Use AI Responsibly
Recognize data bias and harms
Responsible AI means using artificial intelligence ethically, safely, and fairly to help people without causing harm.
Key Points
1. Understand AI’s limitations: AI can complete tasks quickly, but it can make mistakes and may produce biased or inaccurate information.
2. Combine AI with human judgment: AI can help with writing, research, and organizing information, but humans must use critical thinking, creativity, compassion, and logic to make important decisions.
3. Recognize AI bias: AI may treat people unfairly because of biases in its training data. Humans must check AI results to ensure fairness.
4. Protect privacy and security: Avoid sharing sensitive personal information or confidential data with AI tools unless appropriate safeguards are in place.
5. Prevent harm: Always review AI-generated information before using it, especially when decisions affect people’s lives.
Real-Life Example: AI in Hiring
A company uses AI to screen job applications and shortlist candidates.
AI’s role: Quickly reviews resumes and identifies potential candidates.
Human’s role: Checks the results, identifies possible bias, and makes the final hiring decision.
Responsible approach: The company ensures that all candidates are evaluated fairly and that AI does not discriminate against anyone.
Easy Way to Remember
Think of AI as an airplane autopilot. It can help perform routine tasks, but a human pilot must supervise it and take control when important decisions are needed.
Understanding Bias in AI
Artificial intelligence (AI) is becoming an important part of our daily lives. It helps us complete tasks faster, solve problems, analyze information, and make better decisions. For example, AI can help self-driving cars recognize pedestrians, assist doctors in identifying possible health conditions, and recommend useful content based on our interests. However, AI is not always accurate or fair. Sometimes, it produces biased results because of the data it learns from or the way it is designed.
AI Benefits and Limitations
AI offers many benefits, including saving time, improving productivity, and helping people make informed decisions. However, it also has limitations. AI systems can make mistakes, provide misleading information, or treat certain groups of people unfairly. This is why we should not automatically trust every result generated by AI. Human judgment and careful evaluation are necessary to ensure that AI is used appropriately.
Systemic bias refers to unfairness that already exists within society or institutions. When AI learns from historical information containing these inequalities, it may continue to produce similar unfair outcomes. For example, if a company has historically hired more men than women for leadership positions, an AI hiring tool trained on those records might favor male applicants.
Data bias occurs when the information used to train an AI system is incomplete, unrepresentative, or influenced by prejudice. For instance, an AI image generator might repeatedly show men as engineers and women as nurses because its training data contains similar stereotypes. This can reinforce unfair assumptions about people’s abilities and career choices.
Values in AI Design
AI systems are created by people, and their design often reflects the goals, priorities, and values of their developers. This means AI is not always completely neutral. The choices developers make about what information to include, which outcomes to prioritize, and how a system should respond can influence its behavior.
For example, a developer might create an AI tool that encourages people to choose environmentally friendly products. The tool reflects a priority for sustainability through the way it recommends products. Although this goal may be beneficial, it is still important to consider whether the recommendations are accurate, transparent, and fair to users and businesses.
Using AI Responsibly
Using AI responsibly means understanding its limitations, recognizing possible biases, and thinking carefully before accepting its recommendations. We should question AI-generated results, check important information against reliable sources, and consider whether the outcomes could negatively affect individuals or communities.
Human involvement is particularly important when AI is used for decisions about employment, healthcare, education, or access to essential services. AI can support decision-making, but people must remain responsible for evaluating its recommendations and addressing potential harm.
Real-Life Example: AI in Hiring
Imagine a company that uses AI to review hundreds of job applications. The system analyzes applicants’ qualifications and compares them with records of employees who were successful in the past. However, if those historical records mainly represent one gender or a narrow educational background, the AI may favor similar candidates and overlook other qualified applicants.
To address this problem, the company should evaluate the quality of its training data, test the system for unfair outcomes, and ensure that qualified candidates are assessed using relevant criteria. Human recruiters should also review the recommendations rather than relying entirely on the AI system.
This example shows how historical inequalities and data bias can affect AI decisions. It also demonstrates why human oversight is essential for creating a fairer hiring process.
Identifying AI Harms
Artificial intelligence (AI) can make our lives easier by helping us work faster, solve problems, and access information. However, when AI systems are poorly designed, trained on biased data, or used without proper supervision, they can also cause harm to individuals and society. Understanding these risks is an important part of using AI responsibly.
