Uncovering AI: How Machine Learning Works and Its Impact on Our Lives

Ossian Muscad
December 7, 2023
12:18 am
No Comments

Last Updated on December 7, 2023 by Ossian Muscad

Machine Learning revolves around the idea that systems can automatically learn and improve from experience. With this break from conventional programming paradigms, a fundamental change has occurred, enabling computers to analyze vast amounts of data, identify patterns, and make informed decisions autonomously. From recommendation systems in e-commerce to predictive analytics in finance, Machine Learning Works is now the driving force behind countless innovations in our daily lives.

This article will comprehensively understand how machine learning works by examining its essential ideas, main elements, and practical applications. By unraveling the complexities of machine learning, we aim to demystify this technological phenomenon and shed light on its transformative influence on various industries.

Understanding the Basics

Machine Learning is a branch of artificial intelligence that concentrates on creating systems that can learn from experience and get better at it. These systems employ algorithms instead of explicit programming to examine data, identify patterns, and make sound decisions. The fundamental characteristic of machine learning is its capacity to learn from data iteratively, allowing the system to perform better without requiring explicit programming for every task.

The method used to complete tasks is where traditional programming and machine learning diverge most. Precise instructions are given to carry out a particular task in conventional programming. This method works best for issues that have clear rules and definitions.

On the other hand, machine learning involves intentionally programming a model to be trained on data to make predictions or conclusions. Over time, the system improves performance by adjusting to new information and discovering patterns and relationships within the data. Given its adaptability, machine learning is a good fit for challenging issues where explicit programming might not be feasible. Three categories are used to describe the fundamental ideas of machine learning:

Algorithms: These are the mathematical operations or formulas that machine learning systems employ to complete a task. What a model learns from data and predicts is determined by its algorithms. Different algorithms exist, including reinforcement learning for decision-making, supervised learning for labeled data, and unsupervised learning for unlabeled data.
Models: A machine learning system’s understanding of the patterns in the data is represented by a model. The model modifies its parameters during training to reduce errors and produce precise predictions. A model’s quality is important since it directly affects how well the system performs with fresh, untested data.
Data: Machine learning relies on data as its base. It contains both the unknown data that the model predicts and the input data used to train it. The data’s relevance, volume, and quality heavily influence the machine learning system’s efficacy.

The Three Types of Machine Learning

For anyone exploring machine learning, it is essential to understand its various types. An in-depth comprehension of these principles provides a starting point for exploring more complex ideas and applications as the field advances in the near future.

Supervised Learning

A training dataset of input-output pairs is given to the algorithm as part of the learning procedure. The algorithm projects patterns to project new, uncategorized data using the labeled data as a starting point. During training, the algorithm adjusts its parameters iteratively to minimize the difference between its predictions and the actual labeled outputs. Supervised learning is further categorized into two main types:

Regression: In regression tasks, the algorithm learns to anticipate a value within a range for the continuous output variable. One example is predicting home values based on characteristics like location and square footage.
Classification: In classification tasks, the process places inputs into specified classes or labels, and the result is a categorical output variable. Examples are picture classification, in which an algorithm divides photos into different types or spam email detection.

Examples of Supervised Learning Applications

Gaining knowledge of these practical uses might help one appreciate the adaptability and influence of supervised learning in various fields. These applications demonstrate how machine learning can solve complicated problems and increase efficiency across a range of domains by utilizing labeled data.

Email Spam Detection: Building models that can differentiate between spam and non-spam emails based on labeled instances of each is accomplished through supervised learning.
Medical Diagnosis: In healthcare, supervised learning is utilized for medical diagnosis by training models on labeled datasets containing patient data and corresponding diagnoses. The model can then predict potential illnesses or conditions for new patients.
Handwriting Recognition: Optical character recognition (OCR) systems use supervised learning to recognize and convert handwritten or printed text into digital text. Training data includes examples of handwritten characters paired with their correct textual representations.
Credit Scoring: Financial institutions use supervised learning to assess credit risk. Models are trained on historical data that includes information about individuals’ credit histories and whether they defaulted on loans, enabling the algorithm to predict creditworthiness for new applicants.

Unsupervised Learning

Unsupervised learning is a category of machine learning where the algorithm is provided with unlabeled data and must find patterns, relationships, or structures within that data. Unlike supervised learning, no explicit output labels are guiding the learning process. Instead, the algorithm explores the inherent structure of the data on its own.

