Original | Text BFT robot
01
The development of artificial intelligence has ushered in a new inflection point
ChatGPT is awakening global awareness of the transformative potential of artificial intelligence (AI), inspiring an unprecedented wave of attention and creativity.
The technology can mimic human ability to talk and make decisions, bringing us to the first real inflection point in public adoption of AI. Ultimately, everyone, everywhere, will actually feel the disruptive potential of this technology. Just two months after its launch, ChatGPT reached 100 million monthly active users, making it the fastest-growing consumer app ever.
A base model is a generic term for large models that have billions of parameters. Recent advances have enabled companies to build specialized image and language generation models based on these foundational models. The Big Language Model (LLM) is both generative and foundational AI.
Big language models are changing the rules of the market with two advantages. First, such models crack the code of language complexity. Today, machines have unprecedented capabilities to learn language, contextual meaning, and intent to generate and create content independently. Second, after pre-training with large amounts of data (text, images, or audio), these models can be adjusted or fine-tuned for many different tasks. This allows users to reuse the model as is or slightly modified in a variety of ways.
They foresee how big language models and generative AI will fundamentally transform business, academia, and even society itself, opening up new frontiers of capabilities. These new technologies have had a hugely positive impact on human creativity and productivity. For example, Accenture research found that 40% of working time in all industries will be assisted by large language models such as GPT-4.
That's because language tasks account for 62% of total staff hours, and 65% of that time can be boosted by augmentation and automation to increase productivity.
02
History: Milestones in the development of generative artificial intelligence
Machine learning: The analysis and prediction phase
In the first decade of the 21st century, machine learning techniques of all kinds have rapidly advanced to analyze massive amounts of online data, draw conclusions from output information, or learn." Since then, businesses have seen machine learning as an extremely powerful field of AI for analyzing data, discovering patterns, forming insights, building predictions, and automating tasks at a much faster and larger scale than ever before.
Deep learning: the vision and speech processing phase
Entering the second decade, the perception capabilities of artificial intelligence have made great strides, and this field of machine learning is known as deep learning. During this time, deep learning has made breakthroughs. On the one hand, the implementation of computer vision can help search and guide and classify and detect objects in autonomous vehicles: at the same time, it can also support speech recognition, enabling widely used artificial intelligence voice assistants to interact with users in a more natural way.
Generative Artificial Intelligence: A New Stage in Mastering Language
Based on the exponential growth in the scale and capabilities of deep learning models, the next decade will be the era of machines mastering languages. The GPT-4 language model, developed by OpenAl, marks the beginning of a new functional phase for language-based AI applications. Models such as these will have a profound impact on business, as language is inextricably linked to all aspects of a company's day-to-day work – on which institutional knowledge, interactions, and operational processes depend.
03
Use or customization: The popularity and application of generative AI
ChatGPT, Wen Xin Yiyan, Tongyi Qianwen 34DALL· A series of easy-to-use generative AI applications such as StableDiffusion are rapidly driving the adoption of the technology in the business sector and the general public, which will have a profound impact on enterprises.
Because the big language model has the ability to process large-scale data sets, it can "grasp" all the information accumulated by the enterprise for a long time, including the development history, development background, business characteristics and business intentions, and even down to the product, market and customer. Everything that is recorded in words, such as apps, systems, documents, email, chat, video, audio, and more, will be innovated, optimized, and reinvented, ultimately taking it to new heights.
97% of global executives believe that foundational AI models will connect across data types, revolutionizing how and how AI is used.
We are entering the next phase of the technology adoption cycle, with most enterprises starting to buy model-as-a-service business applications. But for many businesses, the greatest value comes from customizing or fine-tuning models with their own data to meet their unique needs.
use
Generative AI and big language model applications are now readily available and available. Enterprises can call these programs through application programming interfaces (APIs) and use prompt engineering techniques such as prompt tuning and prefix learning to customize them to a lesser extent for their specific needs.
customize
But most businesses need to customize models and fine-tune them with their own data to expand their usefulness and value. This enables the model to support specific downstream tasks across the business. In doing so, companies can effectively use AI to leapfrog their performance – empowering employees, improving customer satisfaction, introducing new business models, and sensing upcoming changes.
04
Looking ahead to rapidly changing technology, regulatory and commercial
Businesses will use these models to reinvent the way they work. As it becomes the norm for employees to work collaboratively with AI lieutenants, every role in every business has the potential to be completely transformed, significantly expanding what can be achieved by humans alone. In any given job, some tasks will be automated, some can be assisted, and some will be largely technology-agnostic.
In addition, a host of new tasks await humans, such as ensuring the accurate and responsible use of new AI systems.
