By: Chitrang Negi, Senior Business Analyst, LTIMindtree

In recent times, AI models such as ChatGPT, DALLE, Midjourney, and StableDiffusion, have captured the attention of the online world. Within five days of its launch, more than a million users had logged into ChatGPT. These models are not only used for processing vast amounts of data and making predictions but majorly for creating content from scratch. Right now, the internet is flooded with instances of them being used for writing essays, and poetry, developing apps, creating visual content, passing entrance exams, and many more tasks. In a short span of time, generative AI has created a profound impact across various fields.

Generative AI can produce new text, audio, video, and images without significant human intervention. It also enables generative ai software development. This cutting-edge AI finds the underlying patterns in the input data to create a new and accurate replica with the attributes of the input data set.

Forrester predicts that by this year, 10% of content generated by Fortune 500 enterprises would be through tools using generative AI algorithms. Gartner predicts that by 2025, 30% of marketing messages from companies would be synthetically generated, and IDC predicts that by 2026, 75% of large enterprises will rely on AI-infused processes.

The main growth drivers for this technology are improvements in neural networks, NLP, availability of high computation power, and access to large data sets from sources such as Common Crawl. A host of start-ups have emerged, and VCs have poured in $1.7 billion, with generative ai software development and AI-enabled drug discovery having received the most funding.

Big players such as Microsoft-backed OpenAI, Google, and Meta are also focused on improving their generative AI algorithms. Recently, OpenAI released the latest version of their large language model, GPT-4. Meta showcased its text-to-video model called Make-A-Video, which generates a 5-second video from text prompts. Another text-to-video, Phenaki, can generate video clips from text prompts, still images, and longer clips. Google is also developing a model called DreamFusion, which generates 3D images based on text prompts.

Potential applications of generative AI

Generative AI has caught the attention of enterprises across various industries that are exploring its potential use cases. From healthcare to entertainment, businesses are considering the benefits that can be derived from leveraging generative AI.

Banking and finance

The application of generative AI is on the rise in the BFSI industry. As per Gartner, companies have begun applying GANs (Generative Adversarial Networks) and NLG (Natural Language Generation) in services pertaining to modeling risk factors, detecting frauds, predicting profitable trades, and creating synthetic data.

Fraud detection: Financial fraud and money laundering detection methods rely on databases of human-engineered rules that match suspicious patterns in financial transactions. Using anomaly detection via GANs, models can be trained on historical financial transaction data and then can be used on new financial transactions to predict if they are fraudulent or not.

Trend evaluation: AI/ML technologies help make predictions by detecting patterns. Generative AI allows to conduct deeper analyses beyond conventional calculations.

Synthetic data: </strong Generative AI helps produce synthetic customer data for modelling, analytics and other business activities, thereby protecting privacy of the customer, as the synthetic data cannot be traced back to them.

Healthcare and life sciences

In healthcare, Generative AI can aid in simple use case such as summarizing medical information to complex one such as drug discovery. It can also help to improve patient care by identifying potential tumor and cancer developments.

Drug Discovery: Pharmaceutical companies have begun using technology in drug discovery. Generative AI algorithms can help speed up development by identifying potential candidates and testing their effectiveness. Gartner predicts that generative AI will be used in 50% of drug development activities by 2025

Improved Patient Care: GANs can provide numerous viewpoints of an X-ray picture to show potential tumor development outcomes. It can also identify cancerous developments by comparing images of healthy organs to damaged ones.

Summarizing Information: It can be used to summarize medical conversations from audio recorded during patient visits. This helps save doctor’s time.

Media and entertainment

Gartner predicts by 2030, a blockbuster film would be released with 90% of it being generated by AI. Some other major applications are below.

Movie restoration: Generative AI can upscale the quality of vintage movies and classic cartoons to 4k and beyond, create 60 frames per second rather than the standard 23, reduce noise, and convert black-and-white to color.

Generation of animated models: Along with film, the video game business relies on moving pictures, and generative AI can accelerate the creation and development process. Using generative AI to build 3D models for computer games helps reduces the software developers’ efforts and significantly decreases development time. Video game developers have the option of creating the models entirely from scratch or previously processed 2D photographs.

