Generative AI - Implications and Opportunities for Business

RMIT University - FutureLearn - Course Notes

1. The Basic Anatomy of Generative AI

A Concise History of AI

The concept of a machine mimicking human behavior traces back to early philosophers like René Descartes and Thomas Hobbes, who suggested the brain functioned mechanically. Charles Babbage’s Analytical Engine in the 1830s spurred the idea that machines could potentially possess human-like mental capabilities. Ada Lovelace, his collaborator, hypothesized such a machine could compose intricate, scientific music. Mid-twentieth century brought general-purpose computing and the proposition of a machine capable of creativity and learning. Alan Turing, a pioneer of AI, proposed the Turing Test, a measure of a machine’s artificial intelligence by its ability to imitate human conversation indistinguishably. AI research since has seen two primary ‘traditions’: the classical and the modern approach.

Classical Artificial Intelligence

Classical AI or GOFAI (“Good Old-Fashioned Artificial Intelligence”) relies on logically programmed rules for machine responses to specific inputs. To illustrate, a classical AI translating English to French would need to have rules and exceptions of both languages hard-coded into it, working like a logical tree. Classical AI was prevalent from the 1950s through the 1990s and continues to find applications in fields like medical diagnosis.

Modern Artificial Intelligence

The modern AI approach utilizes deep learning and neural networks to mimic the human brain’s learning structures. In contrast to the classical approach, modern AI learns from vast datasets to recognize patterns, leading to applications like generative AI, image recognition systems, and more. Using the language translation example, modern AI learns the languages’ rules by studying vast amounts of French and English text, making it data-intensive but relieving the developer from knowing anything about translation. The machine learns, discovers grammar rules, and makes decisions, paving the way for the current explosion in generative AI.

Exploring Generative AI

Generative AI represents a significant evolution in the history of artificial intelligence. Here’s a breakdown:

• Neural Nets

  • These structures, invented in the 1940s, aim to mimic organic brain structures. They consist of interconnected nodes or ’neurons'.
  • Neural nets comprise input, hidden, and output layers. They’re ’trained’ on data that shapes the weights in the hidden layer.
  • Early neural nets from the 1950s had a few nodes and one hidden layer. Modern versions, like ChatGPT, possess billions or even trillions of nodes.

• The Deep Learning Revolution

  • Often linked to Geoff Hinton and his team, this revolution is marked by multiple layers in the Artificial Neural Network (ANN).
  • Developed between 2006-12, deep learning operates on pre-trained ANNs, a significant aspect of generative AI.

• Transformers

  • The ‘T’ in GPT stands for Transformers, a new class of deep learning model developed by Google researchers in 2017.
  • Transformers excel in natural language processing tasks like language translation, text generation, and summarization.
  • They process input data to understand word relationships within a sentence or text, generating coherent, natural-sounding text.

• Pre-training

  • The ‘P’ in GPT stands for Pre-trained. Generative deep learning neural net Transformers are pre-trained on vast datasets.
  • Generative AI typically leverages large, open-access datasets, often sourced from common crawl. https://commoncrawl.org
  • Future business models for Generative AI might depend on larger or custom datasets.

It’s important to note that a model reflects the quality of its training data, including biases and property rights, explaining why most large language models train on open access data.

Generative AI Market

Key points about the commercial landscape of generative AI:

• The AI industry is relatively new yet highly competitive. Companies and organizations are creating AI solutions for various industries and applications.

• Openness or closedness of the technology and its associated IP is a distinguishing factor among different AI products and services.

• Open AI platforms (like TensorFlow, PyTorch, Apache Spark) are freely available, often carrying open source licenses. Users can modify and redistribute the software.

  • Examples in generative AI include OpenAI’s ChatGPT, DALL-E models, and the Deep Dream Generator.

• Closed AI platforms (like IBM Watson, Amazon Web Services, Microsoft Azure) are proprietary and controlled by the owning company/organization, often with restrictive licensing terms limiting access to the underlying code and algorithms.

  • Examples in generative AI include Adobe’s Creative Cloud and Nvidia’s StyleGAN/GitHub - StyleGAN.
    • Generative adversarial networks (GANs) are AI algorithms that utilize two competing neural networks: a generator, which produces false data, and a discriminator, tasked with differentiating between fake and real data. Their constant learning from each other leads the generator to create data so realistic that the discriminator fails to label it as fake. This results in GANs’ ability to generate high-quality, realistic data.

• Choosing between open and closed platforms depends on factors like specific application needs, the development team’s expertise and resources, and availability of compatible tools and services.

• Open platforms are flexible and customizable, promoting collaboration and knowledge sharing. However, they might lack security and support as they depend on community maintenance.

• Closed platforms offer better security and vendor support, and potentially superior integration with other proprietary technologies. However, they can be less flexible and customizable and may pose challenges in maintaining IP control over generative AI models.

Generative AI Categories

Generative AI models and techniques have evolved significantly over time, each offering unique capabilities and serving different tasks:

• Rule-based systems (1950s-1960s): Early form of AI, limited by pre-defined rules.
• Markov models (1960s): Used for modeling time-based systems, relying on the probability of different events.
• Variational Autoencoders (VAEs, 2013): Neural network type, representing high-dimensional data in a low-dimensional latent space.
• Generative Adversarial Networks (GANs, 2014): Generate realistic synthetic data via a generator and a discriminator model.
• Auto-regressive models (2015): Generate data sequences by modeling conditional probability of each sequence element based on preceding elements.
• Boltzmann machines (1985): Model complex systems with multiple interacting variables.
• Restricted Boltzmann machines (RBMs, 2006): Boltzmann machines variant, easier to train, used for unsupervised learning.
• Transformers (2017): Neural network architecture, processing data sequences more efficiently than traditional recurrent neural networks.
• Deep belief networks (DBNs, 2006): Neural network type, representing high-dimensional data hierarchically.
• Deep Boltzmann machines (DBMs, 2009): Variant of Boltzmann machines, representing high-dimensional data hierarchically.

Generative AI uses these models to create new data similar to a given dataset, aiming to produce novel outputs for applications like generating new images, music, or text, in diverse industries from entertainment to finance.

Different categories of Generative AI:

• Language Generation: Techniques that generate natural language text, such as sentences, paragraphs, and even entire articles or stories. Examples include GPT-3 by OpenAI, BERT by Google, and T5 by Google

• Image Generation: Techniques that generate images, such as paintings, photographs, or even videos. StyleGAN by NVIDIA, GANPaint by MIT, and BigGAN by Google

• Music Generation: Techniques that generate new musical pieces based on existing music. MuseNet by OpenAI, WaveNet by Google, and BachBot by Cambridge Consultants

• Video Generation: Techniques that generate videos, such as animated cartoons or even deepfake videos. First Order Motion Model by Aliaksandr Siarohin, GANimation by Tencent AI Lab, and Vid2Vid by NVIDIA

• Chatbot Generation: Techniques that generate chatbots, which are computer programs that can simulate conversation with human users. Examples include GPT-3 by OpenAI, Transformer by Google, and DialoGPT by Microsoft

• Style Transfer: Techniques that generate new content based on a combination of existing content and a desired style. Examples include Neural Style Transfer by Gatys et al., CycleGAN by Zhu et al., and MUNIT by Huang et al.

• Speech Generation: Techniques that generate synthetic speech, such as text-to-speech (TTS) systems. Examples include Tacotron 2 by Google, Deep Voice by Baidu, and WaveGlow by NVIDIA

• Recommender Systems: Techniques that generate personalised recommendations for products, services, or content based on user data. Collaborative Filtering by Amazon, Content-Based Filtering by Netflix, and Matrix Factorization by YouTube

• Text Classification: Techniques that assign labels or categories to text data. Sentiment Analysis by Stanford NLP Group, Topic Modeling by Gensim, and TextCategorization by Hugging Face

• Object Detection: Techniques that detect and classify objects within images or videos. Examples include YOLO by Joseph Redmon, Faster R-CNN by Shaoqing Ren et al., and RetinaNet by Tsung-Yi Lin et al.

Assessing Categories

When exploring generative AI tools, you’ll find both paid and free options. While many are user-friendly and require no technical expertise, you should be mindful of potential ethical and societal implications, including bias and privacy issues.

