Five Ways To Teach An Intern

AI Chatbot – Style Responsible AI Chatbot

Generative artificial intelligence is often compared to the world’s sharpest intern, immensely skilled, but lacking life experience. This comparison resonates with many, especially in light of some real-world incidents that highlight both the power and risks of AI.

Take, for example, a lawyer who relied on AI’s persuasive writing and ended up fined for submitting fabricated documents. Or consider the case involving Air Canada, where a customer support chatbot falsely promised a refund, damaging the airline’s reputation.

Public trust in AI chatbots remains low. According to a 2023 Gallup/Bentley University survey, only 21% of consumers believe companies will manage AI responsibly. So, how can we train AI chatbots like we mentor interns, aiming to build responsible, trustworthy digital assistants?

Here are five approaches to consider:

1. Teach Your Bot Basic Manners

Respect isn’t optional. Users instantly sense when their rights are respected or violated. For instance, no one wants to find out via a targeted ad that their teenager might be pregnant.

Studies show that brand transparency influences customer loyalty. One-third of consumers favour companies that clearly explain how they use data. To build trust, apply the “Rule of Three”:

• Be transparent about the bot’s identity and capabilities.
• Share performance data and limitations.
• Clearly explain data usage and privacy policies.

For example, Claude’s privacy policy deletes user data after 30 days, while ChatGPT’s lacks clarity. Such differences matter.

2. Measure and Review Performance Early

AI isn’t a one-size-fits-all tool, making benchmarking tricky. Dr. Catherine Breslin, an AI consultant and former Alexa developer, recommends thorough testing before and after deployment.

Start by teaching bots to distinguish between trustworthy and misleading data. Use diverse datasets to minimise bias. Pedro Henriques, founder of The Newsroom, stresses the need to tailor bots to specific use cases, languages, and user expectations.

Also, use prompt engineering to refine your bot’s responses. An HR assistant chatbot, for instance, should quote the company’s non-discrimination policies when justifying its actions.

Design bots to explain their logic. Involve IT teams early for better testing, integration, and scalability.

3. Train Your Bot in Safety Standards

Just like interns undergo safety training, bots must meet strict security requirements. Jonny Pelter, former CISO of Thames Water, notes that security can’t be an afterthought.

Go beyond basic security. Implement adversarial testing, protect against data poisoning, and use model transparency tools. Regulatory frameworks like the EU AI Act and President Biden’s executive order are now pushing these standards into the mainstream.

4. Keep Bots Legally Compliant

Legal accountability is at the heart of responsible AI. Over 40 governance frameworks can help, each with specific applications.

Regulations like GDPR and the EU AI Act impose binding legal standards. Global frameworks from NIST and ISO promote accountability, fairness, and transparency.

Industries like banking require additional safeguards. The IEEE’s chatbot standards, for instance, leave no room for error when dealing with malicious actors.

5. Embed Ethics and See the Bigger Picture

Just as we expect interns to uphold company values, bots should reflect ethical standards, respect, environmental consciousness, and integrity.

Of course, bots can’t be morally accountable. That’s why humans must remain in the loop. Dr. Breslin recommends assigning clear lines of responsibility and oversight.

Environmental costs also matter. Dr. Nataliya Tkachenko from Cambridge Judge Business School warns that each chatbot interaction consumes significant computing power, especially during live customer support.

As AI adoption accelerates, bots must evolve with responsibility. Organisations already trust young professionals to shape future workplaces. Why not expect the same from our digital interns?

By using these five strategies, you can guide your AI chatbot to behave more like a well-trained intern, respectful, secure, accurate, and aligned with your core values.