I’ve read hundreds of AI reports, and this is the first that strives to make AI more human! It’s about time!
AI is so good it may fool you, but it doesn’t understand human wants or needs.
This is because AI training data sets don’t convey the human experience. Take X’s Grok for example, trained on X data. It will be totally anti-social!
Ipsos, the market research firm, has a solution to that and will Train AI on real human survey answers to give it the human touch in specific domains.
It’s a great idea and may help consumer financial services with product development.
👉TAKEAWAYS:
🔹LLMs should be trained on real human data from actual surveys so that the AI models will convey: Truth, Beauty and Justice!
🔹Truth: The AI model is accurate in reflecting people’s most recent reality
The problem with existing LLM models is that:
Generic training can produce critical knowledge gaps.
Unwritten realities can breed bias and misrepresentation.
Time capsules can fail to capture quick-paced societal change.
🔹Beauty: How explainable an AI is.
Explainable AI allows users to check the output and explain why an AI produced certain results. Survey data helps provide transparency.
🔹Justice: fairness and ethical issues
When trained with surveys, customer training data does not include prior bias, confidential data, or intellectual property.
👊STRAIGHT TALK👊
I think it’s about time that someone tries to make AI more human.
While it may sound trite, it’s big because we rely on AI to give us insights into our society. Those insights may be wrong because of poor, incomplete, or dated data.
By using actual surveys that contain the actual words of humans the AI can better learn how people think of a particular topic. This is big because it helps the AI better represent the human condition which in turn gives humans more faith in AI.
The report uses examples from consumer products, but using this technology in financial services is likely. If banks want to get to know their customers better, they may just want to build an AI that encapsulates their human condition.
My suggestion is that banks should start with their written online complaints! This will get the AI in a perfect mood to answer bankers’ questions!
Thoughts?