AI is latent intelligence, waiting for us to exploit it
Consider a block of unstructured textual data. You want to make the data more legible. You want it in tabular format, so that you can quickly scan it and extract relevant details. Or, you want to pass the information along to your super busy boss.
Now consider manually formatting a table in Excel or Google Sheets: you have a block of text on one side of your screen, and a spreadsheet on the other side of your screen. You start to parse the text for key information. Maybe you’re looking at information about government defense contracts, for example. Key information may be the firm awarded the contract, the dollar amount of the contract, etc. So you start to copy certain parts of the unstructured text, and paste them into individual spreadsheet cells. Here’s a screenshot of what your unstructured data may look like:
This is hard to make sense of. A tabular format would be much better:
This table was, of course, created in ChatGPT. I copied the unstructured data from the website1, and pasted it into ChatGPT’s input window. The prompt was simple: Organize the following text in tabular format.
The technology required to enable this, however, is anything but simple. Quoting from ChatGPT:
To convert the unstructured data you provided into a structured table format, I relied on a combination of natural language understanding (NLU) and manual data organization techniques. Here’s a more detailed look at this process:
Natural Language Understanding (NLU): Part of natural language processing (NLP), NLU helps in understanding and interpreting human language. In your case, it was used to parse the text you provided, identifying key components such a names of organizations, contract amounts, locations, dates, and other pertinent details.
Pattern Recognition: This involves recognizing patterns in the text. For example, recognizing that a string of characters is formatted like a date or a dollar amount. This is crucial for differentiating between different types of data (like distinguishing a contract number from a monetary value).
Data Extraction and Classification: Once the key components are identified, the next step is to extract these elements from the text. This involves classifying each piece of data (like identifying what is a company name, a monetary value, a date, etc.) and extracting it for use in a structured format.
Manual Organization: After extracting the relevant data, I manually organized it into a tabular format. This step is more about applying logical structuring to the dat abased on its nature and the context provided by your request (e.g., organizing it into columns like Contractor, Location, Contract Details, etc.)
Contextual Understanding and Interpretation: Understanding the context of your request (i.e., that you needed the information in a table format) and interpreting how to structure the table based on the nature of the data provided.
The technology used in this process is a combination of NLU and manual data structuring techniques, facilitated by my training as a language model. This training involves large-scale data analysis and pattern recognition, enabling me to understand and organize data in a meaningful way. However, it’s important to note that this process is not entirely automated and can require manual intervention, especially in understanding context and ensuring the accuracy of the structured data.
I relate all of this because we seem to lose sight of all the complex technology that OpenAI has abstracted away in the ChatGPT interface. Once you know what ChatGPT can do, such as turning unstructured data into structured data, using the tool is as simple as writing the correct prompt. This kind of latent intelligence, instantiated only upon an explicit prompt, is as powerful as it is opaque.
Its latent intelligence reflects a significant advancement in AI. It’s akin to having a versatile, multi-disciplinary expert at your fingertimes, ready to dive into a wide array of tasks. As we see above, the system can convert unstructured data to structured data. AI can, in other words, understand, interpret, and reformat information. This capability is crucial in numerous applications, ranging from data analysis to content creation.
Some important caveats to consider about this latent intelligence: the quality of its outputs depends on the clarity and specifity of the input prompts. As a user, you play a crucial role in guiding the AI to deliver the desired outcome, much like a skilled craftsman guides his chisel to hew a sculpture from marble.
Given all of this, it’s worth considering what a future with widespread latent intelligence looks like. The rapid adoption of AI tools with latent intelligence capabilities suggests several significant trends:
Democratization of Expertise: AI tools with latent intelligence can provide expert-level advice and analysis in various fields, which makes expertise accessible to more people. This could lead to a more informed public and potentially flatten hierarchices in knowledge and skill distribution.
Enhanced Productivity and Efficiency: These tools make us more productive by automating routine tasks, analyzing large datasets quickly, and providing insights that would otherwise take us much longer to discover. This allows people to focus on more creative and complex tasks.
Impact on Employment and Skills: These widepsread adoption of these tools will cause significant changes in the job market. Routine roles, such as data-heavy tasks, will be automated. This will increase the demand for skills in managing, interpreting, and leveraging AI tools.
Customization and Personalization: Latent intelligence in AI allows for high levels of personalized services and products. This could lead to tailored solutions for healthcare, education, and consumer products.