Book summary “Human + machine”: summary

Introduction

You face a constant struggle to balance technological disruption against operational efficiency in your organization. Traditional automation forces a false choice between human labor and machine replacement, leaving your business rigid and vulnerable.

Paul R. Daugherty and H. James Wilson argue that you must abandon this binary view. You need to embrace the missing middle where humans and smart machines collaborate as symbiotic partners.

The book guides you through three waves of business transformation, the MELDS framework, and the fusion skills required for enterprise reinvention.

• How do you redesign your core business processes for artificial intelligence?
• What new human roles emerge when algorithms take over routine cognitive labor?
• Which specific fusion skills do you need to master to lead this transition?

1. What You Need to Know about Generative AI

Artificial intelligence has evolved along a continuum consisting of diagnostic, predictive, and generative capabilities. Diagnostic AI analyzes historical data to understand past faults. Predictive AI forecasts future outcomes. Generative AI creates entirely new content such as text, images, code, and synthetic data.

Generative AI acts as a general-purpose technology that transforms cognitive tasks, automates routine labor, and unlocks trillions in economic value through foundation models.

Alphabet's DeepMind developed AlphaFold2 and created a public database accessed by over half a million researchers to study plastic pollution, antibiotic resistance, and food insecurity. AlphaFold2 predicts the 3D structures of almost every known protein from amino acid sequences, outperforming traditional experimental methods that took years per protein. Specifically, it can predict structures for 200 million known proteins, whereas researchers had solved only 170,000 structures in the more than half a century prior.

The AI Continuum

Foundation models derive their adaptability from massive scale, transformer architectures, and growing parameter counts. Transformer neural networks process sequential data in parallel by assigning attention weights based on context. Parameter counts have scaled exponentially from millions to trillions, enabling zero-shot and few-shot learning. OpenAI's GPT-3 debuted with 175 billion parameters, while GPT-4 is estimated to have around 1.76 trillion parameters. By comparison, the core BERT model released in 2018 used 340 million parameters, and OpenAI's GPT-2 unveiled in 2019 used 1.5 billion.

Foundation Models

Economic modeling shows that responsible adoption placing people and innovation at the center can create 10.3 trillion additional economic value in dollars globally by 2038. Analysis of nineteen major industries reveals that banking has the largest opportunity with 72 percent of working hours potentially transformed by generative AI. Insurance follows with 68 percent, and capital markets reach 67 percent. Moreover, 81 percent of companies see generative AI as a main lever for enterprise reinvention, and 86 percent of C-suite executives use it to some degree in their work.

Buy, Boost, or Build

Organizations must choose between three distinct strategic options to deploy foundation models: buy, boost, or build. Buying involves using off-the-shelf models via APIs for quick productivity gains. Boosting involves fine-tuning pre-trained models or using retrieval-augmented generation with proprietary data. Building involves creating a new foundation model from scratch for high strategic differentiation. For example, Bloomberg built BloombergGPT as a custom LLM with 50 billion parameters from scratch using forty years of financial data.

You should pursue a dual-track approach by implementing low-risk no-regrets use cases alongside major strategic bets. A multinational telecom company used a generative AI solution to reduce agent response time by 30 percent and increase its property contract query user experience score by 66 percent.

To capture this potential, you must follow five key principles to scale generative AI and successfully drive organizational reinvention:

• Lead with Value: Start with a business case considering costs, benefits, and readiness to scale across your value chain.
• Understand and Develop an AI-Enabled, Secure Digital Core: Evaluate your technology to complete the journey to cloud computing and develop a secure data foundation.
• Reinvent Talent and Ways of Working: Communicate transparently, invest in continuous learning, and develop fusion skills for human-machine collaboration.
• Close the Gap on Responsible AI: Ensure C-suite accountability and take intentional actions to design and deploy AI to build trust and protect the organization.
• Drive Continuous Reinvention: Adopt a cultural and operational mindset for continuous change powered by a flexible digital core.

