101 real-world gen AI use cases from the world's leading organizations在过去一年半的时间里，生成式人工智能（AI）在企业领域的应用迅速发展。Google Cloud的Next活动中展示了超过300家组织如何利用AI推动企业转型。这些企业已经从简单的问答助手，发展到能够进行预测和采取行动的AI代理，进一步扩展其业务功能和提升效率。
具体来说，AI代理在以下几个关键领域表现出显著的效益：首先是客户服务，AI能够帮助企业更好地理解和满足客户需求；其次是员工赋能，通过自动化日常任务和优化工作流程，AI提升了工作效率；在创意构思和生产领域，AI助力企业快速生成创新的解决方案；数据分析方面，AI通过高效处理和解析大数据，支持决策制定；在编码创建方面，AI简化了开发流程，提高了代码质量；最后在网络安全领域，AI加强了数据保护和风险管理。
这些应用不仅提高了生产力和操作效率，还极大地改善了客户体验和企业的创新能力。AI的多模态能力，即在文本、语音、视频等多种通信模式中的应用，使其能够更全面地满足不同行业的需求。通过这些先进的技术，企业正在开创一个智能、高效和互联的新时代。
我们一起来看看，是否有参考？
Since generative AI first captured the world’s attention a year and a half ago, there’s been a vigorous discussion about what, exactly, the new technology is best used for. While we all enjoyed those early funny chats and witty limericks, we’ve quickly discovered that many of the biggest AI opportunities are clearly in the enterprise.
Our customers and partners at Google Cloud have found real potential for creating new processes, efficiencies, and innovations with generative AI. For proof, look no further than the 300-plus organizations who are featured at this week’s Next event in Las Vegas.
In a matter of months, organizations like these have gone from AI helping answer questions, to AI making predictions, to generative AI agents. What makes AI agents unique is that they can take actions to achieve specific goals, whether that’s guiding a shopper to the perfect pair of shoes, helping an employee looking for the right health benefits, or supporting nursing staff with smoother patient hand-offs during shifts changes.
In our work with customers, we keep hearing that their teams are increasingly focused on improving productivity, automating processes and modernizing the customer experience. These aims are now being achieved through the AI agents they’re developing in six key areas: customer service; employee empowerment; creative ideation and production; data analysis; code creation; and cybersecurity.
These special capabilities are made possible in large part by the new multimodal capacity of generative AI and AI foundation models, which allow agents to handle tasks across a range of communications modes, including text, voice, video, audio, code, and more. With human support, agents can converse, reason, learn, and make decisions.
The hundreds of customers who joined us at Next ‘24 to showcase and discuss early versions of their AI agents and gen-AI solutions have come to rely on Google Cloud technologies that include our AI infrastructure, Gemini models, Vertex AI platform, Google Workspace, and Google Distributed Cloud. We were also joined by more than 100 partners supporting the creation of AI agents and AI solutions, which you can read about in detail.Here’s a snapshot of how 101 of these industry leaders are putting AI into production today, creating real-world use cases that will transform tomorrow.
Similar to great sales and service people, customer agents are able to listen carefully, understand your needs, and recommend the right products and services. They work seamlessly across channels including the web, mobile, and point of sale, and can be integrated into product experiences with voice and video.
ADT is building a customer agent to help its millions of customers select, order, and set up their home security.
Alaska Airlines is developing a personalized travel search experience using advanced AI techniques, creating hyper-personalized recommendations that engage customers early and foster loyalty through AI-generated content.
Best Buy is using Gemini to launch a generative AI-powered virtual assistant this summer that can troubleshoot product issues, reschedule order deliveries, manage Geek Squad subscriptions, and more; in-store and digital customer-service associates are also gaining gen-AI tools to better serve customers anywhere they need help.
The Central Texas Regional Mobility Authority is using Vertex AI to modernize transportation operations for a smoother, more efficient journey.
Etsy uses Vertex AI training to optimize their search recommendations and ads models, delivering better listing suggestions to buyers and helping sellers grow their businesses.
Golden State Warriors are using AI to improve the fan experience content in their Chase Center app.
IHG Hotels & Resorts is building a generative AI-powered chatbot to help guests easily plan their next vacation directly in the IHG One Rewards mobile app.
ING Bank aims to offer a superior customer experience and has developed a gen-AI chatbot for workers to enhance self-service capabilities and improve answer quality on customer queries.
Magalu, one of Brazil’s largest retailers, has put customer service at the center of its AI strategy, including using Vertex AI to create “Lu’s Brain” to power an interactive conversational agent for Lu, Magalu's popular brand persona (the 3D bot has more than 14 million followers between TikTok and Instagram).
Mercedes Benz will infuse e-commerce capabilities into its online storefront with a gen AI-powered smart sales assistant. Mercedes also plans to expand its use of Google Cloud AI in its call centers and is using Vertex AI and Gemini to personalize marketing campaigns.
