Back to Blog

Don’t Trust AI

Bosma Enterprise

I don't trust Artificial Intelligence (AI) and neither should you. Now, I'm not Grandpa Simpson shaking his fists at clouds and yelling about the evils of AI. I'm not here to debate the morality or immorality of AI, or which areas of life should be off-limits to it. Nor am I talking about the prevailing belief that AI will become self-aware, decide that humans are the problem, and then wipe us out (usually with our own weapons of mass destruction). No, television, movies, and books have spent several decades warning of how our hubris with AI might be our own undoing. If it's inevitable, like the comet that killed the dinosaurs, then I guess that is just how it is. 

My distrust of AI stems from a more pragmatic, need I say, practical standpoint. If you have spent any time in the AI micro/macro verse, then you have likely encountered some of the horror stories that are already out there. I'll include a few of the more memorable ones to give you an idea.  

  • A developer asked AI to delete a folder, and then AI deleted the entire server.
  • An AI chatbot gave a customer a discount so large that it threatened the company's very existence, whether through such a large loss or a legal battle that might ensue. 
  • Quite a few stories of AI being exceedingly helpful to individuals who plan out some form of harm. 
  • AI being used to get legal advice, or even case law, that was completely false and/or fabricated. 

In regard to that last one, some law offices are now reporting that over 50% of their work hours are quickly being dedicated to reviewing legal documents they have received from clients and other lawyers to verify that they weren't generated by AI. People are trying to use a shortcut that ends up being a much longer way than doing it without AI, or using AI as I suggest.

You see, I'm not advocating not using AI. AI is here, and it will likely stay and grow as time goes on. Right now, though, it's a highly powerful toddler that some people just aren't willing to learn how to deal with. As my lead developer says, "You must treat AI like your latest intern. Let it do the work, but you must check it." In other words, don't trust it. It's that open trust that AI will do exactly what you want it to that will get you added to the above list. Like an intern, or in this case, a toddler, AI doesn't yet understand some of the nuances that it needs to in order to be given free rein. 
 

AI does exactly what you tell it to, not what you meant for it to do. 


That distinction may seem pretty obvious, but you'll find that a lot of times, AI went off the rails because the people who gave it free rein failed to set proper guidelines or to use instructions the AI understood completely. I'll give you one of my more recent interactions with AI that illustrates this and is slightly comical. 

I asked AI to "generate a simple picture of two people on a couch, enjoying popcorn and watching a movie. Include coffee." AI did that, but with one odd issue. One of the people only had one leg. It wasn't set up so the second leg could possibly be under them, or behind the other person. It just didn't exist. Without giving it too much thought, I replied to the image, "The woman is missing a leg". My thought was that the AI would look, go "oh yeah, she needs another leg." Nope. AI changed the picture so that the one leg the woman did have was now an obvious prosthetic. Highly amusing, but not the outcome I had hoped for. I spent a good bit of time trying to get the AI to fix the issue, only to decide that the image was too far gone and to start over. Luckily, for me, it was just a simple image meant to convey a simple thought. Imagine if the error had been on something more substantial. Something that would forever tarnish the image of the person or company that used it. 

So, my advice to you is simple. Use AI all that you want. Just don't trust it. Check its work. Check its reasoning. And if you are a lawyer, check that it didn't make up the cases it cites for your legal submissions. AI has been known to not just make up citations, but to go a step further and create articles to make those citations look legit. 

AI is a powerful tool, but it’s only as reliable as the people behind it. For expert guidance on integrating AI and navigating your IT or Salesforce needs, visit Bosma Enterprises IT Consulting, or reach out to gwconsulting@bosma.org.