Evaluation of these tools could be based on several criteria:

• Quality of the generated output: Is the output realistic, high quality, coherent and detailed? Do text responses achieve human-like fluency?
• Ability to generate diverse outputs: Can it generate diverse responses to different inputs? Do outputs have different styles and characteristics?
• Capability of real-time output generation: Can it generate responses in real-time?
• Level of human intervention required: To what extent does it require human intervention? Does the output need to be adjusted? Can it be adjusted?
• Complexity of the task: Does it exhibit a deep or nuanced understanding of the subject matter?

Consider these points as a checklist when assessing different generative AI tools.

Examples

Consider using one or more of the following generative AI tools, then use the previous criteria to assess the output. Here are a few options:

• Talk to Transformer: A free tool that uses GPT-2 to generate text from a user-provided prompt. InferKit
• Artbreeder A free tool for generating images through a combination of generic algorithms and deep learning, allowing you to merge and mutate existing images to create unique ones.
• AIVA: A free tool to generate music with an AI-powered composer, where you can choose mood, genre, and tempo to create a custom composition. AIVA
• Deep Dream Generator: A free tool that generates psychedelic versions of images through a neural network algorithm. Deep Dream Generator
• This Person Does Not Exist: A free tool that creates realistic images of non-existent people using a generative adversarial network (GAN) algorithm. Random Face Generator

AI Literacy:

a set of competencies that enables individuals to critically evaluate AI technologies; communicate and collaborate effectively with AI; and use AI as a tool online, at home, and in the workplace.

Prompt Engineering

• A prompt is a specific command or input given to an AI model to direct its output or actions. Prompts can come in various forms like word sequences, statements, or code blocks, with text being the primary method of communication with generative AI models.
• Prompt engineering is about designing and adjusting prompts to elicit the desired output from the model. The instructions or examples in the prompt are vital.
• Depending on the expected outcome, prompts can range from simple to complex. If the goal is to generate poetry, for example, the prompt could include a brief sentence or phrase that establishes the tone, style, type (such as sonnet or haiku), rhyme scheme, or mood.
• The more refined the prompt, the more accurate and consistent the resulting output will be.

Prompt engineering has become a crucial skill with the advent and impact of generative AI models for various reasons:

• It aids in achieving accurate and specific outputs from generative AI models.
• It enhances communication between humans and AI models.
• It assists humans in understanding how AI models organize and categorize information.

Factors that define a good prompt:

• It effectively communicates the expected output to the model, leading to the creation of content that fits the set criteria.
• A poor prompt can result in confusion or imprecise output from the model. For instance, “Write something interesting.” is an example of a vague and unclear prompt.
• A good prompt is specific and detailed. For example, a prompt for language translation like “Translate this sentence into French” can be improved to “Translate this sentence into formal French suitable for a business meeting," which ensures the translation is suitable for a specific context.

Testing and refining prompts is a useful method to understand the data the model uses and how it organizes that information.

Crafting Effective Prompts

• Utilize the latest AI generative tools: It’s beneficial for prompt engineers to use the latest versions of generative AI tools. For example, GPT-4 is more advanced than ChatGPT-3 and offers multimodal capabilities.

• Understand and articulate the problem: An important part of prompt design is understanding the broader problem or question. Break down problems into smaller, manageable prompts that can produce the needed output. Ensure the prompts are clear and simple.

• Explain expectations: Include information about the desired format of output in your prompt. Be specific about what data the model should use to generate its output. For instance, use clear, instruction-based prompts like “a digital representation of a robot” or “use oil painting to create an image of a robot”.

• Simplify prompts: Break down long and complex instruction-based prompts into smaller sections, rather than using one long statement. You can use symbols like “#” or quotation marks (“”) to separate different sections of a prompt.

• Use continual conversation-based prompts: The design of effective prompts might require multiple attempts. Consider treating prompt generation as a conversation with the generative AI tool. If using a text-based prompt, ask probing questions, or refer to previous prompt output to maintain the conversation until the desired output is achieved.

---

2. The Impact of Generative AI on Business

• Understanding AI’s Value Creation: Economic models help identify the origins and flow of value created by AI, which could be the basis for start-ups, improved business operations, or new tools augmenting existing capabilities.
• Strategic Decision-Making Challenges: Implementing new technology in a business context, even with straightforward applications, requires managing operational issues like training, workflow changes, and capability adjustments.
• Generative AI as a Business Tool: Generative AI enhances existing operations and has the potential for substituting certain roles with automation, possibly leading to business process re-engineering opportunities.
• Economic Model for Cost Reduction: A forthcoming model helps consider the impact of generative AI as a significant cost reduction (or capability enhancement) mechanism for various business functions.
• Discovering New Uses for Technology: A second economic model will explore the process of discovering valuable applications for new, general-purpose technology, under conditions of uncertainty.
• In-House vs. Copycat Strategies: Businesses must decide between investing in finding new uses for technology in-house, or observing and copying successful implementations by others.
• Collaborative Innovation Opportunities: Given the accessibility of AI technology via software interfaces, businesses can collaborate on innovation with suppliers and customers using various platforms or toolkits.

Falling Costs

• Economic Insight: The core idea of economic thinking is if something becomes less costly, its usage increases.

• Implication for Generative AI: Generative AI, an automation technology, provides enhanced speed, efficiency, and cost-effectiveness, leading to increased usage and possible economic disruption.

• Effect on Labor Markets: Cost reduction due to AI could impact labor markets. More use of AI could affect employment trends and job roles.

• Strategic Implications: Generative AI can be strategically used to automate expensive or time-consuming tasks. Assessing where these cost reductions occur could identify new opportunities for creating value.

• Value Creation and New Opportunities: The economic approach serves as a template for strategic thinking about creating value and finding new business opportunities with AI.

Falling Costs Due to Generative AI

Generative AI can increase productivity and reduce costs, leading to economic value creation.

• Production Function Model: This economic model demonstrates how new technology, like AI, enhances productivity by producing more outputs with the same resources or maintaining output level while reducing inputs .
• Impact on Market: Adopting new technology is depicted in Diagram 2 by an outward shift of the supply curve. The new market equilibrium may result in a lower price, higher sales, or a mix of both, depending on the demand curve’s slope, which indicates consumer response to price changes.
• Simplified Model Assumptions: The model presumes that AI is a universally accessible generic technology, all firms are identical with no special knowledge, all consumers are the same, and there’s full transparency about AI’s usage. While these assumptions are not entirely accurate, they serve as a useful starting point to explore different scenarios and impacts.

Agrawal, Gans and Goldfarb on the general economics of machine learning

• AI and Machine Learning Model: Ajay Agrawal, Josh Gans, and Avi Goldfarb introduce a model explaining the impact of AI and Machine Learning in their books “Prediction Machines: The simple economics of artificial intelligence” and “Power and Prediction: The disruptive economics of artificial intelligence”.
• Better Predictions: ML allows businesses to make precise forecasts about consumer demand, price change effects, or consequences of events, thereby leading to better decisions, competitive advantages, and improved customer satisfaction. ML can be a powerful tool for enhancing human decision-making, but only when it is used in conjunction with human expertise and judgment.
• Economic Analysis: In economic terms, new machine learning technology is seen as an expansion in the production function, leading to higher output with the same inputs and potentially lower prices, which boosts demand.
• Generative AI: Similar to the insights of Agrawal, Gans, and Goldfarb, Generative AI uses pre-trained models to predict the next expected action
• based on previous data. This method is akin to modeling what an expert human would do next.
• Cost of Prediction: As the cost of prediction decreases rapidly, we will likely use more prediction technologies and rely less on human thought processes.
• Impact on Firms: Each firm will have the same opportunities and is expected to react similarly to this new technology that lowers specific costs. Firms that do not use this technology will face a competitive disadvantage due to higher costs associated with relying on human input instead of fast, cost-effective machines.
• Profitable Adoption of Generative AI: The most profitable adoption of generative AI is predicted to be in areas where the cost of human decision-making and expertise is highest.

Entrepreneurial Discovery

• Potential of Generative AI: The exciting aspect of generative AI lies not only in improving existing processes but also in discovering unprecedented use cases. This can reshape the economy and create new businesses and jobs.
• New Technologies and Value Creation: New technologies, particularly disruptive ones, require experimentation and learning to harness their potential for economic value. This involves managing uncertainty and learning from other industries, businesses, or individuals.
• Entrepreneurial and Experimental Mindset: An entrepreneurial and experimental mindset is vital for uncovering the latent value in these technologies, leading to significant advancements for individuals and societies.
• Role of Startups and Businesses: Startups and businesses can play a crucial role in this process by creating experimental labs within their organizations to explore the potential of these technologies.
• Navigating the Process: While this exploration can be a complex process, strategic thinking and institutional mechanisms can aid in identifying the most valuable, profit-making use cases for this powerful new technology.

