IB Boost Blog

I'm sure it has escaped no one's attention - if you're interested enough in enterprise data to be reading this blog - that Artificial Intelligence (AI) has become a cornerstone of enterprise systems procurement in barely the last 12 months.

It's just one year ago ChatGPT 3 was released and the penny dropped that AI was not some future aspiration, perpetually 40 years away like fusion, but it was here. Now. And it was coming for you whether you were ready or not.

Existential dread aside, it's useful to consider how the landscape has evolved in 2023 as we near the tail end. Just days ago we saw OpenAI's latest iterations of ChatGPT and their - one would imagine - trademark-inspired launch of "GPTs" this week, in addition to a GitHub .

The integration of AI technologies, but in particular generative AI and large language models (LLMs), is reshaping the digital landscape and - perhaps uniquely in the world of technology - the enterprise market is not, for once, lagging in their demand for platforms that are at least "AI-ready" if not "AI-capable".

For the past decade there has been the hum of the Big Data engine in the enterprise data space, but the truth has been that as a proportion of what work is done and what systems actually required such capabilities, it was mostly a well-hyped tip as actually using that data to make money was a different proposition entirely, and often the league of a very rarefied cohort of teams in businesses (at least as a fraction of overall IT spending and work).

In 2023 however, AI has quickly become table-stakes, and there is the increasing demand for relevant and accurate data, organisations are pushing to capture, integrate, and harness data from numerous sources. This, in turn, brings forth significant challenges in data quality and consistency, which are critically important for AI systems to generate meaningful insights and predictions.

One could say this was always the case with Big Data too, however we've quickly gone from statistics pulled together into research teams and batch-generated PowerPoint presentations printed out on paper, to real-time processes at the core of business critical functions. Yes, I'm aware, and have worked with, many organisations that have been doing this too for years but nevertheless a cursory examination of IT procurement habits this past year has shown the change in emphasis in all manner of organisations.

Gartner's top data and analytics trends for 2023 highlight the importance of managing AI risks and optimizing value from AI initiatives and they recently sounded the bell to CIOs everywhere that AI readiness must be a top priority in the next two years. Of course we're on the steepest part of the hype cycle today, but for once it feels, at least to us, to be justified. There are already countless utilities that were simply not possible before the past year or two. It's difficult to imagine a world where any non-trivial enterprise IT system does not at least integrate with AI services, if not provide them themselves.

Yet, there are dragons ahead. For every cherry-picked Xitter post, there exists a hundred failed conversations. I personally have gone in hours-long conversational circles with GPT 4 trying to elicit an (admittedly novel) approach to Groovy metaprogramming, but many of these services will be released or used without an independent arbiter of quality. We all remember Microsoft's hastily-pulled Nazi chat bot.

Whilst alignment, generally, has certainly taken the most vile edges off the conversation, the fundamental problem of correctness remains one requiring yet-to-be-found solutions.

When integrating data from various origins with many potential sources of errors, it will inevitably lead to cascaded influence on AI model-weights in a type of butterfly-effect that will lead to amusing future headlines about how a chat support bot offered customers fictional deals or products that don't exist.

At this point I'd like to have scattered a few more hyperlinks around the place, but this is demonstrative of another future boss-battle to be had with generative AI: sorting insight from artifact. There are countless, rapidly multiplying articles on the use of AI in the enterprise and it's hard to read, say, a Forbes article on enterprise AI and not find yourself asking "...was this written by AI?".

As a large language mod... I mean, human, I'd be offended by such accusations, but this is the world we're coming to where authority can be mimicked and endlessly recycled and when enterprise systems start mindlessly regurgitating the AI model output from others, one starts to conjure images of a snake eating its own tail. Even as a small operator, we have already received CVs with answer leading with "As a large language model".

We know this will get worse, and we'll be forced to abandon some of the efficiencies in our processes to avoid what I could only describe as "employment catfishing". We don't wish to have a company meetup in London only to discover our new colleague Sebastian is actually a GPU running under someone's desk in Grimsby.

Time will tell how this technology changes the world, but we can be assured that in 2023 and beyond, one has to be AI ready to be in the game at all.

Full disclosure: this was written entirely by a human. How can I prove it? I can't. But I'm also yet to meet an AI that likes Vegemite. Perhaps in 2024...