2024년 1월 9일 화요일

Decode the Algorithm: Navigate the World of Machine Learning in Business with 'The AI Playbook'

 

Decode the Algorithm: Navigate the World of Machine Learning in Business with 'The AI Playbook'




Yong Xune Xon PhD. Revision Consulting. Seoul Korea.  

xyxonxyxon@empal.com



In my 30 years of studying, researching, and applying machine learning models for prediction, recommendation and anomaly detection in various business environments, I've consistently felt that there are too few people who understand why they are trying to implement this technology. Whether they are data scientists (now often referred to as data analysts), users or managers who need to use the results of machine learning, or executives who approve the projects, almost all have little to no understanding of what they are creating, why they are creating it, or what results ML should deliver.


'The AI Playbook' is a book about the application of machine learning in real organizations, calmly explaining many very important and fundamental aspects that most data scientists, machine learning engineers, managers, and various practitioners – in other words, the ultimate users of machine learning – do not know or overlook.




https://www.amazon.com/AI-Playbook-Mastering-Deployment-Management/dp/0262048906



Among the many precious stories this book discusses, let's examine a few. Elevating human understanding and leadership is more crucial for the success of machine learning than the technology itself. When business leaders are presented with a model as the outcome of an enthusiastically initiated project, they often reject its deployment. This rejection stems not only from technical limitations but also from a lack of understanding of ML's value within the organization.


The first step of ML is to define the goal. This involves defining the business value proposition and deciding how ML will improve the task (when deployed). However, less than half of the decision-makers in organizations understand data and the stories conveyed through visualization and models. If this issue is not addressed, ML can become meaningless, a waste of resources, or worse, a source of misinformation causing problems within the organization.


I believe that the CRISP-DM, introduced in the 1990s, was a groundbreaking work for its time, offering innovative ideas and well-defined procedures. It was significant because it added a step to the then prevalent methodology of statistical analysis and machine learning, which was technology-centric, to concretize the business objectives and expected outcomes, thereby clarifying why organizations use technology.


However, it's regrettable that even after a considerable time, practices for applying ML in business have not become standardized or universally adopted. 'The AI Playbook' is built on the foundation of bizML, which is the culmination of the author's extensive practical experience aimed at complementing and expanding CRISP-DM.


The philosophy of bizML is that in business, one should focus on the final decision-making and work backwards from there. It emphasizes the need to first define and start with how the business aspect should be changed through the technology of ML. I completely agree that setting the goals of ML deployment through reverse planning can keep the project on the right track. Analysts who have experienced both failures and successes in ML intuitively or instinctively work in this manner, even if they do not directly learn and apply such specific frameworks or methodologies. This is because it is a natural flow for successful ML project execution.


The AI Playbook' meticulously explains how the ML process should be implemented in business, using concrete real-life examples such as UPS's package prediction. Moreover, the book provides crucial details such as when predictions should be obtained for them to hold business value and what the real business implication of what is commonly referred to as 'accuracy' is.


Recently, the spotlight seems to be solely on generative AI like ChatGPT. However, generative AI is also a part of machine learning and shares many potential pitfalls outlined in this book. Therefore, the discussions in this book are essential for those only interested in generative AI.



(Image DALL-E3 in ChatGPT. prompt :: A two-shot image contrasting two scenes: 1) An AI robot lost and wandering in a desert, looking confused and aimless amidst vast sand dunes under a scorching sun, and 2) the same AI robot working productively in an office, efficiently handling multiple tasks with focus and precision, surrounded by computers, files, and office equipment. The image should be in a vibrant, maximalistic pop art style, emphasizing bright, bold colors and dynamic compositions. The shape of the image should be wide.)


This book will be immensely helpful for data scientists and machine learning engineers in developing and deploying truly useful machine learning models more effectively in practice. It will also serve as a significant guide for business leaders, managers, and operational staff who aspire to successfully implement machine learning in business, helping them understand the value of machine learning and the necessary actions to take. Furthermore, for end-users of machine learning models, this book will aid in understanding how the models are developed and how their outputs should be interpreted and utilized. Even if the book feels somewhat challenging, I believe that perusing just a third or half of its contents will be undoubtedly beneficial.





