AI and these brown leaves of islands

The war has always been between what we should become and what we could become. Our unfinished souls remain moored in the sea, our monuments and memories locked in a blue vault. The AI future of these islands is as fluid as the sea that washes over the brown leaves of these islands, causing them to cling to the blue rim of the Caribbean basin.

In Europe, the tradition is to memorialize the past on land using everything dragged from the sea. However, the Caribbean landscape is marked by historical absence. No cenotaph attests to the resistance of the First Nation Peoples against the Encomienda, the savagery of Saltwater Slavery, or the efforts of indentured labourers in Golconda.

With eyes as heavy as anchors, and sinking without tombs, we must write our AI futures, beginning with the salt chuckle of rocks in Stonehaven Bay and their swirling sea pools. Our Chains of Thought (CoT) floating freely in every direction, defying the fixity that is common to land-centered notions of identity and culture. We are the information organisms or inforgs who must live digital lives in the infosphere.

To take us there, we must open the catalogue of AI prompt engineering next practices and learn to use the tools nimbly. It is an ongoing and ever-expanding index that takes us to different places at different speeds. One such tool is the prompting technique known as Chain of Thought (CoT). CoT is about prompting AI to reason in a sequential, deliberate, and logical manner. It resembles a trail of crumbs that leads to a solution. The beauty of CoT is its flow. Each step builds on the last. Avid and experienced users of generative AI and Large Language Models (LLMs) understand the art of CoT prompting techniques that proceed on a stepwise basis. Using CoT, the AI will display the various logical steps that it performs to arrive at a solution. CoT tends to stir generative AI towards better answers.

Using CoT, the AI slows down the process to carefully specify each step. By giving a prompt that explicitly tells the AI to do a CoT, you can direct the AI to methodically and cautiously solve the problem you have framed. The outputs from this technique permit the user to scrutinize the reasoning steps that the AI displays. This is useful as it allows the user to discern if something went amiss in one or more steps. It also helps the user nurture scepticism and doubt in the face of evidence.  CoT also takes some time to process queries and this usually extends the latency. If the generative AI being used is already set up to automatically invoke CoT, then it is best to not ask for CoT in a prompt as this can result in some snags.

Three steps to instruct generative AI to undertake when wanting the AI to proceed on a heightened logic-solving basis include: (1) Logic Extraction: Prompt the AI to identify all the possible logic-based propositions that are embedded in the problem and to display the propositions in a conventional format commonly used to express propositional logic, (2) Using Propositions: Using the extracted propositions, you can then ask the AI to formulate a solution using strident logical reasoning, and (3) Ask the AI to show the propositional logical reasoning that was used during the interaction and to explain what that reasoning consisted of. Another prompt engineering tool is the Atom of Thought (AoT).

AoT dissects a problem into standalone bits or isolated blocks that are independently solved using a distributed problem-solving method, and then snapping the solution blocks together like LEGO bricks. The approach is not unlike the strategy of divide and conquer. Each atom becomes a mini-mission. The modularity of this technique makes it laser-focused with each block being a self-contained problem package. The AI is expected to process the packets in parallel. This technique is a Swiss Army Knife that outputs a highlight reel that is sharp and short.

When comparing the two prompting techniques, it is evident that while CoT prompting solves the problem with a flow of outputs, like a river from the source to the sea, AoT provides all the varied stones one may stumble upon in the mouth of the Sans Souci River. The task is for the problem finder to assemble the stones into countless mosaics shaped by the slow fuse of the imagination. Rearranging the stones in a grounded theory approach generates many solutions.

The difference between CoT and AoT is the question of flow and fragments. With CoT, every step in the chain is critical; skipping a step can disrupt the Chaplet of ideas. AoT allows you to select any block, making it more flexible. CoT acts as the reliable Sherpa for a steep linear climb, not for a lateral traverse across the face of a cliff. AoT is ideal for use in policy labs within municipal corporations. It is suitable for brainstorming sessions, and product owners using agile methodologies to solve real-world problems. It enables bureaucrats to break down large ideas into manageable bits or blocks. It serves as the perfect palette knife for tackling chaos.