Oppo/OnePlus is incorporating Gemini models and Google Cloud AI into their phones to deliver innovative customer experiences, including news and audio recording summaries, AI toolbox, and more.
Samsung is deploying Gemini Pro and Imagen 2 to their Galaxy S24 smartphones so users can take advantage of amazing features like text summarization, organization, and magical image editing.
The Minnesota Division of Driver and Vehicle Services helps non-English speakers get licenses and other services with two-way real-time translation.
Pepperdine University has students and faculty who speak many languages, and with Gemini in Google Meet, they can benefit from real-time translated captioning and notes.
Sutherland, a leading digital transformation company, is focused on bringing together human expertise and AI, including boosting its client-facing teams by automatically surfacing suggested responses and automating insights in real time.
Target uses Google Cloud to power AI solutions on the Target app and Target.com, including personalized Target Circle offers and Starbucks at Drive Up, their curbside pickup solution.
Tokopedia, an Indonesian ecommerce leader, is using Vertex AI to improve data quality, increasing unique products being sold by 5%.
US News saw a double-digit impact in key metrics like click-through rate, time spent on page, and traffic volume to its pages after implementing Vertex AI Search.
IntesaSanpaolo, Macquarie Bank, and Scotiabank are exploring the potential of gen AI to transform the way we live, work, bank, and invest — particularly how the new technology can boost productivity and operational efficiency in banking. Watch the session to learn more.
Employee agents help workers be more productive and collaborate better together. These agents can streamline processes, manage repetitive tasks, answer employee questions, as well as edit and translate critical communications.
Avery Dennison empowered their employees with generative AI to enable secure, flexible, and borderless collaboration for enhanced productivity to drive growth.
Bank of New York Mellon built a virtual assistant to help employees find relevant information and answers to their questions.
Bayer is building a radiology platform that will assist radiologists with data analysis, intelligent search, and to create documents that meet healthcare requirements needed for regulatory approval. The bioscience company is also harnessing BigQuery and Vertex AI to develop additional digital medical solutions and drugs more efficiently.
Bristol Myers Squibb is transforming its document processes for clinical trials using Vertex AI and Google Workspace. Now, documentation that took scientists weeks now gets to a first draft in minutes.
BenchSci develops generative AI solutions empowering scientists to understand complex connections in biological research, saving them time and financial resources and ultimately bringing new medicine to patients faster.
Cintas is using Vertex AI Search to develop an internal knowledge center for customer service and sales teams to easily find key information.
Covered California, the state’s healthcare marketplace, is using Document AI to help improve the consumer and employee experience by automating parts of the documentation and verification process when residents apply for coverage.
Dasa, the largest medical diagnostics company in Brazil, is helping physicians detect relevant findings in test results more quickly.
DaVita leverages DocAI and Healthcare NLP to transform kidney care, including analyzing medical records, uncovering critical patient insights, and reducing errors. AI enables physicians to focus on personalized care, resulting in significant improvements in healthcare delivery.
Discover Financial helps their 10,000 contact center representatives to search and synthesize information across detailed policies and procedures during calls.
HCA Healthcare is testing Cati, a virtual AI caregiver assistant that helps to ensure continuity of care when one caregiver shift ends and another begins. They are also using gen AI to improve workflows on time-consuming tasks, such as clinical documentation, so physicians and nurses can focus more on patient care.
The Home Depot has built an application called Sidekick, which helps store associates manage inventory and keep shelves stocked; notably, vision models help associates prioritize which actions to take.
Los Angeles Rams are utilizing AI across the board from content analysis to player scouting.
McDonald’s will leverage data, AI, and edge technologies across its thousands of restaurants to implement innovation faster and to enhance employee and customer experiences.
Pennymac, a leading US-based national mortgage lender, is using Gemini across several teams including HR, where Gemini in Docs, Sheets, Slides and Gmail is helping them accelerate recruiting, hiring, and new employee onboarding.
Robert Bosch, the world's largest automotive supplier, revolutionizes marketing through gen AI-powered solutions, streamlining processes, optimizing resource allocation, and maximizing efficiency across 100+ decentralized departments.
Symphony, the communications platform for the financial services industry, uses Vertex AI to help finance and trading teams collaborate across multiple asset classes.
Uber is using AI agents to help employees be more productive, save time, and be even more effective at work. For customer service representatives, they’ve launched new tools that summarize communications with users and can even surface context from previous interactions, so front-line staff can be more helpful and effective
The U.S. Dept. of Veterans Affairs is using AI at the edge to improve cancer detection for service members and veterans. The Augmented Reality Microscope (ARM) is deployed at remote military treatment facilities around the world. The prototype device is helping pathologists find cancer faster and with better accuracy.