2023년 12월 31일 일요일

Enterprise Generative AI Innovation 2024: Opportunities, Challenges, and Strategies (Computerworld Korea)

 


Enterprise Generative AI Innovation 2024: Opportunities, Challenges, and Strategies


- Originally published in Computerworld Korea. in Korean. 2023-12-31 https://www.comworld.co.kr/news/articleView.html?idxno=51022)




Yong Xune Xon, CEO of Revision Consulting, Ph.D

(https://www.linkedin.com/in/yong-xune-xon-00796318/)




Technological Advancements in 2023

In 2023, generative AI, including ChatGPT, has shifted the potential scope and level of AI application and utilization for businesses to an entirely different dimension. Developments, which were previously conceived and anticipated for the future, became a reality, though they were not believed to be immediate. According to market research data compiled by ChatGPT, the adoption rate by companies had already exceeded half by the fourth quarter of 2023. This aligns with the surge in interest seen in the "Enterprise AI" Google search index. Compared to the end of 2022, the spread is nearly threefold, making 2023 a memorable year of innovation.


<Figure 1> Estimated Global Adoption Rates of Generative AI and Google Search Index Trends (Enterprise AI)




In 2023, the field of generative AI (Artificial Intelligence) technology witnessed significant advancements across various domains, including improved language models, innovations in image and video generation, advancements in voice generation and conversion, automation and optimization, and applications in creative domains. Language models enabled natural conversations and complex question answering, while image and video generation models supported creative and realistic media creation. Voice generation technology provided natural vocal styles and intonations, finding applications in audiobooks and entertainment. In business, it began significantly aiding in task automation and data analysis. These advancements are expected to continue driving innovation and social impact, playing a crucial role in various industries and job functions.

A key representative of 2023's generative AI innovation was undoubtedly ChatGPT. Its release marked just the beginning, with ongoing innovative updates and the addition of various new features throughout the year.

Improvements in user experience: Users received helpful prompt examples to start conversations and could utilize recommended response features. Convenience features like persistent login and keyboard shortcuts were also introduced.

Functionality expansion: ChatGPT, based on the GPT-4 model, added memory capabilities for previous model selections for paid users. Multi-file upload and a code interpreter for advanced data analysis enabled efficient data analysis and programming tasks. Custom instructions and increased message limits offered personalized user experiences.

Enhanced platform accessibility: ChatGPT mobile apps for Android and iOS were introduced, with improved search functionality on mobile.

Expanded multi-modal capabilities: Image recognition and processing were integrated into ChatGPT, enabling visual data analysis. Integration with DALL-E allowed for text-based visual content creation.

Enterprise version release: The launch of an enterprise version with enhanced data processing and security features opened doors for ChatGPT's formal business applications.

The rapid, innovative, and diverse functional evolution of ChatGPT significantly influenced the trajectory of other generative AI services and providers, expanding the utilization scope of generative AI. However, it still requires ongoing improvements and upgrades, as issues like inaccurate data analysis, failure to follow user prompts accurately, or prolonged operation interruptions persist.


Industry-specific Applications and Business Value Creation

In 2023, the potential presented by generative AI technology was so vast that it led to numerous efforts to explore and experiment with its applications across various industries. A look at industry-specific applications reveals that in finance and banking, Morgan Stanley is utilizing ChatGPT to mine deep insights from large volumes of unstructured data for business analysis. In the insurance sector, Zurich Insurance is using ChatGPT to simplify insurance claim data extraction and enhance efficiency.

In healthcare, Bionic Health, an AI health clinic startup, is leveraging ChatGPT to design preventive care programs and assist patients in understanding complex diagnoses and receiving personalized insights. Massive Bio is integrating GPT-based analytical tools into clinical trial workflows to automate patient data analysis and streamline the trial participation process. In human resource management, Beamery has developed TalentGPT, an advanced AI language model platform, to innovate the recruitment process. It customizes job descriptions, personalizes communication with candidates, and provides career recommendations.

In retail, Walmart offers a helper chatbot for all employees, incorporating features to naturally recommend products during customer chats. Klarna has developed a plugin based on the GPT engine to provide a personalized shopping experience, enhancing product discovery and recommendation through product comparison and link sharing. In the media industry, Koo is integrating GPT models into its social media platform to aid in content creation, helping users create posts more easily and quickly, thus expanding its active user base. In the software industry, there are active efforts to integrate ChatGPT into existing products. Salesforce, a CRM platform provider, has started integrating ChatGPT into various products including its CRM platform, Slack, and Tableau.

