When are you adding AI?

As the Chief Technology Officer (CTO) of LifeRing I often get asked, “When are you adding AI to LifeRing?” The quick answer is, we won’t be, at least not yet. Here’s why.

First off, did I use AI to help write this article? Absolutely. AI tools are incredible, and I would be a fool not to recognise the technology at our fingertips. Nowadays, even using a web browser often involves layers of complex AI (machine learning) models to deliver the results. However, this article is my own work, not just a copy-paste from a complex statistical model. For a deeper dive into what AI like ChatGPT does, you can read this article by Stephen Wolfram: What is ChatGPT Doing and Why Does it Work?.

When people talk about AI today, they’re often referring to machine learning (ML). A decade ago, we used to joke that ML was the code we wrote, while in presentations it was promoted as AI. Back then we could understand every aspect of what our code was doing. Today, it takes years of dedicated study to fully reverse engineer even simple models.

Large Language Models (LLMs) and Their Limitations

LLMs, like those behind AI text generation, are truly impressive. They can summarise a thousand pages into an executive summary with ease, and in any style specified. So why don’t we use AI to summarise Safety Observation Cards or Lessons Learned within LifeRing?

Despite their capabilities, LLMs have significant limitations. They can’t fully comprehend nuances leading to inaccuracies in complex scenarios. Moreover, their inner workings are often opaque and proprietary, making it challenging to ensure accountability, accuracy, and to trust their outputs. LLMs can create fantastical output making up wild ideas and concepts while presenting the output in a style of clear hard facts. Our tests have shown that even simple tasks can result in misleading or incorrect information from LLMs, which doesn’t meet our standards.

The AI Bubble

I’m old enough to have been part of the dot com bubble when something called the Internet started becoming useful to everyone.

In my view, the hype and massive investment into LLMs and AI is just the same as the dot com era. Following the growth and the inevitable bubble pop there will be a realignment, after-which we will have some really useful technologies which will fundamentally change how we do things. We may or may not get to Artificial General Intelligence (AGI) but that is a topic for another day!

We’re witnessing an AI gold rush. In a gold rush if you want to get rich you don’t go mining, you sell shovels. Companies creating the tools (or shovels, in this analogy) are profiting the most (see the share price of NVIDIA), but it remains uncertain whether AI companies will see long-term profitability. As Sam Altman noted, “We will have to monetize it somehow at some point; the compute costs are eye-watering” tweet.

My prediction, the coming months will see a shift in the monetisation of AI. What is available for free today will suddenly become quite expensive or be full or adverts and sponsored content. After all, someone has to pay for all the energy, compute power and shovels used.

We don’t want to introduce technology into LifeRing which will suddenly and unpredictably cost a lot more.

The Financial and Security Considerations

Integrating LLMs into LifeRing isn’t just about acquiring the technology; it involves ongoing running costs and significant monitoring and maintenance. At LifeRing, we prioritise security and data privacy. We’d have to invest heavily to ensure these principles were respected by the LLMs.

The lack of guaranteed Service Level Agreements (SLAs) from LLM providers presents a significant risk. Without assurances on uptime and availability, we can’t confidently meet our commitments to our customers.

Service disruptions, performance issues or errors could lead to outages, reputational damage and financial penalties.

What Could LLMs Bring to LifeRing?

Imagine asking LifeRing to “assess the trends for observation cards and incidents over the last two years and write a plan to make my company safer.”. A LLM could produce a well-written, confident, and compelling plan within a minute. But would it be correct for your company? We can’t be sure.

Instead, we developed LifeRing to assist experienced QHSE professionals to collect better data and aid their decision-making with trends and analysis which can be audited. This approach ensures that our platform remains reliable, accurate and secure while aligning with our commitment to workplace safety and personal well-being.

Will we add LLMs one day?

Maybe. Time will tell.