artificial intelligence

How ChatGPT Factored Human Judgment Into Algorithms

Artificial intelligence dominated the headlines this year after OpenAI launched ChatGPT in late November. The fact that one could type in “write a story about X” and ChatGPT could quickly compose a reasonable essay on the topic spooked a lot of journalists and programmers as well, since ChatGPT can write effective code, too.


We spoke with Avi Goldfarb, a professor of artificial intelligence and healthcare at the Rotman School of Management at the University of Toronto about AI and what it means for marketers. Below are excerpts of that conversation, edited for clarity and length.

Marketing Daily: Is ChatGPT wholly a product of computer science, or are there human biases in its answers?

Avi Goldfarb: It’s not a machine providing judgment. There's still a lot of human input into essentially figuring out which predictions to make and what to do with those predictions once you have the human judgment involved in in a large language models.



What Open AI did in their recent work -- and part of what made ChatGPT so powerful – is, they had many humans judge different identify which were better under very particular definitions.

They ended up fine-tuning the algorithm to reflect that human judgment. More helpful, honest and harmless can go a long way -- to providing good, useful guardrails on what kind of text should come out.

Marketing Daily: What do you see as the primary blind spot with AI right now?

Goldfarb: In many ways, the primary blind spot is on the part of humans using it rather than the AI -- because it's text, and the text is so well written. There's this impression that these are truly intelligent machines. And so, we try to use it the way we’d use a discussion with a human expert.

I see more evolution in that we're going to figure out what this tool is so useful for.

I think the most useful way to understand ChatGPT and other generative models is to think of them as incredibly useful new tools, and to get away from the metaphor of artificial intelligence. Instead, say, "Hey, we have a new tool. Let's think about how to use that tool as effectively as possible to enhance our lives at home and at work."

Marketing Daily: Do you see this as an effective tool right now for for ad agencies?

Goldfarb: It should help you design better copy. It should help you enable the copy to adapt quickly and at scale. So that you can -- once we figure out how to use it, and use it well -- write personalized copy based on the needs of a particular individual at a particular time who's seen your ad or your copy, and respond in real time. Now,  I haven't seen that happen yet.

But I just predict that kind of change will happen soon. The ability to write well is an important part of copy, but there's also the image generation tool that, again, can help deliver more effective, more responsive advertising.

Next story loading loading..