Types of AI Harms
AI can cause different types of harm depending on how it is designed and used. One important type is allocative harm, which occurs when an AI system unfairly denies someone an opportunity or resource. For example, an AI recruitment tool might reject a qualified job applicant because it relies on biased historical hiring data. Similarly, an automated housing system might incorrectly reject a person’s application because of inaccurate background information. These decisions can affect people’s livelihoods and wellbeing.
Another type is quality-of-service harm. This happens when an AI system works well for some people but performs poorly for others. For example, a voice recognition tool might understand some speakers accurately but struggle with people who have speech disabilities, strong accents, or less commonly represented speech patterns. As a result, those users may not receive the same quality of service as others.
Social and Representational Harms
Representational harm occurs when AI reinforces stereotypes, misrepresents certain groups, or contributes to unfair social attitudes. For example, an AI image generator might frequently portray men as business leaders and women as assistants. Such results can strengthen existing stereotypes and influence how people think about different groups.
Social system harm refers to the wider negative effects AI can have on society. One example is the use of AI-generated deepfake videos or fake audio recordings to spread misinformation. If people believe these fabricated materials are genuine, they may make decisions based on false information. This can damage public trust, influence elections, and create confusion within communities.
Interpersonal Harm and Privacy
Interpersonal harm occurs when AI is used in ways that negatively affect individuals or their relationships with others. For example, someone might use AI to create a fake image of another person, spread misleading messages about them, or monitor their activities without permission. These actions can damage reputations, violate privacy, and cause emotional distress.
Protecting personal information is therefore an important part of responsible AI use. People should be careful about sharing sensitive information with AI tools and should avoid using AI to deceive, harass, or exploit others.
Using AI Responsibly to Prevent Harm
Preventing AI harm requires cooperation between developers, organizations, and users. AI systems should be tested regularly to identify errors, unfair outcomes, and weaknesses. Developers should use diverse and representative training data and improve their systems when problems are discovered.
Human oversight is equally important. People should review AI recommendations before making important decisions, especially in areas such as employment, healthcare, education, and housing. Organizations should also provide ways for people to question or appeal decisions that affect them.
By understanding the different types of AI harm, we can make better decisions about when and how to use artificial intelligence. The goal is not to avoid AI completely, but to ensure that it is developed and used in ways that respect people’s rights, protect their privacy, and promote fairness.
My Path to Working in Responsible AI
Artificial intelligence is becoming an important part of our everyday lives, from social media and online shopping to healthcare and education. However, developing and using AI responsibly requires more than technical knowledge. It also requires understanding fairness, inclusion, bias, and the impact AI can have on people and society.
Roles and Responsibilities in AI
Working in responsible AI involves helping ensure that AI systems are developed and used in ways that benefit people without causing unnecessary harm. For example, program managers can coordinate developers, researchers, designers, and product teams to ensure that responsible AI practices are considered throughout a project’s development.
They also need to understand how AI decisions may affect different communities. If an AI system reinforces existing inequalities or excludes certain groups, it can create serious problems. Responsible AI professionals work with teams to identify these risks, establish appropriate safeguards, and ensure that products are developed thoughtfully and fairly.
Understanding Bias and Representation in AI
AI systems learn patterns from data collected from the real world. Because this data reflects human experiences, it may also contain stereotypes, gaps, and historical inequalities. When certain groups are underrepresented, an AI system may perform well for some people but poorly for others.
For example, imagine a facial recognition system trained mostly on images of people with lighter skin tones. The system might identify those individuals accurately but struggle to recognize people with darker skin tones. This creates an unfair experience for users who were not adequately represented in the training data.
To reduce this problem, developers need to evaluate their data carefully and include a broad range of people, experiences, and backgrounds. They should also test AI systems across different groups to identify differences in accuracy and correct problems where possible.
The Importance of Inclusion in AI
Inclusion means ensuring that AI systems consider the needs and experiences of different people rather than being designed for only a limited group of users.
Think of preparing a meal for a group of people. If you consider only one person’s preferences, the meal may not suit everyone. Similarly, if an AI system is developed using information from only one group, it may not work equally well for all users.
Inclusive AI development involves asking important questions: Who is represented in the data? Whose experiences are missing? Could the system disadvantage a particular group? These questions help developers create technology that is more useful, accessible, and fair.