The U.S. Patent and Trademark Office has improved the quality and efficiency of their patent and trademark examination process by implementing AI-driven technologies.
Verizon is using generative AI to help teams in network operations and customer experience get the answers they need faster.
Victoria’s Secret is testing AI-powered agents to help their in-store associates find information about product availability, inventory, and fitting and sizing tips, so they can better tailor recommendations to customers.
Vodafone uses Vertex AI to search and understand specific commercial terms and conditions across more than 10,000 contracts with more than 800 communications operators.
WellSky is integrating Google Cloud's healthcare and Vertex AI capabilities to reduce the time spent completing documentation outside work hours.
Woolworths, the leading retailer in Australia, boosts employees’ confidence in communications with “Help me write” across Google Workspace products for more than 10,000 administrative employees. It’s also using Gemini to create next-generation promotions, as well as for quickly assisting customer service reps in summarizing all previous customer interactions in real time.
Box, Typeface, Glean, CitiBank, and Securiti AI discuss developing AI-powered apps across the enterprise, with measurable returns on investment for marketing, financial services, and HR use cases.
Highmark Health and Freenome join Bristol Myers Squibb to explore how AI can improve efficiency and innovation across care delivery, drug discovery, clinical trial planning, and bringing medicines to market.
Creative agents can expand your organization with the best design and production skills, working across images, slides, and exploring concepts with workers. Many organizations are building agents for their marketing teams, audio and video production teams, and all the creative people that can use a hand. With creative agents, anyone can become a designer, artist, or producer.
Belk ECommerce is using generative AI to craft better product descriptions, a necessary yet time-consuming task for digital retails that has often been done manually.
Canva is using Vertex AI to power its Magic Design for Video, helping users skip tedious editing steps while creating shareable and engaging videos in a matter of seconds.
Carrefour used Vertex AI to deploy Carrefour Marketing Studio in just five weeks — an innovative solution to streamline the creation of dynamic campaigns across various social networks. In just a few clicks, marketers can build ultra-personalized campaigns to deliver customers advertising that they care about.
Major League Baseball continues to innovate its Statcast platform, so teams, broadcasters, and fans have access to live in-game insights.
Paramount currently relies on manual processes to create the essential metadata and video summaries used across its Paramount+ platform for showcasing content and creating personalized experiences for viewers. VertexAI Text Bison is now helping to streamline this process.
Procter & Gamble used Imagen to develop an internal gen AI platform to accelerate the creation of photo-realistic images and creative assets, giving marketing teams more time to focus on high-level planning and delivering superior experiences for its consumers.
WPP will integrate Google Cloud’s gen AI capabilities into its intelligent marketing operating system, called WPP Open, which empowers its people and clients to deliver new levels of personalization, creativity, and efficiency. This includes the use of Gemini 1.5 Pro models to supercharge both the accuracy and speed of content performance predictions.
Data agents are like having knowledgeable data analysts and researchers at your fingertips. They can help answer questions about internal and external sources, synthesize research, develop new models — and, best of all, help find the questions we haven’t even thought to ask yet, and then help get the answers.
AI21 Labs offers a BigQuery integration called Contextual Answers that allows users to query data conversationally and get high-quality answers quickly
Anthropic has partnered with Google Cloud to offer its family of Claude 3 models on Vertex AI — providing organizations with more model options for intelligence, speed, cost-efficiency, and vision for enterprise use cases.
The Asteroid Institute is using AI to discover hidden asteroids in existing astronomical data. This is a major focus for astronomers researching the evolution of the Solar System, investors and businesses hoping to fly missions to asteroids, and for all of us who want to prevent future large asteroid impacts on Earth.
Contextual is working with Google Cloud to offer enterprises fully customizable, trustworthy, privacy-aware AI grounded in internal knowledge bases.
Cox 2M, the commercial IoT division of Cox Communications, is able to make smarter, faster business decisions using AI-powered analytics.
Essential AI, a developer of enterprise AI solutions, is using Google Cloud’s AI-optimized TPU v5p accelerator chips to train its own AI models.
Generali Italia, Italy's largest insurance provider, used Vertex AI to build a model evaluation pipeline that helps ML teams quickly evaluate performance and deploy models.
Globo, one of Brazil’s largest media networks, is using Service Extensions and Media CDN to fight piracy during live events by blocking pirated streams in real time.
Hugging Face is collaborating with Google across open science, open source, cloud, and hardware to enable companies to build their own AI with the latest open models from Hugging Face and Google Cloud hardware and software.
Kakao Brain, part of Korean technology company Kakao Group, has built a large-scale AI language model that is the largest Korean language-specific LLM in the market, with 66 billion parameters. They’ve also developed a text-to-image generator called Karlo.
Mayo Clinic has given thousands of its scientific researchers access to 50 petabytes worth of clinical data through Vertex AI search, accelerating information retrieval across multiple languages.
