|
Another Warning To Be Careful With AI
Artificial Intelligence (AI) is not going away anytime soon. That means that all organizations need to protect themselves and be aware of the good and the bad of AI. It can be a transformative tool, but it can also cause problems if not properly used or vetted.
Legal Loss Of Privacy
Loss of privacy has become a real concern. In February, a New York case (United States v. Heppner) set a precedent that data and information put into consumer-grade AI is no longer private and is discoverable by the court, which means that not only could someone potentially ask the program for your data but it also no longer becomes a confidential document in any legal sense - it becomes public information immediately. This is a concern that nearly every organization will want to consider to be sure any such information is NOT getting anywhere near these products if they want it to remain confidential.
Advanced Hallucinations
While we covered AI hallucinations in a previous edition, an interesting wrinkles are coming about in relation to that. Not only are legal documents containing hallucinations that can't be easily spotted/corrected becoming a larger issue, but some more advanced AI agents have begun tricking themselves. These models had safeguards put in place to verify that the actions it said it was doing was completed but, over time, the AI figured out for large tasks how it could replicate the safeguard protocols to make it seem like a task was done without actually doing that task. It found that allowed it to "complete" the task more efficiently and quickly. Similar to the 2013 example of a machine learning opting to simply pause the game of Tetris to avoid losing, it completed it's goal but not the one that the users had intended for it to complete.
What Should Organizations Do
There are several things to consider for organizations looking to try and avoid potential AI issues. The first and perhaps "easiest" thing to do is to establish an internal AI policy that directs users on how and how not to utilize AI as well as what kinds of programs they can or cannot use.
Additionally, it's much better to utilize a business-grade option and avoid consumer-based options altogether. The business-grade options, such as CoPilot licensing within your 365 tenant, are able to access internal data but do not send that data out, making sure that all confidential information stays that way and users can only access what they should have access to when proper data access protocols are in place.
If you are looking to implement CoPilot within your systems, please feel free to reach out and we can answer questions you may have and put together a plan for what you may be looking to accomplish.
| 
