Navigating the Ethics of Generative AI in Learning and Development

2001: A Space Odyssey, a remarkable sci-fi book and movie released in 1968 about astronauts on a mission to Jupiter to unravel the mysteries of a monolith, creatively showed the world the potential capabilities of AI in the form of a supercomputer-powered robot, HAL 9000. Spoiler alert: in the movie and book, the dark side of AI comes to the fore when HAL takes complete control of the spaceship and disposes of the human crew. It thrilled the audience, leaving them spellbound and with numerous questions about artificial intelligence.

Fifty-five years later, humanity is on a similar type of odyssey, metaphorically speaking, to unravel the deep mysteries of what Generative AI can do. The learning and development industry is pursuing the immense capabilities that Generative AI has revealed for the industry since Open AI unveiled Chat GPT 3 to the world on November 30, 2022. At the time of writing this blog, this industry is already leveraging Generative AI to generate, design, and develop content (text, images, audio, and video), personalized and adaptive learning experiences, chatbots and virtual assistants, simulations and scenarios, automated tools for administrative tasks, and so on. Here’s a small taste of how Generative AI can assist learning designers in automating several of their course development tasks. There’s no telling what unforeseen breakthroughs lie ahead and the new autonomous paradigms for problem-solving that a highly advanced and creative AI might conjure up!

However, the questions about the ethical and responsible use of AI that the movie 2001: A Space Odyssey raised continue to linger. What’s stopping our real-life experiences with Generative AI from becoming warped and rogue like the robot HAL 9000 from the movie if sufficient checks and balances aren’t defined, agreed upon, embraced, and applied rigorously and collectively by the creators and consumers of Generative AI tools?

In this blog, let’s take a closer look at some of the ethical issues surrounding the usage of Generative AI in the learning and development industry, which, if inadequately addressed, will become an albatross around the neck of the industry.

Grey Areas

With the mad race to harness Generative AI to improve work efficiency, here are the top three areas that organizations utilizing Generative AI for learning and development need to ensure they are navigating correctly.

Grey Area 1: Content Authenticity

Content generated by AI tools is often alleged to be plagiarized, as there are no clear citations. Sometimes, on requesting sources, Generative AI tools may throw up fake sources with a semblance of authenticity. On digging deeper, the links are found to be spurious. A New York-based law firm recently came under fire as one of their lawyers unwittingly cited six fake cases as evidence in their case filing. It was revealed that the firm had used ChatGPT for legal research.

Grey Area 2: Copyright Infringement

Many AI image-generating tools have flooded the market. However, earlier this year, major copyright infringement issues came to the fore. Getty Images sued Stability AI (an open-source Generative AI company) for using its image library to train its AI image generator. Also, there continues to be a lot of ambiguity about whether content generated by AI tools can be copyrighted. Artists Sarah Andersen, Kelly McKernan, and Karla Ortiz sued Midjourney and Stability AI for violating artists’ rights. The company owns the AI art-generating app Dream Up, which is trained on the works of millions of artists without their permission or consent.

Grey Area 3: Data Security

One huge security concern is that the AI tools tend to store the data you feed as a prompt. The tool may retain that data (which might be proprietary in nature) and use it to answer similar queries when asked by others. This can severely jeopardize data security.
With its ability to create fake code, data, and images that look incredibly real, Generative AI could lead to a surge in identity theft, fraud, and counterfeiting incidents. Even LLMs aren't immune to being compromised and used maliciously.

The Need for Responsible AI Practices

Clearly, the need of the hour is Responsible AI (RAI). The corporate training industry must be cognizant of various AI risks and responsible AI practices. Organizations using any or a gamut of Generative AI tools need to ensure that the creators of these tools abide by strong ethical principles. What could some RAI practices look like to mitigate some of the potential risks discussed above? Let’s see.

Mitigation 1: Content Authenticity

Challenges:

AI can generate content without clear sources.
Risk of citing fake or non-existent references.

Responsible AI Practices:

Verification Systems: Use other AI models or third-party tools to cross-check the authenticity of content produced by generative AI.
Human Oversight: Include a layer of human review to vet and validate AI-generated content before deployment, especially for crucial applications like legal research.
Source Indication: If possible, design systems to capture potential sources or influences, even if they are broad, that the Generative AI tool draws from.

Mitigation 2: Copyright Infringement

Challenges:

Training AI on copyrighted material without permission.
Ambiguity around copyright ownership of AI-generated content.

