Things to consider before using AI in your business
Artificial Intelligence (AI) is making its way into our lives at an ever increasing rate and that is likely to continue. And there are certainly cases where it is extremely helpful. For me the most obvious is your searches on the web. It used to be when you asked your search engine of choice a “question” you would just get back a bunch of links and you’d have to sort through them one by one, reading bits of the pages, trying to find out if it actually answered what you asked. Now, when I ask Google or Bing a question, it actually comes back with an answer. For example, I asked Bing “When did the Apollo moon landing happen?” and instead of just links, I got an actual answer “July 20, 1969” and references to the sources it used. That’s certainly easier than sorting through links.
That’s all well and good but you have to be careful with AI – – you can’t take its answers at face value.
For example, when I asked Bing if a Dell XPS 8960 desktop has dual monitor support, I got a resounding (and in bold) “YES” in response. Except, that when you looked more closely, that “yes” comes with caveats that you would definitely need to be aware of if you actually wanted to plug two monitors into your brand new XPS desktop. It can’t just be any two monitors, etc. Google did a slightly better job in this case by giving this answer “Yes, the Dell XPS 8960 desktop computer can support dual monitors. It has multiple video ports, including a Dual-Mode DisplayPort 1.4 and an HDMI 2.1 port.” Ok, thanks, that’s great – but then it goes on to say that the Dell XPS 8960 has 3 DisplayPort ports, which it most certainly does not. (Here’s a quick primer on AI “hallucinations”: https://www.cnet.com/tech/hallucinations-why-ai-makes-stuff-up-and-whats-being-done-about-it/.)
How does this apply to your business? There are several areas of concern but I think for small businesses, it boils down to two major concerns when looking at the impacts of AI on your business (and your staff is using it already, whether you like it or not).
Concern #1 – Accuracy
As in my example above AI answers aren’t always all they’re cracked up to be. AI only learns what we teach it – – it doesn’t have any inherent intelligence and it does include downright incorrect information at times. That is very important to remember. I was listening to a speaker at a conference a few months ago who does a lot of research on AI “hallucinations” (as they are referred to) and the most striking story was when she asked the AI in question “How many giraffes are in this picture”. If there were 6 giraffes, the AI got it correct. However, when she asked the AI for the number of giraffes in a picture where there weren’t any giraffes at all, the AI responded with “There is 1 giraffe in this picture.” What?? So, she dug in a bit and she discovered that what was happening was that any time the AI was asked “how many giraffes are in this picture”, there was always at least 1 giraffe. So, the AI assumed that all pictures always have at least 1 giraffe.
Also, the biases of the data that AI ingests will affect its output. For example, various AIs were asked to create images of people in low paying jobs and the images created were disproportionately skewed towards minorities (even more so than in the real world). You asked for an image of a CEO, and you got a white male (most of the time). You asked for an image of a fast food worker, and you usually got someone darker skinned. Here’s an article from Bloomberg that details this issue: https://www.bloomberg.com/graphics/2023-generative-ai-bias/. Another example of where you can see AI biases show up is if you use AI for resume screening – – people have seen issues with racial, gender and age-related biases. You have to ask yourself, is speeding up the resume review process worth the risk of missing out on good candidates or breaking discrimination laws?
The bottom line regarding accuracy is that AI is not ready for prime-time when it comes to being your ultimate source of information for making a decision. You can use it as your first step in doing research and to get you started with what is most likely the answer and the most relevant considerations but you still have to do the factual legwork.
Concern #2 – Privacy
A lot of clients have asked recently about using AI “note taker” add-ins for their meetings. Apps like Otter, Chorus, Leexi and Krisp have all come up over the past year. It seems like a huge help to have an AI attend the meeting, take notes and sum things up for you at the end. (Let’s put aside the accuracy concerns mentioned above for now.) But, you have to remember that this AI assistant has full access to everything said and done in your meetings. So, what is it listening to? Are there legal concerns about what it hears? Are there intellectual property or strategic concerns? It might be ok for the AI to attend weekly staff meetings but is it ok for it to attend your strategic planning meetings with your C-suite? Yes, you can vet the vendors mentioned above and go through their end user agreements but they could also not be abiding by what they promise in those agreements (we’ve seen that way too many times with tech companies over the past decades). Our suggestion at this point in the AI development, and until there is a long history of this data being protected, is that you should assume the data will not stay private. So, if what you are meeting about would be harmful to the company if it were released, don’t invite an AI assistant. If what you are discussing is simply operational updates and standard “meeting” stuff, then you should be ok to enjoy the efficiency benefits of having the AI attend. Also, if you do have an AI attend, be sure that everyone knows that they are being recorded. Usually the AI bots do that but you should be sure – – it is illegal to record someone without their consent in Massachusetts.
In summary, implementing AI at this time requires serious consideration regarding the quality of the data it provides and the privacy of the corporate data to which you are giving it access.