Listen now 57:05 | An interview with USA-based branding & marketing strategist Phil Pallen on his stories of using AI and how he feels about how AI is using people's data and content
Hi Karen, really enjoyed your latest post with Phil Pallen.
Got a question for you. In a conversation someone (who is a lot smarter than I am when it comes to understanding AI) said "AI rewards mediocrity." Their point was that because AI "scrapes" whatever exists in terms of Internet information on a particular topic, it ends up with what is a normal distribution on positions on that topic. In other words, it will share what "most" people are saying about that topic. You will not end up with the outliers which are perhaps more creative, certainly unusual perspectives. Is this correct?
Because another person I know has argued that AI picks up outliers (for example customer reviews) because people tend to write reviews only when they are very satisfied or dissatisfied. Maybe it's both? Wondering what you think?
I want to add here that the "mediocrity" comment was in response to the question "will AI put artists and creators out of business." In the commenter's opinion, no it will not but it DOES raise the bar on what we get paid to create.
Hi Judy, thanks for the question! AI does, statistically, reflect what most of its data tells it. So the first person isn’t wrong. I think the second person is interpreting ‘outliers’ differently, though.
AI training datasets tend to be biased in one or more ways. AI tools can only reflect the data they are given.
In the case of customer reviews, AI sees only the more polarized opinions that are written-up and then scraped as its training data. AI will never see most of the unwritten, less-polarized opinions. So AI cannot reflect them. Those polarized opinions may be outliers (atypical) in the real world. But they are not outliers in terms of the whole dataset the AI tool sees.
This is a self-selection bias, and it’s very common. Example: When an influencer posts a poll on social media, its results won’t reflect the true world view. It will only reflect the views of that influencer’s followers - and ONLY the subset of followers who feel strongly enough about the topic to vote openly.
Does that help? Or does it raise more questions? :)
Hi Karen, really enjoyed your latest post with Phil Pallen.
Got a question for you. In a conversation someone (who is a lot smarter than I am when it comes to understanding AI) said "AI rewards mediocrity." Their point was that because AI "scrapes" whatever exists in terms of Internet information on a particular topic, it ends up with what is a normal distribution on positions on that topic. In other words, it will share what "most" people are saying about that topic. You will not end up with the outliers which are perhaps more creative, certainly unusual perspectives. Is this correct?
Because another person I know has argued that AI picks up outliers (for example customer reviews) because people tend to write reviews only when they are very satisfied or dissatisfied. Maybe it's both? Wondering what you think?
I want to add here that the "mediocrity" comment was in response to the question "will AI put artists and creators out of business." In the commenter's opinion, no it will not but it DOES raise the bar on what we get paid to create.
Hi Judy, thanks for the question! AI does, statistically, reflect what most of its data tells it. So the first person isn’t wrong. I think the second person is interpreting ‘outliers’ differently, though.
AI training datasets tend to be biased in one or more ways. AI tools can only reflect the data they are given.
In the case of customer reviews, AI sees only the more polarized opinions that are written-up and then scraped as its training data. AI will never see most of the unwritten, less-polarized opinions. So AI cannot reflect them. Those polarized opinions may be outliers (atypical) in the real world. But they are not outliers in terms of the whole dataset the AI tool sees.
This is a self-selection bias, and it’s very common. Example: When an influencer posts a poll on social media, its results won’t reflect the true world view. It will only reflect the views of that influencer’s followers - and ONLY the subset of followers who feel strongly enough about the topic to vote openly.
Does that help? Or does it raise more questions? :)