My Responsibility as an AI User
Responsible AI is not only the responsibility of developers and large technology companies. Users also play an important role in ensuring that AI is used appropriately.
Whenever I use an AI tool, I should think critically about the information it provides instead of assuming every answer is accurate or unbiased. I should verify important facts, consider whether the results are fair, and avoid relying entirely on AI for decisions that require human judgment.
Providing feedback is another valuable way to contribute. If an AI tool produces an incorrect, misleading, or unfair result, reporting the problem can help its developers understand where improvements are needed. User feedback, combined with careful testing and evaluation, can contribute to better AI systems.
Real-Life Example: Facial Recognition on Social Media
Many social media platforms use AI to recognize faces and suggest tags in photographs. This feature can save time, but its accuracy depends partly on the quality and diversity of its training data.
If certain skin tones or facial characteristics are poorly represented, the system may incorrectly identify some people or fail to recognize them. This can create frustration and raise concerns about fairness and privacy.
To address these issues, developers should test their systems using diverse datasets, measure accuracy across different groups, and improve performance where weaknesses are identified. Users can also report errors and review their privacy settings before enabling facial recognition features.
This example shows that responsible AI requires cooperation between developers, organizations, and users. Each has a role in identifying problems and helping make AI more inclusive.
AI for social good
Security and privacy risks of AI
Artificial intelligence (AI) has become a useful tool for writing, research, marketing, business management, and many other activities. However, using AI also comes with privacy and security risks. Understanding these risks helps us protect personal information, keep confidential data safe, and use AI tools responsibly.
Privacy and Data Use
Privacy means having control over how personal information is collected, stored, accessed, and used. When we use AI tools, we may share information such as our names, email addresses, medical records, financial details, or workplace documents. If we are not careful, this information could be handled in ways we did not expect.
For example, imagine someone uses an AI assistant to write a business report and includes private customer information. Depending on the tool’s policies and settings, that information may be processed or retained by the service. This is why it is important to understand how an AI tool handles user data before sharing sensitive information.
AI systems may be trained using different types of data, and some services may use user interactions to improve their models. However, data practices vary between providers, so users should review the relevant privacy policies and settings rather than assume all AI tools handle information in the same way.
Security Risks and Protection
Security means protecting information and systems from unauthorized access, misuse, theft, or damage. AI tools can introduce additional security challenges, particularly when people use them to process confidential business information or connect them to workplace systems.
For example, an employee might upload an internal company document to an unapproved AI tool to summarize it quickly. If the tool does not provide suitable protections, the company could expose confidential information to unauthorized parties.
Other risks include phishing messages created with AI, convincing impersonations, deepfake scams, and malicious attempts to obtain sensitive information. These threats demonstrate why individuals and organizations need appropriate security measures when using AI.
To reduce these risks, organizations should use approved AI tools, apply access controls, protect accounts with strong passwords and multifactor authentication, and establish clear rules for handling confidential information.
Best Practices for Using AI Safely
Everyone has a role in protecting privacy and security when working with AI. Before using a tool, read its terms of service and privacy policy to understand what information it collects, how it uses data, and what controls are available.
Avoid entering passwords, financial account details, confidential business documents, private customer information, or sensitive personal records into AI tools unless the use is explicitly authorized and appropriate safeguards are in place. When possible, remove names and other identifying details or use fictional examples instead.
It is also important to verify AI-generated information before acting on it. AI can produce convincing but inaccurate content, and scammers can use AI to make fraudulent messages appear genuine. Be cautious with unexpected requests for money, account access, or confidential information, even when a message sounds familiar.
Finally, stay informed about changes in AI technology, privacy practices, and security threats. Following trusted cybersecurity resources and your organization’s policies can help you make safer decisions as AI tools evolve.
Real-Life Example: AI in a Marketing Team
Imagine a marketing professional who uses an AI tool to create personalized email campaigns. To improve the results, they consider uploading customer names, purchase histories, contact details, and confidential campaign budgets.
Although these details might help generate more specific content, sharing them without permission or proper safeguards could expose private customer information and sensitive company data.
A responsible approach would be to use an approved AI tool, replace real customer details with fictional examples, and provide only the information necessary to complete the task. The marketing professional should also review the company’s data protection policies and check the generated emails before sending them.
