#时事热点头条说#
Source|McKinsey&Company
Compilation|Zhicheng Enterprise Research Institute Cui Shuai
Grasping the future requires understanding the breakthroughs that drove the rise of generative AI, which have been brewing for decades. ChatGPT, GitHub Copilot, Stable Diffusion, and other generative AI tools that have attracted public attention are the result of significant investments in recent years that have advanced machine learning and deep learning while embedding into many of the AI application products and services we use every day.
Back in 2016, AlphaGo, an AI-based program developed by DeepMind, beat a world Go champion, a milestone event worth celebrating, but quickly disappeared from the public eye.
Unlike AlphaGo, generative AI represented by ChatGPT captures the imagination of the world, largely due to its widespread utility — almost anyone can use them to communicate and create — and its supernatural ability to converse with users. The latest generative AI applications can perform a range of routine tasks, such as reorganization and classification of data. Their ability to write, compose, and create digital art has long dominated headlines, convincing consumers to try it on their own, so much so that a wider range of stakeholders are grappling with the impact of generative AI on business and society.
In McKinsey's report, generative AI is defined as an application built using a base model. These models contain extensive artificial neural networks inspired by the billions of neurons connected in the human brain. The base model is part of so-called deep learning, and unlike the deep learning model of the past, the generative AI base model can handle a large number of different unstructured data sets and perform multiple tasks. For example, new base models can process images, video, audio, and computer code, and artificial intelligence trained on these models can perform functions such as classification, editing, summarizing, answering questions, and drafting new content.
Continuous innovation also brings new challenges. For example, the computing power required to train generative AI with hundreds of billions of parameters can become a bottleneck to development. In addition, there is an important initiative – spearheaded by the open source community and spilling over to leaders in generative AI companies – to make AI more responsible, which could increase its cost.
Investment in generative AI, while a small fraction of total AI investment, is significant and growing rapidly – reaching $12 billion in the first five months of 2023 alone. From 2017 to 2022, venture capital and other private outside investments in generative AI will grow at an average compound growth rate of 74% per year. Over the same period, overall investment in AI grew by 29% per year.
The logic of this article:
01 Economic benefits of generative AI
02 Impact of AI on work activities, economic growth and productivity
03 Business and social considerations
04 Conclusion
Friends who want to get the full report, follow the official account and reply to "generative AI" in the account to get it.
(The text is 7800 words, and the reading time is about 13-15 minutes)
01
The economics of generative AI
The first is from a use case perspective, which McKinsey defines as applying generative AI to a specific business challenge and producing one or more measurable outcomes.
For example, one use case in marketing is applying generative AI to generate creative content such as personalized emails, and measurable results could include reduced costs for generating such content and increased revenue from massively improving high-quality content. McKinsey identified 63 generative AI use cases covering 16 business functions that provide a total economic value of $2.6 trillion to $4.4 trillion per year when applied across industries.
From this perspective, economic value could unleash between $11.0 trillion and $17.7 trillion, up from $9.5 trillion to $15.4 trillion in 2017.
From another perspective, generative AI has the potential impact on the work activities required for about 850 occupations. McKinsey models scenarios to assess when generative AI will be able to complete more than 2,100 "detailed work activities" that make up occupations in the global economy, such as "communicating with others about operational plans or activities."
Excluding the overlapping cost reduction from both perspectives, the total economic benefits of generative AI will reach $6.1 trillion to $7.9 trillion per year.
While generative AI is an exciting and rapidly evolving technology, it represents only a small fraction of the overall potential value of AI. Other types of AI are also reaching their full value, such as traditional advanced analytics and machine learning algorithms that are highly effective at performing numerical and optimization tasks such as predictive modeling, and they continue to find new applications across a wide range of industries. But there's no denying that as generative AI continues to evolve and mature, it has the potential to open up entirely new frontiers in creativity and innovation to expand what AI as a whole can achieve.
(1) Potential value by function
While generative AI is likely to have an impact on most business functions, some stand out when measured by the impact of technology as a share of cost. McKinsey's analysis of 16 business functions identified only four functions, customer operations, marketing and sales, software engineering, and R&D, accounting for about 75% of the total annual value of generative AI use cases.
Notably, the potential value of generative AI is much lower when used in several functions where AI use cases stand out, including manufacturing and supply chain functions.
In addition to this, generative AI can drive value across the organization by revolutionizing internal knowledge management systems. For example, generative AI's impressive natural language processing capabilities can help employees quickly make more informed decisions and develop effective strategies by retrieving stored internal knowledge, which can enable teams to quickly access relevant information.
In 2012, the McKinsey Global Institute (MGI) predicted that knowledge workers spend about one-fifth of their time searching for and gathering information, and the benefits would be huge if generative AI could take on these tasks and improve worker efficiency and effectiveness.