Examples of Unsupervised Learning Applications

By understanding unsupervised learning and its uses, one can gain insight into how it can be used to identify hidden patterns in data, giving academics and businesses access to insightful information without requiring labeled instances.

Clustering Customer Segmentation: Unsupervised learning is employed in marketing to identify distinct segments of customers based on their behavior, preferences, or purchasing patterns. This segmentation enables businesses to tailor marketing strategies to different customer groups.
Anomaly Detection in Cybersecurity: Unsupervised learning is used to identify unusual patterns or behaviors in network traffic that may indicate security threats or cyber-attacks. By learning the expected behavior of a system, the algorithm can detect deviations that may signify a security breach.
Topic Modeling in Text Data: Unsupervised learning techniques, such as Latent Dirichlet Allocation (LDA), are utilized to identify topics within large unstructured text data sets. This has applications in content recommendation, document organization, and sentiment analysis.
Recommendation Systems: Unsupervised learning contributes to recommendation systems by analyzing user behavior and preferences. By identifying patterns in user interactions and preferences, these systems can recommend products, movies, or content that align with individual user tastes.

Reinforcement Learning

Reinforcement learning is a machine learning paradigm where an agent learns to make decisions by interacting with an environment. The agent receives feedback in the form of rewards or penalties based on the actions it takes, and its objective is to maximize the cumulative reward over time. Unlike supervised learning, where explicit input-output pairs guide the learning process, reinforcement learning operates in an environment where the consequences of actions influence the learning strategy.

Reinforcement learning’s ability to learn from interactions and adapt to dynamic environments makes it well-suited for scenarios where explicit guidance is limited and optimal decision-making evolves. These real-world examples showcase the versatility and potential of reinforcement learning in various applications:

Self-Driving Cars: Reinforcement learning is employed in training self-driving cars to navigate real-world environments. The agent (car) learns optimal actions, such as steering and accelerating, by receiving rewards for safe and efficient driving and penalties for unsafe behavior.
Game Playing: Reinforcement learning has been notably successful in mastering complex games. AlphaGo, developed by DeepMind, used reinforcement learning to defeat human champions in the ancient board game Go. The agent learned strategies and tactics by playing against itself and improving based on outcomes.
Robot Manipulation: Reinforcement learning is applied in robotics to teach robots tasks such as grasping objects or walking. The robot receives feedback based on the success or failure of its actions, allowing it to adapt and refine its movements over time.
Energy-Efficient Systems: Reinforcement learning is utilized to optimize resource allocation and energy efficiency in smart grids. Agents learn to control energy consumption based on changing conditions and external factors.
Algorithmic Trading: Reinforcement learning is applied in algorithmic trading to optimize trading strategies. Agents learn to make buy or sell decisions based on market conditions, with rewards tied to profitable trades and penalties for losses.

The Machine Learning Process

The machine learning process systematically develops models that can make predictions or decisions without being explicitly programmed. This multifaceted process encompasses several key stages, each crucial in successfully implementing machine learning solutions.

Data Collection and Preprocessing

The foundation of any effective machine learning model lies in the quality of the data it is trained on. High-quality data is essential for accurate predictions and reliable insights. The significance of quality data can be summarized in the following aspects:

Accurate Representations: High-quality data ensures that the training dataset accurately represents the real-world scenarios the model is intended to handle. Biases or inaccuracies in the data can lead to skewed predictions and unreliable outcomes.
Generalization: Well-curated data enables the model to generalize to new, unseen instances. Models trained on diverse, representative data are more likely to perform effectively in various situations.
Reduced Noise: Clean and well-organized data minimizes noise and irrelevant information, allowing the model to focus on relevant patterns and relationships. This, in turn, enhances the model’s predictive accuracy.