Understanding Value Creation through Entrepreneurial Discovery:

The second approach to understanding the value of generative AI involves the economics of entrepreneurial discovery, a model that rejects the assumption that the utility of a new technology is immediately apparent.

• The Information Assumption: Traditional models assume the new technology’s applications are obvious, which is rarely the case, especially with innovative, multifaceted technologies.
• Example of New Technology Adoption: Lasers, invented in 1960, weren’t widely recognized as a general-purpose technology until a decade later when they were applied to optical barcode scanning in supermarkets.
• Challenge of Value Creation: The difficulty lies not only in identifying technical uses but in determining how to leverage the technology to create new value. Many high-value applications of generative AI likely remain undiscovered.
• Discovery of New Use Cases: The discovery of new uses for a technology is a distributed process involving experimentation and learning. Entrepreneurs and managers need to think creatively about how to enable employees, contractors, other firms, customers, and users to explore and discover new applications.
• Cross-Industry Learning: Firms could also learn from the application of the technology in other industries to identify similar problems or value creation processes.
• Scenarios for New AI Applications: The next step involves considering different scenarios in which a firm could discover and learn about new and profitable applications of generative AI.

Competition within firms:

• Internal Discovery of New Uses: Managers within a firm may instruct employees to explore potential applications of generative AI. This process may involve training, dedicated exploration time, team organization, and incentive schemes for high-quality proposals.
• Challenges with Internal Discovery: Monitoring such exploratory activities can be difficult due to the high level of information asymmetry and the potential for shirking, creating potential risks.
• Role of External Consultants: To mitigate these challenges, firms may hire external consultants to guide the process.
• Growth in Generative AI Consultancy Services: The advent of generative AI technology is expected to lead to the emergence of specialized consultancy services, either from existing technology and management consulting firms or entirely new ones.

Competition among workers

• Intra-firm Competition for AI Adoption: The discovery of generative AI applications might emerge from competition among existing workers to automate each other’s tasks.

• Assumptions for this Model:

  • Accessible Generative AI Tools: Tools like ChatGT or Midjourney are affordable, efficient, and discrete, accessible without needing permission much like search engines or social media apps.
  • Open Collaborative Production in Firms: Firms operate as teams where various parts of a process need to work together. This collaborative environment makes each worker’s actions visible to their colleagues.
  • Opportunistic Workers: Each worker may perceive their own work as creative and productive, and others’ work as routine. As a result, workers might believe that they could handle their coworkers’ jobs with the right tools.

• Outcome of the Model: Workers may use tools to automate their coworkers’ jobs, leveraging their access to these tasks (which isn’t always available to workers in other firms due to information constraints). This triggers a cycle of technological advancement, adoption, and creative destruction, driven by workers who are mutually suspicious and technologically opportunistic.

• Entrepreneurs or Managers’ Role: Rather than creating new jobs with generative AI, managers should be attentive to the job creation-destruction arising from workers using new tools to expand their power, responsibility, and potentially their pay. The manager’s role then becomes to validate and implement the discovered opportunities for work automation.

Toolkits for distributed innovation

• Distributed Nature of Generative AI: Generative AI, being easily distributable to any computing device, expands the capability for search and discovery beyond the firm.
• Collaboration in Innovation and Discovery: If a firm is part of a value chain with suppliers, contractors, lead users, etc., providing a platform or toolkit for these entities to participate in innovation and discovery could prove more effective.
• Toolkits Theory: This theory focuses on the organization of innovation from the perspective of expert and local knowledge. It has implications for the industrial organization and business model design for new technologies. Generative AI, at its core, is a toolkit. User toolkits for innovation pdf
• Desktop Publishing as an Analogy: Companies like Microsoft and Adobe disrupted traditional ‘publishing’ by moving production closer to the end user through the provision of new desktop publishing tools. This model might be indicative of how generative AI will likely operate as a business.

Using AI in Strategy

• Strategic Approach to AI Integration: Executives should identify business problems that AI can solve, such as automating routine tasks, improving customer service, and enhancing decision-making. They must also be cognizant of AI’s limitations and potential risks, like job displacement or quality issues.

• Development of AI Strategy: After identifying business problems, an AI strategy should be developed. This includes a roadmap for implementation, plans for talent and resource acquisition, and success metrics. Executives must understand that AI implementation is a long-term process requiring significant investments.

• Data-driven Culture: AI is reliant on data, making it crucial to foster a culture that values data-driven decision-making. This might require investments in data infrastructure, employee data analysis training, and data governance processes.

• Ethical Considerations and Transparency: As AI can significantly impact society, ethical considerations must be prioritized in AI strategy. This includes transparency, bias minimization, and privacy protection. Unintelligible AI systems may erode trust and face legal issues.

• Monitor and Adapt AI Strategy: Once AI is implemented, its impact should be monitored, and the strategy should be adapted accordingly. This might require refining algorithms, altering processes, or investing in new technologies. Adaptability will position organizations to fully utilize AI’s potential to drive innovation, efficiency, and new business opportunities.

Risks Associated with AI in Business Strategy

• Risk of Data Bias and Discrimination: AI systems’ decisions are as good as the data they are trained on. Biased or discriminatory data can lead to unfair or discriminatory decisions, causing reputational damage and potential legal liability.
• Lack of Transparency: Many AI systems operate as ‘black boxes’, making it challenging to understand their decision-making process. This opacity can hinder error and bias identification and rectification.
• Cybersecurity Threats: AI systems can be susceptible to cybersecurity risks, including hacking and data breaches. These risks can intensify if the AI system is linked to other parts of an organisation’s network. Implementing robust cybersecurity measures is essential.
• Legal and Regulatory Risks: AI systems making decisions about individuals, like hiring or lending, may face legal and regulatory scrutiny. Companies must ensure these systems comply with relevant laws and regulations and are transparent and explainable.

Areas of Strategy Where AI May Not Be Appropriate

• Setting Strategic Direction: While AI can offer valuable insights and data, setting the organization’s vision and goals is better suited to human senior executives, who possess a deep understanding of the organization’s objectives and values.
• Making Ethical Decisions: AI can identify ethical considerations and potential issues, but final ethical decisions should rest with human decision-makers. This is due to AI’s inability to fully comprehend the nuances of human values and ethics.
• Building Relationships: AI may not be suitable for building strong relationships with customers and stakeholders. This requires empathy, emotional intelligence, and the ability to understand and respond to complex social dynamics, which is better provided by humans.
• Navigating Complex Social and Political Environments: AI may not be proficient at navigating intricate social and political environments. These environments are influenced by numerous factors like cultural norms, historical events, and power dynamics, which may not be fully captured by AI.

Business Model Transformation

• AI’s Influence on Business Models: AI’s rapid advancements have led businesses to modify or construct their business models around these technologies.
• Role of Business Models: Business models act as a blueprint for how a business creates value, depicting the customer base, operational costs, and revenue generation.
• Impact of AI-Enabled Business Models: Incorporation of AI applications in business models transforms business operations.
• Industry-wide Shift: Businesses across diverse sectors are experiencing a significant shift towards the adoption of AI applications due to the unique opportunities they provide.
• Customization of AI Adoption: While AI’s potential benefits are universal, businesses adopt AI differently, based on their own visions and goals. Many prioritize using AI to enhance customer experience, improve operational efficiency, or augment business intelligence.
• Next Discussion Topic: The next part of the discussion will examine two dimensions of AI that assist businesses in revamping their existing business models.

