Artificial Intelligence - Book review
1 July 2020
Dr. Moria Levy
The book "Artificial Intelligence" is a compilation of essays from the Harvard Business Review series, initially published in 2019. It encompasses various crucial aspects related to artificial intelligence, summarized as follows:
1. Definition and Explanation of Artificial Intelligence
2. Main Applications and Uses of AI
3. Practical Guidelines for Implementing AI
4. Controlling AI Systems
5. The Role of Humanity in the Age of AI
6. The Impact of AI on the Labor Market
7. Predictions on the Advancement of AI
8. List of Articles and Authors
Undoubtedly, Harvard has effectively presented essential and cutting-edge knowledge in a clear and easily understandable manner. The subject matter holds significant importance, and it is increasingly necessary for managers and knowledgeable employees to begin learning and preparing for AI implementation. Organizations that delay embracing this technology may face severe consequences in terms of competitiveness in the future.
In conclusion, this book is highly recommended for those interested in exploring the realm of artificial intelligence and its impact on business and society.
Definition and Explanation of Artificial Intelligence
Artificial intelligence is defined as the intelligence displayed by machines, as stated in the book "Beginning After..." (Wikipedia definition). Undoubtedly, AI stands out as our era's most significant general technology. Different individuals perceive it in various ways, and each perspective holds some truth:
• Managers view AI as a breakthrough technology with disruptive potential.
• Employees fear AI because they believe it might jeopardize their ability to work.
• Consultants promote AI as a universal remedy for various challenges.
• The media portrays AI with a blend of hype and admiration.
The term "artificial intelligence," coined in 1956, has undergone various transformations and advancements over the years. Recently, progress has accelerated with a shift towards reliance on neural network algorithms, commonly called "machine learning" or "deep learning." This shift has led to remarkable achievements, such as machines surpassing human capabilities in tasks like face recognition since 2015.
It is essential to grasp that the current approach to AI differs from what we have been accustomed to for years. Machines no longer seek to imitate human thought processes; instead, they employ pattern-based learning. These machines examine vast databases, learn from examples, and discern what aligns with the defined objectives (e.g., recognizing a cat). This learning process relies on algorithms, statistics, and databases. AI learns what we already know and uncovers hidden insights, offering a window into unexplored aspects we could never have considered.
Today, organizations do not necessarily need to develop everything from scratch. They can adopt pre-existing algorithms from tech giants like Google, Amazon, or Apple and adapt them to address specific business problems. However, despite the enthusiasm surrounding AI, its applicability remains relatively limited and only encompasses some areas of our lives. In this context, human involvement still plays a vital role, which will be discussed further in a separate chapter later in the summary.
Main Applications and Uses of AI
Artificial intelligence has made significant advancements in various areas in recent years. Some of the main applications include:
1. Comprehension: AI excels in image recognition, natural language analysis, and understanding emotions.
2. Cognition (Decision Making): AI solves complex business problems, from games like GO to optimizing server room operations for efficient cooling and electricity savings.
The current market applications are built upon these capabilities. For instance, self-driving cars rely on AI's ability to perceive and comprehend the surroundings, enabling optimal decision-making regarding speed, turns, stops, and more.
Here are some examples of AI applications already in use:
1. Speech recognition for call centers and phone-based personal assistants.
2. AI-powered bots are trading on the stock exchange.
3. Tagging photos.
4. Drug development.
5. Approval of financial requests, such as loans and credit.
6. Fraud detection, particularly in finance and security domains.
7. Intelligent traffic lights to enhance traffic flow.
8. Facial and people recognition in photos.
9. Automated assembly of products like cars.
10. Product design recommendations based on materials, price, and performance.
11. Software project analysis to estimate development costs.
12. Public safety maintenance, identifying crowd movements through cameras.
13. Screening job candidates.
14. Customer service, responding to inquiries, and redirecting complex ones to humans.
15. Casino activity monitoring.
16. Recommendations for machine maintenance based on performance and operation mode.
17. Identification of medical risks based on patient conditions.
18. Wearable AI suggests tailored activities on amusement ships.
19. Wearable intelligence monitoring chronically ill patients (e.g., Parkinson's disease).
20. Personalized recommendations for clothing and fashion items.
21. Emotion-mediating eyeglasses for autistic individuals.
22. Providing support for those experiencing ongoing trauma.
These examples illustrate the diverse and ever-expanding range of applications to which AI technology contributes across various industries and aspects of daily life.
Practical Guidelines for Implementing AI
• As an organization, the ideal time to step into the world of artificial intelligence is now, or even yesterday, as suggested by one of the authors. The technologies have matured enough to be harnessed, but it will take time to realize the benefits within the organization and align its structure and processes accordingly. Delaying entry may result in competitors gaining an advantage.
• Before commencing, individuals should have a basic understanding of AI, its workings at a high level (without delving into complex mathematical details), areas where successful implementations are evident, and areas where caution is needed (such as those involving complex ethical issues).
• Appointing a leader and a supporting team is crucial. Initially, a team for testing and selection (comprising business experts and data-savvy personnel) is needed, followed by a team for full implementation (involving algorithm and computing experts, data scientists, and knowledge engineers). Collaborating with experienced external consultants/partners is also recommended.
• Choosing a pilot project is essential. Opt for 1-2 smaller projects meeting specific criteria:
- Enable quick wins
- Address problems of moderate complexity related to the organization's core content.
- Create value
A good starting point is to focus on automating tasks people perform.
