A higher state of consciousness?
Author: Andy Wood, Founder, Renegade Agency
Disclaimer: AI free. No ai were consulted during the writing of this content. Thoughts expressed here are solely the work of a traditional 100% organic human.
2 weeks ago we ran an event that brought together 50 brand CMOs to discuss the impact of ai on marketing and content. A day later the next iteration of Chat GPT hit the streets and the landscape shifted.
Last week we recorded an interview with The Drum as part of their ‘Ai Asks’ series, and 24hours on the game shifted with the release of the ‘Sparks of Artificial General Intelligence’ research paper detailing what an unrestrained GPT is really capable of.
This video, put together by Ai Explained, is an excellent 12min summary of the 154page Sparks of Artificial General Intelligence paper, which in turn is published by researchers from Microsoft who appear to have had access to an unrestrained GPT months ahead of its latest release just over a week ago.
The video highlights ‘the monumental ability to use tools effectively – an emergent capability not found in ChatGPT’…” from using external APIs to being a true personal assistant, from a Fermi answerer to a Mathlete and a handyman”, and plenty more besides – a must watch.
At the same time, an open letter with signatures from some big names, including Elon Musk, was published on March 29 asking the leading ai labs to pause their ai training for six months, stating that recent advances in AI present “profound risks to society and humanity.”
In the light of these recent developments, we’ve put together a summary update on some of the points we’ve discussed over the past 2 weeks concerning the impact of ai for brands and marketing.
Ability to learn
The ability to use tools is key to human evolution – experiment, fail and learn. This appears to be exactly what an unrestricted GPT can do, without instructions and with minimal demo examples.
According to the research paper GPT can answer complex Fermi questions, engage with the 2022 Maths Olympiad, answer mock code interviews and code its own 3D video game, without having any instructions on these tasks in its database. What’s more impressive/ scary is that these activities also have no clear starting point and often no obvious linear pathway to complete – they’re complicated.
So ,this week we learned ai can already originate code, produce games, and create images from text. Also this week Goldman Sachs published a report projecting as many as 300million at risk jobs from ai.
What does this mean for us?
An important point coming out of the research paper, backing up the discussions we’ve had over the past couple of weeks is the distinction between Incremental and discontinuous tasks. In simple terms ai is still unable to imagine a future outcome and work backwards to design a solution to realise an imagined result – the example given is to write a poem with the same line at the start and at the end. The ai can write a great poem, it just can’t tie the two ends together – yet.
Does this say something important about decision making and our role in using and directing ai tools? Renegades Tracy Wood talks about the need for humans to control key ai inputs and make decisions based on our own imagined & desired outcomes. In short, ai can do the big data work while we set the parameters of a task, prompt engineering, and making decisions about how to best use the ai outputs. This make ai Prompt Engineering a key workplace skill for anyone not wanting to be part of the Golman Sachs 300million at risk jobs – at least for now.
Training your AI
We spent a bit of time playing with GPT using our own Renegade Agency brand name as a topic. At the start of the week GPT didn’t know who we were, but still had a pretty good go at summarising what a ‘Renegade Agency’ might be about and getting pretty close to our own ideas when we founded the business. Over a couple of hours however we were able to train the model, refining and shaping what it thought about Renegade Agency. If you ask it about us today, you’ll get a response that’s much closer to what we’d like you to think and feel about us.
This is something all brands will want to do, but with trillions of data inputs and millions/ billions of humans involved in training ai models that’s not going to be an easy or straightforward thing to do – another key skill for marketers to get to grips with.
Cause for concern?
Training ai is open to abuse and the unconscious bias of the people training it (us). I’m certain there are already undesirable characters training ai models to reflect extreme points of view – think race and gender equality, religious extremism and radicalisation of all kinds.
The Sparks of Artificial General Intelligence paper appears to show unrestricted ai even in its current form can interpret human emotion even when those emotions are contrary to the facts of a situation – think ‘I feel Brexit is a good idea’ balanced against an avalanche of economic data that says it really isn’t. Imagine ai’s ability to support a marketing campaign – crunching trillions of targeting data points and testing creative at a rate only dreamed of a few weeks ago until it gets a desired emotional response. Pretty amazing for brands selling trainers, fast food or games consoles, but with terrifying potential as a propaganda tool.
This amplifies the need for regulation, legislation and transparency in the use of ai tools across all areas of life, something that featuring heavily on the minds of brand owners in our discussions over the past few weeks.
Only time will tell where we go next, but one thing’s for certain we won’t have to wait long to find out. It’s worth noting however, that Apple are yet to release an ai offer. With a track record of fundamentally shifting how technology integrates into daily life (putting the internet in our pocket with the smart phone for example) it seems likely we will see another big leap as Apple release an ‘ai in your pocket’ with app plugins that will almost certainly inject ai into all our daily lives in ways we can’t yet imagine.