An interview with ChatGPT about managing the risks of generative AI

This article is brought to you thanks to the collaboration of The European Sting with the World Economic Forum.

Author: Nelson Novaes Neto, CTO , C6 Bank, Keri Pearlson, Executive Director, Cybersecurity, MIT – Sloan School of Management

• Generative AI has the potential to help workers be more efficient and creative, but is also expected to replace people in some roles.
• AI can enhances operational efficiency, predict market trends and creates new innovations, but also presents business risks.
• We decided to ask online tool ChatGPT about the popular uses and risks of generative AI to see what it said.

Generative artificial intelligence (AI) has the potential to radically change the way we work. In some cases, it is expected to help workers be more efficient and creative. In other instances, however, it is expected to replace workers.

Indian startup Dukaan’s laid off 90% of their support staff following the implementation of an AI chatbot. The tool responded to customer queries instantly, significantly outpacing human staff. This move, while tough, reduced the time to resolve customer issues by 98% and cut customer support costs by 85%.

Around the same time, Samsung engineers thought they had found a new tool to enhance their system design when they used ChatGPT to check for errors and to optimize proprietary code. Instead, as the headlines in the press highlighted, they got more than they expected.

Not only did the tool optimize the code, but the code was used to train their system, leaking it to the outside world. ChatGPT and the other generative AI systems offer incredibly useful insights and suggestions, but managers must understand the new risks associated with its use.

AI enhances operational efficiency, predicts market trends, creates new innovations, finds connections from seemingly unconnected ideas, and much more. These new tools are revolutionizing design, evaluation, generation, management, protection, development and seemingly any task we ask it to do.

But at the same time, these systems are creating new business risk. We decided to ask ChatGPT about the popular uses and risks to see what it said.

Although, of course, it must be noted that large language models such as ChatGPT can have a propensity to ‘hallucinate’ and that users must understand the limitations of what such systems can and cannot do.

What are the most popular uses of generative AI?

We wondered what the most popular use cases were, so asked ChatGPT that very question: “What are the most popular uses of generative AI?”

ChatGPT told us about creative applications, content generation, design and prototyping, data augmentation, personalization, recommendations, virtual assistants, video game development, scientific research, simulation, training, storytelling and narrative generation – and provided detail for each use.

https://cdn.jwplayer.com/players/0r6JhJja-ncRE1zO6.html

ChatGPT concluded by telling us: “Generative AI has a wide range of applications across various domains. These categories represent some of the prominent uses of generative AI, but the field is rapidly evolving, and new applications continue to emerge.”

We wondered if the list would change over time.

What are the risks of using generative AI?

Once we learned about popular uses, we wanted to know about the risks. While generative AI holds great potential, there are serious business risks and challenges to be managed. We asked ChatGPT? “What are the risks of using generative AI?”

ChatGPT listed these risks:

• Misinformation and fake content
• Privacy and data protection
• Ethical implications
• Bias and discrimination
• Intellectual property infringement
• Security vulnerabilities
• Unintended consequences

Without asking for clarification, ChatGPT offer the following: “To mitigate these risks, it is crucial to implement robust security measures, adhere to ethical guidelines, ensure transparent and accountable AI practices, and continuously monitor and evaluate the impact of generative AI systems. It is also essential to promote public awareness and understanding of generative AI to navigate its risks responsibly.”

Discover

How is the World Economic Forum creating guardrails for Artificial Intelligence?

In response to the uncertainties surrounding generative AI and the need for robust AI governance frameworks to ensure responsible and beneficial outcomes for all, the Forum’s Centre for the Fourth Industrial Revolution (C4IR) has launched the AI Governance Alliance.

The Alliance will unite industry leaders, governments, academic institutions, and civil society organizations to champion responsible global design and release of transparent and inclusive AI systems.

How do malicious actors use generative AI?

We were very interested in understanding the new attack vectors that we might see if malicious actors used it. After all, if we can use ChatGPT for good uses, hackers can use it for malicious uses. So we asked ChatGPT: “How do malicious actors use generative AI?”

ChatGPT told us: “Unfortunately, just like any powerful technology, generative AI can be misused by malicious actors for various purposes. Here are some ways ChatGPT told us generative AI can be used maliciously:

• Deep fakes
• Phishing attacks
• Spam generation
• Fake news and disinformation
• Identify theft
• Automated attacks
• Social engineering

Of interest to us was what was left out such as using this technology to generate damaging code, break codes, design malware, and create other invasive threat vectors.

How to reduce the risks from generative AI

Managers can play a crucial role in reducing the risks associated with generative AI. ChatGPT had a few suggestions on how to reduce risk.

The goal of using these strategies, according to ChatGPT, is to mitigate risks and foster ethical deployment of technology.

The strategies suggested by the system are:

• Robust data governance
• Ethical frameworks and guidelines
• Bias mitigation
• Adversarial defence
• Transparency and explainability
• Regular monitoring and evaluation
• Employee education and awareness
• Collaboration and industry standards

Insights gained from utilizing generative AI

ChatGPT had a good understanding of capabilities and risks and outlined popular uses. It also identified potential risks, highlighting the ethical and security challenges with use of the technology today.

The technology swiftly responded to all inquiries, and the speed was remarkable. Yet the responses were generic, lacking much detail. We found the responses to be a useful high-level overview that would seed deeper inquiry. For example, we might have followed up the risk query with a question such as, “What do you mean that bias is a risk?”

The real power of generative AI is its ability to provide more detail in response to more specific queries. Users can ask follow-up questions or request more detailed explanations, and the AI can respond with more specific, detailed, and nuanced information.

https://open.spotify.com/embed/episode/3mHkvGNGnGasuAdEDvVyqB

This iterative process of questioning and answering provides depth, making generative AI a powerful tool for learning and discovery.

The ChatGPT responses were insightful, but they are based on the information used to train the system. This means that answers can change over time, as the system learns. Further, the responses may not be accurate if the training data was inadequate or biased – one of the risks ChatGPT mentioned.

Managers are warned to not just believe the results of generative AI systems straight off. Instead supplement AI-generated insights with human expertise, validate responses, and ask the questions in multiple different ways for the best set of actions.

But overall, we found that the generative AI performed admirably in providing thoughtful examples and practical advice for managers. The easy interface, the speed of response, and the breath of answers were impressive. For our needs, we will use it again.