Generative AI: what it is and what are the business applications of systems like ChatGPT

It's not exactly originality in the "human" sense. But Generative Artificial Intelligence, or Generative AI, a technology that has come to the fore with OpenAI's ChatGPT software, can improve the performance of various business activities, such as the production of standard texts, images, and software code, making them faster and, to what extent, creative thanks to the mix of large amounts of sources and data used.

Generative Artificial Intelligence systems fall into the broad category of General Artificial Intelligence (AGI) and machine learning or Machine Learning (ML). They have the potential to change the way we approach content creation for applications such as design, entertainment, eCommerce, marketing, scientific research, and HR. With opportunities and risks to be carefully evaluated.

“It's clear that generative AI tools like ChatGPT and DALL-E (art output specific tool also developed by OpenAI) could change the way a variety of jobs are done, say McKinsey experts. The full extent of this impact is still unknown, as are the risks, but there are some questions we can already answer."

What does Generative AI mean

As defined by McKinsey, Generative AI describes algorithms that can be used to create new content, including audio, code, images, text, simulations, and video.

Generative AI software starts from requests or descriptions (prompts) formulated in natural language by the user (human or software) and consequently generates texts from texts (Text-to-Text), images from texts (Text-to-Image), or even images from images (Image-to-Image).

Generative AI outputs are combinations of the data used to train the algorithms. Since the amount of data used to train the software is massive – the GPT-3 system on which ChatGPT was built was trained on 45 terabytes of text data – the results can seem “creative”. 

In reality, what they generate is a compilation and resumption of a mix of sources, but given the huge amount of data processed, the output can be new. After all, even reworking can be considered a form of creativity.

There is, clearly, the risk of incorrect or even inappropriate output and intellectual property infringement. But, if the user request is pertinent and human surveillance continues, AI Generative products can be satisfactory. They are also able to improve thanks to user feedback.

Deep Learning: Transformers and GAN Neural Networks

The ChatGPT technology could fall within the scope of GANs or Neural Networks Generative Adversarial Networks (GANs). The question is debated, because according to some experts, ChatGPT is a Transformer (GPT is the acronym of Generative Pretrained Transformer) and not a GAN.

The Transformer is a Deep Learning model used in the field of Natural Language Processing (NLP). GANs, on the other hand, are a type of artificial intelligence algorithm that uses two competing neural networks to generate images, sound, text, and other types of data.

The first network, called the "generator", tries to create fake images or data that look real; the second, called the "discriminator", tries to identify whether the images or data are real or fake. 

The two models compete against each other: the generator tries to generate increasingly realistic data while the discriminator tries to better and better identify whether the data is real or false. 

Over time the generator gets better and better at generating realistic data that fools the discriminator, while the discriminator gets better and better at identifying false data.

The goal of a GAN is to optimize deep learning and avoid superficial generalization errors due to the scarcity of data.

Generative AI: how it improves company performance

For businesses, the opportunity of Generative AI lies in the ability of these Artificial Intelligence tools to produce a wide variety of believable text and images in seconds.

IT and software organizations can use these systems to generate code instantly. Organizations that need short marketing texts or technical manuals also benefit. Generative AI also supports product design, design, and photography. It is currently most effective in producing standard content (such as emails, CVs, or manuals).

The advantages

• Process optimization: Generative AI can be used to optimize business processes, such as production planning or distribution planning.
• Creating New Products: Generative AI can be used to generate new products or to create new product designs.
• Improving Customer Experience: Generative AI can be used to generate personalized content for customers, such as product recommendations or automated responses to customer messages.
• Data analytics: Generative AI can be used to analyze large amounts of data and generate insights that can help businesses make informed decisions.
• Cost reduction: Generative AI can help companies reduce costs by automating some manual processes.

The development of a generative artificial intelligence model is resource-intensive: you need a lot of data and a lot of capital. For now, companies with great availability on both fronts are doing it. Companies that want to use Generative AI can either use the technology as it is, off-the-shelf or train it by inserting their own data and models from which the software learns.

Generative AI: applications and Opportunities

Design

Generative AI offers design firms a faster, more efficient way to create and edit designs. Generative algorithms can be trained on a large set of reference data, such as images of existing products, which are analyzed to then generate new designs and models that meet the established criteria or to modify and customize existing designs, creating new variants and options. Applications range from Fashion design to automotive design to building design and other architectural works.

In the specific field of Product Design, Generative AI is used to generate new ideas and to customize products according to customer preferences.