(2) The value potential of the form
Technology has revolutionized the way businesses do business, and text-based AI is at the forefront of this change. In fact, text-based data is rich, accessible, and easy to process and analyze by large language models (LLMs), attracting strong attention from the initial stages of generative AI development. The current investment landscape for generative AI is also focused on text-based applications such as chatbots, virtual assistants, and language translators. However, McKinsey estimates that nearly one-fifth of the value that generative AI can unlock will take advantage of text-to-text multimodal capabilities.
While most of the initial impetus for generative AI was text-based use cases, recent advances in generative AI have also led to breakthroughs in image generation, OpenAI's DALL· E and Stable Diffusion have amply illustrated this and have also made great strides in audio (both voice and music) and video. These features have obvious applications in marketing to generate advertising materials and other marketing content, and these technologies are already being applied in the media industry, including game design. In fact, some of these examples challenge existing business models in terms of talent, profitability and intellectual property.
The multimodal capabilities of generative AI can also be effectively used in R&D. Generative AI systems can create first drafts of circuit designs, architectural drawings, structural engineering designs, and thermal designs based on prompts that describe product requirements. Achieving this requires training the underlying model in these domains. Once trained, such a base model can increase productivity on a similar order of magnitude to software development.
(3) Industry value potential
Of the 63 use cases analyzed by McKinsey, generative AI has the potential to create $2.6 trillion to $4.4 trillion in value across industries. The exact impact will depend on various factors such as the combination and importance of different functions, as well as the size of the industry's revenue.
For example, it is estimated that generative AI could contribute about $310 billion in added value to the retail industry, including car dealerships, by improving the performance of functions such as marketing and customer interaction. In contrast, much of the potential value of high-tech comes from the ability of generative AI to improve the speed and efficiency of software development.
In banking, generative AI has the potential to improve the efficiency of AI by taking on lower-value tasks in risk management, such as required reporting, monitoring regulatory dynamics, and collecting data. In the life sciences industry, generative AI is expected to make a significant contribution to drug discovery and development.
02
Artificial intelligence for work activities,
Impact of economic growth and productivity
The McKinsey Global Institute began analyzing the impact of automation in 2017 and modeling the scenarios it adopts. At that time, we estimated that half of workers' time was spent on activities that had the potential to be automated, which we call the potential for technological automation. We also simulate a range of potential scenarios to predict how quickly these technologies may be adopted and affect job activity in the global economy.
Large-scale technology adoption won't happen overnight. The potential technical capabilities in the lab do not necessarily mean that they can be immediately integrated into solutions that automate specific work activities, and developing such solutions takes time. Even if such a solution is developed, it may not be economically viable if its cost exceeds the cost of labor. Moreover, even if there are economic incentives for deployment, it will take time to roll it out in the global economy. Thus, by adopting scenarios, combining these factors with the potential for technological automation provides the speed and scale of worker activity over time.
The large-scale shift in work activities and occupational mix is not without precedent. Compare the work of farmers today to a few years ago, where many farmers now access market information on their phones to determine when and where to sell their crops, or download sophisticated weather model models. From a broader perspective, China's share of all labor force employed in agriculture fell from 82 percent in 1962 to 13 percent in 2013. The labor market is also dynamic: in the United States, millions of people leave their jobs every month. However, this does not reduce the challenges faced by individual workers, whose lives are upended by these changes, nor does it reduce the response to organizational or social challenges to ensure that workers have the skills to do the jobs that are needed and that their income is sufficient to improve their standard of living.
Moreover, from a macroeconomic point of view, demographics necessitate such a shift in activity. As global labour force growth slows, an economic growth gap has emerged, and in some major countries, the labor force has shrunk due to an aging population, and labor productivity must accelerate to achieve economic growth and prosperity.
New capabilities of generative AI, combined with previous technologies and integrated into business operations around the world, can accelerate the potential of technology automation of individual activities and adopt technologies that empower the workforce. They may also have an impact on knowledge workers, whose activities will change until the future as a result of these technologies.
(1) Accelerate the use of technological potential to transform knowledge-based work
Based on the development of generative artificial intelligence, it is now expected that the technical performance will match the median human performance and reach the top quartile of human performance earlier than previously estimated across a wide range of capabilities. For example, MGI previously identified 2027 as the earliest possible year for the median technology of human natural language understanding, but in this new analysis, the corresponding point is 2023.
As a result of generative AI's reassessment of technical capabilities, the percentage of total time that can theoretically be automated by integrating existing technologies has increased from about 50% to 60-70%. Due to the acceleration of generative artificial intelligence natural language capabilities, the technological potential curve is quite steep.