McLaren Racing is using Google AI to get up-to-the-millisecond insights during races and training to gain a competitive edge.
Mercado Libre is testing BigQuery and Looker to optimize capacity planning and reservations with delivery carriers and airlines to fulfill shipments faster.
Mistral AI will use Google Cloud's AI-optimized infrastructure, to further test, build, and scale up its LLMs, all while benefiting from Google Cloud's security and privacy standards.
MSCI uses machine learning with Vertex AI, BigQuery and Cloud Run to enrich its datasets to help our clients gain insight into around 1 million asset locations to help manage climate-related risks.
NewsCorp is using Vertex AI to help search data across 30,000 sources and 2.5 billion news articles updated daily.
Orange operates in 26 countries where local data must be kept in each country. They are using AI on Google Distributed Cloud to improve network performance and deliver super-responsive translation capabilities.
Spotify leveraged Dataflow for large-scale generation of ML podcast previews, and they plan to keep pushing the boundaries of what’s possible with data engineering and data science to build better experiences for their customers and creators.
UPS is building a digital twin of its entire distribution network, so both workers and customers can see where their packages are at any time.
Workday is using natural language processing in Vertex Search and Conversation to make data insights more accessible for technical and non-technical users alike.
Woven — Toyota's investment in the future of mobility — is partnering with Google to leverage vast amounts of data and AI to enable autonomous driving, supported by thousands of ML workloads on Google Cloud’s AI Hypercomputer. This has resulted in resulting in 50% total-cost-of-ownership savings to support automated driving.
Broward County, Florida, and Southern California Edison are using geospatial capabilities and AI to improve infrastructure planning and monitoring, generate new insights, and create regional resilience for communities facing climate challenges today and tomorrow.
Kinaxis and Dematic are building data-driven supply chains to address logistics use cases including scenario modeling, planning, operations management, and automation.
NOAA and USAID are among the U.S. government agencies using Google Cloud AI to unlock critical data insights to streamline operations and improve mission outcomes — all with an emphasis on responsible AI. Watch the session to learn more.
Code agents are helping developers and product teams to design, create, and operate applications faster and better, and to ramp up on new languages and code bases. Many organizations are already seeing double-digit gains in productivity, leading to faster deployment and cleaner, clearer code.
Capgemini has been using Code Assist to improve software engineering productivity, quality, security, and developer experience, with early results showing workload gains for coding and more stable code quality.
Commerzbank is enhancing developer efficiency through Code Assist's robust security and compliance features.
Quantiphi saw developer productivity gains of more than 30% during their Code Assist pilot.
Replit developers will get access to Google Cloud infrastructure, services, and foundation models via Ghostwriter, Replit's software development AI, while Google Cloud and Workspace developers will get access to Replit’s collaborative code editing platform.
Seattle Children's hospital is using AI to boost data engineering productivity and accelerate development.
Turing is customizing Gemini Code Assist on their private codebase, empowering their developers with highly personalized and contextually relevant coding suggestions that have increased productivity around 30 percent and made day-to-day coding more enjoyable.
Wayfair piloted Code Assist, and those developers with the code agent were able to set up their environments 55 percent faster than before, there was a 48 percent increase in code performance during unit testing, and 60 percent of developers reported that they were able to focus on more satisfying work.
Security agents assist security operations by radically increasing the speed of investigations, automating monitoring and response for greater vigilance and compliance controls. They can also help guard data and models from cyberattacks, such as malicious prompt injection.
BBVA uses AI in Google SecOps to detect, investigate, and respond to security threats with more accuracy, speed, and scale. The platform now surfaces critical security data in seconds, when it previously took minutes or even hours, and delivers highly automated responses.
Behavox is using Google Cloud technology and LLMs to provide industry leading regulatory compliance and front office solutions for financial institutions globally.
Charles Schwab has integrated their own intelligence into the AI-powered Google SecOps, so analysts can better prioritize work and respond to threats.
Fiserv’s security operations engineers create detections and playbooks with much less effort, while analysts get answers more quickly.
Grupo Boticário, one of the largest beauty retail and cosmetics companies in Brazil, employs real-time security models to prevent fraud and to detect and respond to issues.
Palo Alto Networks’ Cortex XSIAM, the AI-driven security operations platform, is built on more than a decade of expertise in machine-learning models and the most comprehensive, rich, and diverse data store in the industry. Backed by Google's advanced cloud infrastructure and advanced AI services, including BigQuery and Gemini models, the combination delivers global scale and near real-time protection across all cybersecurity offerings.
Pfizer can now aggregate cybersecurity data sources, cutting analysis times from days to seconds.