Data preprocessing is a critical step that involves transforming raw data into a format suitable for machine learning models. The goal is to enhance the quality and usability of the data. Common techniques include:

Handling Missing Values: Strategies for dealing with missing data, such as imputation or removal, are employed to ensure completeness and reliability.
Normalization and Scaling: Normalizing or scaling features helps bring all variables to a standard scale, preventing certain features from dominating the learning process due to differences in scale.
Encoding Categorical Variables: Many machine learning algorithms must convert categorical variables into numerical representations. Techniques like one-hot encoding are commonly used for this purpose.
Handling Outliers: Identifying and treating outliers prevents them from unduly influencing the model’s training. Techniques include trimming or transforming extreme values.
Feature Engineering: Creating or transforming new features can enhance the model’s ability to capture relevant patterns. This involves extracting meaningful information from the available data.
Data Splitting: Dividing the dataset into training and testing sets allows for evaluating model performance on unseen data, ensuring its generalizability.
Addressing Imbalance: In classification tasks, where classes may be unevenly distributed, techniques such as oversampling or undersampling can be applied to balance the dataset.

Model Training

Selecting an appropriate machine learning algorithm is a critical decision that significantly influences the model’s performance. Different algorithms are suited to different types of tasks and data. Key considerations in algorithm selection include:

Nature of the Problem: Whether the task is classification, regression, clustering, or reinforcement learning guides the choice of algorithms. For instance, decision trees are often suitable for classification tasks, while linear regression is used for regression problems.
Data Characteristics: The dataset’s characteristics, such as its size, dimensionality, and the presence of categorical or numerical features, impact the choice of algorithm. Some algorithms perform better with high-dimensional data, while others are more effective in low-dimensional spaces.
Interpretability: Depending on the application, the interpretability of the model may be crucial. Decision trees and linear models are often more interpretable, while complex models like neural networks may provide superior performance at the cost of interpretability.
Computational Resources: The availability of computational resources, including processing power and memory, influences the choice of algorithms. Some algorithms may be computationally expensive and impractical for specific environments.

Once an algorithm is chosen, the model is trained using labeled data. Labeled data consists of input-output pairs, where the algorithm learns to map inputs to corresponding outputs. The training process involves the following steps:

Initialization: Model parameters are often randomly initialized to start the learning process.
Forward Propagation: Input data is fed into the model, and predictions are made based on the current parameters.
Loss Calculation: The difference between the model’s predictions and the actual labels is quantified using a loss function. The goal is to minimize this loss.
Backward Propagation: The model adjusts its parameters to minimize the loss. This is achieved through gradient descent or similar optimization techniques, where the loss gradient with respect to the parameters is calculated and used to update the model.
Iteration: Steps 2-4 are repeated iteratively for multiple epochs until the model converges or reaches a satisfactory level of performance.
Validation: The trained model is evaluated on a separate validation set to assess its generalization to new, unseen data. This helps identify potential overfitting or underfitting issues.
Hyperparameter Tuning: Fine-tuning of hyperparameters, such as learning rate or regularization strength, may be performed to optimize the model’s performance.

Evaluation and Fine-Tuning

After training the machine learning model, evaluating its performance is crucial to ensure its effectiveness in real-world applications. Key metrics and techniques for assessing model performance include:

Accuracy: The proportion of correctly predicted instances among the total cases is a common metric for classification tasks.
Precision and Recall: Especially relevant for imbalanced datasets, precision measures the accuracy of positive predictions, while recall assesses the model’s ability to capture all relevant instances.
F1 Score: The harmonic mean of precision and recall provides a balanced measure of a model’s performance.
Mean Squared Error (MSE): Commonly used for regression tasks, MSE quantifies the average squared difference between predicted and actual values.
Confusion Matrix: A tabular representation of true positive, true negative, false positive, and false negative instances, aiding in understanding model errors.
Receiver Operating Characteristic (ROC) Curve: Evaluates the trade-off between true and false favorable rates, particularly in binary classification.
Area Under the Curve (AUC-ROC): Quantifies the overall performance of a classification model across different thresholds.

Fine-tuning is an iterative process that involves experimenting with different configurations and assessing their impact on performance metrics. The goal is to achieve a well-optimized model that generalizes effectively to new, unseen data and aligns with the desired objectives of the machine learning task.

Hyperparameter Tuning: Modifying hyperparameters such as learning rate, regularization strength, or the number of layers in a neural network can significantly impact performance.
Feature Engineering: Refining input features or creating new features can enhance the model’s ability to capture relevant patterns.
Data Augmentation: Introducing variations in the training data, such as rotating images or adding noise, can improve the model’s robustness and generalization.
Ensemble Methods: Combining predictions from multiple models through ensemble methods, such as bagging or boosting, can enhance overall performance.
Cross-Validation: Using techniques like k-fold cross-validation helps obtain a more robust assessment of the model’s performance by training and evaluating it on different subsets of the data.
Regularization: Applying regularization techniques, such as L1 or L2 regularization, helps prevent overfitting by penalizing overly complex models.
Model Architecture: Adjusting the model’s architecture, such as adding layers or nodes in a neural network, can impact its capacity to learn complex relationships.