Interestingly, the time frame between early and late scenarios has been compressed compared to the expert assessment in 2017, reflecting greater confidence that higher levels of technical capabilities will arrive at a given period.
(2) Acceleration of automation processes
McKinsey's analysis of adoption scenarios takes into account the time required to integrate technological capabilities into solutions capable of automating individual work activities, the cost of these technologies compared to the human costs of different occupations and countries around the world, and the time it takes for technology to diffuse throughout the economy.
As the potential for technology automation enabled by generative AI accelerates, so does the adoption of automation. Given that the speed at which solutions are developed and adopted will vary depending on decisions based on factors such as investment, deployment, and regulation, these scenarios encompass a wide range of outcomes. But they show how much variation can occur in the activities that workers engage in on a daily basis.
Teachers, for example, whose specific work activities include preparing for exams and assessing students' work. With the enhanced natural language capabilities of generative AI, more of these activities can be done by machines, perhaps initially by teachers creating edited first drafts, but ultimately requiring less human editing. This frees up time for these teachers to spend more time on other tasks, such as directing class discussions or tutoring students who need extra help.
Previous modeled adoption scenarios suggest that 50% of the time spent on work activities in 2016 will be automated sometime between 2035 and 2070, with around 2053 in between. Today, given the development of generative AI, McKinsey's updated adoption scenario simulates that 50% of the time spent on work activities in 2023 will be automated between 2030 and 2060, with a midpoint of 2045 – an acceleration of about a decade compared to previous estimates.
In developed countries, automation technology may also be adopted more quickly, where wages are higher, so the economic viability to adopt automation appears earlier. Even if the potential of technological automation for a particular work activity is high, the cost of doing so must be compared with the cost of human wages. In countries with lower wage rates, such as China, India and Mexico, automation adoption has been slower than in countries with higher wages.
(3) The potential impact of generative AI on knowledge work
Previous generations of automation were particularly effective at collecting and processing data-related tasks. The natural language capabilities of generative AI increase the automation potential of these types of activities to some extent. But it has a much smaller impact on physical labor activity, which is not surprising since it is fundamentally designed to perform cognitive tasks.
As a result, generative AI is likely to have the greatest impact on knowledge work, especially activities involving decision-making and collaboration, which previously had the lowest potential for automation.
Generative AI's ability to understand and use natural language in a variety of activities and tasks goes a long way towards explaining why the potential for automation has risen so dramatically. In economic activity, about 40% of jobs require at least a moderate level of human understanding of natural language.
As a result, many work activities involving communication, supervision, documentation, and interaction with the average person have the potential to be automated through generative AI, accelerating job transformation in occupations such as education and technology, where previously expected automation potential to emerge in the future.
Labor economists often point out that the deployment of automation tends to have the greatest impact on workers with the lowest skill levels (measured by education), known as skills bias. Generative AI, however, has the opposite trait – it is likely to have the greatest incremental impact by replacing the activities of some of the more educated workers.
Previous generations of automation have typically had the greatest impact on occupations with declining wages in the income distribution. Job automation becomes more difficult for lower-paying occupations because the potential benefits of automation are competing with lower-cost manpower. In addition, some tasks performed in low-wage occupations are technically difficult to automate, for example, manipulating fabrics or picking delicate fruits. Previous models suggest that job automation is likely to have the greatest medium-term impact on low- and middle-income populations. However, the impact of generative AI is likely to change the jobs of high-wage knowledge workers to the greatest extent as their activities advance in the potential of technological automation that were previously considered relatively unaffected.
(4) Generative AI can drive higher productivity growth
Global economic growth from 2012 to 2022 was slower than in the previous 20 years, and while the pandemic was an important factor, long-term structural challenges – including declining birth rates and an aging population – remained obstacles to growth.
Declining employment is one of those obstacles. The CAGR of the global workforce declined from 2.5% in 1972-1982 to 0.8% in 2012-2022, mainly due to ageing. In many large countries, where the size of the labor force is already declining, productivity measures output relative to inputs, or the value of goods and services produced, divided by the amount of labor, capital, and other resources needed to produce those goods and services, and is the main engine of economic growth in the 30 years from 1992 to 2022. Since then, however, productivity growth has slowed along with slowing job growth, leaving economists and policymakers baffled.
The deployment of generative artificial intelligence and other technologies can help accelerate productivity gains, partially compensate for employment problems, and boost overall economic growth. According to our estimates, automating individual work activities enabled by these technologies could provide the global economy with a productivity boost of 0.2% to 3.3% per year from 2023 to 2040, depending on the rate of automation adoption – generative AI contributes 0.1 to 0.6 percentage points to this growth – but only if individuals affected by technology move to other work activities that at least match the level of productivity in 2022. In some cases, workers will continue to work in the same occupation, but their mix of activities will change, and in some countries, workers will need to switch occupations.