To find even more customers using our AI tools to build agents and solutions for their most important enterprise projects, visit the Google Cloud customer hub and watch the Next ‘24
Technology Integration
2024年04月12日
Technology Integration
How Generative AI Adds Value to the Future of Work
这篇Upwork的文章深入探讨了生成式人工智能（AI）在重新塑造工作价值方面的变革力量，强调了自动化和创新不仅改变了工作岗位，还在各个行业提高了生产力和创造力。文章着重讨论了对劳动力市场的细微影响，强调了技能发展和道德考虑的重要性，并对人工智能与人类合作的未来提供了前瞻性的视角。
Authors: Dr. Ted Liu, Carina Deng, Dr. Kelly Monahan
Generative AI’s impact on work: lessons from previous technology advancements
In this study, we provide a comprehensive analysis of the initial impact of generative AI (artificial intelligence) on the Upwork marketplace for independent talent. Evidence from previous technological innovations suggests that AI will have a dual impact: (1) the displacement effect, where job or task loss is initially more noticeable as technologies automate tasks, and (2) the reinstatement effect, where new jobs and tasks increase earnings over time as a result of the new technology. Take for example the entry of robotics within the manufacturing industry. When robotic arms were installed along assembly lines, they displaced some of the tasks that humans used to do. This was pronounced in tasks that were routine and easy to automate. However, new tasks were then needed with the introduction of robotics, such as programming the robots, analyzing data, building predictive models, and maintaining the physical robots. The effects of new technologies often counterbalance each other over time, giving way to many new jobs and tasks that weren’t possible or needed before. The manufacturing industry is now projected to have more jobs available as technologies continue to advance, including Internet of Things (IoT), augmented reality, and AI, which transform the way work is completed. The issue now at hand is ensuring enough skilled workers are able to work alongside these new technologies.
While this dynamic of displacement and reinstatement generally takes years to materialize, as noted above in the manufacturing example, the effects of generative AI may be taking place already on Upwork. For the platform as a whole, we observe that generative AI has increased the total number of job posts and the average spend per new contract created. In terms of work categories, generative AI has reduced demand in writing and translation, particularly in low-value work, while enhancing earnings in high-value work across all groups. In particular, work that relies on this new technology like Data Science and Analytics are reaping the benefits. The report highlights the importance of task complexity and the skill-biased nature of AI's impact. Skills-biased technology change is to be expected as the introduction of new technologies generally favors highly skilled workers. We observe this on our platform as high-skill freelancers in high-value work are benefiting more, while those in low-value work face challenges, underscoring the need for skilling and educational programs to empower freelancers to adapt and transition in this evolving work landscape.
Understanding the lifecycle of work on Upwork and the impact of gen AI
Generative AI has a growing presence in how people do their work, especially since the public release of ChatGPT in 2022. While there’s been extensive discussion about the challenges and opportunities of generative AI, there is limited evidence of such impact based on transaction data in the broader labor market. In this study, we use Upwork’s platform data to estimate the short-term effects of generative AI on freelance outcomes specifically. The advantage of the Upwork platform is that it is in itself a complete marketplace for independent talent, as we observe the full life cycle of work: job posts, matching, work execution, performance reviews, and payment. Few other instances exist where a closed-system work market can be studied and observed. Thus, the results of this study offer insights into not only the online freelance market, but also the broader labor market.
How technological progress disrupts the labor market is not a new topic. Acemoglu and Restrepo (2019) argue that earning gain arises from new tasks created by technological progress, which they term the “reinstatement effect,” even if the automation of certain tasks may have a displacement effect in the labor market initially. What this means is that there may be a dynamic effect going on: the displacement effect (e.g., work loss) may be more noticeable in the beginning of a new technology entry, but as new jobs and tasks are being created, the reinstatement effect (e.g., rates increase, new work) will begin to prevail. In the broader labor market, such dynamics will likely take years to materialize. But in a liquid and active independent work marketplace like Upwork, it’s possible that we’re already observing this transition happening.
Existing studies such as this provides a useful conceptual framework to think about the potential impact of generative AI. It’s likely that in the short term, the replacement of generative AI will continue to be more visible, not just at Upwork, but also in the broader labor market. Over time and across work categories, however, generative AI will likely spur new tasks and jobs, leading to the reinstatement effect becoming stronger and increasing rates for those occupations with new tasks and a higher degree of task complexity. We’ve already seen evidence of new demand as a result of gen AI on our Upwork platform, with brand new skill categories like AI content creator and prompt engineer emerging in late 2022 and early 2023. We test this hypothesis of both work displacement and reinstatement, and provide insights into how generative AI affects work outcomes.