This example illustrates the difference between privacy and security. Privacy concerns how information is collected and used, while security focuses on protecting that information from unauthorized access. Both are essential when using AI in the workplace.
Bias, drift, and knowledge cutoff
A key part of being a responsible AI user means knowing how AI works—and where it can fall short. Every AI tool has limitations. Recognizing these limitations can help you evaluate outputs more effectively and use AI in a way that is fair, accurate, and ethical.
Understanding data bias
Data bias can be a foundational challenge for many AI tools. It occurs when the data used to train an AI model is skewed, incomplete, or reflects historical or societal biases. Because the tool’s output is a direct reflection of its training data, it can learn and even amplify these biases in its responses.
As a responsible user, your role is to guide the tool toward a fair and impartial output. Here are a few ways you can mitigate bias when using AI tools:
Be specific about the output you want from your prompt.
Include important context about your intended audience and their needs.
Provide fair and balanced references for the tool to follow.
Use follow-up prompts to correct any output that seems biased or inaccurate.
The constraint of knowledge cutoff
Knowledge cutoff is the concept that an AI model is trained at a specific point in time. AI models have a knowledge cutoff because their core knowledge is based on the data they were trained on.
You might notice that some AI tools can provide information about very recent events. Some may do this by performing a live web search to find current information and supplement their answer. It’s helpful to think of this as the difference between what the model knows from its training versus what it can look up in the moment. The model’s core knowledge isn’t being updated, which is why the concept of a knowledge cutoff remains a critical limitation to be aware of.
Responsible AI use requires you to verify time-sensitive information. Always use a search engine or other reliable sources to fact-check statistics, news, or any information about recent events.
Pro Tip: Some AI tools tell you their knowledge cutoff date if you ask them directly.
Drift in an AI tool’s output
Drift is the gradual decline in an AI tool’s accuracy and relevance as the real world changes. You might observe drift in two ways:
Factual drift: This is when the AI’s becomes less accurate over time because of its knowledge cutoff. For example, an AI’s advice on “current fashion trends” may become less useful the further you get from its training date.
Behavioral drift: This refers to changes in an AI tool’s behavior over time. As developers update models, you might notice that the tool’s formatting, tone, or conversational style changes, even when you use the same prompts.
Here are a few ways to manage and mitigate both kinds of drift:
Provide accurate and up-to-date context in your prompts, especially for topics that change quickly like market trends or technology.
Keep chats focused by starting a new conversation for each specific task. This also helps to reset the context if a conversation becomes too long or the output starts to feel off-topic.
Be explicit with clear and specific instructions in your prompts.
Being the human-in-the-loop
Being the “human-in-the-loop” means you are always the final judge of an AI’s output. This is especially important when using more automated AI assistants or agents. While these tools can handle tasks more independently, they are still subject to the same limitations of bias, knowledge cutoff, and drift. Understanding these limitations is critical for setting up and supervising automated tools responsibly to ensure they continue to operate in a safe and helpful way over time.
Develop AI That Works for Everyone
Artificial intelligence (AI) is becoming an important part of our lives, helping us learn, work, communicate, and solve problems. However, AI should not be designed to benefit only certain people or groups. Responsible AI development means creating systems that are fair, inclusive, accessible, and useful for people with different backgrounds, abilities, and experiences.
Fairness and Inclusion in AI
Fairness means ensuring that AI systems do not unfairly disadvantage people because of their background, gender, disability, or other personal characteristics. Inclusion means considering the needs of different people so that everyone has a fair opportunity to benefit from technology.
For example, an AI-powered learning platform should support students with different learning needs, including students with visual, hearing, or other disabilities. If the platform only works well for students who can see images or hear audio, it may exclude others. Developers can make the system more accessible by including features such as screen-reader compatibility, captions, and alternative ways to access information.
AI should also be able to discuss sensitive topics responsibly. Avoiding every sensitive subject can prevent people from receiving useful educational information or appropriate support. Instead, AI systems should respond carefully, respectfully, and safely, depending on the context.
Addressing AI Bias and Data Representation
AI bias occurs when a system produces unfair or unbalanced results. One common cause is the use of training data that does not adequately represent different groups of people.
For example, a voice recognition system trained mainly on the voices of people without speech disabilities may struggle to understand users who speak differently. This can make the technology less useful for those who need it.
