-1.png?width=5000&height=1375&name=MicrosoftTeams-image%20(87)-1.png "Ai in the Workspace Part 2")
Against AI
Let’s take the examples above and place “what-if…?”, “but…” and “that will never work!” against these specific use cases.
The first example was the mirror for my thoughts, but when I submit this to an online AI engine, these data can then be utilised in the large language model (LLM) and influence the chat engine’s responses for another user.
If this information contained confidential data or intellectual property, this would be a huge security risk and potentially devalue the content I am creating or maybe even the company I work for because the confidential information has leaked into the ‘outside world’.
There is also the potential for AI hallucinations; a scenario whereby the AI provides what appears to be coherent and reliable information, but when checked is nothing more than a concoction of various of data sources, potentially with no common point of reference, that AI has joined together. There is a recent occurrence of this type of hallucination where a lawyer, claiming to be unaware of the potential for unreliable data in AI, relied on it to provide a court brief. Although the content seemed reliable and read as a perfectly formed court argument, the precedents mentioned were fictitious. More can be read on the BBC website about this here.
Recently, I have tried to use chat bots to find part number cross-references for one of my hobbies. The response provided from the AI chatbot looked sensible, had valid model numbers and data pertaining to the type of question I was looking for, as well as manufacturer part numbers for the components I sourced. To the uninitiated, this looked like a perfect response; the results however were anything but this. The model numbers referenced were correct, but 50% of the product codes led to a product that matched the description but not the dimensions of the item I was looking for; in short, the entire result was wrong and unable to provide me with the answer I needed. I took more time checking the results than I would have spent looking for the answer manually in the first place!
This is one of the biggest risks with using AI: reliability of the response.
We have the human checker, the user, who can read the responses from AI, but this is little-to-no help when looking at large-scale implementations.
My colleague Rob Sims focuses on the enablement of organisations to utilise closed-loop AI systems and specific LLMs, to improve the relevance of data in AI responses by narrowing down the quantity and quality of data being utilised to create responses. You can find Rob’s articles on the CDW news site here.
Moving on to the writer’s block example; there is the risk that the content AI suggests is not unique or is based on biased information. What if you were to use and share content created like this? If it was tested (by AI) and shown not to be human-created content, do you lose reputation, do you remove the value you provide as an individual?
There is a real challenge in discerning human content from AI content. I have seen content that was human-created being scored as 78% AI-generated and I have seen the opposite where AI content was tested as being almost 90% ‘human’.
However, if we were to utilise a human or AI to create documentation around a structured process, such as service management, the quality and accuracy, especially if using a closed loop LLM defined for this purpose, would deliver the same value and probably be indistinguishable from each other. When I write content, I want it to be clear that I wrote it, that it is my thought process, investigation, and insight. With ever-more content being AI-created, I want to ensure mine is discernible as unique and human-generated.
I foresee a very real scenario where the information we rely on to make decisions and the content we read daily is nothing more than an amalgamation of all the available information stream. Over time, this will be biased by having more AI- than human-generated content; resulting in a narrative that is AI-led. Now, this isn’t like in film where the computers take over physical machines, but it can result in directing human thought processes - intentional or otherwise.
‘Facts rule the world, emotions run the world’, is a phrase my father taught me, and it is worth bearing in mind when we consider the influence AI has on the content we digest, especially on the impressionable. A machine cannot experience an emotion (yet), but it can elicit one; are we ready to let AI have a controlling capability in the content we read, and, in turn, influence our decisions and emotions?