Anyone can use AI! Right? Well, yes, anyone with a device and data connection can. Technically, at least.
Artificial Intelligence has the ‘reading level’ of a post graduate student. In other words, there are not many terms or phrases that it does not understand. In giving our employees Gen AI tooling, we are expecting them to communicate with and work at the same level as a technology that operates at a far higher level of education. In the 2024 OECD PIAAC (Survey of Adult Skills) conducted in the UK, results indicated that around 55% of a workforce have skills associated to that of GCSE grades A-C, and that only 25% possess graduate level skills. Even if you were lucky enough to have a good education, or are naturally intelligent, you are still communicating with a technology that knows something about everything. We are not talking about quality or accuracy, just that AI is a kind of ‘know-it-all’ technology. There is the first general mismatch.
Now let’s think on how we communicate with it. Interacting with AI is most certainly not the same as interacting with a fellow colleague. Albert Mehrabian, Professor Emeritus of Psychology at the University of California, conducted research into the communication stack and developed a model explaining how various parts of the stack contribute. If I were talking to you face-to-face, we would be sharing 100% of our communication stack, including facial expressions, gestures, body language, and intonation, etc. but that is not how we communicate with AI.
Mehrabian concluded that 55% of what we communicate in person came from body language, 38% from voice and tone, and only 7% from the actual words. That means that when we type in a prompt or request to AI, we are using our very worst form of communication.
The picture that is being formed here is that at the top of the intelligence stack is AI running at this post-grad level. Humans are, at best, operating at university level, but most likely falling in the 55% mentioned earlier, and from that we are using our worst form of communication. But don’t worry, all is not lost, we can now speak to AI… but does it work like that?
Well, the picture is unclear. When you type out your prompt, it is typically more structured than if you just say what is on your mind. From a technical perspective the spoken words are captured and translated into text, so it is just a verbal input, but the question about structured versus a more fluent stream of words begs deeper exploration.
English is a complex language, anyone who has learned it as a second or third language will confirm that for sure. Whilst we don’t assign genders to everything as in other languages, we use a whole range of words to blend sentences together, we use the same word for a multitude of meanings and with a multitude of pronunciations, we use a lot of double negatives, colloquialisms, formal and informal terms and much more. So, what impact does it make when LLMs are typically built on English, when that is not your first language, and how does that impact the output. The short answer is, I don’t know. But I do know that even changing the language from formal to casual changes the response you get.
The other thing to raise at this point is intonation and sarcasm, Britain was, after all, founded on sarcasm. There are of course companies working on this, AI is being trained on ‘prosody’ where the pattern and rhythm is compared to the words being used. Exaggerated or flat tones, changes in speed and volume all go to indicate if you are using some form of intonation in your prompt. For example, if text detects the words “I’m so happy” but the audio indicates that the pitch is low and slow, the incongruity as a high probability of sarcasm.
Progress is being made, but time will tell, AI does however struggle to detect context-dependent sarcasm. This is referred to as ‘the Family Dinner Test’ where the joke relies on a shared history between people rather than just the tone of voice. I assume that will rely heavily on a persistent memory developing over time, but one key question remains. Will AI be trained how to use sarcasm back at us? Either way, businesses of all shapes and sizes will need to focus far more on ensuring that their employees have a strong command of their chosen language. It will be the number one skill for any employee in, or entering, the workplace.