2. The Self-Aware Factory Floor

Digital twins, collaborative robots, and generative design revolutionize industrial operations, supply chains, and manufacturing processes.

Generative AI significantly impacts working hours in the industrial sector by automating routine tasks while augmenting human roles. Economic modeling shows that generative AI affects nearly a third of industrial working hours, freeing personnel to focus on higher-value tasks while requiring greater human involvement. 32 percent of the working hours in the industrial sector affected by generative AI include 26 percent subject to automation and 6 percent subject to augmentation. Nearly half of industrial production managers working hours are affected, with 30 percent for automation and 19 percent for augmentation.

Product Development

Generative design software allows engineers to combine human goals with cloud computing to consolidate multiple parts into a single, optimized component. Engineers provide software with goals and constraints such as materials, budgets, and manufacturing methods, allowing the application to evaluate all permutations and generate lightweight, strong components.

You can follow a structured approach to generate optimized parts:

• Provide generative design software with design goals and constraints including materials, budget, and manufacturing methods.
• Let the application evaluate all possible design permutations and recommend a solution.

General Motors worked with Autodesk to redesign a car seat bracket, replacing eight welded pieces with a single stainless-steel piece created by generative design. The software generated 150 alternative designs for the seat bracket that were 40 percent lighter and 20 percent stronger than the previous version. Generative AI enables the rapid creation of specialized, one-of-a-kind components that are significantly lighter and stronger than human-designed counterparts. NASA research engineer Ryan McClelland used commercially available AI software at Goddard Space Flight Center to pioneer specialized components for space missions called evolved structures. Generative AI executed 30 or 40 iterations in just an hour to produce structures two-thirds lighter than components designed solely by humans and ten times less subject to stress than traditional designs.

Robotics and Digital Twins

Modern robotic arms equipped with deep reinforcement learning can learn tasks autonomously and share their knowledge across a network of robots. Using trial and error based on a picture of a successful outcome, robotic arms become highly accurate quickly without explicit manual programming, freeing human experts for complex tasks. Fanuc partnered with Preferred Networks in Japan to develop robotic arms that take eight hours to become at least 90 percent accurate through deep reinforcement learning. Eight arms working together for one hour can learn as much as one working alone for eight hours. Equipping robots with sensors and compliance allows them to work safely alongside humans without physical barriers. Traditional fast-moving robots required segregation behind protective barriers, but collaborative robots use elastic actuators and back-drivable joints to absorb energy upon contact. Rodney Brooks founded Rethink Robotics to build collaborative robots equipped with sensors that recognize objects and avoid injuring people, while iRobot sold 40 million autonomous robots as of 2021. Advanced machine-vision algorithms allow warehouse robots to handle irregularly shaped packages and instantly measure shelf space. Symbotic robots zip around warehouse floors at speeds of 25 miles per hour, immediately removing products from incoming pallets and eliminating the need for traditional storage spaces and conveyor belts.

Hypercustomization

Digital twins model complex systems and physical assets on a computer, enabling reimagined maintenance, product development, and operations. By collecting field data and simulating real-world conditions, digital twins allow companies to predict part failures, improve thermodynamics understanding, and holistically optimize entire networks. General Electric uses digital twins to monitor wind turbines, discovering that slowing down the front turbine increases the energy generation of backend turbines across a wind farm by 20 percent, providing $100 million of value over the lifetime of a 100-megawatt wind farm. Neural networks and machine-learning algorithms improve demand planning and reduce supply chain uncertainties. A leading health-food company utilized machine-learning capabilities to analyze demand variations during promotions, achieving a 20 percent reduction in forecast error and a 30 percent reduction in lost sales. Precision agriculture combines internet-of-things sensors with machine-learning engines to maximize crop yield and reduce resource waste.

You can operationalize precision farming with connected systems:

• Collect granular data using a vast network of diverse internet-of-things sensors.
• Process IoT-sensor data with a machine-learning engine to provide feedback.
• Route feedback directly to farmers or deploy recommendations automatically through digital work-management systems.

