The Misconceptions of AI: Separating Fact from Fiction

As I prepared for my trip to India, I found myself struggling through a buggy online visa application process. After finishing, I decided to test ChatGPT’s capabilities by asking it about the process. Unfortunately, most of the points it provided were partially or completely wrong. This experience highlights the misconceptions surrounding AI and its limitations.

The Unfair Test

I asked ChatGPT about the visa application process, knowing that it would struggle to provide accurate answers. This was an unfair test, as AI models are not databases and do not produce precise factual answers. They are probabilistic systems, not deterministic, and cannot guarantee accurate answers. However, this test is relevant because it demonstrates the importance of understanding AI’s limitations.

The Misconception

Many people assume that AI models are useless because they cannot provide precise answers. This is a misunderstanding. AI models are excellent at providing answers that look correct, even if they are not entirely accurate. There are use cases where “looks like a good answer” is exactly what you need, and others where “roughly right” is “precisely wrong.” The key is to understand when to rely on AI and when to seek alternative solutions.

The Science Problem

One approach to improving AI is to treat it as a science problem. The models will get better with time, but how much better? The answer lies in the ongoing debate among machine learning scientists. However, this approach may not be enough to solve the problem.

The Product Problem

The alternative approach is to treat AI as a product problem. How can we build useful mass-market products around models that will inevitably provide incorrect answers? This requires a shift in focus from the AI model itself to the product design and user experience.

The Product Design Problem

The current product design misleads users by communicating certainty when the model is inherently uncertain. The solution lies in creating a product that communicates the limitations of AI and guides users to ask the right questions.

The Alternative Approaches

There are two alternative approaches to building AI products. The first is to contain the product to a narrow domain, creating a custom UI that communicates what the model can and cannot do. The second approach is to abstract the AI away, making it an enabling technology that users are not even aware of.

The Future of AI

As we look to the future of AI, we must consider what native experiences we can build that take advantage of this technology. We need to unbundle the general-purpose technology into single-purpose tools and experiences. This might involve creating products that users do not even realize are powered by AI.

The future of AI is uncertain, but one thing is clear: it will change the world.