Enhanced customer experience

• Customer Experience Definition: This refers to all the interactions a customer has with a business, including cognitive, affective, sensory, and behavioural aspects across all interaction phases.
• AI’s Role in Enhancing Customer Experience: AI deployment enhances customer experiences in several ways:
  • Marketing for Potential Customers
    • dentifying and predicting trends from unstructured data for improved targeted marketing.
    • Automating consumer segmentation to launch personalized advertising initiatives.
    • Generating personalized content using consumer profile data.
    • Deploying AI agents for faster marketing content generation.
    • Facilitating targeted marketing with intelligent engines.
  • Customer Experience Improvement
    • Predicting customer needs using social media data analysis.
    • Launching virtual product try-ons and demos (e.g., recommendations for clothing, make-up, sunglasses).
    • Offering personalized services and products by analyzing customer search patterns.
    • Providing ubiquitous and accessible customer support (e.g., voice assistants, chatbots, self-service checkouts).
• Real-world AI Application in Enhancing Customer Experience: KLM, an airline company, improved customer experience using an AI-powered chatbot, Blue Bot (BB), which assists customers with flight booking, check-in, and provides personalized travel recommendations.

Improved operational excellence

• Operational Excellence Definition: Operational excellence refers to the activities a business undertakes to produce goods or services, including organizing and planning tasks to achieve the necessary outcomes for business operations.
• AI’s Role in Enhancing Operational Excellence: AI can enhance operational excellence in several ways:
  • Optimization of Layout Planning: AI applications can generate and test layout plans under varying parameters, such as community demographic data, consumer segment, audience size, etc.
  • Staff Allocation Design: AI can help design scenarios for staff allocation across various business units, including production and customer service.
  • Real-Time Business Unit Monitoring: AI enables real-time monitoring of business units, supporting proactive strategy development.
  • Staff Safety Enhancement: AI can improve staff safety through augmented reality-assisted devices, useful in high-risk areas like mining operations.
  • Support Task Automation: AI can automate various support tasks, including HR, accounting, and administrative duties.

AI-Enabled Digital Business Models

AI in Start-ups and New Business Models: AI is not only enhancing existing business models but also fostering the creation of new start-ups based on AI-enabled digital business models.

Subscription based

• Subscription-Based Models with AI:
  • Definition: Subscription-based business models provide products or services to customers for a recurring fee, creating predictable revenue.
  • Role of AI: AI augments these models by enabling personalization of offerings based on customer preference analysis, and designing omnipresent customer support systems like chatbots or voice assistants.
  • Example - Spotify: Music streaming service Spotify leverages AI to analyze user data (listening history, search queries, user-generated playlists) and provide personalized music recommendations. The algorithms continuously learn from user behavior and can predict future music interests, facilitating discovery of new artists and songs.
  • AI and Customer Churn Analysis: AI can help analyze subscription data to identify customer churn rates, allowing companies to detect patterns that signal potential customer loss. Companies can then proactively implement retention strategies, such as targeted promotions or incentives. More about Spotify’s AI use

Platform Based

• Platform-Based Business Models and AI:
  • Definition: Platform-based business models provide a digital platform to enable interaction among two or more groups (e.g., social media interactions, financial transactions, or collaboration).
  • Role of AI: AI applications, like predictive analytics, assist groups in accessing relevant content, businesses, or partners for collaboration.
• Preparing for AI Deployment in Business Models:
  • Understanding AI’s Value Proposition: Businesses should understand the value proposition AI offers them before integrating it into their models. This evaluation can be done through observing AI use cases, which demonstrate the impact of AI transformation in business units (like improved customer satisfaction or increased sales).
  • Assessing Digital Infrastructure: A thorough evaluation of the business’s digital infrastructure is necessary, as insufficient infrastructure can hinder AI system components (like data capturing, cleaning, or secure storage), jeopardizing the successful launch of AI-enabled business model transformations.
  • Upskilling the Workforce: Workforce education and training are crucial in this transformation process, and it should involve all relevant staff (system designers, marketers, sales representatives, production units, etc.).
  • Understanding AI-Deployment-Associated Risks: Businesses should understand the potential risks associated with AI deployment and develop strategies for risk mitigation.

Workplace Transformation

• Definition of Disruptive:

  • Refers to technologies that change business models or practices in a fundamental way, leading to different implications and uses.

• Generative AI’s Disruptive Potential:

  • In the Workplace: Generative AI is seen as a disruptive force, particularly in ideation within the workflow. It helps in bringing fresh perspectives and making certain practices more efficient, like producing forms and creating content.
  • Impact on Workflow Efficiency: Generative AI can change workflow efficiencies and the role of human input in content or output production.

• Boundary Breaking Nature of Generative AI:

  • Job Roles: AI technologies can blur job boundaries as they enable tasks previously allocated to specific roles to be carried out more universally. This leads to role disruption, changing how and with whom work is produced.

• Impacts of Disruptive AI:

  • Organizational Structures: Disruption can lead to streamlining of organizational structures and roles, where some roles might no longer be filled by humans. This presents opportunities for reshaping value addition within an organization.

• AI Deployment and its Implications:

  • AI has dramatically changed the workplace with implications for both businesses and employees.
  • While AI could lead to some job losses, it also offers potential for new human-robot collaborations.
  • Introduction of an AI-augmented workforce, where AI technologies enhance human workers’ capabilities.

The Future of Work:

• Utilization of machines in the workforce is not new.
• As highlighted by David De Cremer and Garry Kasparov in the Harvard Business Review, AI Should Augment Human Intelligence, Not Replace It :

“The history of work — particularly since the Industrial Revolution — is the history of people outsourcing their labor to machines.”

• AI-Augmented Workforce: This concept brings together the strengths of human and AI to accomplish more efficient and effective outcomes. Some advantages include:
  • Automation: AI enables automation of repetitive and mundane tasks in businesses such as manufacturing and service delivery.
    • Improved Efficiency: AI applications have an inherent feature of processing large volumes of unstructured data quickly.
    • Accuracy: AI applications, under supervised learning, can generate accurate outputs, aiding business leaders in decision making.

The future of the workplace sees a logical collaboration of human and AI-based intelligent machines, termed as an AI-augmented workforce.

AI-augmented workforce

• AI-Augmented Workforce Benefits:
  - Enhanced Productivity and Efficiency: AI can take over routine tasks, freeing human agents for more creative tasks that require human judgment. - Error-free Output: The use of AI can minimize the chance of mistakes, particularly in settings like factory work where AI systems provide manufacturing instructions.
• Open Loop System:
  - AI systems provide recommendations to human operators who use their judgment to accept or request additional solutions. - The role of AI in such systems is to aid human operators in making better decisions.
• New Diversity:
  - The workforce of the future will consist of both humans and machines. - Leaders in such environments will need skills in understanding AI systems and integrating the abilities of both human and machine team members.
• Addressing Potential Issues:
  - Despite the potential for misunderstanding and mistrust between human and machine colleagues, properly managed teams can minimize these issues. - Leaders of such teams must understand AI systems and educate others about their workings.
• Upskilling the Workforce:
  - Training and upskilling are essential for the workforce to optimize the use of AI technologies. - Upskilling is a proactive approach to address concerns about job losses due to AI. AI-enabled jobs aim to enhance the workforce’s skills rather than eliminate human jobs. - Balancing the future of work involves improving the human role in high-skilled jobs while providing efficient and effective job execution solutions.

The Role of a Prompt Engineer

3 new and emerging jobs you can get hired for in 2023 | World Economic Forum

• Prompt Engineering as a Career:
  - Recognized by the World Economic Forum as one of three emerging jobs. - Applicable not only to coding or software development, but also to marketing, essay writing, fashion designing, and any job requiring knowledge workers.
• Role of a Prompt Engineer:
  - Design prompts to generate the most relevant response from a Large Language Model (LLM). - Craft appropriate prompts and develop a contextual understanding of prompts and their output. - Assess the accuracy of the output generated, correcting inaccuracies.
• Nature of Prompt Engineering:
  - Combination of science (logic) and art (creativity). - Requires understanding of the purpose of the prompt (logic) and ability to modify the wording for optimal outcomes (creativity). - Also involves artistic judgment in assessing responses for accuracy, bias, and feasibility.
• Skills Required for a Prompt Engineer:
  - Understanding the purpose of the prompt. - Ability to develop clear and concise prompts. - Ability to iterate prompts using both logic and creativity. - Ability to test and evaluate the effectiveness of prompts and their outputs. - Understanding the model’s limitations. - Knowledge of machine learning and natural language processing. - Ability to collaborate within transdisciplinary teams. - Staying updated with the latest developments in the field. - Willingness to learn from other fields and apply that knowledge.