The Indian nonprofit Akshaya Patra combines AI, blockchain, and IoT sensors to forecast food demand and monitor cooking processes for school meal programs serving 2.1 million children in over 24,000 schools out of 795 million people without enough food worldwide.

3. Accounting for Robots and Corporate Functions

Artificial intelligence automates anti-money-laundering checks, optimizes insurance underwriting, and accelerates corporate recruitment and software development.

Financial Services and Compliance

You can radically streamline anti-money-laundering operations using advanced analytics and machine learning algorithms. A large global bank deployed advanced analytics for anti-money-laundering detection, which reduced false positive alerts by 30 percent and cut investigation costs by 40 percent. Up to ten thousand staffers engaged in identifying suspicious transactions at a large global bank before network analysis optimized alert thresholds.

Insurance companies are utilizing generative and traditional AI to remake underwriting, increasing close rates by up to 20 percent and decreasing leakage by up to 10 percent. 100 percent of work time for underwriters has potential for task augmentation through catastrophe modeling and automated document generation.

HR and Corporate Functions

Unilever implemented a two-stage digital hiring process using neuroscience games and HireVue video analysis, which doubled job applications to 30,000, reduced average hiring time from four months to four weeks, and increased university representation from 840 to 2,600. This workflow achieved a 75 percent drop in recruiter application review time.

Virgin Trains installed the inSTREAM machine-learning platform with NLP capabilities to manage customer complaints, decreasing manual work by 85 percent and increasing correspondence by 20 percent. Morgan Stanley went live with a GPT-4-based generative AI assistant giving financial advisers access to approximately 100,000 research reports and documents. Developers using AI coding assistants completed tasks more than 55 percent faster in one studied case.

4. The Ultimate Innovation Machine

Traditional drug discovery is lengthy and prone to high failure rates. It typically takes ten years and $1 billion on average to bring a drug to market, with biologic drug discovery facing a 90 percent failure rate. Next-generation platforms combine deep-learning target identification with generative chemistry and knowledge graphs to rapidly hypothesize, synthesize, and test potential molecules.

Generative chemistry and machine-learning platforms compress drug discovery timelines and uncover novel scientific knowledge from dark experimental data.

Insilico Medicine used its Pharma.AI platform alongside AlphaFold structures to discover and synthesize a potential drug for hepatocellular carcinoma in just thirty days after synthesizing only seven compounds. These platforms use tools like PandaOmics and Chemistry42 to navigate complex biological pathways.

Biotech company Absci used zero-shot generative AI to create de novo antibodies that worked in the lab right out of the computer, potentially reducing clinic entry time from as much as six years to eighteen to twenty-four months. This approach bypasses the slow and costly phase of lab optimization by producing unique antibody designs that successfully bind target proteins straight out of the computer.

Data-Driven R&D

Researchers often discard or ignore failed or dark experimental attempts. By extracting properties from both successful and unsuccessful trials and processing them through decision-tree models, machine learning algorithms can predict outcomes with higher accuracy and provide transparent logic.

Chemistry researchers at Haverford College used machine learning on nearly 4,000 crystal-producing chemical reactions over a decade, with nearly 300 properties attributed to each reaction. The algorithm achieved an 89 percent crystal prediction rate compared to 78 percent by researchers using intuition alone.

Advanced reverse engineering and forward simulation platforms test vast arrays of patient health records without relying on preexisting assumptions or human guidance. GNS Healthcare used its Reverse Engineering and Forward Simulation (REFS) platform on anonymous data from about 200,000 patients, churning through approximately 45 quadrillion hypotheses to successfully recreate the results of a two-year study in three months.

5. Say Hello to Your New Front-Office Bots

Front-office bots handle customer inquiries and retail scheduling while creating new branding challenges through platform disintermediation.

Front-Office AI

Generative AI chatbots in contact centers handle routine inquiries, allowing human customer service representatives to focus on complex issues and upskilling into low-code AI developers. By automating common questions, chatbots reduce wait times, provide 24/7 self-service, and enrich jobs as agents are trained to become low-code AI developers.

