AI Tools Provide Some Fun, But Far Too Few Facts
Everyone is talking about AI and how it will change our lives. Some of you have perhaps even played around with a variety of text generating AI tools or spent more hours than you ever expected creating images using AI. They are undoubtedly fun, but in their current incarnation many of these products are fairly flawed and render subpar results.
Arguably, the delta between the perfect results we’re expecting and the passable (at best) outputs we’re receiving from these tools is a byproduct of various companies trying to move fast while simultaneously innovating very much in the public eye. The vast majority of these AI businesses are simply building on top of widely available models and algorithms that are open source, but also open to disaster.
It seems many of these companies should focus more on the quality rather than the commerciality of their products. Dragomir Radev, a noted computer science professor at Yale, said more than six months ago, “I don’t think that there will ever be a time where computers will be as good as humans at every language task. There are certain tasks that will always be beyond their limits.” That’s not to say we won’t get better, but currently we’re not even close enough.
The recent announcement of a nine-figure Series A for an AI company that generates content prompted me to test that software out of professional curiosity about how it may affect the SEO industry. (To be sure, I’ve previously kicked the tires on many other AI text tools and they’ve universally disappointed.) I also decided to horse around with those highly touted text-to-image AI programs.
Both popular and highly valued products – which I will not name, out of respect – ask users to input simple terms to help yield results. And both failed miserably.
For the text generator, I simply added my name “Michael Lewittes.” And since AI models are based on what’s publicly available on the web, I was interested in seeing what that content creating program would produce about me since there’s actually quite a lot of information about me on the web, including but not limited to a few bios such as the one of my company Ranktify’s website, as well as the one here on grIP.
NOTE: There are no lawyers in the world named “Michael Lewittes.”
Below is word-for-word of the first three paragraphs that the program generated based on the input, “Michael Lewittes”:
Michael Lewittes is a well-known American lawyer. He has worked on some of the most high-profile cases in the country, including the O.J. Simpson murder trial and the impeachment of President Bill Clinton. In addition to his legal work, Lewittes is also a prolific writer, and has authored two bestselling books on law and politics.
Lewittes was born in New York City in 1965. He received his undergraduate degree from Columbia University in 1987, and his law degree from Harvard Law School in 1990. After graduation, he clerked for Judge Abner Mikva on the U.S. Court of Appeals for the D.C. Circuit. He then went into private practice, first at law firm of Kirkland & Ellis in Chicago, and then at the New York firm of Davis Polk & Wardwell.
In 1994, Lewittes left private practice to join the O.J. Simpson defense team as a special advisor. He played a key role in Simpson's acquittal, and his work on the case was chronicled in Jeffrey Toobin's bestselling book, The Run of His Life: The People v. O.J. Simpson.
While that person sounds very impressive, he does NOT exist. And the text that was generated is most assuredly NOT about me. Among the AI program’s many mistakes are these clunkers:
I am not a lawyer
I did not work on the impeachment of President Bill Clinton
I was not born in 1965
I neither attended Columbia nor Harvard
I never clerked for anyone on the U.S. Court of Appeal for the D.C. Circuit nor elsewhere
I didn’t leave private practice to join the O.J. Simpson defense team as a special advisor nor play “a key role in Simpson's acquittal”
I’ve never authored any bestselling books, let alone two on law and politics
I was never written about in a bestselling book by Jeffrey Toobin (side note: nor have I ever been on a Zoom with him)
These wholly fictitious “facts” were all generated from an AI company that’s now valued at $1.5 billion.
The text-to-image generator did not fare much better. I gave the instructions to create “a centaur with Bryce Harper's face on a horse's body.”
Admittedly, it’s a weird prompt but that’s what these image-generating programs are supposed to do -- marry images of one entity [Bryce Harper] to another entity [horse].
What resulted is the image above.
Unquestionably, AI will power the future, but using it in its present state – with all its flaws – can be as scary as that faceless image of Harper holding a horse’s head next to… I-don’t-know-what. And you can trust me, because I played a “key role in Simpson's acquittal” -- no, I did NOT!