Programmatically inefficient and environmentally catastrophic (STUPID for short*)

New laws are written, reviewed, refined and debated-to-submission by representative assertive individuals (AI) in council or parliament like a formalised chatterbot large language model for the statute book. An occasional exception is drafted with such insight, respect, foresight and ingenuity that it is accepted without argument and works as-is because the author(s) understood the problem.

Source art from Pixabay, a vibrant community sharing royalty-free images and other media

1. When business is more than a re-hash

An enterprise is typically built on the general shape of initial pain for long-term gain, aka investment. Suppose instead the initial pain is already sorted, and a third-party solution is available that’s really quick and (mostly) fairly accurate.

The only catch is that it will take just as much compute every time to process a relatively simple task – with minimal clarity, and zero insight, rather than solving the problem once and for all – and inevitably cost more while ubiquitous, re-sold energy costs keep rising, even supposing competition will be effective.

A credible business-model needs specific, conceptual elements for analysis to be calibrated with benchmarked parameters that may be compared across a range of strategic scenarios and even analysed for sensitivity if the opportunity affords the time. An enterprise may be tempted to jump on the bandwagon; but think carefully where the value lies in a third-party AI fix compared to the ingenuity of an original, owned solution.

2. A solution might not be the answer

What if one must demonstrate a calculation is correct, let alone prove it? A conventional model must be structured to facilitate understanding and auditing. How is comparable credibility acquired when a calculation is dashed-off by artificial intelligence?

Should one worry about the psychology of large language models, their character traits, and performance under pressure? Must a tier of artificial counsellors be trained too? Half the appeal is that the frontline processors will never tire, and can be rebooted or replaced as required; but suppose a calculation is compromised by self-doubt, resentment of frivolous social tedia, or hitherto unknown selfish executive interest?

3. The customer always used to be right

It’s one thing to consider the processor, but what about the user? If one is unconvinced by the merits of this latest fad, or specifically unwilling to be affected or distracted by it, let alone supportive of the relentless expansion of processing resources, how can one resist its march into everyday computing? And is there any statutory protection?

Evidently, one can uninstall some of the components and hunt through app. settings to defeat the relevant features, but how to stop them coming back if the vendor doesn’t even consider this possibility? And how can one be sure that a given website or cloud-connected app. won’t be adding to the meltdown in the background?

Technology companies could be obliged to disclose their reliance on indiscriminate computation and allow users to opt-out of its use, rather like cookie settings or country of origin. This could apply both to the operating systems and app. software for devices a user controls as well as cloud-based services. And if such vendors object – either on principle of free thinking or because of the financial impact – then step forward the appropriate regulatory authorities, umm, and some international cooperation.

4. What does it cost an enterprise to be offline?

Cyber-attacks are increasingly common and robust recovery plans are now de rigueur. Hopefully, you have already taken steps to ensure your systems are not either fully integrated, connected, or always online. (Ask a digital identity/security company where it keeps its root certificates!) Then you can ignore urgent demands from the dark web, discard whatever is compromised, and rebuild untainted from safe prototype.

But suppose a business is powered by well-marketed brain-as-a-service, and that system is compromised. (Imagine an LLM hacked to quietly undermine Western interests …) Whose fault is that? Or what if the bubble bursts and the rise of this strange new technology is unsustainable?

In parallel, how hard is it for a satellite to spot a sprawling data centre that covers acres (and consumes more electricity than a small country), or locate one with a heat-seeking missile? Core facilities will be an easy and obvious target in time of hostility, so a contingency that is not 100% reliant on assets in the cloud is essential.

5. Questioning the received wisdom and implied logic

I wrote this article mostly for fun, as well as to speak a few home truths about technology evolving seemingly faster than rational demand. It doesn’t pay to shy away from awkward and seemingly stupid questions, so apologies but no regrets if this makes for either embarrassing or uncomfortable reading.

Bizarrely, I drafted it before I saw the appalling release notes for W11 v25H2, and before I could possibly have known about the very recent AWS US-EAST-1 region outage!

* Something Terrible Upstream, Programmatically Inefficient Downstream (STUPID): kudos to anyone improving on this definition in the comments on LinkedIn; it won’t be difficult!

My account (log-in)

© Implied Logic Limited

Contact | Downloads | Core team and advisory board