What’s the future of generative AI? An early view in 15 charts

| Article
Three flowers in a vase made out of polygon shapes. Parts of the flowers and vase are distorted and stretched.
Three flowers in a vase made out of polygon shapes. Parts of the flowers and vase are distorted and stretched.

Since the release of ChatGPT in November 2022, it’s been all over the headlines, and businesses are racing to capture its value. Within the technology’s first few months, McKinsey research found that generative AI (gen AI) features stand to add up to $4.4 trillion to the global economy—annually.

The articles and reports we’ve published in this time frame examine questions such as these:

  • What will the technology be good at, and how quickly?
  • What types of jobs will gen AI most affect?
  • Which industries stand to gain the most?
  • What activities will deliver the most value for organizations?
  • How do—and will—workers feel about the technology?
  • What safeguards are needed to ensure responsible use of gen AI?

In this visual Explainer, we’ve compiled all the answers we have so far—in 15 McKinsey charts. We expect this space to evolve rapidly and will continue to roll out our research as that happens. To stay up to date on this topic, register for our email alerts on “artificial intelligence” here.

Gen AI finds its legs

Generative AI has been evolving at a rapid pace.

Image description:

A calendar displays actual AI events along a timeline. The events are as follows:

Nov 30, 2022: OpenAI’s ChatGPT, powered by GPT-3.5 (an improved version of its 2020 GPT-3 release), becomes the first widely used text-generating product, gaining a record 100 million users in 2 months;

Dec 12: Cohere releases the first LLM that supports more than 100 languages, making it available on its enterprise AI platform;

Dec 26: LLMs such as Google’s Med-PaLM are trained for specific use cases and domains, such as clinical knowledge;

Feb 2, 2023: Amazon’s multimodal-CoT model incorporates “chain-of-thought prompting,” in which the model explains its reasoning, and outperforms GPT-3.5 on several benchmarks;

Feb 24: As a smaller model, Meta’s LLaMA is more efficient to use than some other models but continues to perform well on some tasks compared with other models;

Feb 27: Microsoft introduces Kosmos-1, a multimodal LLM that can respond to image and audio prompts in addition to natural language;

Mar 7: Salesforce announces Einstein GPT (leveraging OpenAI’s models), the first generative AI technology for customer relationship management;

Mar 13: OpenAI releases GPT-4, which offers significant improvements in accuracy and hallucinations mitigation, claiming 40% improvement vs GPT-3.5;

Mar 14: Anthropic introduces Claude, an AI assistant trained using a method called “constitutional AI,” which aims to reduce the likelihood of harmful outputs;

Mar 16: Microsoft announces the integration of GPT-4 into its Office 365 suite, enabling broad productivity increases;

Mar 21: Google releases Bard, an AI chatbot based on the LaMDA family of LLMs;

Mar 30: Bloomberg announces an LLM trained on financial data to support natural-language tasks in the financial industry;

Apr 13: Amazon announces Bedrock, the first fully managed service that makes models available via API from multiple providers in addition to Amazon’s own Titan LLMs.

End of image description.

The road to human-level performance just got shorter

Due to generative AI, experts assess that technology could achieve human- level performance in some capabilities sooner than previously thought.

Image description:

A Gantt chart shows estimated ranges for AI technology to achieve human-level performance, by technical capability area, and how those estimates have shifted sooner rather than later when comparing 2017 (pre-generative AI) estimates with 2023 (post-recent generative AI developments) estimates. The latest median estimates show, by technical capability: coordination with multiple agents, around 2035, down from around 2045; creativity, around 2023, down from around 2048; logical reasoning and problem solving, around 2023, down from around 2043; natural-language generation, no change; natural-language understanding, around 2025, down from around 2055; output articulation and understanding, no change; generating novel patterns and categories, around 2020, down from around 2023; sensory perception, no change; social and emotional output, around 2031, down from around 2048; social and emotional reasoning, around 2033, down from around 2050; and social and emotional sensing, around 2030, down from around 2037. Comparisons were made on the business-related tasks required from human workers.

Source: McKinsey Global Institute occupation database; McKinsey analysis.

End of image description.

And automation of knowledge work is now in sight

Advances in technical capabilities could have the most impact on activities performed by educators, professionals, and creatives.

Image description:

A series of bar charts compare the automation potential of generative AI (gen AI) among different occupation categories, as a percentage of total. There are 2 bars per category: 1 before gen AI, and 1 after gen AI. The occupations are ordered by biggest differential, with the biggest change in effect on automation at the top, to the lowest change at the bottom. The largest percentage-point shift was about 40, and the smallest was 4. From top to bottom, the occupations are: educator and workforce training, +39 percentage points; business and legal professionals, +30; STEM professionals, +29; community services, +26; creatives and arts management, +25; office support, +21; managers, +17; health professionals, +14; customer service and sales, +12; property maintenance, +9; health aides, technicians, and wellness, +9; production work, +9; food services, +8; transportation services, +7; mechanical installation and repair, +6; agriculture, +4; and builders, +4. The overall average percentage point increase is about +8.

Exhibit includes data from 47 countries, representing about 80% of employment around the world.

