Powerful Technology Requires Responsible Use
Generative AI can help people learn, work, create and solve problems.
However, the same capabilities that make Generative AI useful can also create risks.
A responsible AI user needs to understand both sides.
The goal is not to avoid AI completely. The goal is to understand when AI is appropriate, how its output should be checked, and what safeguards are necessary.
Risk 1: Inaccurate Information
One of the most important limitations of Generative AI is that it can produce information that sounds convincing but is incorrect.
A language model is designed to generate likely sequences based on its learned patterns and current context. It is not automatically a fact-checking system.
This means a confident-sounding answer should not automatically be treated as proof that the information is true.
AI Hallucinations
The term hallucination is commonly used when an AI system generates information that is false, unsupported or fabricated.
For example, an AI system might provide a citation to a document that does not exist or confidently state a fact that is incorrect.
Hallucinations can be particularly dangerous when users assume that fluent language means reliable information.
How to Reduce the Impact of Errors
Users can reduce the impact of AI errors by:
- Checking important facts against reliable sources.
- Providing accurate context.
- Asking the AI to explain its reasoning or assumptions where appropriate.
- Using trusted external information sources.
- Having a qualified person review high-impact outputs.
Risk 2: Bias
AI systems learn patterns from data.
If the training data contains biases, limitations or imbalances, those patterns can influence the behaviour of a model.
Bias can also be introduced through the way data is collected, labelled, selected or used.
Why Bias Matters
Bias may produce unfair or inappropriate results.
This can become especially important when AI is used in areas such as hiring, lending, education, healthcare, insurance or other decisions that affect people.
High-impact applications require careful evaluation and appropriate safeguards.
Risk 3: Privacy
AI applications may process information provided by users.
That information could include personal information, customer information, company documents or other sensitive data.
Users should understand how an AI service handles the information they provide.
Do Not Share Sensitive Information Carelessly
Users should be cautious about entering passwords, confidential business information, private customer records, financial information or other sensitive material into an AI system.
Whether information can be safely provided depends on the particular service, its configuration, the organization's policies and applicable requirements.
Risk 4: Security
Generative AI systems can create security risks in several ways.
An AI-generated program may contain a security vulnerability.
An AI application may also be manipulated through malicious instructions or carefully crafted input.
Organizations therefore need security controls around AI systems, just as they do around other software.
Prompt Injection
When an AI application processes information from users or external documents, an attacker may attempt to include instructions designed to manipulate the model.
This type of attack is commonly called prompt injection.
Applications that use AI with external tools or private information need appropriate safeguards to limit what the model can access and do.
AI-Generated Code
AI coding assistants can increase developer productivity, but generated code is not automatically secure.
A developer should review generated code, run appropriate tests, check dependencies, and consider security implications before deploying it.
Risk 5: Misinformation
Generative AI can make it easier to create large amounts of convincing text, images, audio and video.
This can be useful for legitimate communication and creative work.
It can also be used to create misleading content.
Synthetic Media
AI-generated or AI-modified media can sometimes look or sound realistic.
Such content can be used for entertainment, education and accessibility, but it can also be used to impersonate people or create deceptive material.
Users should consider the source of important media rather than assuming that something is authentic simply because it looks realistic.
Deepfakes
The term deepfake is commonly used for realistic synthetic or manipulated media involving a person.
Deepfake technology can have legitimate creative applications, but it can also be misused for impersonation, fraud, harassment or misinformation.
Consent and responsible use are therefore important considerations.
Risk 6: Copyright and Intellectual Property
Generative AI can create content that raises questions about copyright and other intellectual-property rights.
The legal treatment of AI-generated material can depend on the jurisdiction, the specific circumstances, the source material and how the content is used.
Organizations and creators should understand the rules that apply to their particular situation.
Using Other People's Material
Users should not assume that AI makes copyright restrictions disappear.
When working with copyrighted material, users should consider whether they have the right to use the material and whether their intended use is permitted.