Impact on Industries

Integrating machine learning technologies has brought about transformative changes across various industries, revolutionizing how businesses operate and make decisions. The impact of machine learning extends to sectors ranging from healthcare and finance to manufacturing and beyond.

Healthcare

The integration of machine learning in healthcare has ushered in a new era of precision medicine, significantly impacting disease diagnosis, treatment, and drug discovery. Here are two key areas where machine learning is making a substantial impact:

Disease diagnosis and treatment

Machine learning algorithms are invaluable in disease diagnosis and personalized treatment plans, offering healthcare professionals powerful tools to enhance decision-making. Key contributions include:

Medical Imaging Analysis: ML excels in analyzing complex medical images, such as MRIs, CT scans, and X-rays. Algorithms can detect subtle patterns or anomalies that might be challenging for the human eye, aiding in early and accurate diagnosis.
Predictive Analytics: By leveraging patient data, machine learning models predict the likelihood of diseases or complications. This enables healthcare providers to proactively intervene, leading to preventive measures and improved patient outcomes.
Treatment Personalization: ML algorithms analyze patient data, including genetic information, to tailor treatment plans based on individual responses. This approach, known as precision medicine, optimizes therapeutic outcomes and minimizes adverse effects.

Drug discovery

The drug discovery process is known for its complexity and time-intensive nature. Machine learning accelerates this process by leveraging computational methods to analyze vast datasets and identify potential drug candidates. Key contributions include:

Target Identification: ML models analyze biological data to identify potential drug targets, such as specific proteins or disease-related genes. This knowledge guides researchers in developing targeted therapies.
Compound Screening: ML algorithms assist in the virtual screening of chemical compounds, predicting their potential efficacy and safety. This accelerates the identification of promising drug candidates while minimizing the need for extensive laboratory testing.
Drug Repurposing: ML techniques analyze existing drugs and their interactions, uncovering opportunities for repurposing drugs for new therapeutic applications. This approach can significantly reduce the time and resources needed for drug development.

Finance

The financial sector has undergone a significant transformation with the integration of machine learning, introducing advanced tools for fraud detection and algorithmic trading. Here’s a detailed look at how machine learning is reshaping the landscape of finance:

Fraud Detection

Machine learning plays a pivotal role in fortifying the security measures of financial institutions by employing sophisticated algorithms for fraud detection. Key contributions include:

Anomaly Detection: Machine learning models analyze patterns in transaction data to identify anomalies that deviate from typical customer behavior. Unusual transaction amounts, locations, or frequencies trigger alerts, allowing rapid intervention.
Behavioral Analysis: ML algorithms establish baseline behavior profiles for individual account holders. Any deviation from these profiles, such as unexpected spending patterns or unusual transactions, raises red flags for potential fraudulent activity.
Predictive Modeling: By leveraging historical data, machine learning models predict the likelihood of fraudulent transactions. This proactive approach allows financial institutions to take preventive measures before fraudulent transactions occur.

Algorithmic Trading

Algorithmic trading, enabled by machine learning, has revolutionized the financial markets, automating the execution of complex trading strategies and decision-making processes. Key contributions include:

Market Analysis: Machine learning models analyze vast datasets of market prices, trading volumes, and other relevant factors to identify patterns and trends. This analysis informs trading algorithms about potential market movements.
Predictive Analytics: ML algorithms predict price movements and market trends based on historical data, news sentiment, and macroeconomic indicators. This predictive capability enables algorithms to make informed decisions in real time.
High-Frequency Trading (HFT): Machine learning facilitates high-frequency trading by executing many trades at extremely high speeds. Algorithms can exploit arbitrage opportunities and respond to market fluctuations faster than human traders.

