vinci presented Bajirao Peshwa with his masterpiece painting shown above.

Before you start Googling and sending me hate mails, let me admit that this is indeed a false story. However, the above painting is “The Shaniwar Wada” rendered in the style of Leonardo Da Vinci by an AI algorithm. This is a new style of algorithms called Generative AI that is changing the way we live and do Business.

What is this Technology about?

Generative AI has recently gained a lot of attention for both positive and negative reasons, with techniques like Stable Diffusion, GPT3 and ChatGPT gaining major media attention. We have ChatGPT used by people to create poetry, write book chapters, suggest funny titles, and even generate defect-free code. All it takes is a text “prompt” and it does rest of the magic. If you have not yet tried it just go to the link and register for free and start asking it questions: https://chat.openai.com/chat . Besides the creative uses, enterprises are actively discussing how their business model will change with this game-changer technology and how to get ready for this disruption.

Generative AI refers to a class of artificial intelligence systems that can create new and unique outputs based on some input data. The outputs can range from text, images, audio, or any other form of media. The main goal of generative AI is to mimic human creativity and generate new and diverse outputs. Stable Diffusion is a generative AI model that uses deep learning algorithms to generate images based on a specific style. It learns from a dataset of images and uses this knowledge to generate new images that resemble the style of the input data. The goal of this process is to create images that are so close to the input data that they cannot be distinguished from real images. GPT-3 (Generative Pretrained Transformer 3) is a language generation AI model developed by OpenAI. It is one of the largest language models to date, with over 175 billion parameters. GPT-3 can generate human-like text based on a prompt, answer questions, write stories, or summarize articles. GPT-3 is trained on a massive dataset of text and uses a transformer architecture, which allows it to handle the large amounts of data required for language generation tasks. ChatGPT is a conversational AI model based on the GPT-3 architecture. It is designed to generate human-like text in response to a user's questions or statements. ChatGPT is trained on conversational data and can understand and respond to natural language inputs in a conversational manner. This makes it useful for applications such as customer service chatbots, virtual assistants, or dialogue systems. Just last week OpenAI has upped their game and released GPT-4 the next version of the model which is multi-modal and can accept image and text as input. Google is also joining this model-as-a-service race by releasing their PaLM (Pathways Language Model) as an API, although staying slightly conservative and not releasing to general public - most probably due to fears of biased results. Meta is also actively researching this space and will soon release their LLaMA (Large Language Model Meta AI) model as API.

What will this Technology change?

As you must have imagined, the implications of this technology are vast. This is the first time an AI technology has come seriously close creating artifacts indistinguishable from those created by highly skilled humans. Generative AI for images and videos will drastically improve the world of art, animation, advertisement, and cinema. We are already seeing aging Hollywood stars like Bruce Willis use this technology (called Deep Fakes) to perform physically challenging scenes. In Sept 2022, generative AI took the art world by storm when an AI-generated painting “Edmond de Belamy” was sold for USD 432,000 in New York City. Online portals like Amazon are getting crowded with eBooks that are written in whole or part by GPT-3 and ChatGPT.

Some non-obvious use cases are in areas like Computer programming. Generative AI tools are actively being used by Software development companies to write fresh code, correct coding errors, and generate comments for legacy code. Machine learning models like OpenAI Codex trained on millions of lines of source code in several languages can help programmers be highly efficient and productive.

Generative AI can aid in drug discovery by helping to speed up the process of identifying new potential drugs. Virtual Screening models can be trained on existing drug and protein data to predict the likelihood of a molecule binding to a specific protein target. This can help to identify potential new drugs more quickly and effectively. De Novo Design models can be used to generate novel molecules with specific properties, such as high potency or specificity towards a particular target. This can be particularly useful in discovering new drugs for diseases where existing treatments are inadequate. Models can also be used to optimize existing drug candidates to improve their efficacy, safety, or other desired properties.

Generative AI can help with synthetic data by creating realistic and diverse datasets that can be used for training and testing machine learning models. This can be done by using algorithms such as Generative Adversarial Networks (GANs) or Variational Autoencoders (VAEs) to generate new data samples that are similar to the original dataset. The resulting synthetic data can be used for a variety of purposes, including increasing the size and diversity of training data, improving the performance of machine learning models, and reducing the cost and time required to collect and label real data. Additionally, synthetic data can also be used to test and validate models in situations where real data is not available or feasible, such as in medical imaging or autonomous driving.

What are some problems and risks?

One major issue that has come up repeatedly is around safety and security risks. These systems can create malicious content, such as fake news or hate speech, leading to safety and security risks for individuals and organizations. These models can perpetuate and amplify existing biases, leading to unfair outcomes and discrimination against certain groups. AI models may generate content that is harmful, but it can be difficult to identify who is responsible for such content.

Moreover, there is always the issue of copyright and ownership of content. Can we use freely available images of paintings from great artists to train models which then generate similar content but no attribution to the source. Like the Shaniwar Wada painting I was trying to sell you in beginning of this article. Many have called these AI models as “stochastic parrots”, as they tend to memorize and recreate patterns from millions of training examples. However, unlike a human artist, the model has no context of the content it is generating, it does not know the significance of the subject but only the geometric shape.

It is important for the AI community to be mindful of these issues and work to address them in a responsible and ethical manner. That is why most advanced technology companies like Persistent have a dedicated Responsible AI program. We need to be conscious of the amount of autonomy we give to AI systems and when to bring a human in the loop to validate the outputs.

Finally, to sum up in the words of the great Marathi author VP Kale

???????? ???? ?????? ?????? ?????. ?? ???????????? ????? ??????? ?????.

(Technology can win over Machines. To win Hearts you need special Magic.)