Other opportunities open up in product optimization. For example, generative algorithms can be used to analyze product performance data and generate design changes to improve them.

eCommerce and Marketing

In the retail sector, generative artificial intelligence is used for the personalization of products and contents: emails or product recommendations, promotional content (advertisements and posts), website design, and mobile applications. Generative AI can also generate descriptive text for each product in a massive list of items for sale on an e-commerce site.

Changing the visual characteristics of the products or their description on video is another field of application. It goes beyond the 360° video of a product: generative AI can do automatic rendering with a large variability of parameters (angle, size, colors, modifications, settings).

"It is evident how artificial intelligence and machine learning technologies can be very interesting for marketing strategies - wrote Enzo Mazza, CEO of FIMI (Federation of the Italian Music Industry) - providing, for example, essential information, recommendations, and strategies for communication and promotion to artists and record companies. In the music sector, communication is particularly integrated today and Big Data and Social Media are strategic elements ”, continued Mazza, but obviously the application of Generative AI in Marketing goes beyond the music industry.

Scientific research

Generative AI can be used in many areas of scientific research to generate new ideas, test hypotheses and accelerate discovery, and also for scientific writing, as Microsoft intends to do, which uses ChatGPT thanks to the close collaboration with OpenAI, in which it has invested 1 billion dollars (but could soon invest another 10 billion).

Application fields include Bioinformatics for the identification of new proteins and discovery of potential drugs through the generation of protein models and multiple simulation scenarios; Astronomy for generating images of galaxies and the simulated universe to better understand the evolution of the universe; Physics for the generation of artificial materials and the discovery of new materials through simulations based on AI.

With Generative AI it is also possible to conduct simulations in the medical field in support of 3D technologies to pre-visualize prostheses and molecular organisms.

Entertainment industry

The use of Text-to-Image technology is already being used for creating visual content for movies, games, and other marketing and multimedia tools. The June 2022 cover of Cosmopolitan, for the first time in the history of newspapers, was made entirely with DALL-E 2 artificial intelligence. The project was born from a collaboration between the editors of Cosmopolitan, the OpenAI specialists, and the digital artist Karen X. Cheng, who found the perfect image by typing as a prompt: “A young woman's hand with nail polish holding a Cosmopolitan cocktail”; “Closeup of a woman dressed as fashionably as Wes Anderson would”; “A woman who wears an earring that is a portal to another universe”.

The same experiment had been conducted a week earlier by The Economist for its cover. DALL-E 2 also allows Image-to-Image generation: starting from existing images to improve their quality or imagine previously non-existent outlines and contexts.

The HR

Reverse, an international headhunting and human resources company, has launched a series of experiments to apply the potential of ChatGPT to the personnel search sector.

These are above all writing to support recruiters, such as summarizing CVs in a less schematic form, helping in writing job advertisements, pre-setting positive or negative feedback emails for candidates already interviewed, writing suggestions to attract passive candidates, and finally getting help to deepen and better understand the technical aspects of the roles sought.

“It is a tool that can certainly support the world of recruitment on various fronts – says Daniele Bacchi, CEO and Co-Founder of Reverse -. It allows us to speak to the machine no longer by clicking on the screen but through our language. This is the second major revolution in the sector after LinkedIn; one of the aspects to manage is certainly data protection”.

Generative AI: How to avoid harmful outcomes

Organizations that rely on generative AI models must consider that there are reputational and legal risks associated with the unintentional publication of biased, offensive, copyrighted, or privacy-protected content. From this point of view, the – opposite – experience of two online newspapers is emblematic: Cnet has started using ChatGPT to write entire articles, running into serious problems of information correctness, including plagiarism, while Buzzfeed is experimenting with ChatGPT to generate quizzes and other very schematic and limited content, and always under the guidance of journalists, with much more positive results.

Risks can therefore be mitigated and managed, as McKinsey explains. First, it is critical to carefully select the initial data used to train the AI ​​models to avoid including toxic or biased content. Next, rather than employing an off-the-shelf generative AI model, organizations can use smaller, more specialized models or even customize a general model using their own data to meet their needs and minimize risk.

Equip yourself with a real-life supervisor

A human supervisor is also advisable, a human being who checks the output of a generative AI model before proceeding with publication or use. Finally, avoid using generative AI models for critical decisions.

“It cannot be stressed enough that this is a new field – writes McKinsey -. The landscape of risks and opportunities is likely to change rapidly in the coming weeks, months, and years. New use cases are tested monthly and new models are likely to be developed. As generative AI becomes increasingly and seamlessly embedded in business, society, and our personal lives, we can also expect a dedicated regulatory framework to take shape.”

It is therefore right to experiment and create value with Generative AI, but with initially defined and supervised projects, while continuing to monitor the results and technological and regulatory evolutions.