Companies should pay special attention to the impact of AI on these roles:
Content Creation:
Generative AI will be an indispensable creative companion, revealing new ways to reach and engage audiences, but also bringing unprecedented speed and innovation in production design, design research, visual identity, name generation and testing, and real-time personalization.
Enterprises are introducing the most complete artificial intelligence system DALL· E, for social media promotion. DALL· E creates realistic images and artwork based on text descriptions, handles up to 12 billion parameters when converting text into images, and creates images that can be shared on Instagram and Twitter.
Write the code:
Software coders will be greatly productive with the help of generative AI—quickly translating one programming language to another, mastering programming tools and methods, automating code writing, predicting and pre-empting problems, and managing system documentation.
Accenture is experimenting with the OpenAl Big Language Model to increase developer productivity by automatically generating documentation. For example, the rationale for the configuration of the SAP system and the setting of various functions or technical parameters. This solution enables users to submit requests through Microsoft Teams chat conversations while working; The correct combination of documents is then quickly returned, a typical example of how to enhance the ability to complete specific tasks and automate them without changing the entire job.
Automation:
Generative AI's proven understanding of historical context, next best moves, summaries, and predictive intelligence will usher in a new era of hyper-efficiency and hyper-personalization in both back-office and front-office environments, taking business process automation to a transformative new level.
A multinational bank is using generative artificial intelligence and big-language models to change the way it manages a large number of post-trade emails, such as automatically drafting messages with recommendations for action and sending them to recipients. This not only reduces the workload, but also makes customer communication smoother.
Security protection:
Over time, generative AI will support organizations in strengthening governance and information security, preventing fraud, improving regulatory compliance, and proactively identifying risks by establishing cross-domain connections and inference capabilities both inside and outside the organization.
In a strategic cyber defense system, the big language model can provide a variety of useful functions, such as explaining malware and quickly classifying websites. But in the short term, businesses are likely to see hackers using the expertise of generative AI to generate malicious code or write the perfect phishing email.
Moments like this are not common. Investments in generative AI, big language models, and foundational models will be huge in the coming years. Unlike in the past, technology, regulation, and commercial applications will evolve in parallel and at an ever-increasing pace. In the past innovation curve, technology has typically moved faster than adoption and regulation.
Technology stack
The complex technologies that support generative AI are expected to evolve rapidly at every stack level, with wide-ranging business implications. The amount of computation required to train top-tier AI models is growing exponentially—according to reports, which now doubles every 3.4 to 10 months. As a result, cost and carbon emissions have become core considerations for the adoption of energy-intensive generative AI.
"The hottest new programming platform is napkins. Paul Daugherty – Global President and Chief Technology Officer, Accenture Technology Services
He refers to the fact that entrepreneurs are using OpenAl to build work websites based on creative sketches drawn on napkins.
Risk and regulatory environment
Businesses will have thousands of ways to apply generative AI and foundational models to maximize efficiency and enhance competitive advantage. It's clear that companies are gearing up on this new track. Companies need to start from a holistic strategy, in addition to generative AI and large-language models, must consider all types of AI and the related technologies they intend to use.
ChatGPT has further triggered people's thinking about the healthy development and normative application of artificial intelligence. As technology evolves and adopts faster than legislation, businesses in particular should pay close attention to any legal, ethical and reputational risks they may face.
It's important to note that generative AI technologies, including ChatGPT, are designed with accountability and compliance in mind to ensure that such models and applications don't pose an unacceptable risk to the business.
As an industry leader in responsible AI, Accenture defined and implemented the principles of responsible AI back in 2017 to integrate them into our business practices and customer service. Responsible AI is a practice of designing, building, and deploying AI that follows clear principles to empower business while safeguarding the public interest and benefiting society. As a result, companies can also give AI full trust and expand the scope of AI use with confidence.
AI systems need to be "refined" with diverse and inclusive input datasets that reflect broader business and social responsibility, fairness, and transparency. If AI can be designed and implemented within an ethical framework, it can accelerate the potential of responsible, collaborative intelligence tools that bring human intelligence and technology closer together.
This not only lays the foundation of trust for consumers, professionals and society at large, but also improves business performance and opens up new sources of growth.
The scale at which businesses adopt generative AI
To create the value of AI, companies must transform the way they work. Business leaders need to start now, design jobs and tasks with new ideas, and reskill people. Ultimately, every role in the enterprise is likely to be reshaped, and today's work will be broken down into a set of tasks that can be automated or powered by artificial intelligence, and the future of human-robot collaboration will be reimagined.
As we learn more about generative AI, it will disrupt traditional working models and usher in a new era of cooperation between humans and AI. Most workers will have competent "assistants" that fundamentally change how and what work is done.