Risk 7: Over-Reliance on AI
Another risk is becoming too dependent on AI-generated answers.
If users accept every answer without checking it, errors can pass into documents, software, decisions and communications.
AI should support human thinking rather than eliminate the need for judgment.
Critical Thinking
Responsible AI use requires users to ask questions such as:
- Is this information accurate?
- What evidence supports it?
- Could important information be missing?
- What assumptions is the answer making?
- Is this task appropriate for AI?
- Does a qualified person need to review the result?
Risk 8: Lack of Transparency
Users may not always know exactly how a particular AI system produced an output.
Models can be extremely complex, and the internal reasoning behind an output may not be easy to inspect.
This can create challenges when organizations need to explain or audit important decisions.
High-Impact Decisions
AI should be used particularly carefully when its output can significantly affect a person's life or rights.
Examples can include employment, financial decisions, healthcare, education and access to important services.
In such situations, appropriate human oversight, documentation, testing and governance are important.
Responsible AI Principles
Responsible use can be organized around several practical principles:
- Accuracy: Verify important information.
- Privacy: Protect personal and confidential information.
- Security: Test and secure AI applications.
- Fairness: Look for harmful or discriminatory bias.
- Transparency: Explain how AI is being used where appropriate.
- Human oversight: Keep people involved in important decisions.
- Accountability: Make sure someone remains responsible for the outcome.
Responsible Use Is Context-Dependent
There is no single rule that determines whether every AI use is safe or appropriate.
The right approach depends on the task, the information involved, the potential consequences of errors and the people affected.
A creative brainstorming session may have relatively low consequences if an idea is imperfect.
A medical or financial decision may require much stronger safeguards.
Human Oversight
Human oversight means that a person has an appropriate opportunity to review and influence an AI-assisted process.
The level of oversight should match the potential impact of the task.
A low-risk draft email may need a quick review.
A high-impact decision may require detailed evaluation by a qualified professional.
Testing AI Systems
Organizations should test AI applications before relying on them in important workflows.
Testing can include:
- Accuracy testing.
- Security testing.
- Bias and fairness testing.
- Privacy testing.
- Adversarial testing.
- Performance testing.
Monitoring After Deployment
Testing should not stop when an AI system is launched.
Real-world usage can reveal problems that were not visible during initial testing.
Organizations should monitor important systems and provide a way to identify, investigate and correct problems.
AI Policies
Organizations can establish policies that explain how employees should use AI.
A policy might define:
- Which AI tools are approved.
- What information employees may enter.
- When human review is required.
- How AI-generated content should be identified.
- Who is responsible for final decisions.
A Simple Responsible AI Workflow
A practical workflow is:
Define the Goal → Assess the Risk → Provide Appropriate Information → Generate → Verify → Review → Use
This does not guarantee that every AI result will be perfect, but it provides a structured way to reduce avoidable problems.
The Benefits Still Matter
Understanding risk does not mean Generative AI should be avoided.
When used appropriately, AI can improve productivity, support learning, help people communicate and enable new creative possibilities.
The objective is to capture these benefits while managing the risks.
The Big Picture
Generative AI is powerful because it can produce useful content quickly and flexibly.
It is also imperfect.
Responsible use means understanding the technology's limitations, protecting information, checking important outputs, considering fairness and maintaining appropriate human responsibility.
What You Should Remember
- Generative AI can produce inaccurate or fabricated information.
- AI systems can reflect bias in their data and design.
- Privacy and security must be considered when using AI.
- AI-generated code should be reviewed and tested.
- Generative AI can be used to create misleading synthetic media.
- Copyright and intellectual-property questions require careful consideration.
- Users should avoid blindly trusting AI output.
- High-impact decisions require appropriate human oversight.
- Organizations should test, monitor and govern important AI systems.
What Comes Next?
We have now examined both the capabilities and risks of Generative AI.
In the final lesson of this module, we will bring everything together and build a complete mental model of Generative AI from training through real-world use.