The term generative AI refers to a category of artificial intelligence that leverages machine-learning algorithms to create new outputs (text, images, videos, audio, and even computer language code). The technology relies on machine-learning methods to train models on large data sets of existing text, audio files, and images to be able to create the new content. Since the foundation for this technology was laid in 2006–12, generative AI has reached a stage at which larger models and data sets are enabling the introduction of many tools and apps. Initially, the most significant opportunities are likely to be for the technology giants, which have access to the funds necessary to bring applications to market. Companies already poised to move forward will have an advantage.
With generative AI, machine-learning-based AI models create new content, including text, images, videos, and/or audio. However, the process also requires human involvement to input prompts (words, numbers, images) in the beginning, evaluate, and edit the content before use.
The foundation for this technology was laid in the first decade of the twenty-first century (Exhibit 1). In 2006, Nvidia released CUDA, Nvidia’s programming language allowing use of graphics processing units (GPUs). Three years later, researchers at Stanford University introduced ImageNet, a collection of labeled images. Building on these innovations, AlexNet initiated the deep-learning revolution in 2012 by creating the best visual classifier at the time.
Since then, the advent of (general adversarial networks (GANs) in 2014 and transformers in 2017 (Google Brain) further advanced image and text generation. These all preceded the largest such generative AI models: OpenAI’s GPT-3 and GPT-3.5 (GPT stands for generative pre-trained transformer), CLIP (contrastive language-image pretraining), and Google’s LaMDA (Language Model for Dialogue Applications)
With a plethora of tools across text-to-text, text-to-image, image-to-image, text-to-speech, audio and video generation, code generation, and synthetic data generation generative AI is set to transform creative industries with automation and democratization of content production while allowing greater quality and a wider variety of content, including increased personalization.
Already in 2022, generative AI made headlines with $1.4 billion in total funding across 78 deals—seven times the amount in 2020 (Exhibit 2), and six start-ups reached unicorn status. Image generation tools (DALL-E, Stable Diffusion, and Midjourney) flooded the market. In November, OpenAI’s NLP-based chatbot, ChatGPT, was released and gained a million users in its first week1. In addition, DALL-E has 1.5 million users, Stable Diffusion has 10 million daily users, and Midjourney has over two million users2. Microsoft’s $10 billion in funding3 of OpenAI put generative AI on the map, with many calling it the beginning of the “summer of AI.”
The state of the art now includes a strong foundation of sophisticated algorithms for natural-language processing (NLP) and machine learning (ML), larger models (such as LaMDA and GPT-3.5), and bigger data sets (for example, GPT-3, OpenAI’s large language model chatbot, trained on 45 terabytes of data4). With these capabilities in place, generative AI is expected to exceed an estimated $100 billion by 20305 as more and more tools and use cases enter the market.
The momentum is likely to continue. Leading venture capitalists express excitement, and we see sweeping usage across tools—for example, 100 million users for ChatGPT in its first two months6 and code-writing tool Github Copilot used by more than 1.2 million software engineers7. Big Tech also is making strides. Microsoft announced a new version of Bing and Edge to be powered by GPT-3.58, and Alphabet’s LaMDA-based rival, named Bard, released in February 20239.
Widespread adoption of an overall set of generative AI tools would require addressing multiple concerns—plausibility of generating deep fakes, limited control and predictability over the output, high setup costs given the computational requirements, copyright ambiguities of generated content, the system’s lack of originality and creativity to come up with content on its own, and risks related to possible misuse of fakes.
Much of the recent attention has been directed to ChatGPT, a text-to-text chatbot capable of engaging in human-like dialogue to respond to customer queries, requests, and comments. It is based on GPT-3 (more recently upgraded to GPT-3.5), a state-of-the-art natural-language processing model. Within one week of the ChatGPT prototype launched, nearly million users began dabbling with it (Exhibit 3) to generate human-like text, growing to 100 million users two months later.
Their experiences point to applications spanning customer service, e-commerce, digital content curation, email marketing, education and training, and personal assistance (Exhibit 4).
While ChatGPT’s meteoric rise has spawned an AI revolution, with more recognition and awareness for generative AI, wide application of the technology will require addressing its limitations. These include information inaccuracy and ambiguity, limited knowledge of recent events, IP issues on underlying training data, concerns around unwanted biases and ethical and plagiarism concerns in education.
We have six core beliefs on the future of generative AI. First, we think the extraordinary momentum of 2022 is but the tip of the iceberg; we anticipate compounding growth in the years ahead. Second, we also expect that Generative AI will challenge some major “myths” about AI and show applications for tasks that not just involve logic, pattern recognition, and decision-making, e.g., producing creative works. A third belief is that monetization of generative AI will rely on sales to business customers; in the B2C space, sellers will need to build demand with freemium business models. Fourth, what will differentiate the winners in this new technology will be their access to funds for capital expenditures. This will fuel an arms race between the’ big tech players.