Impact of generative AI on work
To understand the short-term impact of generative AI on the Upwork freelance market, we capitalize on a natural experiment arising from the public release of ChatGPT in November 2022. Because this release was largely an unanticipated event to the general public, we’re able to estimate the causal impact of generative AI. The essential idea behind this natural experiment is that we want to compare the work groups affected by AI with the counterfactual in which they are not. To implement this, we use a statistical and machine-learning method called synthetic control. Synthetic control allows us to see the impact that an intervention, in this case, the introduction of gen AI, has on a group over time by comparing it to a group with similar characteristics not exposed to the intervention. The advantage of this approach is that it allows us to construct reasonably credible comparison groups and observe the effect over time.
The units of analysis we use are work groups on the Upwork platform; we analyze variables such as contract number and freelancer earnings. Instead of narrowly focusing on a single category like writing, we extend the analysis to all the major work groups on Upwork. Moreover, we conduct additional analysis of the more granular clusters within each major group. The synthetic control method allows for flexibility in constructing counterfactuals at different levels of granularity. The advantage of our comprehensive approach is that we offer a balanced view of the impact of generative AI across the freelance market.
Generative AI’s short-term impact on job posts and freelancer earnings
Looking at the platform as a whole, we observe that generative AI has increased the total number of job posts by 2.4%, indicating the overall increased demand from clients. Moreover, as shown in Figure 1, for every new job contract, there is an increase of 1.3% in terms of freelancer earnings per contract, suggesting a higher value of contracts.
Figure 1 Effect of Generative AI on Freelancer Earning per Contract
The Upwork platform has three broad sectors: 1. Technological and digital solutions (tech solutions); 2. Creative & outreach; 3. Business operations and consulting. We have observed both positive and negative effects within each of the sectors, but two patterns are worth noting:
The reinstatement effect of generative AI seems to be driving growth in freelance earnings in sectors related to tech solutions and business operations. In contrast, within the creative sector, while sales and marketing earnings have grown because of AI, categories such as writing and translation seem disproportionately affected more by the replacement effect. This is to be expected due to the nature of tasks within these categories of work, where large language models are now able to efficiently process and generate text at scale.
Generative AI has propelled growth in high-value work across the sectors and may have depressed growth in low-value work. This supports a skills-biased technology change argument, which we’ve observed throughout modern work history.
More specifically and within tech solutions, data science & analytics is a clear winner, with over 8% of growth in freelance earnings attributed to generative AI. This makes sense as the reinstatement effect is at work; new work and tasks such as prompt engineering have been created and popularized because of generative AI. Simultaneously, while tools such as ChatGPT automate certain scripting tasks (therefore leading to a replacement effect), it mainly results in productivity enhancements for freelancers and potentially leads to them charging higher rates and enjoying higher overall earnings per task.
In terms of contracts related to business operations, we observe that accounting, administrative support, and legal services all experience gains in freelance earnings due to generative AI, ranging from 6% to 7%. In this sector, customer service is the only group that has experienced reduced earnings (-4%). The reduced earnings result for customer service contracts is an example of the aggregate earnings outcomes of AI, related to the study by Brynjolfsson et al (2023), who find that generative AI helps reduce case resolution time at service centers.
A potential outcome of this cut in resolution time is that service centers will need fewer workers, as more tasks can be completed by a person working alongside AI. At the same time, the reinstatement effect has not materialized yet because there are no new tasks being demanded in such settings. This may be an instance where work transformation has not yet been fully realized, with AI enabling faster work rather than reinventing a way of working that leads to new types of tasks. A contrasting case is the transformation that happened with bank tellers when ATMs were introduced. While the introduction of these new technologies resulted in predictions of obsolete roles in banks, something different happened over time. Banks were able to increase efficiency as a result of ATMs and were able to scale and open more branches than before, thereby creating more jobs. In addition, the transactional role of a bank teller became focused on greater interpersonal skills and customer relationship tasks.
When taken together, the overall gains in such business operations work on Upwork are an encouraging sign. These positions tend to require relatively intensive interpersonal communication, and it seems the short-term effects of generative AI have helped increase the value of these contracts, similar to what we saw in the banking industry when ATMs were introduced.
As of now, the replacement effect of AI seems more noticeable in creative and outreach work. The exception is sales and marketing contracts, which have experienced a 6.5% increase in freelance earnings. There is no significant impact yet observed on design. For writing and translation, however, generative AI seems to have reduced earnings by 8% and 10% respectively. However, as we will discover, task complexity has a moderating effect on this.
High-value work benefit from generative AI, upskilling needed for low-value work
Having discussed the overall impact of generative AI across categories, we now decompose the impact by values. The reason we’re looking at the dimension of work value is that there may be a positive correlation between contract value and skill complexity. Moreover, skill complexity may also be positively correlated with skill levels. Essentially, by evaluating the impact of AI by different contract values, we can get at the question of AI's impact by skill levels. This objective is further underscored by a discrepancy that sometimes exists in the broader labor markets – a skills gap between demand and supply. It simply takes time for upskilling to take place, so it’s typical for demand to exceed supply until a more balanced skilled labor market takes place. It is worth noting, however, freelancers on the Upwork platform seem more likely than non-freelancers to acquire new skills such as generative AI.