A study of over 5100 customer support agents measured a 14 percent average increase in productivity by issues resolved per hour.

John Hancock partnered with Microsoft to implement chatbots on customer-facing websites, which moderated call volumes and enabled agents to give full attention to difficult cases.

Retail and Brand Disintermediation

Retailers use AI technologies like smart mirrors and scheduling software to personalize the in-store shopping experience and optimize floor staffing. Smart mirrors recognize items via radio-frequency identification, translate languages, and adjust lighting, while store analytics help optimize schedules to eliminate manager biases and boost sales.

Survey data shows that 70 percent of surveyed customers report wanting in-store recommendations. Store analytics reveal that 53 percent of the time stores were overstaffed, and 33 percent of the time stores were understaffed. Percolata scheduling recommendations delivered a 10 to 30 percent boost in sales.

Ralph Lauren partnered with Oak Labs to create connected fitting rooms with smart mirrors, while a global Japanese clothing retailer used Percolata software to optimize scheduling across twenty US locations.

Voice assistants and AI portals create brand disintermediation by inserting a tech platform between customers and traditional brand companies. When companies allow customers to interact with them through platforms like Amazon Alexa, the platform owner controls the fundamental interface and information exchange, capturing valuable data and diluting the direct brand relationship.

Amazon customers use Echo speakers and Alexa to order pizzas from Domino's, check bank balances at Capital One, or get flight statuses from Delta, shifting interface control to Amazon.

6. Rearing Your Algorithms Right

Deploying responsible AI requires specialized human roles including trainers, explainers, and sustainers to govern system behavior and maintain user trust.

As AI adapts to you rather than you adapting to computers, specialized professionals are needed to shape system behavior. Yahoo algorithms achieved 80 percent accuracy when trained to detect sarcasm on social media. Fanuc trained 47,000 people to use its equipment.

You can rely on specific new roles to guide your systems:

• trainers who teach AI systems how to perform tasks, adopt empathy, and exhibit appropriate personalities;
• explainers who bridge the gap between technologists and business leaders by auditing opaque algorithm decisions;
• sustainers who ensure that AI systems function safely, ethically, and in alignment with human values.

Trainers

Data hygienists and prompt engineers shape bot behavior and prevent biases. Empathy trainers, personality trainers, and worldview and localization trainers help systems interact naturally.

Apple sought a creative individual to rewrite Siri as a sassier character. Startups like Koko developed machine-learning systems to train chatbots to respond with sympathy and depth.

Explainers

Advanced machine-learning models can be opaque. Algorithm forensics analysts, transparency analysts, and explainability strategists evaluate accuracy versus explainability to justify automated decisions.

Zest AI uses complex algorithms to evaluate credit risk for underbanked applicants. The system provides clear explanations of the scoring factors used in lending decisions.

Sustainers

Context designers, AI safety engineers, ethics compliance managers, automation ethicists, and machine relations managers act as watchdogs. They monitor system safety, address legal breaches, and conduct performance reviews.

Anthropic developed Constitutional AI using a written set of rules like the UN Universal Declaration of Human Rights. This approach keeps its chatbot Claude in line with human values.

7. Super Results from Everyday People

Workplace augmentation operates through amplification, interaction, and embodiment to deliver exceptional productivity and growth.

AI augmentation in the workplace operates through three primary categories of human-machine interaction:

• amplification lifts human productivity and creativity using tools like generative design and analytics;
• interaction employs advanced natural language interfaces to facilitate communication at scale;
• embodiment utilizes physical spaces, sensors, and robotics for collaborative work in tangible environments.

Generative design software acts as an amplifier by letting algorithms iterate through vast design spaces while humans act as curators directing the aesthetic and engineering parameters. Autodesk designers used Dreamcatcher software's generative design features alongside engineers to create the Elbo Chair, which required 18 percent less material than the original model.

