A Proposed AI Contribution Scale

2Bobs with David C. Baker and Blair Enns







Download

Categorizing AI-generated copy as “human vs AI” is not as binary as with AI-generated images or video. So Blair has developed an AI Contribution Scale to help us all be more transparent when using LLMs in the creation of written communication intended for prospects, clients, and peers.

[This episode’s notes are AI-0]

‍ ‍

Links

AI Contribution Scale

“The Coming Collapse of ‘Thought Leadership’” by Blair Enns for winwithoutpitching.com

Curt Jaimungal’s 5-minute YouTube video on why you shouldn’t let AI write for you

Transcript

David Baker: All right, Blair, today we're going to tackle another AI topic.

Blair Enns: This is not the AI episode [chuckles].

David: One of these days, maybe in 2029, we'll talk about AI and be willing to stand behind it. So far we're just terrified of our own thoughts about this.

Blair: Dancing around it.

David: Yes, really, but I'm interviewing you on a proposed AI contribution scale, the idea being that we should probably identify, self-identify how much we used AI. I'll tell you, it's like I got an email from a competitor of mine, yesterday-- Well, a friendly competitor. They do slightly different things, but somebody might hire them instead of me. I've always gotten their emails and I always loved getting them, because I would get some interesting ideas in there. This one just smacked of AI, so I copied it and pasted it into the AI detection thing, and it said immediately, "100% AI."

I just wanted to unsubscribe and never read anything from that person again, which is an overreaction. Is that what prompted this, or am I--

Blair: Yes, that's part of what's going on here. I want to contrast that with something else. We were recording this September 1st, last week. I didn't read it, but I think it was The Wall Street Journal, Stanley Druckenmiller wrote an op-ed. Now, Stanley Druckenmiller, if you're not a finance or an investment geek, is probably one of the world's foremost macro investors. "Macro" meaning he basically invests in economies of countries or currencies, and he's famous for-- I think his first big trade in the '80s was being long on the Deutschmark at the time of German reunification when everybody else was short.

He made a few billion, and then he shorted the British pound a couple of years later, made a few billion more, and he just kept going after opportunities. Anyway, so he worked for George Soros. Scott Bessent, who is the Treasury secretary. Is that right? I think he's the Treasury secretary.

David: Yes.

Blair: Of the United States.

David: At least this morning he was. [laughs]

Blair: Yes. He was Druckenmiller's protégé. The geeky stuff is, Bessent's been buying US Treasuries--

David: With borrowed money.

Blair: Yes. US Government is printing a whole bunch of debt, as it does, along with every other government in the Western world, and then Bessent is buying it to keep the interest rates low because they've got to refinance $10 trillion in debt in the next 12 months or so. Then Druckenmiller, his former mentor, wrote this op-ed, again, I think it was in The Wall Street Journal, saying, "This is a bad move." He wasn't really criticizing Bessent. He was essentially criticizing-- and this is somebody else's take on it that I just heard last week, and it makes sense. He was essentially criticizing the US Congress for putting Bessent in this position where he had to do this.

He publishes this substantive piece that says, "Hey, listen, we're headed towards a cliff here." The consensus reaction was not about the substance of his piece. It was about the fact that it was so clearly written by AI. People started saying, "AI wrote this," and he said, "Yes, I had AI write it." He had no qualms about it. Now there's this big discussion about, should you write an opinion piece in one of these big papers using AI? The answer is probably not. In Druckenmiller's case, I want to make the distinction between that piece of thought leadership that you read, and you were so disappointed by the author, because in that example, you were using that writing as a proxy for that person's thinking and their intelligence.

You're making some mistake, because that's the world we're coming from, where somebody's writing was a proxy for their thinking. The output was a proxy for the input, for what's going on in the brain of this person. In Druckenmiller's case, he's already regarded as one of the most accomplished investors of all time, and he's simply trying to communicate a point. Getting the AI to better communicate that point, I give him full props for that and for owning it. I want to make the distinction between two different types of writing. There is the writing to communicate, and that is writing that you might do internally.

Even my marketing writing, the marketing writing that we do, we use a fair amount of AI to help us in that writing. Then there's the writing that you do to think, to further your thinking, to deepen your thinking skills, to find out what it is that you actually know and understand. When you publish that form of writing, that is thought leadership. Let's make the distinction. I'm not saying AI is bad for writing, but I'm with you. There are certain types of writing where people are effectively lying, using AI to lie about the thinking that they did not do.

There's a lot of discussion in the world right now, and I think the EU is mandating watermarking of AI-created works, including writing. We can go down that rabbit hole, but that makes sense for images and video, perhaps. It doesn't make sense for writing, because there are a lot of valid different ways that you might have AI contribute to your writing. As long as you voluntarily disclose how you used AI in your writing, I think that's the answer. It's not watermarking. The problem with watermarking is it's binary, this is AI or this is human.

I created this AI contribution scale of 0 to 5, so six different levels, where you can voluntarily disclose the extent to which you used AI, and then you can let people decide for themselves whether or not the thinking is yours or the machine's.

David: I read through these six levels, from 0 and then 1, 2, up through 5, and I'm not sure I would list everything I've written in the last two and a half years as AI-1. In other words, it's AI-researched. To expand on that, you say that's written entirely by the author, with AI contributing research. AI might be used after the fact for proofreading and formatting or identifying awkward passages without suggesting corrective language. No substantive AI-generated language was retained unless cited. That's an exact description of how I use AI and have for the last two and a half years, and every single thing I write would fit in that category.

My first thought as a user, like if I didn't know you and I was just reading this suggested contribution scale, I would just have on my website this statement that says, "Everything you read here is AI-1, and AI-1 means this." Then I wouldn't call it out every time I wrote something. You're suggesting that people might use AI to different degrees, and so they might identify each use if it's different, right?

Blair: Yes, that's right. I published this about a week or so ago, and I have committed to, on my public blog posts, so not in the Win Without Pitching Academy, where I don't use any AI, but on my public blog posts and on LinkedIn, I will just list the AI contribution scale. I will say, "This is AI-0, human," or, "AI-1, AI-researched," and then there's "AI-2 is AI-assisted," "AI-3 is human/AI collaboration," "AI-4 is AI drafted," and "AI-5 is AI-generated." There are some governing principles at the top of the scale that we can talk about. If you want to look at the scale in detail, you can go to winwithoutpitching.com/aiscale, and we will put a link in the show notes.

David: I read about that. I didn't read it myself, but I think even if the content was really good, I imagine I would have quit reading early on, because everything that I have run through that little tester that has been 100% AI is too long, and it bores me to death. Maybe that's just me. Even if you're using AI to generate something that's really useful and good, it might still be very boring. I just don't understand why you would even do that.

Blair: I'm giving Stanley Druckenmiller the benefit of the doubt here, because he's quite an accomplished human being. His track record speaks for itself, and when he speaks, you should probably listen. If he chooses to use A