Education

Incorporating machine learning in education is reshaping traditional teaching methods, introducing personalized learning experiences and automated grading systems. Here’s an in-depth exploration of how machine learning is impacting the field of education:

Personalized Learning

Machine learning is revolutionizing education by tailoring learning experiences to individual student’s unique needs and preferences. Key contributions include:

Adaptive Learning Platforms: Machine learning algorithms analyze students’ performance, learning styles, and preferences to adapt educational content dynamically. This ensures that each student receives a customized learning path suited to their strengths and weaknesses.
Intelligent Tutoring Systems: ML-powered tutoring systems provide real-time feedback and assistance to students. These systems adapt to the student’s pace of learning, offering additional explanations or practice exercises when needed and advancing to more challenging content when appropriate.
Personalized Content Recommendations: Recommender systems powered by machine learning suggest learning materials, resources, and activities based on a student’s learning history and preferences. This enhances engagement and encourages self-directed learning.

Automated Grading Systems

Machine learning automates the grading process, providing educators with efficient tools for assessment and feedback. Key contributions include:

Grading Automation: ML algorithms analyze students’ responses to assignments and assessments, automating the grading process. This saves educators time, ensures consistency, and reduces grading biases.
Natural Language Processing (NLP): Machine learning. NLP is used to assess written assignments, essays, and open-ended questions. Algorithms can evaluate the quality of written responses, providing detailed feedback on language usage, structure, and content.
Pattern Recognition: ML models recognize patterns in correct and incorrect answers, adapting to variations in student responses. This enables automated grading systems to handle diverse inputs and assess complex problem-solving skills.

Ethical Considerations

Integrating machine learning technologies brings about transformative opportunities but raises important ethical considerations that must be carefully addressed. As we delve into the ethical landscape of machine learning, several key aspects come to the forefront:

Bias in Machine Learning

Despite their great capacity, machine learning algorithms can be biased with the data they were programmed on. Resolving and reducing algorithmic bias is essential to the development of ethical AI. Let’s examine methods for reducing algorithmic bias and delve into its complexities:

Understanding algorithmic bias

When systematic and unjust discrimination appears in the results generated by machine learning algorithms, it is referred to as algorithmic bias. This bias may take many different forms, such as socioeconomic, racial, or gender biases.

Training Data: Historical data reflecting preexisting social biases and discrimination can often be the source of bias. The model may reinforce or intensify established biases if the training data has been skewed or unbalanced.
Feature Selection: Additionally, inaccurate feature selection may factor in algorithmic biases. The model may produce skewed predictions if some features are improperly highlighted or disregarded.

Addressing bias in AI systems

To ensure fairness and plurality, addressing algorithmic bias needs a multifaceted strategy that takes both technological and ethical factors into account:

Data Collection: Aim for representative and diversified datasets that accurately capture the range of racial and cultural characteristics of the population the model will affect.
Data Audits: Audit training data frequently to find and fix biases. This means assessing how data are distributed among various demographic categories.
Define Fairness Metrics: Provide precise indicators to ensure fairness in model results. This involves establishing the fairness standards and assessing the model’s performance in relation to these criteria.

Privacy Concerns

The ethical application of machine learning requires monitoring data protection and balancing data utility and privacy. Here is an in-depth look at privacy concerns:

Data protection in machine learning

Machine learning concerns regarding privacy are centered on handling sensitive data responsibly, and this calls for careful consideration at every stage of the project’s life cycle:

Data Minimization: To mitigate the risk of privacy violations, only acquire the minimum amount of data required for the intended purpose.
Anonymization and Pseudonymization: Use anonymity to preserve the dataset’s usefulness while protecting individual identities.
Secure Data Storage: Set robust safety safeguards for the storage and management of datasets to stop illegal access and data breaches.
Data Access Controls: Restrict access to sensitive data, allowing only authorized personnel to handle and process information.
Informed Consent: Get the consent of those whose data is used in machine learning projects clearly and informally, outlining the intended use of their data.
Data Encryption: Protect data while in transit and at rest by using encryption techniques to ensure it stays private even if it gets intercepted.
Regular Audits: Conduct routine audits of data processing procedures to find and fix any vulnerabilities and maintain continued adherence to privacy regulations.