---

3. Generative AI in Industry

Implications for Generative AI Across Industries

• The Principle of Automation of Knowledge:
  - Automation is not just about replacing tasks or jobs, it’s about automating knowledge. - The disruption lies in how knowledge is used by businesses.
• Working with Automated Human Knowledge:
  - The challenge isn’t about a robot replacing jobs, but about interacting and leveraging machines that hold human knowledge. - It’s not just about working with machines but working with machines that have access to all human knowledge.
• Industry-specific Disruptions:
  - Different industries like law, health, coding, and architecture will have different use cases and applications of this disruption.
• The Role of Entrepreneurs and Businesses:
  - Each business and entrepreneur needs to figure out how to develop this new capability.
• Example: Aged Care Sector:
  - Using generative AI with social media content to create memories for aged residents to alleviate loneliness. - This is not a job replacement but a new use case for AI.
• Potential for the Metaverse:
  - Generative AI might be the ingredient needed to make the Metaverse more exciting. - Possibility of interacting with intelligent avatars powered by models like Chat GPT.
• International Influence of Generative AI:
  - The role of AI in different countries’ operations is uncertain and worth observing. - Example: Impact of generative AI on business process outsourcing in India and its effects on Australian firms and local AI capabilities.

Creativity: from originality to imagination

Gaming

• Adoption of AI in Gaming Industry:
  • Gaming studios face the decision to either build or buy AI models for their games.
  • Lower entry barriers due to foundational models such as ChatGPT, Stable Diffusion, and Midjourney, which are open to all.
  • Risks of oversaturation and legal issues regarding copyright with use of foundational models.
• Horizontal vs Vertical Integration:
  • Horizontal integration: Use open-source AI models to create games.
  • Vertical integration: Build custom datasets in-house for unique game designs, likely pursued by AAA studios and indie developers due to high costs.
• Potential Strategies:
  • Acquiring old games to build large training datasets.
  • Purchasing datasets with appropriate rights or creating in-house datasets to avoid legal issues.
• User Generated Content (UGC):
  • Generative AI could reduce costs and increase the speed of content generation.
  • Potential shift towards games heavily reliant on UGC (e.g., Roblox, Minecraft, Grand Theft Auto 5).
  • AI could assist users in generating content to enhance gaming experiences.
• Future of AI in Gaming:
  • Potential for AI-generated extensions of existing games and personalized recommendations and experiences.
  • AI could generate different dialogues, new quests, alter NPC interactions, and recommend games based on individual player’s preference.
  • Future AI in gaming will lead to unique experiences every time a game is played.

The Generative AI Revolution in Games | Andreessen Horowitz Generative AI Will Change Gaming and Entertainment Forever

Marketing

• Impact of Generative AI on Marketing:
  - Bringing in new tools and techniques, improving efficiency, reshaping how marketers work.

• Copywriting, Email Marketing, and Captions:
  - Generative AI makes marketing tasks like copywriting, email marketing, and captioning more efficient. - Enables rapid creation of engaging subject lines, captions, and entire marketing campaigns. - Generates content faster than humans, using machine learning algorithms to understand language nuances, tone, and context.

• Social Media Insights and Campaign Generation:
  - Generative AI uses insights from social media platforms to generate relevant marketing campaigns. - Can suggest improvements by analysing successful campaigns, identifying trends, and patterns for more effective marketing strategies. - Helps in keyword generation for SEO, leading to more targeted campaigns, higher engagement, conversions, and better ROI.

• Visual Content, Website Design, and Logo Creation:
  - Revolutionizes visual content creation, impacting industries like social media marketing, product merchandising, advertising, and product demos. - Enables rapid logo generation and website design, saving time and money. - Analyses website traffic data to suggest improvements and creates personalised content by understanding consumer behavior and preferences.

• Process Improvement and Efficiency:
  - Complements skilled marketers in creating effective campaigns, streamlining workflows, and improving output. - Frees up marketers’ time for strategic initiatives, allowing creation of more targeted campaigns. - Personalises content, driving engagement through the analysis of consumer behavior and preferences. - Requires marketers to understand how to interact with AI to remain competitive and maximize the benefits of these tools.

Generative AI for Creative

• Generative AI in Creative Industries - Generative AI is transforming creative processes and lowering the barriers to enter creative fields. - Raises intellectual property (IP) and ownership concerns.

• Adoption of Generative AI Tools - Foundation models such as DALL-E, Midjourney, and Stable Diffusion, along with derived tools, have gained popularity among creatives. - They offer a range of applications for Web3 Artists, Graphic Designers, Content Creators, Photographers, Videographers, Film Producers, and Architects.

• Idea Generation with AI - Generative AI models streamline the idea and draft generation process. - Creators use AI to generate initial drafts, refine ideas, or get inspiration, based on a generated prompt or supplied images. - The same approach is applicable in various creative domains, from designing logos to scripting YouTube videos and generating movie plots.

• Democratization of Creative Tools - New AI tools enable smaller creators to access features previously only affordable to large studios, such as storyboarding.

• Re-defining Artist Role with Generative AI - Generative AI has made becoming an artist more accessible through prompt engineering, similar to the advent of photography. - Critics are concerned about the impact on traditional art, but history suggests that AI will create a new intertwined form of art. - Arushi Kapoor, ARTSop CEO, suggests AI could be a tool for creators to enhance their works, rather than replace them: >Human creativity is what art is all about…The more optimistic view is that artificial intelligence evolves into a greater tool for existing creators to enhance, discover and replicate their works. We all hope for a world where our technologies help us, and not replace us.”

  Is Artificial Intelligence Set To Take Over The Art Industry?

• Addressing Intellectual Property and Copyright Issues - Artists are expressing concerns regarding generative AI using existing online content to produce outputs, leading to copyright and IP disputes. - Matthew Butterick, who filed a lawsuit against Stable Diffusion, criticizes AI systems for using copyrighted works without consent, credit, or compensation.

  Stable Diffusion litigation

• Proposed Solutions Using Blockchain Technology - One potential solution involves using blockchain technology to link each data piece in an AI model to a Non-Fungible Token (NFT). - This approach would provide transparency about which data was used for AI output and could ensure proper compensation for creators, fostering a more equitable economy. - Such methods could help resolve IP and copyright issues associated with generative AI models.

Artist Compensation

• AI Artworks and Compensation Controversy - Obvious, a French art collective, created an AI artwork “Edmond de Belamy” that sold for $432,500 at a Christie’s auction in 2018. - The artwork was generated using a Generative Adversarial Network (GAN) trained on 15,000 portraits from the 14th to the 20th century. The generator creates a new image, and the discriminator tries to differentiate between human-made and generator-created images. - Critics argued the AI artwork lacked originality and creativity due to its heavy reliance on the training data, and raised concerns about the borrowed code used by the artists. - This case highlights the IP and copyright concerns associated with AI-generated art, particularly whether artists should be compensated whenever AI uses their work to generate prompts.

  AI Art at Christie’s Sells for $432,500 - The New York Times

Business: from improvement to transformation

Generative AI for Enterprise

• Impact of Generative AI on Enterprises - Generative AI is introducing a new era for businesses, necessitating a rethink of work procedures. - AI’s greatest potential lies in process streamlining, which allows for more time on tasks requiring human input like strategic planning, decision-making, relationship building, qualitative research, and creativity. This could lead to cost savings through increased productivity. - Microsoft’s CEO, Satya Nadella, has expressed optimism that the next AI generation will trigger productivity growth and renew the joy of creation. - Enterprises are still figuring out how to effectively and safely utilize the array of new generative AI tools, a transition that analyst Rowan Curran indicates will take time as quoted in a CNN article. - Companies are already finding ways to use generative AI to enhance productivity, though specific implementations vary.

• Sales and Customer Service - Generative AI can significantly transform the sales industry, specifically in prospecting and lead qualifying, by providing personalized experiences for clients and streamlining processes.

  • AI can assist in prospecting, lead generation, and pipeline growth, producing personalized emails, videos, and virtual product walkthroughs.
  • This allows sales reps to focus more on converting leads to sales rather than initial outreach.
  • AI can respond to basic inquiries from leads based on past questions, maintaining lead engagement and enabling the sales team to focus on quality and personalization.
    • Generative AI can function as a virtual assistant, drafting emails, scheduling meetings, and preparing notes for future interactions, thereby better preparing the sales team for important discussions and upselling opportunities.
    • In customer service, generative AI can automate processes like chatbots and virtual assistants.
  • AI can leverage past case and inquiry data to generate appropriate responses, enabling self-service portals and chatbots to provide quick and accurate responses.
  • This automation frees up customer service representatives to handle more complex issues.
    • As AI models improve with more data and advanced training, the quality of these services will increase, though human oversight will be needed to counteract biases and errors.

