The first two weeks of the Generative AI for Everyone went through an introduction of AI terminology and functions and then how AI tools function.
In week 3, Andrew Ng discusses the application of AI tools in real world business and society environments.
He discusses using a general-purpose AI tool as a writing assistant or editor for brainstorming, summarizing text, or writing code.
An interesting point he makes is related to the concern that jobs will be lost. But then he points out that it isn't likely that AI tools will be used to automate jobs but rather to automate tasks. I think this is a great way to think about it, since AI isn't going to cause jobs to go away (or at least very many of them) but it will take certain tasks away. This will cause more people's jobs to change but will not likely actually take the jobs themselves. In particular, AI can be used to augment tasks, which is where the AI assists and makes the job easier, vs. automation which is where a task is actually completely given to the AI to perform. This may be a process in some situations, to augment initially, while keeping a human in the process, and eventually as the system is trained those tasks may be trusted enough to be run automatically.
If you're unsure if an LLM can complete a task, experiment. Ask it to do the task, and see what happens. He pointed out last week that the internet won't blow up if you ask an LLM to perform a task and it can't for any reason. the answer may not be yes or no but not yet, meaning some fine-tuning or training could help it learn how to do so. The question is how much time is taken on the task now and how much value is created by using AI to do it faster, cheaper, or more consistently?
An interesting concept Andrew calls out is that we usually think of the iconic part of a particular job role, such as a lawyer arguing a motion in court or a doctor performing surgery. What we don't think about are all the more mundane tasks that have to be done outside of the most iconic pieces. An AI tool could be used to take a difficult process faster and shorter to do the same basic task. Or it can be used to do a deeper analysis, which may take the same amount of time as it would initially but end up with a better result by doing more thorough testing and analysis. He shares a great quote by Curtis Langlotz, which is that "AI won't replace radiologists. But radiologists that use AI will replace radiologists that don't." In the past, technology has been used to create more jobs than it has destroyed, since the cost savings potential of new technology is limited while the growth potential of new technology is unlimited.
Data on the internet which are used to train data models represent our past and present. We may be hopeful for something different in the future. Fine-tuning and reinforcement learning can be used to create a fairer, less biased, and more just future.
In the conclusion, Andrew calls out how human intelligence is expensive. It takes a lot of time and money to train a wise human being, and thus only the wealthiest people can afford to hire the most intelligent people. Artificial intelligence, however, is much less expensive, so AI can be used to give everyone the ability to hire intelligence at low cost.