Seeing Like an Algorithm — Remains of the Day

Seeing Like an Algorithm — Remains of the Day

In today's digital landscape, machine learning algorithms have become a driving force behind the success of many companies. Whether it's TikTok's recommendation algorithm or the chatbot ChatGPT, understanding how these algorithms achieve their accuracy is crucial for businesses across all industries. While some may doubt that TikTok has made groundbreaking advances in machine learning recommendations, the true magic lies in its closed loop of feedback.

To design an app that effectively serves users, it is essential to consider how best to help an algorithm "see." TikTok's algorithm-friendly design is a prime example of this concept. From the moment a video begins playing, every action and interaction serves as a signal of the user's sentiment towards that video. Prior to being sent to your phone, a human on TikTok's operations team has already watched the video and added relevant tags or labels. Additionally, TikTok's camera filters are designed to track human faces, hands, or gestures, further utilizing vision AI.

One common design element found in our largest social networks is the infinite vertically scrolling feed. Instead of displaying one story at a time, these apps present multiple items on the screen simultaneously. While this may provide a lower friction scanning experience for users, it poses challenges for algorithms in accurately judging sentiment. Positive feedback mechanisms dominate, leading to a potential mismatch between a user's true interests and the content derived from their social graph. This divergence is inevitable if the algorithm fails to "see" signals of a user's growing disinterest.

However, algorithm-friendly design does not have to be user-hostile. It simply requires a different approach to serving the user's interests. The goal of any design should be to help the user achieve their desired outcome, even if it means introducing friction. In the software era, true competitive advantages are increasingly illusory, as most features or UI designs can be easily copied. The real magic lies in the interconnectedness of every element and process within TikTok's design, creating a dataset that allows the algorithm to train itself to peak performance.

Moving beyond TikTok, let's explore another fascinating chatbot called ChatGPT. While search engines provide reliable results by retrieving information from the internet, ChatGPT offers a more flexible approach. Its objective is not solely focused on factual accuracy, allowing it to generate information or provide convoluted yet precise answers. This flexibility comes with a tradeoff; ChatGPT can make up information as easily as it can provide accurate responses.

One notable difference between search engines and language models like ChatGPT is their approach to information retrieval. Search engines rely on a list of links that approximate the user's query, while language models have the ability to generate responses based on the given prompt. This distinction is important as search engines won't make things up since they search for existing information. However, they may still display fake news or unreliable sources that can be fact-checked.

It's crucial to recognize that the limitations often associated with language models like GPT-3 are not a reflection of their capabilities but rather our own shortcomings in formulating adequate prompts. Our ability to harness the true potential of these models is hindered by our own understanding and utilization. As we continue to explore the possibilities of AI-powered technologies, it is essential to push our own boundaries and leverage these tools to their fullest extent.

In conclusion, understanding how algorithms perceive and process information is crucial in today's digital landscape. Whether it's TikTok's closed loop of feedback or ChatGPT's flexibility, these algorithms have the potential to revolutionize industries. To make the most of these advancements, here are three actionable pieces of advice:

1. Embrace algorithm-friendly design: Consider how your app or platform can best serve the algorithm to optimize user experiences. Find ways to provide relevant and accurate signals to guide the algorithm's decision-making process.

2. Strive for alignment and interconnectedness: Ensure that every element and process within your design aligns with a single purpose and goal. By creating a cohesive ecosystem, you can maximize the algorithm's performance and effectiveness.

3. Push the boundaries of AI capabilities: Don't limit your usage of AI-powered technologies based on preconceived notions or limitations. Continuously explore, experiment, and challenge the capabilities of these tools to unlock their true potential.

By incorporating these principles into your approach, you can navigate the evolving landscape of machine learning algorithms and leverage their power to drive success in your industry.