• Generative AI in Forecasting, Scenario Generation, and Reporting - Generative AI can analyze vast amounts of data to generate accurate predictions in forecasting and scenario generation.

  • Businesses can gain insights into customer behavior, market trends, and future product and service demand.
  • For instance, insurance companies can leverage Generative AI to predict future claims and adjust pricing strategies.
  • Generative AI can also generate multiple scenarios based on different variables, aiding businesses in making informed decisions.
    • Generative AI has a strong potential in automating business reporting, leading to significant time saving and efficiency improvements.
  • It can rapidly and accurately analyze vast data volumes, reducing the likelihood of human errors.
  • Generative AI can tailor reports for individual users and scale to accommodate growing data volumes.
    • Overall, Generative AI provides a cost-effective solution for efficient, accurate, and personalized reporting, thereby enabling businesses to make better decisions and maintain competitiveness.

Legal Compliance and RegTech

• Challenges of Increasing Regulatory Compliance and Role of Generative AI

  • There has been a consistent growth in regulatory constraints over the past 50 years.

• This increase is due to a mix of factors such as changes in perceptions of social costs, political priorities, and rent seeking.

  • Significant events or the emergence of new technologies often lead to regulatory interventions.

• For instance, the global financial crisis of 2007-2008 resulted in new financial regulations such as increased capital requirements for banks, tighter restrictions on lending, and new rules around financial instrument trading.

• Similarly, the rise in data collection by companies, particularly digital platforms, led the EU to introduce the General Data Protection Regulation (GDPR).

  • Generative AI is now being incorporated to help businesses and regulators navigate these increasingly complex regulatory landscapes.

• RegTech and the Role of Generative AI

  • The rise in regulation complexity and quantity has increased the compliance burden on organizations.

• This includes costs associated with obtaining legal advice, developing compliance procedures, and hiring compliance staff and auditors.

• For financial crime compliance alone, the global cost was estimated at $274 billion in 2022 by LexisNexis.

  • Regulatory Technology (‘RegTech’) refers to tech that aids in meeting regulatory requirements more effectively or efficiently.

• It’s useful in areas where regulatory environments are complex, risk-based regulatory approaches can be improved, better monitoring is enabled, and more uses of data for regulatory compliance can be unlocked.

  • Generative AI has multiple potential applications in RegTech, including:

• Generating compliance reports: Automating the interpretation of reporting requirements, data extraction and classification, report generation, and analysis. Artificial Intelligence for reporting

• Automating compliance processes: Reducing human error and inconsistencies in reporting, such as real-time monitoring of financial transactions, reducing false positives in Anti-Money Laundering compliance. AI and Regtech

• Automating complaint procedures: Streamlining the collection, analysis, and categorization of complaints, and generating responses. However, the method of prioritizing complaints is under scrutiny by resourcing constrained regulators. Gone in 38 seconds: Regulator using AI to reject serious criminal complaints

• Predicting future risks: Identifying patterns and predicting potential compliance risks, generating reports that highlight issues for further action, and predicting and reporting on future trends. Artificial Intelligence in Risk Management

Technology: from automation to innovation

Coding

• Generative AI Revolutionizing Coding
  • The swift evolution of generative AI has spiked interest in transforming the coding process, and is expected to be a universal tool in software development.
  • Key ways generative AI is currently reshaping software development include:
• Code translation: Also known as source-to-source translation, this tool translates code between programming languages. It allows developers to avoid learning multiple languages and aids in migrating legacy code to more modern languages. An example is TransCoder by Meta. Deep learning to translate between programming languages
• Code autocompletion: This involves generative AI predicting and suggesting code snippets, functions, or methods based on the context and syntax of the written code. It enhances the speed and accuracy of coding, reduces errors, and improves consistency, which is beneficial in large software projects. GitHub Copilot is a notable generative AI-based code completion tool. GitHub Copilot · Your AI pair programmer
• Natural language to code: Translates human language into code, enabling programmers to code in a more intuitive way without needing extensive knowledge of programming languages. For instance, a command like “create a function that calculates the square root of a number” would generate the corresponding code. OpenAI Codex, which powers GitHub Copilot, is an example of this technology. Powering next generation applications with OpenAI Codex
• Code review: Analyses code and provides feedback on its quality, readability, and maintainability. It’s particularly helpful for large codebases, where manual review can be tedious and prone to errors. It can identify potential security vulnerabilities and bugs by analyzing code patterns, catching potential issues early, and reducing the risk of security breaches.

Case Study: GitHub Copilot

  - GitHub Copilot is a code generation tool that employs OpenAI's GPT model to produce code based on the context of a developer's ongoing work. The tool has access to a multitude of code snippets and generates code considering the code structure, function names, and even comments.
  - It has garnered positive reception from developers for its ability to swiftly and accurately generate code. It suggests variable names, function names, and other elements of code structure, facilitating developers in writing consistent and easy-to-understand code.
  - A study by GitHub found that developers using GitHub Copilot experienced a 55.8% boost in productivity and a 60% rise in satisfaction. This is due to generative AI automating monotonous and time-consuming tasks, allowing developers to concentrate on more intricate and inventive tasks.
  - As stated in KPMG's Game Changer report, generative AI is transforming software development by offering new methods to code more efficiently and effectively, likely becoming an invaluable coding partner for many developers.

https://advisory-marketing.us.kpmg.com/html/speed/pdfs/8956-Gen-AI.pdf

Generative AI and Web3

• Generative AI and Web3 Technology - Web3 signifies the next generation of the internet, leveraging blockchain, decentralized networks, and smart contracts for a more open, equitable, and transparent online experience. The fusion of Generative AI with Web3 technology is anticipated to revolutionize internet operations.

  • Combating Biases with Verifiable Data

• Blockchain can mitigate biases in Generative AI by offering transparent and immutable data tracking. Non-fungible tokens (NFTs) can represent data inputs in an AI model, allowing users to track data utilized in training models, including data source, access history, and modifications.

• Blockchain enables decentralized data collection, ensuring data diversity and preventing skewness towards a specific population subset. This can help counteract biases often introduced when data is collected from a restricted sample.

• Infrastructure - Current foundational models like Chat-GPT, Stable Diffusion, and Midjourney are owned by Web2 companies such as Facebook, Google, and Microsoft, sparking concerns about data safety and governance. - Web3 aims to solve these issues by using blockchain technology infrastructure, such as smart contracts, tokens, and Decentralised Autonomous Organisations (DAOs), fostering a decentralized, democratized internet that incentivizes and compensates contributions. - This infrastructure could enable AI models to be community-governed, similar to DAOs like Uniswap, addressing Web2 issues and facilitating a fairer internet. - Novel organizations might emerge, pooling data for generative AI models to access for research. Conversely, groups could form DAOs to collectively purchase data and associated IP for diverse applications.

• Case Examples

  • For instance, fans of a discontinued video game could use generative AI and appropriate training data to create additional content or alternative endings.
  • Similar approaches could be applied to movies, TV shows, books, art, music, etc.

• New Incentive Structures for Training Data - Blockchain technology could incentivize individuals and organizations to contribute high-quality data to generative AI models, possibly through cryptocurrency rewards. - Microsoft has explored using blockchain and cryptocurrencies to establish new incentive structures, similar to decentralized oracle services like Chainlink, rewarding high-quality data submissions while disincentivizing poor ones.

Key Takeaways

* Generative AI can be used in a variety of industries, with gaming being particularly interesting due to its complexity and participant diversity. The potential for fostering a vibrant cottage industry of creative games is significant.
* Software, another area deeply affected by generative AI, is ubiquitous in our world. It represents a form of automation, converting human tasks into machine tasks.

• Implications of Generative AI in Software
  • The bottleneck in software development has traditionally been the need for developers and engineers. However, generative AI technologies can not only aid developers but potentially turn anyone into a developer, marking a revolutionary change.
  • This shift could vastly increase our capabilities to build across diverse industries, opening up exciting possibilities.

