9 Ethics of building and using LLMs

 

This chapter covers

  • How LLMs’ abilities to perform many tasks also create unanticipated risk
  • The question of LLMs’ misalignment with human values
  • The implications of LLMs’ data use on content creation and building future models

Although the discussion of ethics may remind some of you of the dull readings from an entry-level college class, there are critical considerations when implementing algorithms that have the potential to affect humanity. Given the rapid growth in LLM use and their scope of capabilities, we must be aware of and attend to many evolving concerns. If you are unaware of these concerns, you will have no voice in their resolution.

Exploring the ethics of building and using LLMs is an incredibly complex topic that is challenging to represent completely. As a result, this chapter will present what we believe to be common concerns about building LLMs and the related ethical questions. Throughout the chapter, we’ll reference materials that round out this conversation so you can investigate further if you wish.

We’ll cover three main topics:

9.1 Why did we build LLMs at all?

9.1.1 The pros and cons of LLMs doing everything

9.1.2 Do we want to automate all human work?

9.2 Do LLMs pose an existential risk?

9.2.1 Self-improvement and the iterative S-curve

9.2.2 The alignment problem

9.3 The ethics of data sourcing and reuse

9.3.1 What is fair use?

9.3.2 The challenges associated with compensating content creators

9.3.3 The limitations of public domain data

9.4 Ethical concerns with LLM outputs

9.4.1 Licensing implications for LLM output

9.4.2 Do LLM outputs poison the well?

9.5 Other explorations in LLM ethics

Summary