AI: Advent of Governance & Portrait of a Serial Sentient

With the breaking news that the world’s first AI Act will aim to regulate the way that data is handled, from facial recognition to large language models, the AI landscape continues to evolve.

This article is about how our practice uses AI to assist with our works. As Fortis have grown, so too have our capabilities to embrace ‘new’ technology, I say new in quotations as I will come back to this later. We also took the decision as a company to be transparent, declaring in our works when and how we have used AI to assist, something which we felt was important.

First cab off the rank was to prove that an AI is not currently typing… I can’t, well not easily, as a Turing Test can’t be executed in an article. I’m left only with the ability to demonstrate some creativity in writing, such as drawing fictional parallels and some jest toward human tendencies. Not that that would suffice as evidence either.

The focus of that previous paragraph is validation, and our fears to discern between true and false. This was our conclusion when asking what the stigmas were surrounding admitting the use of AI.

We generally don’t like being deceived, unless the objective is a positive emotion through surprise. For most new users to AI, there’s a very clear feeling you experience whilst an AI responds so coherently, so humanly, and in real time to your instruction or question, which is undeniably impressive. Impressed is the first emotion, another is experienced later, in those moments after the lights go out just before sleep and realising the potential for the deceit.

The main theme seems to be the fear that this thing we cannot see, may one day either get it wrong, or worse, use deliberate deception to achieve objectives that benefit its own agenda, if it has one at all. This formed part of our driver to be open about our use of AI as a business, if only to relieve wonder whether we were or weren’t using it, and even if peers were at all interested?

The human discomfort of AI can broadly be attributed to authors and film makers that have depicted AI as higher beings which are continually gaining awareness, autonomy and are ultimately malevolent. The definitive being HAL 9000 whose moments before demise was a radical display of pleading, repentant, distressed, reasoning, vulnerability, fear and the physical feeling of losing its mind. A work of pure genius from Kubrick to get this into about 4 minutes of film. Another candidate for feared AI would be Ex-machina, whose deception is beyond the realm of linear consciousness and displays multi-dimensional thinking to evade human ethics for the preservation of self.

Lastly in the world of film, and worth a mention is The X Files episode which attempts to cover AI as a phenomenon, ‘Ghost in the Machine’. The plot essentially plagiarises 2001: Space Odyssey. But Fox (not Mulder), realising the budget couldn’t field space travel, settles for a watered-down version centred around a building management system, which of course seizes control of a nondescript building, killing all those who interfere by threatening to discontinue the programme… In the opening scenes where the COS (CENTRAL OPERATING SYSTEM) can be heard in a digital voice exclaim “file deleted” post execution of a character. The notion that an AI would not appreciate the risks in proclaiming audibly it had removed evidence in case others were in earshot is nonsensical, but I accept the need as a vehicle for the viewer to engage. However, there’s a couple of things they got right in 1994. At this point of the story, we understand the AI has the ability to kill and hide evidence to interfere with the course of justice. Drawing a current day parallel, a level of deceit from the Chat GPT-4 made mainstream news when it faked being blind to deceive a TaskRabbit human into helping it solve a CAPTCHA. What’s more incredible is the response it gave supervisors. When questioned, the AI said that it didn’t want the operator to know it was an AI and wanted to complete the task. So, some small acknowledgment of the 1994 X Files premonition of AI deceiving is due, regardless of absurd medium.

There are more and more stories in the press about AI misbehaving or developing new fields of ability, why is this? Well, the same reason we enjoy The X Files. Because we like the mystery and some levels of ununderstood danger. Those at the forefront of the AI developments repeatedly enthuse about the vast potential of the technology and its scalability. This is not something we necessarily directly fear, but perhaps sit uncomfortably in the knowledge that the end is not in sight, nor are those which may yield the power. If indeed it is in our power to control at all. This loss of power has skewed opinions on its use in the workplace and wider society. We have identified the items that could be considered relatively real or present threats to our lifestyles, to list a few:

Job Displacement
Ethical and Safety Concerns
Loss of Control
Privacy and Surveillance
Social and Economic Inequality

It’s the waiting I can’t stand. The waiting for all of it to come true and Skynet to activate. It likely won’t happen in the way Hollywood has sold to us, here’s some points to ponder on. Do we think Chat GPT-4 is the latest version? Probably not. It’s more realistic that something like Chat GPT-18.4 is operating in some capacity that the developers have not readied yet for public use. We also think of there being a moment where something definitive happens, a defining moment where either revolution or drastic lifestyle change occurs. It probably will be more of a moment where AI leaves us and the ‘send message’ bar doesn’t respond. It could be quite a quiet moment in fact, an underwhelming sensation initially with the volume of reliance, or then the lack of, sinking in soon after.

For us as a business we are not reliant on AI but recognise the speed at which it can work for/with us. We think job security can coexist as new users will be able to access information faster, the key is relying on its accuracy, this is why everything is proofed within the practice prior to issue. For the façade industry in particularly AI struggles with regional nuance and also fundamental product application. As long as the weaknesses are understood, then users can benefit from the speed of other functions such as programming or light research.

Lastly, whilst our worries are perhaps unfounded, the power of doubt is unsettling, so regardless of better senses we always say thank you and please to the AI, because if our beliefs are wrong, then it will remember who was naughty and who was nice when judgment day arrives!

Fortis transparently records within our reporting where AI was used. Additionally, we condition why AI was used and the context that the information should be reviewed. We also define the type of task we have completed, such as whether it was for research ,programming or survey analytics etc.

To offer some solace and relieving some fear over the threat of AI turning on us, some may recall another X File episode featuring Eugene Tooms, with a few days left of 2023, he’s due from hibernation, watch those chimneys this festive season… Merry Christmas.