Audio synthesis: Generative AI can be used for synthesizing audio and improving sound quality. AI-enabled audio enhancers can be used in cinematography and video gaming to create ambient noises, voiceovers, and other audio effects. Generative AI makes it possible to create voices that resemble humans. Microsoft has a text-to-speech AI model, VALL-E, that can simulate a person’s voice with just a 3-second recording. DeepComposer is another AI that can turn a short melody into a complete song.

Manufacturing

Generative AI is impacting the manufacturing industry by optimizing the design process.

Generative design: It is possible to imitate an engineer’s design process using generative AI. Designers or engineers can input design criteria into algorithms, which subsequently create all possible results. Producers can quickly generate hundreds of design options for a single product using this technique.

Technology

Generative AI assists users while writing code and can help advance search engines.

Software development: Generative AI can support programmers while writing code, suggesting snippets of software that might help fill a gap. Generative AI software development would highly improve a developer’s productivity and make them more efficient. GitHub Copilot, is an example of an AI-based pair programmer.

Search engine services: With Generative AI, search engines would not be limited to fetching links based on users prompts, but they can generate answer, and also produce images and videos based on text prompts. Microsoft has released the new Bing, which gives direct answers to the user’s queries.

Challenges to address

Generative AI models such as ChatGPT also have some limitations. They predict the next word based on the dataset they have been trained on. This is usually text from the internet, and as a result, it can also generate misleading information. They can be used to spread misinformation and spam and generate toxic content.

Requirement of a massive amount of unbiased training data –  Generative AI algorithms require a massive amount of training data to train the models. They rely on the internet for freely available data, but since not all data on the internet is vetted, it can produce biased and fraudulent information. Also, there are concerns regarding data privacy and copyright infringement.

Unexpected outcomes – It is difficult to manage the behavior of some generative AI models. They sometimes fabricate responses when they do not have sufficient information. Hence it is necessary to verify the information they provide.

Deep fakes – Some of the generative AI algorithms are good enough to produce content that is difficult to tell apart from the original one. This can be misused to commit fraud under someone else’s name, like withdrawing cash from a bank. These outcomes might be employed for retaliation, blackmail, coercion, or extortion.

Massive energy usage and high compute costs – In order to train and run these models, massive computation power is required. This leads to higher energy consumption and increased costs. Sam Altman, CEO of OpenAI, described the costs of keeping ChatGPT free to be “eye-watering”.

In order to minimize the false information, some companies are trying Reinforcement Learning from Human Feedback (RLHF) to make the output more truthful. Other limitations could be addressed by having high-quality and unbiased training data, and checks should be in place to detect and remove any misleading information. Guidelines and regulations should be developed to prevent misuse, and energy-efficient algorithms need to be developed.

Preparing for generative AI

Generative AI is viewed as a future differentiator and a way to obtain a competitive advantage as AI becomes a ubiquitous technology for businesses and consumers worldwide. Technology service providers should evaluate the current state of the research and create a proof of concept to assess the potential of generative AI for various industry use cases.

Organizations can collaborate with their customers to identify projects, rank them according to their importance and complexity, and build a minimum viable project strategy. They must create a comprehensive plan to capitalize on the potential benefits of this technology. Moreover, it is necessary to establish an architecture and team structure. Businesses should involve their executives in their AI initiatives, provide them with the required training, and ensure that their AI activities align with their strategic goals.

Conclusion

The mediums to express creativity through data are constantly evolving and expanding. With challenges and technology abuse gaining ground, there is a dire need to take on the responsibility to embrace this revolution and, at the same time, ensure that the generative technology that drives innovation is grounded in ethical practices. While the technology holds a lot of promise, genuine concerns about the societal impact, including the generation of deep fakes, have led to multiple streams of innovation around digital forensics to mitigate related risks. Hence, AI budgets are only increasing with time and will witness a substantial increase year-over-year within the next decade.

A McKinsey survey found that the number of organizations with AI budgets of more than 5% of their digital budgets has increased from 40% in 2018 to 52% in 2022. Some survey respondents also attributed at least 5% of their organization’s EBIT (earnings before interests and taxes) to AI. As enterprise adoption of AI continues to progress, companies will move towards more dynamic AI models. Such models that can help them create content can be a big differentiator. Due to competition, other companies may be forced to use generative AI to stay relevant.