Almost all jobs will be affected, and many new types of jobs will continue to emerge. Companies that take immediate steps to break down work into tasks and invest in training employees to work with machines to work differently can leapfrog performance beyond short-sighted competitors.
Nearly six in ten businesses intend to use ChatGPT for learning purposes, and more than half plan to pilot it in 2023. More than 40% of companies are willing to invest in this.
05
Embracing the Age of Generative AI: Six Key Points for Technical Applications
Business-driven
Even with the many advantages of innovative technology, rolling it out across the organization is not easy, especially when it will revolutionize existing ways of working.
Companies can start the transformation and reskilling agenda by trying the many capabilities of generative AI, accumulating early success, gaining support from change advocates and opinion leaders, increasing employee acceptance of new technologies, and creating the conditions needed for further adoption.
Companies must try both prongs. First, focus on accessible opportunities and realize returns quickly using consumable models and applications. Second, focus on using models tailored to your own data to reinvent your business, customer negotiations, and products, and services. Business-driven thinking is key to defining and successfully building application patterns.
As enterprises deepen their exploration of AI to reshape their businesses, they will reap tangible value, identify the most suitable types of AI for various application scenarios, and clarify the scale and complexity of investment. They can also test and improve approaches to data privacy security, enhance model accuracy, prevent bias, protect fairness, and know when "human in the loop" protections are needed.
98% of global executives agree that foundational AI models will play an important role in their business strategy in the next three to five years.
A bank uses an enhanced search tool to give employees the information they need
A large European banking group uses Microsoft Azure cloud platform and GPT-3 language models to help employees retrieve electronic documents. This initiative allows users to quickly get answers to their questions, saving time and improving accuracy and compliance.
To further upskill its workforce, the bank has built a three-year innovation program that will follow with generative AI in areas such as contract management, conversational reporting, and bill classification. This not only upgrades the internal knowledge base, helps employees get the information they need, but also helps advance their goal of becoming a data-driven organization.
Putting people first
For generative AI to succeed, companies need to focus as much on people and training as they do on technology. Therefore, they should significantly increase their investment in talent to address two different types of challenges: creating artificial intelligence and using artificial intelligence. This means developing talent in technical capabilities such as AI design and enterprise architecture on the one hand, while training people across the organization to effectively work with AI-based processes.
For example, in our analysis of 22 job categories, we found that large language models affect all categories, ranging from a minimum of 9% per workday to a maximum of 63%. Five out of 22 occupations can revolutionize more than half of the hours worked using big-language models.
In fact, an economic study conducted by an independent body shows that companies are seriously underinvesting in helping employees keep pace with AI developments, which requires more integrated cognition and judgment-based task setting. Even real-world experts in various fields who are well-versed in how to apply data (e.g., doctors interpreting patient health data) lack sufficient technical knowledge to understand how these models work and believe that technology can be "working partners."
Companies will also create entirely new roles, including linguistic specialists, AI quality controllers, AI editors and prompt engineers. For the most promising areas of generative AI, companies should first break down existing efforts into a portfolio of basic tasks. Then assess the extent to which generative AI might impact each task – fully automated, augmented, or unrelated to it.
Prepare proprietary data
To customize the underlying model, enterprises need to use domain-specific enterprise data, semantics, knowledge, and methods. Before the advent of generative AI, enterprises can derive value from AI through an AI approach centered on application patterns without having to modernize their data architectures and assets. Now, however, the situation is quite different. The underlying model requires a lot of well-organized data to learn, so solving the data challenge has become a top priority for every business.
Organizations need a strategic, disciplined approach to acquiring, developing, refining, protecting, and deploying data. Specifically, build a modern enterprise data platform with a set of trusted, reusable data products based on a cloud environment. With the cross-functional nature of such platforms, enterprise-grade analytics tools, and storing data in a cloud warehouse or data lake, data can be used across the enterprise without organizational silos. Enterprises can then analyze all business data uniformly in one location or through distributed computing strategies such as data meshes.
Invest in a sustainable technology base
To fully meet the large-scale computing needs of big language models and generative AI, enterprises need to consider whether they have the right technology infrastructure, architecture, operating model, and governance structure, while keeping an eye on costs and sustainable energy consumption. They must find ways to evaluate and compare these technologies with other AI or analytics tools in terms of costs and benefits, which may be better suited to a particular application model and at a fraction of the cost.
As the use of AI increases, so will the carbon emissions generated by the underlying infrastructure. Therefore, companies need to establish a strong green software development framework that considers energy efficiency and material-related emissions at all stages of the software development lifecycle. AI can also play a broader role in making businesses more sustainable and meeting environmental, social and governance (ESG) goals. Our research found that 70% of companies that have successfully reduced emissions in their production and operations use AI.