In South Korea, several companies have announced cases of applying generative AI. Yanolja, Baedal Minjok, and others have disclosed cases of applying ChatGPT in their operations. POSCO announced its integration of ChatGPT into internal operations to accelerate digital transformation. The company introduced the P-GPT (Private GPT) platform and integrated it with Microsoft’s collaborative platform 'Teams', creating an environment where employees can freely and creatively utilize generative AI services. POSCO DX is combining ChatGPT with its RPA solution to enhance development convenience and functionality.

Thus, in 2023, both domestically and internationally, there were considerable efforts to review and try integrating generative AI into business operations across various industries.

From a corporate perspective, the value of new technologies ultimately depends on their impact on business performance. Therefore, it's necessary to examine the influence of generative AI in terms of business function optimization, cost reduction, and revenue increase. Based on major global companies that have started actively utilizing AI in various operations, it's estimated that in terms of business process optimization, improvements in processes like production, supply chain management, and HR operations have led to an operational efficiency improvement of approximately 10-20%. Cost reductions achieved through improved process efficiency and reduced repetitive tasks are estimated to have reduced operational costs by about 5-15%. Additionally, in terms of revenue increase, it's speculated that these companies have seen a 10-25% impact through creating new business opportunities and enhancing existing products or services. However, these figures should be taken as rough estimates, as there are not many reliable, concrete research data available yet. These estimates are based on research conducted using ChatGPT's search capabilities and are therefore limited in accuracy and reliability.

<Figure 2> Estimated Business Performance Impact by Category of Generative AI Adoption




Of course, the extent of these impacts will vary by industry and the level of AI adoption and integration into business strategies. The actual results are still not as significant as the potential of generative AI, considering it's still in the early stages of adoption. The relatively lower impact on cost savings might be due to companies not prioritizing cost reduction in their AI adoption strategies. In the case of Korean companies, there was a more conservative atmosphere of observation and review in 2023, rather than bold adoption. Concerns about potential risks and trial and error seem to be the main reasons for this cautious approach.


Generative AI Adoption in Corporate Sectors for 2024


In 2024, the spread of generative AI utilization among corporations is expected to be relatively slower compared to other areas or small-scale businesses. This is attributed to several issues including data privacy and security concerns, lack of interpretability and transparency, scalability constraints, limited application fields, cost and ROI considerations, and regulatory and ethical considerations. Corporations handle sensitive confidential data, and "black box" generative AI models pose risks in terms of accountability and compliance due to their interpretability challenges. Moreover, unlike small-scale applications, scaling up for large-scale corporate use introduces technical and system integration difficulties. Thus, adoption in complex operational areas like manufacturing or supply chain management could be slower compared to simpler applications like marketing content creation. The necessary large investments and uncertainty over long-term benefits of AI projects lead to a general cautiousness in investments.

Meanwhile, Amazon Web Services (AWS) offers alternatives for complex and critical tasks in manufacturing, where engineers analyze large datasets to improve safety, generate simulation datasets, and accelerate product launches. The Amazon Bedrock service provides access to various foundation models through APIs, offering customization and scalability for specific tasks in manufacturing. Such approaches enable secure data utilization and effective large-scale applications. Ultimately, the level of technological advancement and solutions to data privacy, interpretability, and scalability issues provided by major suppliers like AWS will determine the extent and speed of generative AI adoption in core business functions.

Nationally, industrially, and functionally, the pattern of generative AI adoption and spread is expected to vary significantly. In the U.S., a global leader in AI technology, active adoption of generative AI, including ChatGPT, is anticipated across various industries, especially in financial services, healthcare, retail, and manufacturing. The increase in AI adoption rate in the U.S. is attributed to its technological infrastructure, abundance of venture capital, high demand for innovation, and a strong R&D environment. The U.S. education system and corporate culture encourage quick adaptation and integration of new technologies, further boosting AI adoption rates.

South Korea, already highly digitalized, with strong government support policies for AI and strategic AI ecosystem development, is also expected to see a steady increase in AI adoption. Manufacturing, financial services, and IT are expected to be active in AI adoption, with particular prominence in advanced technology sectors like semiconductors. South Korea's stringent stance on ethical AI use and data protection will also influence the rate of AI adoption spread.