Responsible AI Practices:

Clear Licensing: Use only datasets with clear licensing terms that allow for training generative models. Avoid datasets with copyrighted or proprietary information unless explicit permissions are obtained.
Copyright Education: Educate users on the potential legal and ethical concerns of using AI-generated content. Encourage them to seek permission when necessary.
AI-generated Content Rights: Advocate for clear legal frameworks that address the copyright status of AI-generated content. This can be done through collaboration with legal bodies and industry groups.

Mitigation 3: Data Security

Challenges:

Risk of AI tools retaining sensitive or proprietary data.
Generative AI's potential to create fake or misleading data/code that can be used maliciously.

Responsible AI Practices:

Data Retention Policies: Establish strict data retention policies for AI tools. Ensure that data used as prompts are not stored longer than necessary or used without explicit consent.
Differential Privacy: Implement differential privacy techniques to ensure that AI models do not inadvertently leak information about the individual data entries they were trained on.
Regular Audits: Conduct security and data audits on AI systems to ensure compliance and detect potential vulnerabilities.
User Education: Educate users on the risks of feeding sensitive or proprietary information into AI tools.

Best Practices for the Ethical Usage of Generative AI for Learning and Development Purposes

Currently, the primary use cases of how organizations might leverage Generative AI for learning and development – that also might have ethical implications – can be clubbed under:

Content generation – generating raw textual content such as for the use of soft-skill courses or using generative AI to generate audio and video that can be used in training.
Chatbot creation – creating a chatbot powered by an enterprise-specific large language model (LLM) trained on datasets specific to your organization.

Keeping these primary use cases in mind, here are some best practices:

Best Practice 1: Avoid Over-reliance

While AI can personalize and adapt training content, human oversight is crucial. Ensure there's a balance between AI-generated content and human-curated material.
Provide channels for human trainers or mentors to engage with trainees on nuanced or complex topics.

Best Practice 2: Transparent Disclosure

Clearly inform trainees when they are interacting with or benefiting from AI-generated content.
Clarify the objectives and expected outcomes of training modules powered by generative AI.

Best Practice 3: Bias and Fairness

Regularly evaluate the AI models for biases that might inadvertently favor or disadvantage specific groups of trainees.
Use diverse training datasets to avoid perpetuating stereotypes or prejudices.

Best Practice 4: Feedback Mechanisms

Enable trainees to provide feedback on AI-generated content and the overall training experience.
Use this feedback loop to refine the AI models and ensure they meet the desired training standards.

Best Practice 5: Accessibility and Inclusivity

Ensure that AI-generated training materials are accessible to all employees, including those with disabilities.
Implement features like text-to-speech, subtitles, or multilingual support as necessary.

Best Practice 6: Continuous Learning and Updates

Enterprise-level LLMs and their outputs should be regularly updated to reflect the latest industry standards, knowledge, and best practices.
Engage subject matter experts in the loop to ensure the content remains relevant and accurate.

Best Practice 7: Ethical Use of Simulations

If using AI to generate simulations or role-playing scenarios, ensure they are respectful, do not perpetuate harmful stereotypes, and are rooted in real-world applicability.

Best Practice 8: Informed Consent

Secure explicit consent from trainees if their data will be used to refine or adapt AI-driven training modules.

Best Practice 9: Data Privacy and Protection

If personal data or performance metrics of employees are used to train or tailor AI models, ensure they are anonymized and de-identified to protect individual privacy.
Use encryption and secure data storage methods. Adhere to global data protection regulations, such as GDPR or CCPA.
Ensure trainees understand how their data will be used and the benefits of the AI-driven approach.

Best Practice 10: Accountability

Establish a clear line of accountability for the outcomes and potential issues arising from AI-driven learning and development.
Be prepared to intervene, modify, or even halt AI-driven modules if they are found to be ineffective or ethically questionable.

These measures can help your organization leverage Generative AI to innovate and automate your learning and development activities and offerings while promoting the ethical usage of AI to create a responsible and a fair learning environment for your employees.

About Encora

At Encora, we've been focusing on developing functionalities that assess the effectiveness and efficiency of Generative AI throughout the digital engineering software development process. Additionally, we've collaborated closely with hyperscalers to implement Generative AI's benefits across various industries we cater to, such as HiTech, FinTech & InsurTech, HealthTech, Supply Chain & Logistics, Telecom, and more! Click here to find out more.