The Future of Machine Learning: A World of Frustration and Diffused Responsibility

In the near future, machine learning technologies will increasingly frustrate us and obscure accountability. Companies are diverting customer service to large language models (LLMs), making it harder to reach human agents. These models will lie, make unkeepable promises, and complicate problem resolution. The concept of 'agentic commerce' suggests new advertising methods, dark patterns, and confusion.

## Customer Service

I've spent a significant part of my life dealing with customer service issues, from insurance denials to billing errors. Machine learning will make these interactions more annoying. Companies view customer service as a cost to minimize, using offshoring and scripts to reduce expenses. Now, they're pushing support requests to LLMs, which will soon handle phone calls too. These machines, while polite, will often do stupid things and lie, making the process infuriating for those with complex issues.

## Arguing With Models

LLMs will extend beyond support to handle 'fuzzy' tasks like parking fines or insurance assessments. They don't need to be accurate, just cost-effective. This will create new drudgery as people try to outsmart these systems. Doctors might need to phrase requests carefully to get necessary procedures approved, and consumers may have to game systems to get fair prices.

## Diffusion of Responsibility

Machine learning models will harm innocent people, as seen in cases where facial recognition led to wrongful arrests. These aren't just failures of technology but of sociotechnical systems. Human oversight often fails to catch absurdities, and ML systems can encode biases, presenting them as objective truths. The complexity of these systems diffuses responsibility, making it hard to hold anyone accountable.

## Market Forces

'Agentic commerce' involves LLMs making purchasing decisions, potentially reducing ad revenue as they replace human buyers. Companies will try to influence LLM behavior, leading to an arms race similar to SEO battles. This could warp the internet and LLMs in unpredictable ways. As LLMs negotiate prices, consumers might find themselves in bizarre exchanges, with companies exploiting these interactions for profit.

In this future, ordinary people may have to run personal LLMs to navigate commerce, leading to exhaustion and frustration. Despite the potential for self-limiting behaviors, the widespread adoption of LLMs may force everyone to participate, creating a complex and biased commercial landscape.