APRIL 2024 IN
RICK ASTER’S WORLD
The latest AI tools are tools of imitation at heart. This limited scope of the AI tools will limit their life span.
The name of one current popular AI tool, Copilot, looks so much “Copier” that readers regularly confuse one word for the other. This resemblance speaks volumes about the ultimate value of the current generation of AI tools. In a very real sense, AI is the Xerox copier of today.
The term “generative AI” is a little too generous sometimes, as AI tools copy so freely from their unnamed sources that plagiarism has become a major concern. There are unresolved legal questions about the extent to which AI users may be liable for their use of a tool’s content copied from a published source.
On the other side of things, AI-generated content can never be protected by copyright or patent, as laws stipulate that only human-created work is eligible. An AI-generated logo might similarly be impermissible for use as a trademark, if it plagiarizes existing visual art works or if it imitates existing trademarks.
Setting legal concerns aside, it is not necessarily a good look if an expert’s AI-generated “5 tips” are the same as those listed somewhere else years earlier.
Going deeper, we may soon collectively come to appreciate the limited value of rehashed content that an AI tool pieced together without any deep understanding of the subject.
In this sense, AI is not so different from the Xerox machine. The technology of photocopiers has never been more precise or more accessible, yet the number of photocopy pages printed has declined to less than a tenth of what it was at its peak around 1990. The decline is so large that Xerox’s home page no longer mentions copiers, except in the clever marketing line, “We used to be copiers. Now we're innovators.”
In the same way, I expect that the use of AI tools to generate imitative content will hit an early peak, then lose its prominence as the not-so-original content loses its appeal.
All it really does is copy, and in a few years we’ll start to run out of things that are valuable to copy.
Most of the use cases of general-purpose imitative AI will be taken up by special-purpose algorithmic tools that don’t have the same copyright and plagiarism questions. Why not have a program specifically to generate a background vocal group for a song, if it it is not a featured part of the song, or to detect style mishaps in a blog post or book manuscript?
One of the first places where this switch will happen is in corporate information technology. Here it is security and accuracy that will drive the changes. The insistence on well-controlled processes in the corporate world will lead it to create and test specialized tools, many of them simple scripts, to replace the requests that ask the AI tool to “Do something like A, but in the context of B.”
It won’t take long for these special-purpose tools to outshine the general-purpose tools in every obvious use case, even if they start out by imitating the effects of AI tools.
We are already seeing these tools in other contexts, photo editing, for example. Specialized editing tools already exist to remove objects from photos or automatically balance colors for a particular purpose.
As another example, in fraud detection, specialized tools detect patterns of fraudulent transactions many times faster than the previous generation of general-purpose tools.
AI tools might inspire a million such specialized tools. That might sound like a lot of work for software developers, but it really isn’t. For comparison, consider fill-in forms. They were one of the best use cases for the photocopier in its classic era. Now those forms are online. It took 50 million fill-in forms online, covering everything from personality tests to product configurations, to take away the need for the copier, but it was still a cost-saving transition. This shift offers a hint at the limited life span of the current AI tools.
Fish Nation Information Station | Rick Aster’s World | Rick Aster