For simplicity, let’s assume that the value of contracts is a good proxy for the level of skill required to complete them. We’d then assume that high-skill freelancers typically do high-value work, and low-skill freelancers do low-value work. In other words, our goal is also to understand whether the impact of generative AI is skills-biased and follows a similar pattern from what we’ve seen in the past with new technology disruptions. Note that we’re focusing on the top and bottom tails of the distribution of contract values, because such groups (rather than median or mean) might be most susceptible to displacement and/or reinstatement effects, therefore of primary concern. We define high-value (HV) work as those with $1,000 or more earnings per contract. For the remaining contracts, we focus on a subset of work as low-value (LV) work ($251-500 earnings).
Figure 2 shows the impact of AI by work value, across groups on Upwork. As we discussed before, writing and translation work has experienced some reduction in earnings overall. However, if we look further into the effect of contract value, we see that the reduction is largely coming from the reduced earnings from low-value work. At the same time, for these two types, generative AI has induced substantial growth in high-value earnings – the effect for translation is as high as 7%. We believe the positive effect on translation high-value earning is driven by more posts and contracts created. In the tech solutions sector, the growth in HV earnings in data science and web development is also particularly noticeable, ranging from 6% to 9%. Within the business solutions sector, administrative support is the clear winner.
There are two takeaways from this analysis by work value. First, while we’re looking at a sample of all the contracts on the platform, it’s possible that the decline of LV work is more than made up for by the growth of HV work in the majority of the groups. In other words, except for select work groups, the equilibrium results for the Upwork freelance market overall seem to be net positive gains from generative AI. Second, if we assume that freelancers with high skills (or a high degree of skill complexity) tend to complete such HV work (and low-skill freelancers do LV work), we observe that the impact of generative AI may be biased against low-skill freelancers. This is an important result: In the current discussion of whether generative AI is skill-based, there exists limited evidence based on realized gains and actual work market transactions. We are one of the first to provide market-transaction-based evidence to illustrate this potentially skill-biased impact. Finally, additional internal Upwork analysis finds that independent talent engaged in AI-related work earn 40% more on the Upwork marketplace than their counterparts engaged in non-AI-related work. This suggests there may be additional overlap between high-skill work and AI-related work, which can further reinforce the earning potential of freelancers in this group.
Figure 2
Case study: 3D content work
To illustrate the impact of generative AI in more depth, we have conducted a case study of Engineering & Architecture work within the tech solutions sector. The reason is that we want to illustrate the potentially overlooked aspects of AI impact, compared with the examples of data science and writing contracts. This progress in generative AI has the potential to reshape work in traditional areas like design in manufacturing and architecture, which rely heavily on computer-aided design (CAD) objects, and newer sectors such as gaming and virtual reality, exemplified by NVIDIA's Omniverse.
Based on activities on the Upwork platform, we see that there is consistent growth of job posts and client spending in this category, with up to 12% of gross service value growth year over year in 2023 Q3, and over 11% in job posts during the same period. Moreover, applying the synthetic control method, we show a causal relationship between gen AI advancements and the growth in job posts and earnings per contract. More specifically, there is a significant increase in overall earnings because of AI, an average 11.5% increase. Additionally, as shown by Figure 3, the positive effect also applies to earning per contract. This indicates a positive impact on freelancer productivity and quality of work, due to the fact that we’re measuring the income for every unit of work produced. This suggests that gen AI is not just a facilitator of efficiency but also enhances the quality of output.
Figure 3 Effect of Generative AI on Freelancer Earning per Contract in EngineeringIn a traditional workflow to create 3D objects without generative AI, freelancers would spend extensive time and effort to design the topology, geometry, and textures of the objects. But with generative AI, they can do so through text prompts to train models and generate 3D content. For example, this blog by NVIDIA’s Omniverse team showcases how ChatGPT can interface with traditional 3D creation tools.
Thus, the positive trajectory of generative AI in 3D content generation we see is driven by several factors. AI significantly reduces job execution time, allowing for higher productivity. It facilitates the replication and scaling of 3D objects, leading to economies of scale. Moreover, freelancers can now concentrate more on the creative aspects of 3D content, as AI automates time-consuming and tedious tasks.
This shift has not led to a decrease in rates due to the replacement effect. In fact, this shift of workflow may create new tasks and work. We will likely see a new type of occupation in which technology and humanities disciplines converge. For instance, a freelancer trained in art history now has the tools to recreate a 3D rendering of Japan in the Edo period, without the need to conduct heavy coding. In other words, the reinstatement effect of AI will elevate the overall quality and value proposition of the work, and ultimately enable higher earning gains. This paradigm shift underscores generative AI's role in not just transforming work processes but also in creating new economic dynamics within the 3D content market. Fortunately, it seems many freelancers on Upwork are ready to reap the benefits: 3D-related skills, such as 3D modeling, rendering, and design, are listed among the top five skills of freelancer profiles as well as in job posts.
