Can we trust artificial intelligence to make decisions for us? | Bradley Hayes | TEDxMileHigh

Summary

In this talk, the speaker explores the growing role of artificial intelligence (AI) in our daily lives and the challenges of trusting autonomous systems. While AI systems like recommendation engines and self-driving cars are already making decisions for us, the next wave will involve even more critical decisions, such as medical diagnoses or guiding visually impaired individuals. The speaker advocates for explainable AI, emphasizing the importance of understanding how and why these systems make decisions to build trust. Only through transparent communication between AI and humans can we ensure that autonomous systems benefit society responsibly and safely.

Takeaways

• 😀 AI is increasingly present in our daily lives, from vacuum cleaning robots to hotel concierge robots.
• 😀 The next generation of AI will not only automate physical tasks but also make critical decisions that impact our lives.
• 😀 Current AI systems make decisions for us in areas like Netflix recommendations, YouTube, and Google Maps, but they often lack transparency in how they make these decisions.
• 😀 To trust AI, we need to understand how and why it makes decisions, which is where the field of explainable AI comes in.
• 😀 We trust humans to make important decisions for us because they can demonstrate their abilities and explain their decision-making processes.
• 😀 AI systems, unlike humans, don’t share experiences or common sense, which makes their decision-making process harder for us to understand.
• 😀 Explainable AI helps bridge the communication gap between AI systems and humans by providing insights into AI’s decision-making logic.
• 😀 Machine learning allows AI systems to learn from data and experience, but it can be difficult to understand what the machine has learned or how it learned it.
• 😀 A well-known example of a flawed AI system is one that learned to identify wolves based on the snow in the background of photos rather than the actual animal.
• 😀 For critical systems like robotic seeing-eye dogs, it’s vital that AI systems can explain their decisions, ensuring they make the right choices in real-world scenarios.
• 😀 Creating trustworthy AI requires broad participation, with everyone asking questions and ensuring that AI systems work for all, not just a select few.

Q & A

What is explainable AI and why is it important?

Explainable AI refers to AI systems that can explain their decision-making process in a way that humans can understand. It is crucial because as AI systems take on more critical tasks, like medical diagnoses or driving vehicles, we need to understand how and why they make decisions to trust them fully.

How do we currently trust humans to make important decisions for us?

We typically trust humans through two methods: demonstration and explanation. Demonstration is seeing someone perform a task successfully (e.g., parallel parking), and explanation is when someone clearly describes the rationale behind their decisions (e.g., a doctor explaining a diagnosis).

What is the challenge with understanding AI's decision-making process?

The challenge lies in the fact that AI systems, especially those using machine learning, learn from data rather than being explicitly programmed. This makes it hard to trace the exact rules or patterns the AI has learned, often resulting in a 'black box' situation where we can't easily understand how it arrived at a decision.

What does the example of the AI system identifying wolves and dogs illustrate about AI's learning process?

The example demonstrates that AI can make incorrect associations if the data it learns from is biased or limited. In this case, the AI mistakenly learned to identify wolves based on the presence of snow in the background, not the characteristics of wolves or dogs themselves.

Why is it dangerous to assume a robot understands a concept based on just a few examples?

Assuming that a robot understands a concept based on limited examples can be dangerous because the AI might learn to associate concepts with superficial features, like how a person looks, rather than the true nature of the object or concept. This could lead to errors in decision-making.

How does explainable AI contribute to the development of trustworthy systems?

Explainable AI helps us understand the reasoning behind an AI system's decisions, allowing us to verify if it is making the right choices. If we can trust how an AI reaches its conclusions, we can confidently deploy it in critical applications like autonomous vehicles and healthcare.

What are the potential benefits of trustworthy AI systems in society?

Trustworthy AI systems can bring about significant benefits, such as providing mobility to people who can't drive, improving efficiency and safety in manufacturing, and revolutionizing medicine by reducing costs and enhancing the quality of care.

What role does collaboration play in advancing explainable AI?

Collaboration is essential to advancing explainable AI because it involves input from a broad range of people and perspectives. We need diverse participation to shape AI systems that are transparent, understandable, and trustworthy, ensuring they benefit society as a whole, not just a select group of researchers or programmers.

How does explainable AI improve communication between humans and autonomous systems?

Explainable AI improves communication by allowing AI systems to express their decision-making processes in terms that humans can understand. This requires AI to use high-level concepts, such as 'person' or 'crosswalk,' that align with human experiences, ensuring that we can interpret their actions correctly.

What can we learn from the analogy of a robot's perception of humans based on their shoes?

The analogy highlights that AI systems can develop narrow, context-specific interpretations of the world based on the data they are trained on. In this case, a robot might identify humans based on features like sandals without understanding the broader concept of a person, which could lead to errors in decision-making if the context changes.