Source: McKinsey Global Institute analysis.

End of image description.

Apps keep proliferating to address specific use cases

There are many applications of generative AI across modalities.

Image description:

A table displays a list of generative AI use cases across modalities. From top to bottom, they are:

Text, content writing, with 2 bullet points: marketing, for creating personalized emails and posts; and talent, for drafting interview questions, job descriptions. Text, chatbots and assistants, with 1 bullet point: customer service, using chatbots to boost conversion on websites. Text, search, with 2 bullet points: making more natural web search; and corporate knowledge, for enhancing internal search tools. Text, analysis and synthesis, with 2 bullet points: sales, for analyzing customer interactions to extract insight; and risk and legal, for summarizing regulatory documents.

Code, code generation, with 1 bullet point: IT, for accelerating application development and quality with automatic code recommendations. Code, application prototype and design, with 1 bullet point: IT, for quickly generating user interface designs. Code, data set generation, with 1 bullet point: generating synthetic data sets to improve AI models’ quality.

Image, stock image generator, with 1 bullet point: marketing and sales, for generating unique media. Image, image editor, with 1 bullet point: marketing and sales, for personalizing content quickly.

Audio, text to voice generation, with 1 bullet point: trainings, for creating educational voiceover. Audio, sound creation, with 1 bullet point: entertainment, for making custom sounds without copyright violations. Audio, audio editing, with 1 bullet point: entertainment, for editing podcast in post without having to rerecord.

3-D or other, 3-D object generation, with 2 bullet points: video games, for writing scenes, characters; and digital representation, for creating interior-design mock-ups and virtual staging for architecture design. 3-D or other, product design and discovery, with 2 bullet points: manufacturing, for optimizing material design; and drug discovery, for accelerating R&D process.

Video, video creation, with 2 bullet points: entertainment, for generating short-form videos for TikTok; and training or learning, for creating video lessons or corporate presentations using AI avatars. Video, video editing, with 3 bullet points: entertainment, for shortening videos for social media; e-commerce, for adding personalization to generic videos; and entertainment, for removing background images and background noise in post. Video, video translation and adjustments, with 3 bullet points: video dubbing, for translating into new languages using AI-generated or original-speaker voices; live translation, for corporate meetings, video conferencing, and voice cloning or replicating actor’s voice or changing for studio effects such as aging. Video, face swaps and adjustments, with 4 bullet points: virtual effects such as enabling rapid high-end aging, de-aging; cosmetic, wig, and prosthetic fixes; lip syncing or “visual” dubbing in postproduction; editing footage to achieve release in multiple ratings or languages; face swapping and deep-fake visual effects; and video conferencing, for real-time gaze corrections.

End of image description.

Some industries will gain more than others

So understanding the use cases that will deliver the most value to your industry is key

Generative AI could deliver significant value when deployed in some use cases across a selection of top industries.

Image description:

A table shows different industries and key generative AI use cases within them. The first example is banking, with an estimated total value per industry of $200 billion to $340 billion, and a value potential increase of 9–15% of operating profits based on average profitability of selected industries in the 2020–22 period. The second example is retail and consumer packaged goods, including auto retail, with an estimated total value per industry of $400 billion–$660 billion, and a value potential increase of 27–44% of operating profits based on average profitability of selected industries in the 2020–22 period. The third example is pharma and medical products, with an estimated total value per industry of $60 billion–$110 billion, and a value potential increase of 15–25% of operating profits based on average profitability of selected industries in the 2020–22 period.

End of image description.

Despite gen AI’s commercial promise, most organizations aren’t using it yet

Commercial leaders are already leveraging generative AI use cases—but most feel the technology is underutilized.

Image description:

A series of 2 pairs of comparative bar graphs show support, among senior executives in significant global B2B and B2C sales and marketing organizations across a wide range of industries and company maturity levels, for machine learning and for generative AI (gen AI). They were asked: “To what extent is your organization using machine learning/generative AI solutions?” And, “How much do you think your organization should be using machine learning/generative AI solutions?” One bar shows that they currently use gen AI, and the other shows where respondents feel gen AI should be used. The data indicate that 65% of respondents currently use machine learning sometimes or rarely, and that most respondents feel that it could be used often or almost always, at about 90%. For gen AI, about 60% of respondents say that it is used rarely or never at their organizations, and 90% of respondents say that it should be used often or almost always.

End of image description.

Marketing and sales leaders are most enthusiastic about three use cases

Commercial leaders are cautiously optimistic about generative AI use cases, anticipating moderate to significant impact.

Image description:

A horizontal bar graph comprising 11 bars illustrates the views of respondents at commercially leading organizations regarding the impact of generative AI (gen AI). Respondents are senior executives in significant global B2B and B2C sales and marketing organizations across a wide range of industries and company maturity levels. They were asked, “Please share your estimated ROI/impact these tools would have if implemented in your organization.” The areas where respondents indicated the impact of gen AI would be significant, by use case, are, from most impacted to least impacted: lead identification, 60%; marketing optimization, 55%; personalized outreach, 53%; dynamic content on websites and marketing materials, 50%; up-/cross-selling recommendations, 50%; success analytics, 45%; marketing analytics, 45%; dynamic customer-journey mapping, 45%; automated marketing workflows, 35%; sales analytics, 30%; and sales coaching, 25%.