Both countries are expected to see an increase in generative AI adoption rates, but while the U.S. is likely to witness widespread adoption and rapid changes, South Korea is anticipated to experience a more balanced expansion. Globally, technologically centric countries are expected to adopt and spread generative AI rapidly, while countries that are passive or lagging in technology adoption will likely experience a slower response, widening the gap in the utilization of generative AI technologies.

Industrially, information technology and platform sectors are expected to lead, with distribution, finance, and healthcare following closely. In agriculture, differences in interest between large and small farms, regional variations, and technical, economic, and social barriers will influence adoption. In construction, the complexity of the work environment, safety regulations, and initial investment costs will make AI adoption challenging. In manufacturing, compatibility with existing processes, high costs, and a lack of technology may pose constraints. Common reasons for relatively lower adoption and spread in agriculture, construction, and manufacturing include high initial investment costs, complexity in technology integration, issues in interaction with skilled personnel, and stringent regulations and safety standards. The figure compares approximate estimates of industry-specific adoption rates for U.S. companies between 2023 and 2024.


<Figure 3> Estimated U.S. Corporate Sector ChatGPT Adoption Rates and Growth Forecast

Source: ChatGPT Research



Looking ahead, generative AI adoption rates categorized by business functions are expected to see significant increases, particularly in Information Technology (IT) and customer service sectors. In the IT sector, the use of ChatGPT is anticipated to expand for coding, problem-solving, and IT support. As of 2023, over 60% of developers are reportedly using generative AI for coding tasks. In customer service, automated chatbot responses and customer consultation support are expected to be prevalent. In marketing and sales, the utilization of AI is projected to increase for content creation, digital marketing, and customer analytics. In the Human Resources (HR) domain, the use of AI for resume screening, job posting writing, and employee inquiries is also expected to grow. In the areas of management and strategic planning, AI is forecasted to be progressively used for data analysis, market analysis, and decision support.


<Figure 4> Estimated Changes in Adoption Rates by Business Function


Source: ChatGPT Research


Management and strategic planning involve high-level analysis, creative thinking, and complex decision-making processes, requiring deep industry knowledge, understanding of corporate culture and values, and consideration of long-term visions and goals. Therefore, there are limitations in the current level of AI technology application in these areas. Moreover, strategic decisions often involve sensitive data and confidential information, necessitating caution.


Key Changes Expected from Major Suppliers


The utilization of generative AI in corporations is largely influenced by the supply capabilities of the providers. Major suppliers include OpenAI, the developer of ChatGPT, and Microsoft, among others.

OpenAI: OpenAI is not only providing cutting-edge functionalities but also rapidly integrating and upgrading features. It is expected to continuously impact businesses in areas like customer service automation, content creation, and data analysis.

Microsoft and Amazon Web Services (AWS): As major cloud-based providers, they are anticipated to become primary alternatives for corporate AI use by integrating existing machine learning tools with generative AI.

Facebook AI Research: In the social media platform domain, it could offer new tools for marketing, consumer analysis, and enhancing customer engagement. It is expected to provide strengthened services based on generative AI for data-driven insights, user behavior analysis, and marketing.

Other notable entities include NVIDIA in hardware, IBM’s Watson platform, Google and Google DeepMind, and Adobe. All are likely to intensify efforts to enhance generative AI service functionalities and expand their customer base in 2024.

Particularly, there is significant interest in the upcoming upgrades of ChatGPT. Rumors and speculations about an upgrade to GPT-5 or GPT-4.5 are increasing, although OpenAI has not confirmed these rumors. This suggests that regardless of GPT-4.5’s release, some form of upgrade or improvement is anticipated.

Projected improvements for ChatGPT in 2024 include enhanced contextual understanding capabilities, facilitating more natural conversations and consistent responses. It is expected to be upgraded to handle more complex queries and provide more accurate and comprehensive answers.

Furthermore, the integration of multi-modal functionalities is likely to be enhanced, understanding and generating a wider range of input types, including additional media types like 3D images, videos, and audio. This would significantly improve productivity and efficiency across various industries, particularly manufacturing, engineering, and healthcare, enabling interactive and intuitive interactions with schematics, blueprints, or genomics.