• Verify compatibility by assessing whether a simple algorithm could provide a viable alternative. Jumping into complex AI solutions solely because they are possible is only sometimes advisable.
• Gaining organizational support from managers is vital. Ensuring they understand the potential value created by a successful AI project is essential.
Coaching (Machine Learning and Examples):
• The machine learning process may start slow but later picks up pace and stabilizes.
• It is essential to anticipate failures, as they are part of both the solution development and the machine's learning process. Human supervision is necessary.
• Evaluating the machine's success should always be based on whether it aligns with the customer's business goals rather than solely assessing algorithm performance.
• In the operational phase, new and current data are required, distinct from the data used during the training phase. This data is essential for decision-making and continually improving the learning process, as the environment is constantly changing.
• A separate chapter deals with control and will be described below.
• Celebrate successes at the end of the pilot project when significant milestones are achieved during the process.
Note: Apart from the technology-specific training phase, the stages of an AI project are similar to those of other subjects, such as knowledge management or machine learning.
Controlling AI Systems
As previously mentioned, machines are not flawless, and there are three risks associated with their functioning:
1. Machines can exhibit biases based on the content they learned when making decisions initially. These biases might reflect racial, gender, or other ethical considerations.
2. Machines rely on statistics and may struggle to handle edge cases. In critical scenarios, such as nuclear facilities or safety decisions, this poses a risk.
3. Detecting and correcting machine mistakes (which are likely to occur at some point) can be highly challenging.
Given the possibility of errors in machine operations, ongoing control measures are necessary, including:
• Control of input information, restricting it to validated data.
• Testing algorithms for typical biases related to gender, race, age, and other content-based factors.
• Proactively searching for potential failure scenarios and devising workarounds.
• Having a contingency plan for a less "smart" response in case of system failure.
• Establishing a prepared channeling strategy to address failures during the operational phase, including issuing apologies if needed.
The Role of Humanity in the Age of AI
The book emphasizes the importance of human involvement in the AI process. This role extends beyond algorithm specialists, data scientists, and software workers. People play a crucial role in various aspects:
1. In processes where AI cannot provide a complete solution:
- Complex tasks involving diverse functions (e.g., marketing).
- Tasks requiring creativity (e.g., brainstorming, sales meetings, system characterization).
- Tasks involving emotion-based decision-making (e.g., handling at-risk customers).
2. In processes where AI offers a partial solution, human-machine collaboration is vital:
- Processes comprising mechanized and delicate tasks (e.g., assembling luxury cars).
- Processes involving routine responses by machines, with exceptions handled by humans (e.g., customer service).
3. In processes fully leveraging AI, people are required for various critical functions:
- Defining the business problem and solution.
- Making decisions regarding data usage.
- Defining data requirements and ensuring data quality.
- Providing initial guidance and understanding of desired goals.
- Training the AI system.
- Explaining AI-generated outputs in especially crucial fields like law and medicine where explanations are essential.
- Monitoring mechanized activity.
- Addressing issues when failures occur.
To facilitate optimal integration between people and machines, organizations are advised to follow these recommendations:
1. Redefine organizational processes with a focus on long-term goals and aspirations.
2. Involve employees extensively, drawing on their expertise to determine what will be effective.
3. Formulate a clear AI strategy for the organization.
4. Define clear responsibilities for data collection.
5. Redesign work processes to incorporate AI capabilities alongside employee skills seamlessly.
By following these guidelines, organizations can harness the potential of AI while ensuring that human contributions remain an essential and strategic component of their operations.
The Impact of AI on the Labor Market
Predictions labor market vary, and five possible scenarios are presented:
1. The labor market will become a battleground between machines and humans, with machines eventually emerging victorious.
2. Intelligent machines will gradually take on more tasks, leading to overall economic well-being but reduced demand for human labor.
3. Machines will enhance labor productivity, resulting in economic growth, but the benefits will be unevenly distributed. People will need to focus on continuous education and invest in technology.
4. Despite the advanced capabilities of machines, productivity improvement will be only marginal.
5. Productivity gains will occur in specific sectors and successful companies while new professions emerge. However, manual workers may face challenges transitioning to these new roles. Research and planning are required to address this issue.
Recommendations for managers to prepare for the future labor market include:
1. Leveraging technologies to automate routine tasks and achieve cost savings. Redefining the organization's operating model is essential.
2. Identifying opportunities for redefining roles by determining which tasks can be automated, integrated, or performed solely by humans. The organizational structure should be adapted accordingly.
3. Engaging employees as partners in intelligent enterprise planning, involving them in decision-making processes related to integrating AI and automation.
Predictions on the Advancement of AI
Despite the exciting AI capabilities already in use, laboratories are actively working on the next generation of solutions, which will encompass:
1. Emotions: Advancing AI to recognize, process, and even imitate human expressions.
2. Accuracy: Continuing efforts to enhance predictive accuracy in understanding people's desires and preferences, leading to innovative sales business models (e.g., shipping items deemed desirable to your home and avoiding unwanted ones).
3. Data: Developing AI that can achieve comparable results using less data, reducing costs, and expanding the application of AI to various business problems and other domains.
4. Conceptual comprehension: Improving AI's ability to comprehend concepts, facilitating more advanced learning stages.
5. Handling edge cases and exceptions: Addressing challenges from unusual situations or circumstances.
6. Common sense: Enhancing AI's capability to apply logical reasoning to objects, places, and actors, including humans.
7. Improved assessment: Enabling better decision-making under conditions of uncertainty.
The reality of AI's progress surprises us repeatedly, and its continuous advancement is indeed admirable! (M.L.)