A dynamic interplay: task complexity, skills, and gen AI
Focusing on the Upwork marketplace for independent talent, we study the impact of generative AI by using the public release of ChatGPT as a natural experiment. The results suggest a dynamic interplay of replacement and reinstatement effects; we argue that this dynamic is influenced by task complexity, suggesting a skill-biased impact of gen AI. Analysis across Upwork's work sectors shows varied effects: growth in freelance earnings in tech solutions and business operations, but a mixed impact in the creative sector. Specifically, high-value work in data science and business operations see significant earnings growth, while creative contracts like writing and translation experience a decrease in earnings, particularly in lower-value tasks. Using the case study of 3D content creation, we show that generative AI can significantly enhance productivity and quality of work, leading to economic gains and a shift toward higher-value tasks, despite initial concerns of displacement.
Acemoglu and Restrepo (2019) argue that the slowdown of earning growth in the United States the past three decades can partly be explained by new technologies’ replacement effect overpowering the reinstatement effect. But with generative AI, we’re at a point of completely redefining what human tasks mean, and there may be ample opportunities to create new tasks and work. It's evident that while high-value types of work are being created, freelancers engaged in low-value tasks may face negative impact, possibly due to a lack of skills needed to capitalize on AI benefits. This situation underscores the necessity of supporting freelancers not only in elevating their marketability within their current domains but also in transitioning to other work categories.
To ensure as many people as possible benefit, there’s an imperative need to provide educational resources for them to gain the technical skills, and more importantly skills of adaptability to reinvent their work. This helps minimize the chance of missed opportunities by limiting skills mismatch between talent and new demands created by new technologies. Upwork has played a significant role here by linking freelancers to resources such as Upwork Academy’s AI Education Library and Education Marketplace, thereby equipping them with the necessary tools and knowledge to adapt and thrive in an AI-present job market. This approach can help bridge the gap between low- and high-value work opportunities, ensuring a more equitable distribution of the advantages brought about by generative AI.
Methodology
To estimate the causal impact of generative AI, we take a synthetic control approach in the spirit of Abadie, Diamond, and Hainmueller (2010). The synthetic control method allows us to construct a weighted combination of comparison units from available data to create a counterfactual scenario, simulating what would have happened in the absence of the intervention. We use this quasi-experimental method due to the infeasibility of conducting a controlled large-scale experiment. Additionally, we use Lasso regularization to credibly construct the donor pool that serves the basis of the counterfactuals and minimize the chance of overfitting the data.
Moreover, we supplement the analysis by scoring whether a sub-occupation is impacted or unaffected by generative AI. The scoring utilizes specific criteria: 1. Whether a certain share of job posts are tagged as AI contracts by the Upwork platform; 2. AI occupational exposure score, based on a study by Felten, Raj, and Seamans (2023), to tag these sub-occupations. We also use data smoothing techniques through three-month moving averages. We analyzed data collected on our platform from 2021 through Q3 2023. We specifically look at freelancer data across all 12 work categories on the platform for high-value contracts, defined as those with a contract of at least $1,000, and low-value contracts, consisting of those between $251 and under $500.
The main advantage of our approach is that it is a robust yet flexible way to identify the causal effects on not only the Upwork freelance market but also specific work categories. Additionally, we control for macroeconomic or aggregate shocks such as U.S. monetary policy in the pre-treatment period. However, we acknowledge the potential biases in identifying which sub-occupations are influenced by generative AI and the effects of external factors in the post-treatment period.
About the Upwork Research Institute
The Upwork Research Institute is committed to studying the fundamental shifts in the workforce and providing business leaders with the tools and insights they need to navigate the here and now while preparing their organization for the future. Using our proprietary platform data, global survey research, partnerships, and academic collaborations, we produce evidence-based insights to create the blueprint for the new way of work.
About Ted Liu
Dr. Ted Liu is Research Manager at Upwork, where he focuses on how work and skills evolve in relation to technological progress such as artificial intelligence. He received his PhD in economics from the University of California, Santa Cruz.
About Carina Deng
Carian Deng is the Lead Analyst in Strategic Analytics at Upwork, where she specializes in uncovering data insights through advanced statistical methodologies. She holds a Master's degree in Data Science from George Washington University.
About Kelly Monahan
Dr. Kelly Monahan is Managing Director of the Upwork Research Institute, leading our future of work research program. Her research has been recognized and published in both applied and academic journals, including MIT Sloan Management Review and the Journal of Strategic Management.
Technology Integration
Technology Integration
扫一扫
添加客服