For instance, in manufacturing, multi-modal AI could make interactions with complex machinery and processes more efficient. Workers could control machinery via voice commands while receiving visual feedback, speeding up problem-solving and adapting quickly to new production lines. In healthcare, especially in genomics and medical imaging, substantial advancements are anticipated. Physicians and researchers could interact with complex genomic data using natural language processing to easily identify patterns and anomalies, and visual data combined with AI-based analysis could enable faster and more accurate diagnoses.

Additionally, the introduction of continuous learning frameworks will better reflect user needs and respond more quickly to current events, along with ongoing collaboration with other AI models and continuous improvements in user interfaces. The exact trajectory of ChatGPT and AI technology will vary based on technological advancements, market demands, and the regulatory environment, but ChatGPT is expected to remain a key focus in the tech world for some time. As 2024 approaches, ChatGPT and similar AI technologies are anticipated to become mainstream and more sophisticated in businesses, offering value across various domains, from operational efficiency to strategic market insights.


Strategic Considerations for Corporate Generative AI in 2024

The adoption and utilization of generative AI in corporations are shadowed by concerns and uncertainties about stability, reliability, ease of integration, and management. This encompasses issues such as model drift (where generative AI models become less accurate over time due to changes from the initially trained data patterns or environments) and data contamination (inclusion of incorrect, inaccurate, or biased data in AI systems' training datasets, degrading model performance). Functional risks like resource waste and confidentiality breaches, operational risks, and legal risks including copyright violations and bias-induced discrimination are also significant. Despite ongoing advancements, it is unlikely that all these challenges will be resolved by the end of 2024, necessitating corporations to carefully consider the following strategic decision points:

Elastic Adoption Pace: Considering the imperfections and uncertainties of generative AI technologies, a careful calibration of adoption speed is crucial. Expanding small-scale pilot projects while intensively evaluating the reliability and limitations of AI technology is important. Understanding the division of roles with AI in complex decision scenarios is vital. Continuous evaluation to adjust strategies and being prepared to modify or halt AI integration processes if necessary are essential. Preparing contingency plans and clearly understanding the evolving potential and limitations of technology is critical.

Scalability and Integration: In corporate settings, generative AI must be scalable enough to handle large volumes of data and transactions and seamlessly integrate with existing corporate infrastructures, including databases, CRM systems, and other operational tools. Deciding whether to opt for more powerful generative AI functionalities or easier-to-integrate alternatives is essential.

Competitive Differentiation: As generative AI becomes a ubiquitous tool, differentiation becomes critical. Especially for large corporations, customizing AI solutions to fit their industry and operational characteristics, developing proprietary algorithms, or integrating unique data sources with AI to achieve results unavailable to competitors is key. However, such differentiation requires substantial investment and thus needs careful moderation. For smaller, follower companies, prioritizing basic functionalities at a lower cost might be a priority.

Risk Management: Regular audits of AI systems to review ethics compliance, data privacy, and security are necessary. Establishing clear guidelines and policies for AI use within the organization, educating employees on best practices, and responding to regulatory changes are essential.

Monitoring and Evaluation: Continuously monitoring and evaluating AI implementations is vital. Establishing metrics to measure AI's impact on business performance and regularly reviewing them enables evidence-based, rational decisions on expanding or adjusting AI applications.

Investment in Talent and Training: Effective utilization of AI requires having the right talent, thus investing in continuous education and development of existing employees and maintaining a balance with the recruitment of new talent with specialized AI skills is important.

Partnerships with Vendors and Data Providers: Collaborating with external vendors and data providers offering platforms or models is crucial for enhancing the quality of generative AI outputs or ensuring stable operations. However, minimizing dependency risks is also important.

Collaboration with External Entities: Supplementing internal capabilities through collaboration with AI experts, research institutions, or other companies is also necessary. Joint investments for new AI application development, partnerships with AI startups, or participating in industry consortiums for standard development are options to consider.

In 2024, corporations will face rapid technological advancements, changing market environments, various risks, and uncertainties in adopting and utilizing generative AI. The advancement of generative AI technology is both an opportunity and a competitive threat. Active responses and maximizing potential while managing costs and risks require adequate preparation and planning.