This content was originally featured in the May 1st, 2024 newsletter found here: INBOX INSIGHTS, MAY 1, 2024: AI ETHICS, MODEL TUNING
HELPFUL, HONEST, HARMLESS
Are you using Generative AI?
That sounds like the start of a sales pitch, doesn’t it?
It’s not. I promise.
I genuinely want to know if you’re using Generative AI. Not only do I want to know, but I want to know if you understand how the system you’re using decides what responses to give you.
I’m not going to get into technical details, that’s not why you’re here. There are academic papers and other articles that get into the weeds. For context, at a high level, the guiding principles for how companies, like OpenAI and Anthropic, are training Large Language Models (LLM) are HHH. HHH stands for Helpful, Honest, and Harmless.
Sounds good, right? Who wouldn’t want to use a system that is helpful, honest, and harmless?
Well, let’s not get ahead of ourselves.
There isn’t one singular definition of helpful.
There isn’t one singular definition of honest.
There isn’t one singular definition of harmless.
You see where I’m going with this.
Generative AI is a great tool to integrate into your workflow. There are a lot of reasons why marketers would want to optimize their efficiencies. But this is where I am going to encourage you to read the fine print. The thing that we all say we do, but don’t.
Ok, I’m actually going to ask you to do more than just that. Before you sign up for a generative AI tool to integrate into your workflow, I want you to go through a simple exercise. The goal is to determine what you will and won’t accept from a Large Language Model.
I’d like you to start by outlining your company values. When we think about helpful, honest, and harmless you should be able to tie those into what your company stands for.
As an example, here are the values that we outlined for Trust Insights:
- We reject deception and secrecy. We are transparent and honest.
- We reject laziness and stupidity. We are committed and smart.
- We reject obfuscation and bullshit. We are clear and direct.
- We reject discrimination and bias. We are fair and just.
- We reject ego and selfishness. We are humble and generous.
- We reject pigheadedness and willful ignorance. We are cooperative and aware.
- We reject gloomy and dramatic. We are cheerful and agreeable.
- We reject thoughtless acceptance of the status quo. We do better.
When I go through the exercise of selecting a piece of software, like generative AI, I want to have those values front and center. Why? Because the way that the model is trained may not align with your values. For instance, what I think is fair and just, may not resonate with you.
No, this is not normally a step you need to take when assessing software vendors. You want to take this extra step because of how companies are training the models. Unless you’re getting into the code (which they won’t share with you) you don’t know what the companies consider helpful, honest, or harmless. You have to do your due diligence and make those judgments for yourself.
Once you have a shared understanding of your value, go ahead and read the fine print, also know as the terms and conditions. Make sure you know what you’re signing up for and that you’re comfortable with the software. Generative AI is rapidly evolving. So quickly that most of us feel like we can’t keep up, let alone know exactly what it entails.
This is an important time in our industry to be skeptical and questioning. Someone who isn’t you is deciding what is helpful. Someone who isn’t you is deciding what is honest. Someone who isn’t you is deciding what is harmless. You can’t control that. But you can control whether or not to use their software.
Our friends over at the Marketing AI Institute are doing a lot of work trying to understand and educate on this topic. Be sure to follow them to stay up to date as well as following our content.
Are you clear on your values? Reply to this email to tell me or come join the conversation in our Free Slack Group, Analytics for Marketers.
– Katie Robbert, CEO
Newsletter
Podcast
