AI’s Awkward Acne and Crackly Voice Phase

We’re in an awkward in-between period of the evolution of generative AI, and one day we’ll look back and chuckle at ourselves. In the meantime, let’s try our hardest not to make fun of AI’s pimples and crackly voice.

Human nature is such that we often use new technologies in ways that are similar to those of previous technologies, because we haven’t yet figured out the native benefits of new technologies.?

Early television shows were direct adaptations from radio shows like “The Lone Ranger” and “Amos n’ Andy”. Early automobiles looked like horse drawn carriages. In the early days of computers, we would send textual letters to each other like physical mail, except we would call it “email”.?

Oh, wait… I guess that one somehow never died.

We’re doing the same thing today with generative AI, and it’s a period that we’ll ultimately look back on as a tiny sliver of time in the history of AI. I’ll call it the Human period of AI.

It affects both sides of AI. Today we train our AI with data that was originally created for human consumption: PDFs, articles, message boards, books, photos, etc. And much of what we use AI to generate is created for the purpose of human consumption: answers, summaries, writing, chatbots, translation, art, advertisements.

Feeding AI the wrong food:

This is highly inefficient. Information created for human consumption is just not well suited to ingestion by computers for AI training. First, it’s challenging for computers to understand it. What’s more, there’s a dwindling amount of high enough quality human-intended content, and with the growth of generative AI that’s unlikely to improve. We don’t feed spinach to robots, or talk to trees. We’d be much better off training AI with information that was created for the sole purpose of training AI.

Asking AI to do the wrong stuff:

The same is true on the backend. Using AI to create content for human consumption is very short-sighted and of limited use. There’s so much more value in utilizing AI to create information for the purpose of ingesting into other AI.

A marketing example:

Let’s use marketing content as an example. There are lots of companies offering generative AI to create marketing and sales content. This is the equivalent of a car that looks like a horse drawn carriage, or a TV show that’s a visual radio show.

Let’s take B2B marketing specifically. In 5 or 10 years, will companies make procurement decisions based on a couple humans reading a couple dozen articles or emails that were written by AI? That’s so 2024.

Rather, in the near future, companies will use AI to help them make these decisions. AI will ingest hundreds or thousands of sources of various types, and score options against custom criteria specific to a company’s needs. And it’s that scoring that humans will review. So B2B marketing content will ultimately need to be what feeds into these types of systems most effectively, and that won’t look like slick 5-paragraph thought leadership blog posts or pithy sales emails.

And that’s not even taking into account the high risk of mediocrity in AI-created marketing, an issue I’ve been documenting for well over a year, and has been subsequently backed up by academic research.

Native AI use cases:

Use cases for AI that are intended for ingestion by computers are the ones I see as being more scalable and sustainable:

• Code
• Pharmaceutical drug discovery
• Synthetic data

This list is incomplete because we don’t know all the native uses for AI yet. Just like no one could have known at the advent of 4G wireless technology that it would lead to ride-sharing or Snapchat.

In summary, I recommend getting excited about AI use cases that are focused on feeding scalable data into other systems, rather than those that are trying to recreate what humans did before generative AI. And one of the best ways for us to get there will be to search for training data that was purpose-built for the training of AI, not re-purposed from what was originally intended to be human consumption.