Considerations
Before diving into using these tools, keep in mind the limitations that any AI tool will have in the near future — with many lasting far into the future. We’re in the early days of this artificial intelligence transformation, so of course there are going to be flaws.
The key is to remember these flaws and adjust. That’s all it really takes.
Data Quality
Whatever data you have in the system, that’s what all the systems are going to summarize. This is the biggest limitation of these tools going forward considering this is an issue in all environments. Organizations are going to be cleaning up a lot of their data to get a better understanding of what they have.
To this point, I won’t be listing data quality and accuracy in this chapter. This information is outdated quickly, and it’s best to judge each assistant on their intent and nothing else when selecting one.
Hallucinations
Simply put, response accuracy is going to vary. AI Assistants can sometimes produce incorrect or misleading information, but so can humans. Because of this, there always has to be a human in the loop for now and for quite a while.
I prefer to call it “trust and assess:” view it as valuable information, but then assess what you have to check to make sure it’s accurate. Some of the AI Assistants do a great job of showing where it found the information so you can assess on your own.
Bias
The data these models are trained on reflect human biases, full stop.
All humans are biased, so you can make the case about all data being biased. It’s easy to correct, but there will always be a lens to the answer.
Complexity
Across the board, the engines struggle with complex prompts. The more complex the prompt, the less relevant some of the answers are. There seems to be a “just right” level that works best, and this differs wildly by model so you’ll have to experiment.
Privacy and Data Security
These are critical challenges, as these tools often process sensitive information, raising concerns about data breaches and use as training data. If your company has rules around their data outside of the organization, using these tools might violate them.
If you’re not familiar already, familiarize yourself with what data is legally required to be protected, then go a step further and consider what other data you have a moral obligation to protect. Any legally protected, sensitive data should not be shared with AI Assistants.
Power Consumption
All of these tools take up power — a lot of it. There have been many articles written about the computing power needed to power the AI models. It is a concern like how cloud power consumption is high for streaming services or the power consumption consumed by bitcoin miners.
It’s a problem, but I also believe that the engineers at Nvidia and the other chip manufacturers are working to get the power consumption down. This will happen over the next few years because energy cost will be a driver.