Accelerate ecosystem innovation
Creating a base model can most likely be a complex, costly, computationally intensive endeavor. With the exception of the world's top enterprises, almost all organizations cannot accomplish this task on their own, which is beyond their capabilities and methods. It's exciting to see that thanks to massive investments from hyperscale cloud service providers, tech giants, and start-ups, enterprises can now harness the power of emerging ecosystems. In 2023 alone, global investment in AI startups and growth-stage companies is expected to exceed $50 billion. These partners bring best practices honed over years and provide valuable insights into how to use the underlying model efficiently and effectively in specific application patterns. Having the right network of partners—including technology companies, professional services providers, and academic institutions—will be key to navigating rapid change.
Improve your own level of responsible artificial intelligence
The rapid adoption of generative AI presents a new imperative for all businesses: establishing a robust and responsible AI compliance system. This includes two aspects – establishing a control process to assess the potential risks of generative AI applications at the design stage; Develop clear measures to embed a responsible AI approach across the business. Accenture research shows that most companies still have a long way to go. Our 2022 survey of 850 executives worldwide shows that respondents generally recognize the importance of responsible AI and AI regulation. However, only 6% of companies believe they have a robust foundation for responsible AI.
Corporate principles of responsible AI should be defined and led at the top and translated into an effective risk management and compliance governance architecture, including organizational principles and policies, as well as applicable laws and regulations. Responsible AI use must be led by the CEO, starting with enhanced training and awareness development and expanding to a focus on execution and compliance. Accenture pioneered this approach to managing responsible AI several years ago, setting a CEO-led agenda and now taking a formal compliance program into place. Our own experience shows that a principles-driven approach to compliance provides both guardrails and is flexible enough to evolve with rapid technology to ensure that businesses are not always struggling to "catch up".
To be accountable by design, companies need to move from a reactive compliance strategy to a proactive development of well-established responsible AI systems. This must be done through an integrated framework that covers: principles and governance measures, risk management, policies and controls, and technology and enabling factors. Cultural and training work.
Timing is everything. In a recent Accenture Technology Trends Survey, 72% of 225 Chinese executives surveyed were very, or extremely excited, about the new capabilities enabled by foundational AI models, but slightly below the global average, leaving room to further explore the potential and applications of generative AI. For the benefits of AI big language models, Chinese companies have more positive expectations than globally in areas such as rapid and large-scale analytics capabilities, improved employee skills, new AI applications and service development, communication, processes and talent, but lower expectations in accelerating innovation, improving customer experience, and making quick decisions.
Nevertheless, more and more Chinese companies are actively exploring generative AI technologies and are beginning to apply large language models to achieve more innovation and efficiency gains. To this end, we have sorted out the methods and application suggestions applicable to the local deployment of Chinese enterprises.
In China, there are three main ways to apply big language models: SaaS, private cloud deployment, and localized deployment.
At present, the SaaS deployment method is the most mature, with foreign Azure OpenAl services as the benchmark. But in the domestic market, Baidu's Wen Xin Yiyan and Ali's Tongyi Qianwen are participating in fierce competition. Compared with the services provided by Azure, the SaaS services provided by domestic vendors have more advantages in data security and compliance, although the comprehensive capabilities need to be strengthened.
Serve relatively professional clients, leveraging the industry knowledge provided by clients while not being available to competitors. A more flexible server usage strategy also makes this approach much lower than the upfront investment of localization. On the whole, this is the most feasible way for domestic vertical industry customers to achieve at present.
There are many options for localizing your deployment. Academia offers ChatGLM from Tsinghua University, Alpaca from Stanford, Dolly from Databricks from commercial companies, and big language models from Scale.ai image expertise. Compared with the above two methods, the localization deployment method has two problems: high cost and uncertain usage effect. Therefore, it is currently at a very early stage and it remains to be seen whether it can be used further.
In general, the large language model is in a stage of rapid development, and its future form cannot be predicted, but it is certain that large-scale application must be an inevitable trend. Whether in scientific research, business or civil applications, large language models have a wide range of application prospects, and continuous innovation and progressive technology also provides a broader development space for its future applications.
Companies need to invest as much as they invest in technology, in evolving business operations and people skills training. Reimagining the way work is done and helping employees keep up with technology-driven change will be two of the most important factors in realizing the full potential of AI technology leapfrogging.
At present, Chinese enterprises are in a critical period of breakthroughs in artificial intelligence. AI will not only reshape business but also entire industries. The future is promising.
Source: Accenture
Report Editor: Intelligent Robotic Systems
For more exciting content, please pay attention to the public number: BFT robot
This article is an original article, the copyright belongs to BFT robot, if you want to reprint, please contact us. If you have any questions about the content of this article, please contact us and we will respond promptly.