---

4. Challenges in Adoption and Use

• Ethical considerations and concerns about deep fakes, fake news, privacy, and intellectual property rights arise when using generative AI.
• Integrating new technologies like generative AI into existing institutions involves navigating ethical and moral considerations, particularly in relation to privacy and handling large-scale human data.
• Ensuring safety, effectiveness, and value production from generative AI poses ongoing challenges that require the establishment of legal, regulatory, moral, and business codes and practices.

Best Practices for Adoption and Use

Effective Use of AI

• The development of more advanced AI brings excitement and opportunities for end users and businesses.
• It is important not to rely solely on AI to solve all problems, as we are still in a phase where the quality of input affects the output.
• Connecting different AI components and ensuring their proper functioning is crucial for delivering effective solutions.
• Considerations should be given to handling failures and building redundancies in case of component failures or internet outages.
• Prompt engineering is a new skill that needs to be developed and shared within the AI community.
• Interacting effectively with AI systems presents unique challenges due to their black-box nature and alien intelligence.
• Developing skills and knowledge-sharing in AI usage, along with effective interaction with AI systems, will be key to successful implementation.

Despite the numerous benefits that AI offers for businesses, it is crucial to approach its deployment with careful consideration of various factors. AI systems have the potential to revolutionize industries, but they also come with security, safety, and ethical concerns that must be addressed to ensure responsible and effective use.

One significant challenge is the potential for AI systems to malfunction or produce hallucinations, providing incorrect or misleading solutions. When AI does not perform as intended, it can undermine the purpose of using AI in the first place and erode trust in these technologies. For example, the testing of autonomous vehicles has faced scrutiny and controversy after an accident involving a pedestrian, raising legal and ethical questions about the deployment of self-driving cars.

Similarly, AI-powered tools like COMPAS, used for criminal risk assessment, have shown biased results toward individuals of different racial backgrounds. These instances highlight the critical importance of designing and deploying effective AI systems that address such challenges and produce reliable and unbiased outcomes.

So, what does it mean to have an effective AI system? Here are some key characteristics:

1. Efficiency: An effective AI system should increase productivity and optimize existing processes, delivering desirable outputs more efficiently than traditional methods.

2. Accuracy: The output generated by an effective AI system must be accurate, both in terms of quantity and quality. It should avoid false, unreal, or biased outcomes, leveraging large volumes of data to achieve greater accuracy.

3. Fairness: An effective AI system should depict fairness and equality for all groups affected by its output. It should consider diverse datasets to ensure that decisions are unbiased and treat everyone equally.

4. Reliability: Consistency is crucial for an effective AI system. When employing a similar machine learning model and utilizing a comparable dataset, the system should consistently produce the same results, ensuring reliability.

5. Ethical: Ethical considerations play a vital role in effective AI systems. These systems should adhere to moral and human values, respecting individual rights and societal norms. Ethics in AI encompass multiple dimensions and involve various stakeholders.

6. Transparent: Transparency is integral to effective AI. It involves informing relevant stakeholders about the processes used in designing, developing, and deploying AI systems. Transparency promotes trust and understanding.

7. Explainable: An effective AI system should be explainable, going beyond transparency to provide insight into the reasons behind specific processes or decisions made during design, development, or deployment phases.

8. Accountable: Accountability is crucial for an effective AI system. As AI systems are the result of human and machine interactions, mechanisms must be in place to ensure proper functioning. In case of malfunctioning, clear legal and regulatory frameworks should hold responsible parties accountable.

9. Sustainable: Sustainability is a key consideration for effective AI systems. This encompasses various aspects, such as power consumption, hardware requirements, cost-effectiveness, and the ability to upskill the workforce. AI-enabled solutions should be economically, environmentally, and socially feasible.

10. Integrated: An effective AI system must be integrated into various components of the business model. A fragmented or piecemeal approach to AI deployment may hinder its potential benefits. A comprehensive AI strategy that aligns with the overall business strategy is essential to maximize value.

When deploying AI, organizations need to carefully evaluate its significance to their value proposition. They must consider technical, social, and organizational factors to build effective AI systems that enhance organizational values while adhering to social and regulatory principles. By addressing these considerations, businesses can unlock the transformative potential of AI while ensuring responsible and beneficial outcomes for all stakeholders involved.

Generative AI Errors

The mistake highlights the biggest problem of using AI chatbots to replace search engines — they make stuff up: Google’s AI chatbot Bard makes factual error in first demo - The Verge

The hilarious & horrifying hallucinations of AI - Sify

We asked Bing with ChatGPT to review the Galaxy S23 Ultra — and it got a ton wrong

Ethical Considerations

• Deep Fakes: - Deep fakes created using generative AI can spread misinformation, fake news, and damage reputations. - Regulation and transparency are needed to address this ethical concern, along with the development of deep fake detection technologies. - Example: The viral deep fake video of Nancy Pelosi highlighted the potential harm caused by generative AI in spreading misinformation. Doctored Nancy Pelosi video highlights threat of “deepfake” tech - CBS News

• Bias: - Generative AI trained on biased data can perpetuate and amplify existing biases in its output. - Diverse and representative data sets should be used to train generative AI systems, and measures should be in place to detect and address bias in their output. - Example: Amazon’s facial recognition software exhibited bias against people of color and women due to being trained on a predominantly white male dataset. - Amazon facial-identification software used by police falls short on tests for accuracy and bias, new research finds - The Washington Post - The racist hijacking of Microsoft’s chatbot shows how the internet teems with hate | Paul Mason | The Guardian - Female historians and male nurses do not exist, Google Translate tells its European users - AlgorithmWatch

• Data Privacy: - Generative AI systems require large amounts of data, raising privacy concerns. - Strict data privacy regulations and anonymization of data used to train generative AI systems are necessary to protect sensitive information. - Example: The use of generative AI in healthcare raises concerns about the privacy and security of personal medical data. - Data privacy and artificial intelligence in health care | Reuters

• Harmful Content: - Generative AI can be used to generate discriminatory remarks, hate speech, or other harmful content. - Systems should be developed to detect and prevent the generation of harmful or discriminatory content, and ethical guidelines should be promoted for the use of generative AI. - Example: Microsoft’s chatbot Tay started spewing offensive comments on Twitter due to a lack of contextual understanding.

These ethical concerns highlight the need for responsible use and development of generative AI, including addressing deep fakes, bias, data privacy, and preventing the generation of harmful content. Establishing regulations, promoting transparency, and ensuring diverse and representative training data are essential for harnessing the potential of generative AI while mitigating its risks.

The “Black Box” of Generative AI

• Generative AI: - Generative AI works through large language learning models and uses tokens to make predictions. - It can be seen as a black box to users, making it important to understand what goes into and comes out of generative AI.

• Explainability: - Two aspects of explainability for generative AI: what data is being input and what results are being generated. - Data inputs include textual and visual information in various digital formats. - The output of generative AI can be seen as a hallucination, as it can provide wrong or made-up information.

• Evolution and complexity: - Generative AI models become more complex and nuanced as they iterate and consume more information. - Adding different data sets or additional data can impact the output and have implications for business strategies.

Understanding what data is inputted into generative AI and how it affects the generated output is a fascinating aspect of explainability.

• Generative AI tools: - Described as “black boxes” due to their opaque internal workings. - Trained on large datasets and employ complex algorithms to generate outputs. - Difficult to understand why a specific output was produced or identify biases and errors. - Lack of transparency raises ethical and legal concerns about accountability and responsibility.

• Transparency and explainability: - Ongoing efforts to develop methods to increase transparency and explainability of generative AI systems. - Important for addressing the challenges associated with accountability and real-world consequences.

The opaque nature of generative AI tools and the challenges in understanding their internal workings emphasize the need for increased transparency and explainability in order to address ethical and legal concerns.

Safety Guardrails

• Generative AI tools have inbuilt guardrails that regulate user interaction with the technology.
• These guardrails function as a form of “regulation through code” proposed by Professor Lawrence Lessig: Codev2 - LESSIG
• Understanding the risks, benefits, and costs of guardrails is important for businesses utilizing generative AI tools.
• “AI safety” overlaps with “Responsible AI,” encompassing the safe and ethical operation of AI systems.
• Safety risks associated with generative AI tools include producing inaccurate or harmful content.
• Guardrails are implemented to promote safe and responsible use of generative AI tools.
• Examples of guardrails include bias mitigation, content moderation, content warnings, and transparency.
• Guardrails aim to address issues such as bias, legality, appropriateness, and user trust.
• Drawbacks of AI guardrails include potential impacts on performance, user experience, subjectivity, and freedom of expression.
• Consideration of unintended consequences and balanced implementation is necessary when employing guardrails.