Heads-up displays and augmented reality tools significantly boost field worker productivity by overlaying instructions directly onto the physical workspace. At a global industrial services company, using an augmented-reality headset for wiring wind-turbine control boxes improved worker performance by 34 percent on first use. Boeing achieved 25 percent efficiency improvements, while similar cases averaged 32 percent productivity improvement.

Three Types of Augmentation

Natural language processing allows conversational AI assistants to scale customer service and handle routine tasks while routing complex issues to human experts. Swedish bank SEB deployed a virtual assistant named Aida to handle customer inquiries as the first point of contact for one million customers, learning from human experts when encountering unresolved questions.

Collaborative robots, or cobots, enable physical embodiment by working side by side with humans in manufacturing and warehousing to handle heavy lifting and repetitive tasks. Mercedes-Benz shifted away from maximum automation to collaborative setups where workers guide part-automatic robots while processing 1,500 tons of steel daily and producing 400,000 vehicles annually. BMW plants utilized lightweight robotic arms to improve ergonomics and prevent wrist strain during door panel assembly, achieving 85 percent more productive human-robot interactions compared to humans or robots working alone in an MIT study. Zipline utilized drones to deliver blood and health care products in remote areas such as Rwanda and Ghana, later expanding into commercial deliveries in partnership with Walmart.

From Tasks to Process Change

Deploying AI augmentation requires a fundamental reimagining of entire business processes rather than merely automating isolated tasks. When combined with human-centric roles, AI technologies unlock hidden efficiency gains, enhance worker safety and fulfillment, and enable the creation of entirely new business models. Emerging technologies such as spatial computing, advanced wearables, autonomous agents, and quantum computing promise further expansion of human-machine collaboration. Humane released the AI Pin wearable device to handle tasks like searching and translating, while the Rabbit R1 utilizes a proprietary Large Action Model to execute tasks based on user intent.

8. A Leader’s Guide to Reimagining Process

Implementing advanced AI requires going beyond traditional IT methods by using the MELDS framework, which encompasses mindset, experimentation, leadership, digital core, and skills. The MELDS framework addresses advanced AI issues that are typically neglected, such as corporate culture, responsible AI, ethics, consumer trust, and employee trust.

Leaders must leverage the MELDS framework to break down jobs, conduct structured experiments, build a secure digital core, and enforce responsible AI governance.

To reimagine processes effectively, you must adopt a mindset that ruptures with status-quo methods by breaking down jobs into constituent tasks and evaluating their reliance on language, knowledge, and creative reasoning. This approach avoids mere incremental improvements and targets step-level performance gains.

• Break down the job under consideration into the tasks that it entails for a particular role, from customer service to manufacturing to data science.
• Determine whether the task involves the intensive use of language (natural, computational, or mathematical) for completion.
• Assess how knowledge is used to perform a particular task, looking at ambiguity, collaboration needs, and subject-matter validation.

Best Buy partnered with Google Cloud and Accenture to use Gemini models and Vertex AI for a virtual assistant helping customers on BestBuy.com and an AI assistant supporting in-store frontline employees.

Experimentation and Digital Core

Experimentation in the age of AI requires structured and rigorous testing of use cases from the perspective of the missing middle, rather than a chaotic approach. Amazon opened Amazon Go in Seattle on December 5, 2016, to test just-walk-out technology using cameras, sensors, and deep learning, later scaling it to 29 stores and grocery locations. Walmart launched Store No. 8 in March 2017 as a tech incubator, creating the Intelligent Retail Lab in Levittown, New York, in 2019 before closing the hub in January 2024 after capabilities were fully embedded organization-wide.

Establishing an organizational culture that promotes responsible AI requires intentional governance, continuous risk assessment, and systematic mitigation of potential harms. Although executives understand the importance of responsible AI, a 90 percent gap exists between those who recognize this need and those systematically operationalizing it. 92 percent of automation technologists do not fully trust robots, and 76 percent suggest visual output dashboards as a solution to opacity.

