After 6 months of extensive research and development using the tools, AI might not be taking the form you think it is.
As we kick off our quarterly series of Newsletters, it seems appropriate that the subject we’ll address surrounds the topic of AI. There’s no more talked-about change in our industry over the past decade, and certainly no technology that has the power to transform how we work and impact the way we interact with clients, than artificial intelligence.
When we speak about AI with anyone, we rapidly hear things like “they’re coming for our jobs!” Indeed, anyone with a white-collar job today worries about the AI’s ability to do things better. And even if the AI can’t do it “better,” it only needs to do it faster or cheaper, and we (the workers) come out on the losing side of the equation. Are we holding our collective breath for the ‘killer app’ that will create, compose and administer an entire survey on any topic, using the appropriate methodology, with just a few inputs? That seems to be the question most of us lose sleep over. Does our industry have a future with ‘humans’ driving the industry forward…or will the computers supersede us?
If this is you, then let me set your mind at ease…a bit. While we’ve seen several companies leveraging AI to develop and create surveys, after 6 months working with the technologies, I don’t believe it will be the most prevalent early use of AI. But to understand my logic, I think it’s useful to briefly examine what AI is currently doing well.
What do large language models do?
Quick definition of LLM: A large language model (LLM) is a type of artificial intelligence (AI) algorithm that uses deep learning techniques and massively large data sets to understand, summarize, generate and predict new content.
If you’ve ever played or experimented with a LLM AI, there are already things the systems can do extremely well:
1) Writing for a non-writer:
A friend of mine recently confessed that he was struggling to write a concise half-page article surrounding a topic for his job in the healthcare field. I asked if he had ever leveraged AI? He immediately pushed-back! “I don’t want AI to write this – I want these to be MY ideas.” I told him, “That’s exactly what AI does best!” and proceeded to walk him through a test. Having already spent 3 hours on this task, we put his thoughts into a tool and asked the system to summarize his ideas in an 8-sentence paragraph. After a 5 second iteration, we had a concise, well-written summary of his ideas. Within another 10 minutes, he had edited the AI’s summary, and was done…undoubtedly saving several additional hours of time.
2) World’s greatest editor:
Have you ever been writing a document on a short topic and needed your summary to be shorter or slightly longer? As a former English major, we often were told “get your ideas down first, then summarize and edit later.” This is precisely what AI can help us do! My wife, who heads up a technical communication team, will always insist that good communication uses plain language. “Use your ‘weekend words’”, she often says, “the way you speak to friends in conversation if you want to be truly understood!” With that in mind, AI is built for this technique.
3) Synthesize and summarize complex information in simple terms:
Leveraging AI to summarize complex ideas is a bullseye of this technology. Take, for example, the topic of string theory. Asking the systems, “What are the main ideas behind string theory?” results in a 6 to 8-point definition that is eye-watering and verbose…perhaps not clarifying anything. Instead, ask AI to “Summarize string theory in simple terms that a 10th grader would understand.” Now, you have a concise summary that perhaps more closely matches your understanding for an introduction to a complex topic.
I believe these strengths have big potential to change the way we approach research.
The current dangers of AI:
AI is not like a typical search engine. In fact, it has a propensity to “hallucinate” in certain circumstances, and it will do so with confidence. What is an AI hallucination? It is an artifact between the LLM and the facts that underlie the response. These AI models are primarily focused on the language, drawing consensus from numerous inputs on the same topic. However, if the model contains false or misleading inputs, the results can be dubious. I’ve seen an AI model tell me that the winner of last year’s Super Bowl was, in fact, the losing team – without a hint of hesitation or doubt. What this means for us (especially those that are developing tools early) is that we need to check the outputs extensively before assuming the AI model did something right. This is the primary reason I believe LLM’s are not currently as successful at creating surveys. Hallucinations abound!
What can we expect from AI in the next couple of years?
As we can see, synthesizing, summarizing and describing are key strengths of AI using LLM. Looking into a crystal ball is always challenging, but after experimenting and even developing our own AI-driven models for the past several months, it’s clear that the systems have an innate ability to extract insights better than anything else. In other words, simply leveraging the technology to do what AI currently does well – examine language (potentially larger amounts of information than we’re currently looking at) to summarize and deliver themes. Of course, this sounds easier than it is. The models that drive these tools conduct calculations in the billions, not the millions. Analysis can’t be performed with desktop computers – the models won’t load, or crash out in seconds. True development requires a confident and highly technical team that is comfortable with both ambiguity and rapidly evolving tools. And yet, how wonderful would it be to spend less time conducting analysis – summarizing themes – and more time drawing conclusions? With the same number of researchers, we could deliver more projects in far less time, offering insights to customers within days rather than weeks. In this way, leveraging AI will surely be a boon to our industry, extending our ability to turn projects faster and more efficiently, ultimately allowing us to generate more revenue with the same number of employees. Sound exciting? We think so!
Ryan Jay – Founder & CEO
Stay tuned for more information about AI tools coming from Outsized Insights in the coming months. Need help with your current projects? We’d love to help – drop us a line at Bids@outsizedinsights.com