The power of generative AI greatly expands the scope of work activities, accelerates the pace of automated deployment, and expands the types of workers it affects. Like other technologies, AI's ability to take on routine tasks and work can increase human productivity, which has been growing at a below-average rate for nearly 20 years. It could also offset the effects of ageing, which is beginning to erode labor force growth in many of the world's major economies. But to realize these benefits, large numbers of workers will need to fundamentally change their jobs, either in existing occupations or in new ones. They also need support as they transition to new activities.
03
Business and social considerations
History shows that new technologies have the potential to reshape society. AI has already changed the way we live and work, for example, it can help our phones understand what we say, or draft emails. However, most of the time, AI remains behind the scenes, optimizing business processes or making recommendations for the next product to buy. The rapid development of generative AI is likely to significantly enhance the overall impact of AI, generating trillions of dollars of added value each year and changing the nature of work.
But the technology could also present significant new challenges. Given the speed at which AI can be generated, stakeholders must act quickly to prepare for opportunities and risks. Risks have surfaced, including concerns about what generative AI systems produce: Will they infringe intellectual property rights because of "plagiarism" of the training data used to create the base model? Is the answer given by the LL.M. when questioned accurate and explainable? Is AI-generated content fair, or biased in ways that users don't want, such as generating content that reflects harmful stereotypes?
There are also challenges on the economic front: the scale and scope of the workforce transition is considerable, and about a quarter to one-third of work activity in the intermediate adoption scenario could change over the next decade. The task before us is to manage both the potential positive and negative impacts of the technology. While balancing our enthusiasm for the potential benefits of the technology with the new challenges it could bring, we need to address some key questions.
(1) Companies and business leaders
How can companies act quickly to capture potential value while managing the risks posed by generative AI?
How will generative and other artificial intelligence change the careers and skill sets required by company employees in the coming years? How will companies implement these shifts in hiring programs, retraining programs, and other HR areas?
Do companies have a responsibility to play a role in ensuring that the technology is not deployed in "negative use cases" that could harm society?
How can companies transparently share their experiences with scaling the use of generative AI within and across industries, as well as with government and society?
(2) Policymakers
What will the future of work look like on an economic level in terms of careers and skills? What does this mean for workforce planning?
How can you support employees whose activities have changed over time? What retraining programs can be implemented? What incentives are needed to support private sector investment in human capital? Are there programs that earn while learning, such as apprenticeships, that allow people to continue to support themselves and their families while being retrained?
What can policymakers do to prevent generative AI from being used in ways that harm society or vulnerable groups?
Can new policies be developed and existing ones modified to ensure human-centered AI development and deployment, including human oversight and differing perspectives and societal values?
(3) Individuals as workers, consumers and citizens
How worried should individuals be about the emergence of generative AI? While companies can assess how this technology will impact their bottom line, where can citizens get accurate, unbiased information about how it will impact their lives and livelihoods?
How can individuals as workers and consumers balance the convenience of AI and its impact on the workplace?
Do citizens have a say in decisions that make decisions to deploy and integrate generative AI into the fabric of their lives?
04
epilogue
Technological innovation can inspire the same level of awe and attention. Both reactions could be amplified when this innovation seems to take shape overnight and spread widely. The arrival of generative AI in the fall of 2022 is the latest example of this phenomenon due to its unexpectedly rapid adoption and the subsequent rush of companies and consumers to deploy, integrate, and use it.
All of us are at the beginning of our journey to understand the power, scope, and capabilities of this technology. If the past eight months are any guide, the next few years will take us on a roller coaster, characterized by fast-paced innovation and technological breakthroughs that force us to recalibrate our understanding of the impact of AI on our work and lives. It is important to properly understand this phenomenon and predict its impact. Given the speed at which generative AI has been deployed so far, the need to accelerate digital transformation and retrain the workforce is enormous.
At a time when the global economy is considering the enormous costs of adapting to and mitigating climate change, these tools have the potential to create significant value for the global economy. At the same time, they are likely to be more unstable than previous generations of AI. They are able to possess the most human abilities, language, which is an essential requirement for most work activities related to professional knowledge and knowledge, and a skill that can be used to hurt feelings, create misunderstandings, cover up the truth, incite violence and even war.
We want to better understand the ability of generative AI to add value to company operations, drive economic growth and prosperity, and its potential to dramatically change the way we work and our social goals. Companies, policymakers, consumers, and citizens can work together to ensure that generative AI delivers on its promise to create significant value while limiting its potential to disrupt lives and livelihoods.
Cover image source: Pixabay
—END—