To reduce bias, developers should use diverse and representative datasets, test AI systems with different groups of users, and correct problems when they are discovered. They should also consider how algorithms make decisions and how the system presents information through its user interface.
Reducing AI bias requires more than changing the training data. Developers must evaluate the entire system to ensure that it works fairly and does not create unnecessary barriers for particular users.
Transparency in AI Systems
Transparency means helping users understand how an AI system works, what information it uses, and what its limitations are. When AI systems provide clear explanations about their capabilities and limitations, people can make better-informed decisions about whether to trust their results.
For example, if an AI tool recommends a candidate for a job, employers should understand which relevant criteria influenced that recommendation. They should not simply accept the AI’s decision without considering whether it is accurate, appropriate, and fair.
Although some AI systems are difficult to explain fully, developers should provide meaningful information about their purpose, limitations, and evaluation wherever possible.
The Importance of Critical Thinking
AI can be a useful assistant, but its answers are not always correct. It may misunderstand a question, produce inaccurate information, or reflect biases in its training data. Therefore, users should evaluate AI-generated content rather than accepting it automatically.
Critical thinking involves checking important facts, comparing information with reliable sources, asking questions about possible bias, and considering alternative perspectives. When an AI system makes a recommendation that could affect someone’s life, appropriate human review is especially important.
Using AI in this way allows people to benefit from its speed and capabilities while maintaining responsibility for their decisions.
Real-Life Example: AI Hiring for People with Disabilities
Imagine a company that uses AI to screen job applications. The system analyzes resumes and ranks candidates based on their qualifications and experience.
However, the AI may disadvantage applicants with disabilities if it relies on narrow historical hiring data or uses assessment methods that are not accessible. For example, an automated interview system might incorrectly evaluate a candidate who communicates differently because of a disability.
To address this issue, the company should evaluate the system using diverse and representative data, provide accessible assessment methods, and check whether candidates are being treated fairly. Hiring managers should also review the AI’s recommendations and provide reasonable accommodations where needed.
Use AI for Positive Change
Artificial intelligence (AI) is changing the way people learn, work, solve problems, and understand the world. When used responsibly, AI can help address important challenges, improve access to information, protect the environment, and support scientific discoveries. The goal is not simply to develop more advanced technology, but to use it in ways that make a positive difference in people’s lives.
The Role and Purpose of Internet Evangelism
A chief internet evangelist is someone who promotes the benefits of the internet and encourages more people to gain access to it. The role focuses on helping people understand how internet connectivity can create opportunities for education, communication, business, and access to essential information.
Although the internet has connected billions of people, many communities still face barriers such as poor infrastructure, high costs, and limited digital skills. Expanding internet access can help bridge this digital divide and give more people the opportunity to participate in the digital world.
AI can build on this progress by helping people access information, learn new skills, and discover opportunities. However, these benefits are only meaningful when people have reliable internet access, suitable devices, and the knowledge needed to use digital tools effectively.
AI Applications in Crisis Management and Science
One important way AI can create positive change is by helping people respond to emergencies. For example, AI systems can analyze satellite images, weather information, and sensor data to help detect signs of wildfires. Early warnings can give emergency teams more time to assess the situation, protect communities, and plan responses.
AI is also making important contributions to scientific research. Researchers at DeepMind developed AlphaFold, an AI system that predicts the three-dimensional structures of proteins. Understanding protein structures helps scientists investigate how biological processes work and can support research into diseases and potential treatments.
These examples demonstrate how AI can assist people in addressing complex challenges. However, AI predictions are not always perfect, so experts must validate important findings and use appropriate human judgment before taking action.
Understanding AI Literacy
AI literacy means understanding what AI can do, where it has limitations, and how to use it responsibly. It does not mean that everyone needs to become a programmer or understand every technical detail behind an AI model.
Consider driving a car. Most drivers do not need to understand how every engine component works, but they should know the basic rules of the road, recognize potential dangers, and understand how to operate the vehicle safely.
The same principle applies to AI. People should understand that AI can generate useful ideas and analyze information, but it can also make mistakes, reflect biases, and produce misleading results. Developing AI literacy helps people make informed decisions about when to use AI and when to rely on human expertise.