• Establish AI governance and principles with sponsorship from the CEO and C-suite accountability.
• Conduct continuous AI risk assessment by inventorying AI usage and pinpointing potential risks.
• Enforce systematic responsible AI mitigation and testing through ongoing technical capabilities.
• Continually monitor AI for compliance post-deployment with internal policies and external regulations.
• Assess AI’s impact on value delivered, workforce, environment, privacy, and security.
• Build a responsible AI culture by employing appropriate role-based training and communications.

A modern digital core relies on treating data not as static silos, but as a dynamic end-to-end data supply chain that captures, cleans, integrates, curates, and stores information for AI applications. AI trainers spend approximately 90 percent of their time on data preparation and feature engineering rather than writing algorithms. Ducati Corse used up to one hundred IoT sensors on race bikes connected to an analytical engine and data-visualization tools to test bikes more effectively.

• Choose a simple outcome to start, focusing on well-defined, small-scale early projects with data.
• Ensure the data supply chain includes disparate, accessible structured, semistructured, and unstructured sources.
• Embed cybersecurity practices early in the life cycle across technology.

9. Extending Human + Machine Collaboration

Thriving in the missing middle requires eight novel fusion skills that blend human and machine capabilities without needing technical programming expertise. You must master these skills to successfully operate within AI-enhanced processes, handle machine uncertainty, and continuously reinvent workflows.

Thriving in the missing middle requires mastering eight novel fusion skills that blend human judgment with machine capabilities.

• Master intelligent interrogation by knowing how best to ask an AI assistant questions across levels of abstraction.
• Practice bot-based empowerment to collaborate with intelligent assistants and punch above your weight at work.
• Engage in reciprocal apprenticing by teaching AI assistants new skills while undergoing on-the-job training.
• Develop holistic melding by building mental models of AI agents that improve collaborative outcomes.
• Focus on rehumanizing time to amplify time available for distinctly human tasks and learning.
• Pursue responsible normalizing by shaping the purpose and perception of human-machine collaborations.
• Apply judgment integration to choose courses of action amid machine uncertainty.
• Embrace relentless reimagining to think of novel ways to overhaul work, processes, and business models.

A power plant maintenance worker uses digital twin software and augmented reality to identify turbine wear, interrogate the system across levels of abstraction, and prevent a costly outage. This intervention achieves a 95 percent confidence level of the maintenance software in its repair assumptions, saving 12 million dollars by preventing an unplanned turbine outage.

Continuous Reinvention

You must continuously reinvent work and business models rather than simply automate old tasks. Leading companies establish centers of excellence to build AI-driven customer tools and manage organizational transformation.

• Ensure deployment of AI for good and not harm as a publicly traded company.
• Comply with laws and regulations such as GDPR and the EU Artificial Intelligence Act.
• Think through possible unintended consequences that can create brand and public relations issues.

10 Key Ideas

1. Generative AI acts as a general-purpose technology that transforms cognitive tasks, automates routine labor, and unlocks trillions in economic value through foundation models.
2. Digital twins, collaborative robots, and generative design revolutionize industrial operations, supply chains, and manufacturing processes.
3. Artificial intelligence automates anti-money-laundering checks, optimizes insurance underwriting, and accelerates corporate recruitment and software development.
4. Generative chemistry and machine-learning platforms compress drug discovery timelines and uncover novel scientific knowledge from dark experimental data.
5. Front-office bots handle customer inquiries and retail scheduling while creating new branding challenges through platform disintermediation.
6. Deploying responsible AI requires specialized human roles including trainers, explainers, and sustainers to govern system behavior and maintain user trust.
7. Workplace augmentation operates through amplification, interaction, and embodiment to deliver exceptional productivity and growth.
8. Leaders must leverage the MELDS framework to break down jobs, conduct structured experiments, build a secure digital core, and enforce responsible AI governance.
9. Thriving in the missing middle requires mastering eight novel fusion skills that blend human judgment with machine capabilities.
10. You must continuously reinvent work and business models rather than simply automate old tasks.