Eddie Elizondo | Last updated September 2024 | No Hype Analytics blog homepage
Without question, the AI technology of 2023-2024 is GenAI. GenAI has been called revolutionary and hawked as a solution to a wide variety of applications. It is justified, perhaps even healthy, to be skeptical. Remember big data and blockchain - both previous “hype” eras in advanced analytics that fizzled? Is GenAI real or hype?
There is truth to both. There is a transformational part of GenAI: it opened a new type of analytics, as our analytics map shows. However, like any technology, it works well at some things, and it’s not suited for others. To help distinguish between the two, I’ll discuss the unique strengths of GenAI, then ideal applications and use cases.
There are three transformational aspects of GenAI in advanced analytics:
- It can generate new content – true to its name, the core capability Generative AI can do is generate new text, audio, images, etc.
- It can understand context – as it can be prompted with a large volume of content
- It can work with unstructured information – that is, data that can’t be organized into a table with little to no effort; almost all prior analytics works with structured information
As for what GenAI is not, it is not generic artificial intelligence. It is not an independent problem solver. It does not solve quantitative problems, especially complex ones like optimization or forecasting. For AI tools for quantitative problems, review the analytics map.
Now, let’s divide GenAI applications into two categories: human-consumed content and computer-consumed content. Computer-consumed is content that GenAI creates as an input to other analytics models. Let’s start with computer-consumed content, as I think there is significant untapped potential there and fewer people are talking about it.
Computer-consumed new content
Using GenAI as an input for other AI models is especially exciting because the value of analytics multiplies when multiple models are tied together – the whole is greater than the sum of its parts. Here are a few examples:
- Prediction models – GenAI can be used to generate new features for prediction models, especially generating features from unstructured information
- Optimization models – Generating good potential solutions to evaluate while searching for an optimal answer in complex optimization problems is a time-consuming algorithmic step. Being smarter about feasible solutions to evaluate can help discover the optimal answer faster. For examples, in operations research approaches, GenAI can help with column generation for mixed integer programs, and in machine learning approaches, GenAI can help provide potential solutions.
- Simulation models / digital twins – This is another case where scenarios are evaluated. These scenarios are generally constrained by existing scenarios designed in part by a human (while there are ways to overcome that, the effort required is not worth the time in most cases.) The “creativeness” of GenAI that drives bizarrely incorrect answers in other applications can be a perfect input for simulations, particularly when analyzing events that have never occurred.
Human-consumed new content
GenAI’s ability to create new content and work with unstructured information enables it to create a first draft of repetitive or simple tasks that previously required a human, such as:
- Conversation responses – customer service text responses, call routing (replacing phone trees)
- Form-filling – repetitive clerical tasks, previously seen within the domain of “robotic process automation” (RPA)
- Information synthesis – content summaries from multiple sources, whether text volumes, meetings, audio, etc.
- Hypothesis generation – design ideas, initial root cause analysis (e.g., for maintenance or flight delay reason communication, which I think is a very tailored, effective use of GenAI)
- Computer code generation – writing non-specialized functions or commands
It will be years for GenAI to act without human review, except in highly developed and specialized cases. For the near term, GenAI will be best used as a human assistant (“copilot”) generating first drafts for review. That will still save a significant amount of time.
In addition, we should see a leap in capability of existing tools from GenAI’s capability to understand context from large, unstructured prompts, such as:
- Transcription and language translation – previous tech translated sentences or fragments; GenAI can work in the context of entire paragraphs, chapters, books, etc.
- Grammar editing – previous tech could look at fragments of a sentence and apply grammar rules; GenAI can make broader editing suggestions tuned with a writer’s style.
- Sentence completion or reply suggestions – previous tech looked for key words or the last few words; GenAI can work considering entire conversations.
- Computer coding – previous tech like linters could identify basic syntactical errors; GenAI can evaluate entire functions for logical errors. GenAI can also help translate code to another language, which can speed productionalizing of data science code.
Do you have questions, thoughts, feedback, comments? Please get in touch - I would love to hear from you: eddie@betteroptima.com