Gen AI Ain't Your Buddy - Neither Is Your Lawnmower

We humans tend to anthropomorphize anything we can. We give our vehicles names and buy them gifts, or threaten them when they don't work the way we need them to. We get emotionally invested in our homes, computers, yards, even our grills and smokers. Now enter Generative AI, a technology that can interact with us almost like an actual person - via plain language text and audio. It is enough to

We've dealt with this in fiction numerous times, from Asimov's writings to Philip K. Dick's replicates, to R2-D2 HAL and KITT, all the way through Raj's fantasies about Siri. We know better. But yet we're falling in love anyway - we can't seem to help it.

And like a person who sees red flags as a reason to pursue a partner all the more, we seem to be really good at overlooking Generative AI's shortcomings, figuring that it won't hurt me, or that it will be different this time, or that I really won't put myself in a place to be hurt again yet we go crashing right into a situation we objectively know isn't going to end well. Hey, we're only human. But our AI is not.

Sarcasm: the Bane of Gen AI

No, I'm not being sarcastic. Mostly. But Gen AI can't seem to figure out when we humans are spouting facts or saying something that isn't necessarily true and factual to make a point. It isn't just sarcasm, it is all of the quirks in our language: hyperbole, jokes, cynicism, exasperation - anything that isn't a straightforward statement of facts and truths. Anything that has emotional content to it that is expressed not in the words, but in the delivery of the words. the phrase "I really love you" can have several different meanings given context, body language, vocalization, emphasis, etc. can confound Gen AI solutions.

Sure, AI can get jokes, in much the same way a seven year old who just bought a joke book at the school book fair gets jokes - they can parrot the content but they don't necessarily comprehend the nuance and concepts that make the joke funny. That shouldn't be too surprising though: think about how you'd explain the concept of humor and comedy to a non-human in a way that doesn't ultimately collapse into "you just have to experience it to get it."

The School of Hard Knocks

There's nothing like life experience to teach us invaluable lessons. Did you learn how to balance on your bicycle by reading about it in a manual, or did you just have to go do it and get a few skinned knees and elbows before you got good? Do you really consult a manual before you do your laundry?

How many of the things we know - like not to put glue on a pizza - are things that have to be explicitly stated somewhere? Did you have to be told not to shout "movie" in a crowded firehouse? Do you have a moral compass? Did you learn that "I don't know" is many times the best response to a question? Do you get emotionally involved in a project and the completion of it?

Gen AI has none of these advantages, and hasn't learned any of these lessons. (Yes, Gen AI isn't about to go riding bicycles any time soon, but please follow the metaphor, OK?)

Lesson Plans

Most of us humans - at least those who are likely to read this article - were beneficiaries of a school curriculum, one that focused on teaching us how to discern fact from fiction, how to investigate and learn, and how to communicate thoughts and ideas. We were trained to cite our sources, though many of us have conveniently forgotten that lesson. There's a reason that accredited colleges and universities hand out advanced degrees and Reddit doesn't.

Now remind me again where Gen AI has "learned" from. Generally, the Internet. You know, the place that houses both the New York Times and your Uncle Paul's conspiracy theory spewing website. (Not to disparage your Uncle Paul, I'm sure he's a really nice guy in person) The place where Wikipedia and 4Chan both exist. A place where facts take a back seat to agendas, and deep investigation generally stops at about the 2nd page of Google search results.

Why do Gen AI solutions train so broadly? Well, there's no larger singular repository of information in our world than the Internet today. If you don't want to spend time creating a detailed training plan, and validating the content the AI trains on, well, you just set it loose.

Limits and Abilities

No need to hold that against Gen AI, but these sorts of things begin to define the abilities and limits of these systems and solutions. Understanding these abilities and limits can assist you in making good choices when interacting with Gen AI solutions.

This is the same as understanding the limits of a table saw. Your table saw, for example, has a really dangerous spinning blade. If it is new enough, this blade has a really effective brake system to prevent harming you, but you'll have to replace the blade because the brake damages it. If it isn't new enough, that blade isn't going to stop for nuthin but a power outage. Table saws are extremely good for dealing with wood sheet materials like plywood and paneling. However it usually isn't a great choice for sheets of metal, drywall, or insulation. Your table saw is fantastic for ripping lumber down to thinner dimensions. It's not terribly good at cutting lumber to length however - you can do it, but a handheld circular saw or a power miter saw is better for that task. But if you want to do any of those tasks without using electricity you'll want a hand saw or perhaps a water-wheel powered lumber mill.

Understanding this about any tool you use is extremely valuable. This doesn't mean that in a pinch you can't use the table saw to cut something it isn't great for, but you can lower your expectations for the quality or efficiency of the cut. Just like there are different table saws there are different Gen AI solutions. Some are better focused on helping you create good code. Others are focused on sharing "phacts" that map to the owner/creator's agenda. Still others are better at assisting you in web searches.

Personalities and Bias

OK, pardon the anthropomorphization (I know, I railed against it just a few paragraphs back) but every Gen AI has biases based on the organization that developed it and their particular methodologies. From a company that developed their solution to "make search better," to the company that said "hey, we've got all this great Reddit data we should train an AI from it," every Gen AI is different. But none of them have a conscience. None of them have a sense of loyalty or privacy. They're all voyeurs, looking in every window they can because they never know where the next answer might be hiding. By that very definition they know nothing about privacy, and we need to understand that. That tangles us up in all sorts of copyright issues and the like, and I hope those get litigated quickly so we can address the concerns and move on to a better model one way or another.

Don't think there's bias in these tools? OK, just type ANYTHING that Grok can see and you're likely (at the time of this writing) to get told about a nonexistent white genocide in South Africa. If that isn't an agenda or a bias I don't know what is.

Wow, Does AI Have No Redeeming Qualities?

Of course it does. AI does lots of fantastic things. Finding the needle in the cybersecurity log haystack, driving like a drunk 14 year old who has always wanted to take dad's nice car out for a joy ride (yeah, ok, this capability needs more work), and analyzing large volumes of well defined data for trends and conclusions far more quickly and efficiently than humans ever could. But these aren't the kinds of "AI" that people know about or think about in their daily lives. No, they're thinking of Large Language Model Generative AI solutions.

However Gen AI is a different animal. This is the coke snorting, night-club party boy that is selling an image of success with less substance to it than he's letting on, trying desperately to make sure you envy him and want to emulate him, thinking that if he can convince you he's the way things should be then that's what they'll become.

Don't let Gen AI be the face of AI for you. Don't mistake it for your friend either.