End of image description.

Software engineering, the other big value driver for many industries, could get much more efficient

Generative AI can increase developer speed, but less so for complex tasks.

Image description:

A negative bar graph shows the potential percentage reduction in task completion time for development tasks using generative AI (gen AI) compared with task completion without gen AI. From most-impacted to least-impacted they are: code documentation, 45–50% reduction; code generation, 35–45% reduction; code refactoring, 20–30% reduction; and high-complexity tasks, less than 10% reduction.

End of image description.

And gen AI assistance could make for happier developers

Generative AI tools have potential to improve the developer experience.

Image description:

A series of 3 bar graphs are grouped into 2 segmented bars each. Each pair of bars is under a different topic, with data representing developer respondent’s feelings with and without the involvement of generative AI in their work. The metrics are whether respondents “felt happy,” were “Able to focus on satisfying and meaningful work,” and were “in a flow state.” In all cases, the more positive responses were, on average, doubled among those using generative AI.

End of image description.

Momentum among workers for using gen AI tools is building

But organizations still need more gen AI–literate employees

Job postings for fields related to tech trends grew by 400,000 between 2021 and 2022, with generative AI growing the fastest.

Image description:

A series of 15 technology trends and their associated jobs postings comparing 2021 and 2022 in thousands of available positions per annum each on its own graph, and the percentage change between each. Out of 150 million surveyed job postings. Job postings are not directly equivalent to numbers of new or existing jobs. The graphs are ordered in most numerous job numbers in 2022 to least numerous, in 3 rows of 5 graphs. They are: applied AI, 6%; next-generation software development, 29%; cloud and edge computing, 12%; trust architecture and digital identity, 16%; future of mobility, 15%; electrification and renewables, 27%; climate tech beyond electrification and renewables, 8%; advanced connectivity, 7%; immersive reality technologies, 10%; industrializing machine learning, 23%; Web3, 40%; future of bioengineering –19%; future of space technologies, 16%; generative AI, 44%; quantum technologies, 12%.

Source: McKinsey’s proprietary Organizational Data Platform, which draws on licensed, deidentified public professional profile data.

End of image description.

Organizations should proceed with caution

Inaccuracy, cybersecurity, and intellectual property infringement are the most-cited risks of generative AI adoption.

Image description:

A pair of stacked horizontal bar graphs show generative AI related risks that organizations consider relevant and are trying to mitigate. Only respondents whose organizations have adopted Al in at least 1 function were asked. For both risks considered relevant and risks mitigated, n = 913. The first graph shows factors that organizations consider relevant risks as a percentage of respondents, and the second graph shows the same percentage of respondents from organizations that are actively working to mitigate those same risks. The top 5 relevant risks, according to respondents, are inaccuracy, 56%; cybersecurity, 53%; intellectual property infringement, 46%; regulatory compliance, 45%; and explainability, 39%. The top risks being actively mitigated, according to respondents, are cybersecurity, 38%; inaccuracy, 32%; regulatory compliance, 28%; intellectual property infringement, 25%; and explainability, 18%.

Source: McKinsey Global Survey on AI, 1,684 participants at all levels of the organization, April 11–21, 2023.

End of image description.

Gen AI could ultimately boost global GDP

Generative AI could contribute to productivity growth if labor hours can be redeployed effectively.

Image description:

A series of graphs show predicted compound annual growth rates from generative AI by 2040 in developed and emerging economies considering automation. This is based on the assumption that automated work hours are reintegrated in work at today’s productivity level. Two scenarios are shown for early and late adoption of automation, and each bar is broken into the effect of automation with and without generative AI. The addition of generative AI increases CAGR by 0.5 to 0.7 percentage points, on average, for early adopters, and 0.1 to 0.3 percentage points for late adopters. In the overall average for global growth, generative AI adds about 0.6 percentage points by 2040 for early adopters, while late adopters can expect an increase of 0.1 percentage points. Data are based on 47 countries, representing about 80% of world employment.

Source: Conference Board Total Economy database; Oxford Economics; McKinsey Global Institute analysis.

End of image description.

Gen AI represents just a small piece of the value potential from AI

Generative AI could create additional value potential above what could be unlocked by other AI and analytics.

Image description:

A waterfall graph shows the potential additional value that could be added to the global economy by new generative AI uses cases. An initial $11.0 trillion–$17.7 trillion could come from advanced analytics, traditional machine learning, and deep learning. And additional $2.6 trillion–$4.4 trillion of incremental economic impact could be added from new generative AI use cases, resulting in a total use-case-driven potential of $13.6 trillion–$22.1 trillion. If all worker productivity was enabled by AI, including use cases, an additional $6.1 trillion–$7.9 trillion could be added (this includes revenue impacts conservatively translated into productivity impact as difference between total impact and cost-isolated impact, leaving a total AI economic potential of $17.1 trillion to $25.6 trillion.

End of image description.