AI in Education, Creativity, and Research
AI has the potential to transform education by helping students learn at their own pace. AI-powered tutoring tools can explain difficult concepts, provide practice questions, and adapt explanations to different learning needs. Teachers can also use AI to develop learning materials and organize educational activities, while continuing to provide human guidance and support.
In creative work, AI tools can help people generate ideas, develop stories, create images, and explore new approaches to design. These tools can support creativity, but human imagination, personal experience, and judgment remain important in producing meaningful work.
AI can also assist researchers by analyzing large amounts of information, identifying patterns, and helping explore new scientific questions. When combined with human expertise and careful testing, these capabilities can support discoveries that might otherwise take much longer.
Real-Life Example: AI for Wildfire Detection
Imagine a region where forests become increasingly vulnerable to wildfires during hot and dry weather. Emergency teams need to identify potential fires quickly because even a small fire can spread and threaten nearby communities.
An AI-powered monitoring system can analyze satellite images and sensor readings to identify possible signs of a fire. When the system detects a potential threat, it can alert trained personnel, who then verify the information and decide what action to take.
This approach can help emergency teams respond more quickly, protect homes and wildlife, and reduce environmental damage. However, the system must be tested carefully to limit false alarms and missed detections. Human responders remain responsible for confirming threats and making emergency decisions.
This example shows how AI can be used for positive change when technology, reliable information, and human expertise work together.
Using AI Responsibly for Social Good
To ensure that AI benefits society, developers and users must consider both its opportunities and its risks. AI systems should be designed with fairness, privacy, accessibility, and safety in mind. Their results should be evaluated carefully, especially when they influence decisions affecting people’s lives.
It is also important to make AI education available to more people. When individuals understand how AI works and recognize its limitations, they are better equipped to use it effectively and avoid potential harm.
Responsible AI is not only about what technology can achieve. It is also about who benefits from it, whether its advantages are accessible to different communities, and how its risks are managed.
Checklist for using AI responsibly
When you use a generative AI tool, you’re in the driver’s seat. The tool provides computational power, but you provide direction, judgment, and critical thinking. Using AI responsibly means being thoughtful and intentional with your prompts to ensure the quality, safety, and accuracy of the outputs.
This reading provides a set of key practices to help you make responsible choices each time you use AI.
Principles of responsible AI use
Applying the following four principles helps you use AI tools both responsibly and effectively.
1. Be clear and specific in your prompts
Vague or ambiguous prompts can lead to generic outputs. The more specific you are in your instructions, the more you guide the AI tool away from unhelpful assumptions and toward the nuanced output you actually want. Being specific is a key part of using AI responsibly.
2. Protect sensitive information
This is a crucial responsibility for every AI user. Public AI tools are not the place for personal, confidential, or sensitive data. Before entering any prompt, it’s a good idea to pause and ask yourself: Am I including data in my prompt that someone else might expect me to keep private?
Here are a few practical ways to protect sensitive information when prompting:
Use placeholders by referring to people, projects, or places with generic names.
Frame prompts around the task you need done, not the people involved.
Input only relevant context needed to complete the task, rather than entire documents.
Pro tip: Check out how to adjust your privacy settings and manage your conversation history in Gemini.
3. Provide high-quality information
The quality of an AI tool’s output can reflect the quality of the references you provide. If you give the tool examples or data that are biased, outdated, or contain errors, the tool is likely to reproduce those same flaws in its response.
You should always try to provide inputs that are as accurate and unbiased as possible. Before using a piece of text or data as an example, ask yourself: Does this information represent a fair and balanced perspective? Is it from a source I trust?
By providing fair and factual inputs, you guide the tool toward creating a more responsible and accurate output.
4. Always verify the output
No matter how good your prompt is, the final responsibility for the output is yours. You are the essential human-in-the-loop who must be the final judge of accuracy, tone, and appropriateness.
Here is a simple workflow to apply this principle:
Review the output to ensure it meets your goals.
Fact-check any claims or data by cross-referencing the information.
Revise the content yourself by treating the AI-generated text as a first draft.
Pro tip: Some AI tools can help you fact-check their own responses. You can learn more about how to use the double-check feature in Gemini to evaluate its responses.
Ultimately, being a responsible AI user means combining the capabilities of the tool with your own critical thinking and judgment. By practicing each of these principles, you do more than just get better outputs—you contribute to a safer and more ethical use of AI.