Fifth, we believe this new technology will be a prime threat to Google Search. Generative AI tools like ChatGPT, DALL-E, and Midjourney are adopting a question-and-answer approach to provide users with a more convenient and context-based way of obtaining search results (Exhibit 5). This could pose a significant threat to Google's search business as Microsoft's Bing-ChatGPT combination is already showing indications of being a serious competitor. Downloads of the Bing app surged following the announcement of ChatGPT integration10, indicating a potential shift in user behavior. Google risks experiencing a Kodak moment if it fails to respond quickly and launch its own suite of generative AI tools.
Finally, generative AI will be a unique opportunity for conventional industries like the industrial sector to leapfrog in AI adoption/ use. They will be faster in adoption this time around compared to other technological advances.
Looking ahead to 2025, analysts expect industrials to lag many sectors (IT and communications, financial services, education, public sector, and consumer goods) on tech adoption.11 This trailing behavior on tech adoption is associated with several structural drivers. These include manufacturing’s widely dispersed footprint (operating in multiple geographies), decentralization (silos even within businesses), poorly integrated IT assets, shortage of tech talent, and poorly developed processes and controls.
Generative AI, however, presents an opportunity for the industrials sector, especially given its multiple potential use cases across functions. Industrials companies can reap benefits in sales and marketing (scaling the e-commerce business, accelerating aftermarket revenue growth, and improving sales conversion), customer service (improving customer response time, enhancing customer experience), manufacturing (improving standard operating procedures, improving quality control, and reducing labor and machine downtime), and supply chain management (improving supplier engagement, increasing yield, and improving quality).
Although generative AI technology is revolutionary and presents many applications for the industrials sector, significant risks are present. To succeed under these conditions, companies and their investors should employ a five-step methodology: (1) prioritize use cases; (2) understand build-versus-buy alternatives; (3) set up organizational capability; (4) adopt and outcome-based approach; and (5) invest in data, analytics, and digital infrastructure (Exhibit 6). Companies that proceed thoughtfully can position themselves to gain a real advantage.
Last year’s surge of interest in generative AI supports the view that this promising technology will transform the way creative work gets done. Generative AI tools have potential for nearly every business function. The technology offers ways to add efficiency, design better products, customize messages and services, and retain customers and employees. Given the potential, our core beliefs about the future of generative AI assume rapid change in the industry: compounding growth, freemium business models for B2C markets, challenges to “myths” about AI, increasing significance of capital expenditures, an existential challenge to Google, and conventional industries capitalizing on the opportunity. Companies that begin now to identify applications worth pursuing and ways to get involved stand to gain an early advantage.
1 Daniel Ruby, “ChatGPT Statistics for 2023: Comprehensive Facts and Data,” Demand Sage, February 8, 2023, https://www.demandsage.com/chatgpt-statistics/.
2 Louis Rosenberg, “Generative AI: The Technology of the Year for 2022,” Big Think, n.d., https://bigthink.com/the-present/generative-ai-technology-of-year-2022/.
3 “Microsoft Confirms Its $10 Billion Investment Into ChatGPT, Changing How Microsoft Competes With Google, Apple And Other Tech Giants,” Forbes, January 27, 2023, https://www.forbes.com/sites/qai/2023/01/27/microsoft-confirms-its-10-billion-investment-into-chatgpt-changing-how-microsoft-competes-with-google-apple-and-other-tech-giants/?sh=787b1b803624.
4 Priya Shree, “The Journey of Open AI GPT Models,” Walmart Global Tech Blog, November 10, 2022, https://medium.com/walmartglobaltech/the-journey-of-open-ai-gpt-models-32d95b7b7fb2.
5 Generative AI Market Size: Global Industry, Share, Analysis, Trends and Forecast 2022–2030, Acumen, December 2022, available at https://www.acumenresearchandconsulting.com/generative-ai-market.
6 Ruby, “ChatGPT Statistics for 2023.”
7 Poulomi Chatterjee, “Generative AI Startups Are the New VC Favourites as Hype Grows,” Analytics India, October 27, 2022, https://analyticsindiamag.com/generative-ai-startups-are-the-new-vc-favourites-as-hype-grows/.
8 James Vincent, “Microsoft Announces New Bing and Edge Browser Powered by Upgraded ChatGPT AI,” The Verge, February 7, 2023, https://www.theverge.com/2023/2/7/23587454/microsoft-bing-edge-chatgpt-ai.
9 George Glover, “Microsoft, Google, and Baidu Are Putting the Final Touches on Their Answers to ChatGPT,” Business Insider India, February 8, 2023, https://www.businessinsider.in/investment/news/heres-how-tech-giants-microsoft-google-and-baidu-are-taking-on-chatgpt-and-what-that-could-mean-for-investors/articleshow/97622643.cms.
10 Sarah Perez, “Bing’s App Sees a 10x Jump in Downloads after Microsoft’s AI News,” TechCrunch, February 8, 2023, https://techcrunch.com/2023/02/08/bings-previously-unpopular-app-sees-a-10x-jump-in-downloads-after-microsofts-ai-news/.