Balancing data utility and privacy

Finding a way to extract valuable insights from machine learning models while upholding people’s right to privacy involves establishing a balance between data utility and privacy:

Federated Learning: Distribute model training across decentralized devices to avoid centralized data storage. This approach maintains privacy by keeping data localized.
Privacy Impact Assessments: To weigh the advantages and disadvantages of machine learning initiatives, conduct privacy impact assessments, and ensure privacy concerns are incorporated into the planning and execution.
Ethical Use of AI: Clearly define the rules and regulations for the ethical use of AI, focusing on treating personal information with respect and responsibility.
User Control and Transparency: Give users the authority to decide what happens to their data by giving them clear descriptions of how it will be used, opt-out procedures, and alternatives for data deletion.
Algorithmic Transparency: Attempt to improve algorithmic transparency for machine learning so that people can see how their data is used and decisions are made.

Future Trends in Machine Learning

The field of machine learning is ever-changing owing to technological breakthroughs and the growing need for intelligent solutions. The following trends will influence how machine learning develops in the future:

Advancements in Deep Learning

Artificial intelligence (AI) has grown in recent years due to the tremendous improvements in deep learning. Deep learning has seen unparalleled success in tasks like speech recognition, image recognition, and natural language processing with the development of deep neural networks and complex architectures like convolutional and recurrent neural networks.

These developments have boosted AI systems’ precision and effectiveness while opening the door for difficult problem-solving and judgment breakthroughs. As researchers continue to explore innovative architectures and training methodologies, the future of deep learning holds promising opportunities, with potential applications ranging from healthcare and finance to autonomous systems and beyond.

Integration of AI in Everyday Devices

Artificial intelligence is finding its way into increasingly prevalent services, influencing how we use technology daily. These days, AI algorithms are used by wearables, home appliances, and smartphones to improve user experiences, automate repetitive chores, and provide tailored recommendations.

Natural language processing and machine learning-powered voice assistants are already widely used, providing consumers with easy-to-use interfaces. Incorporating artificial intelligence (AI) into daily items not only improves convenience but also paves the way for a more intelligent and connected future in which technology will quickly adjust to the requirements and preferences of each individual.

Emerging Applications and Industries

The impact of machine learning transcends conventional fields, resulting in the creation of new applications and domains that capitalize on its possibilities. From personalized medicine and predictive maintenance in manufacturing to precision agriculture and climate modeling, machine learning drives innovation in various industries.

The ability of ML algorithms to uncover patterns, make predictions, and automate decision-making has found applications in finance, education, and cybersecurity, among others. As machine learning develops, we see technology being incorporated into unexplored industries, providing new solutions to challenging problems and redefining how we approach many aspects of our lives.

Frequently Asked Questions (FAQs)

Q1: What distinguishes machine learning from artificial intelligence?

Artificial intelligence (AI) is a broader category that includes devices or systems capable of carrying out operations that generally call for human intelligence. A subset of artificial intelligence called machine learning concerns how well machines can learn from data.

Q2: In machine learning, what does overfitting mean?

When a machine learning model learns the training data too well, it is said to be overfitting because it captures noise or unimportant patterns that do not transfer to fresh, unobserved data. This can lead to poor performance on new datasets.

Q3: Is machine learning limited to math and programming experts?

Although having a background in mathematics and programming is advantageous, people from various backgrounds can work with machine learning principles thanks to the availability of user-friendly tools and platforms. Many instructional resources are also accessible to study machine learning from the ground up.

Take Advantage of Machine Learning with DATAMYTE

DATAMYTE is a quality management platform with low-code capabilities. Our Digital Clipboard, in particular, is a low-code workflow automation software that features a workflow, checklist, and smart form builder. This tool also incorporates machine learning capabilities that allow you to automate your day-to-day activities effectively.

DATAMYTE also lets you conduct layered process audits, a high-frequency evaluation of critical process steps, focusing on areas with the highest failure risk or non-compliance. Conducting LPA with DATAMYTE lets you effectively identify and correct potential defects before they become major quality issues.

With DATAMYTE, you have an all-in-one solution for embracing Machine Learning Technologies in your organization. Book a demo now to learn more.

Conclusion

Understanding the intricacies of machine learning is not merely a technical need; it holds the key to unleashing artificial intelligence’s full potential. It also empowers systems to evolve, adapt, and refine their capabilities over time, ushering in an era where applications can learn, mirroring the complexity of human cognition.

The future of machine learning looks filled with even more breakthroughs as we find ourselves at the intersection of innovation and ethical considerations. It is a journey that invites continuous exploration and learning, ensuring that machine learning’s development is consistent with the principles of our society.