YouTube - Australian AI Ethics Principles

Legal Risks

• Legal considerations are important when using generative AI in business.
• Three major legal risks are intellectual property infringementWhat is Intellectual Property?, misleading and deceptive conduct, and privacy violations.
• Generative AI models can potentially infringe on intellectual property rights, leading to legal action by copyright owners.
• Misleading or false content generated by AI models can violate consumer protection laws and result in legal action.
• Privacy laws and confidentiality obligations must be considered when using generative AI, as personal data may be involved.
• Managing legal risks includes implementing quality control, data security and privacy policies, staying informed about legal developments, and seeking legal advice.
• These strategies can help businesses mitigate legal risks associated with generative AI.

Mitigation Issues

• Alongside the benefits of AI, organizations must consider and mitigate the risks associated with its use.

• Risks can be analyzed and addressed at three tiers: application-level, business level, and national level.

• Application-level risks include performance issues, security, control, and biased output.

• Business-level risks include organizational reputation, financial performance, resource sustainability, power concentration, and workforce challenges.

• National-level risks impact society, including unemployment, biased outcomes, inequality, discrimination, and threats to human rights.

• Mitigation guidelines exist at different levels: organizational, national, regional, and external international bodies.

• Organizations should formulate standardized processes, monitor data privacy and governance, and adhere to regulatory guidelines.

• Governments release frameworks to regulate AI, ensuring responsible deployment.

• Regional and international bodies like the OECD, EU, and Microsoft offer AI governance frameworks and guidelines. FATE: Fairness, Accountability, Transparency & Ethics in AI - Microsoft Research

• Business leaders must stay updated on AI advancements, monitor organizational operating procedures, and proactively follow regulations to mitigate risks.

• Google: Established ethical AI principles to guide development and deployment, focusing on social benefit, fairness, accountability, and privacy. Google AI Principles – Google AI

• Microsoft: Formed an AETHER committee to develop guidelines for responsible AI use, emphasizing transparency, human rights, and risk mitigation. Our approach to responsible AI at Microsoft

• IBM: Created ethical design principles for AI, prioritizing transparency, fairness, privacy protection, and user empowerment. AI design ethics overview

• Salesforce: Established a Center for Ethical and Humane Use of Technology, developing guidelines that emphasize transparency, diversity, inclusion, and human rights. Why Salesforce Aims to Build Products That Are ‘Ethical by Design’ - Salesforce News

Intellectual Property

• Ownership of training data is a significant intellectual property concern in generative AI.

• Companies and individuals need to establish ownership rights and restrictions on the use of their training data through legal means.

• Methods to protect training data include technology solutions, governance solutions, and legal solutions.

• There are arguments around “fair use” rules permitting the use of copyrighted works as training data for generative AI models.

• Ownership of input (prompts) and output (content) data in AI models is a complex issue.

• Ownership may belong to the creator or owner of the AI model, the user of the model, or be shared between them.

• Obtaining patent protection for output data generated by AI models can be challenging due to the novelty requirement.

• There is ongoing debate about whether AI itself can own intellectual property such as patents.

• Clarity over intellectual property ownership and understanding of terms of use are crucial for businesses.

• Employers may own intellectual property created by employees, but this varies depending on jurisdiction and circumstances.

• Steps to mitigate legal risks should be followed to ensure compliance.

AI inventors: can AI own intellectual property rights? - Raconteur

Terms of Use for Generative AI Tools

• OpenAI’s terms of use govern the use of their services and define the rules and requirements.
• Content is defined in the terms of use as the text, images, audio, and any other material generated by the Services.
• OpenAI retains ownership of the Input and Output generated by the Services.
• Content generated by the Services cannot be used for commercial purposes unless explicitly allowed by OpenAI.
• Multiple users may receive the same or similar Output from the Services.
• The accuracy of the Output generated by the Services is not guaranteed, and users are responsible for evaluating its accuracy themselves.

Terms of use - OpenAI Sharing & publication policy

The Good, The Bad, and the Controversial

• Observation about AI’s transformative power, similar to electricity and computers in the past. AI, specifically generative AI, is viewed as a fundamentally disruptive technology, changing the face of business automation.

• Recent experiences with AI tools like ChatGPT and Midjourney indicate potential use cases where AI could substitute human roles. More of such exploratory use cases are expected to emerge.

• The discussion then pivots to second order effects and deeper consequences of AI adoption. It is suggested that AI’s role won’t be limited to replacing human tasks; instead, AI would work with humans as an ‘amazing coworker’.

• AI tools like ChatGPT, using large language models, can draw upon a vast corpus of human knowledge to assist in novel tasks and idea generation.

• The conversation then shifts to Aaron’s perspective on challenges, particularly in law and regulation. The ‘pacing problem’ is discussed where regulation struggles to keep pace with rapidly advancing technology.

• Historically, regulations have lagged a decade or so behind technological advancements, as observed during the advent of personal computing and the internet. This lag is expected to pose challenges in the coming decade.

• Regulatory catch-up issues around intellectual property are raised, both for the outputs from generative AI tools and the inputs into those tools, such as language models. Concerns about copyrights, patents, and designs on these sources are discussed.

• A key challenge with AI technology is the ‘pacing problem’ where law and regulations lag behind the rapid technological advancements. This issue might lead to the emergence of private ordering of AI technology, with industry codes and internal policies set by businesses.

• A scenario is proposed where initial regulations may come from industries themselves, similar to how agreements among businesses handled the advent of electricity. This could eventually evolve into national infrastructure frameworks and market rules.

• Regulators and legislators typically take time to understand new technology before providing regulatory certainty. However, a balance is needed between regulation and fostering innovation for consumer benefit.

• The discussion turns to the topic of algorithmic bias and data training sets, questioning if the market alone should handle this or if high-level codes of practice are needed.

• The existing AI governance framework in Australia is referenced, but with the rapid advances in AI technology, a review of the framework is needed.

• Notably, global AI governance frameworks are inconsistent, which poses challenges for businesses exporting AI-based products. What might be legal in one country may not be in another, requiring adaptation to meet different regulatory requirements.

• Innovation often precedes regulation. Businesses, driven by the excitement of AI advancements, are pushing ahead without fully considering governance, privacy, and security implications in the rush to be the first mover in their sectors.

• There is a dynamic relationship between strategy, new technology, and regulatory challenges. This balance becomes particularly important when discussing rapidly evolving technologies, such as large language models.

• Notable players in this field include both large tech companies and newer, smaller entities, such as OpenAI.

• There is an emerging trend of smaller firms leveraging these large language models for specific industry applications.

• The regulatory and legislative landscape remains fluid, necessitating a close collaboration between policy makers and innovators.

• The failure of big companies in the tech space, such as Google, highlights the risks and challenges of pushing products to market without thorough consideration of the possible implications, including potential data leaks.

• Looking forward, main concerns revolve around the swift evolution of technological capabilities and the legislative landscape.

• These technologies operate globally, involving multiple jurisdictions and requiring strategic attention to both local and international regulations.

• Key legislative issues relate to intellectual property, misleading advertising, data privacy, and cybersecurity.

• Concerns:

• Massive disruption caused by this technology could leave certain sectors or populations behind. This requires a concerted effort towards public education about this technology.

• The capacity of generative AI to ‘hallucinate’ or generate false information is a cause for concern.

• Excitements:

• Large potential productivity gains and wealth creation from automation.

• The possibility for marginalized groups to gain access to complex and expensive knowledge through these technologies.

• The transformative potential for areas of life including work, education, and entertainment.

• In education, AI can complement student studies and potentially shift the focus of teaching towards critical thinking.

• Risks:

• Companies rushing to adopt AI technologies may encounter regulatory problems or unintended consequences, potentially slowing the overall progress of the industry.

• In sensitive sectors like healthcare, mistakes caused by AI misdiagnosis could result in serious harm and backlash, potentially halting the application of AI in the sector.

• Data breaches and privacy concerns are significant issues.

Reflection on the history of technology, with the example of nuclear technology illustrating how fear can impede technological potential. The hope is that AI will not follow the same path and will continue to be explored and developed fully.

---