Sparking reactions ranging from fear to excitement, the oft anxiety-inducing technology is here to stay, according to Marshall, and it’s time to get — responsibly — used to it. 

To start the discussion on one of the major evolving technologies of today, Marshall began his talk on AI at Dynamic Earth with a quick history lesson. 

The Great Horse Manure Crisis of 1894 — where major cities across the world were overrun by horse manure from carriages — required an international conference to attempt to solve this major sanitation and infrastructure problem. 

But the conference ended in just three days with no solution – there just wasn’t the technology to solve the issue. 

The solution came “completely out of the blue,” Marshall said: the motor car.

When the motor car first arrived though, this solution was met with fear. 

“Immediately when there’s new technology, even when it solves a lot of problems, people react badly to it,” Marshall explained. 

He went on to mention the first patent for a tube of paint – this tiny technological invention enabled a revolution in art, sparking the impressionist movement. The ability for artists to paint outside and on the go made art more accessible, and transformed the art world completely. But even this movement was met with derision. 

AI may seem revolutionary, and indeed, it is. But we’ve seen these revolutions, at multiple scales, before. 

“Does generative AI change things? Yes it does,” Marshall conceded. “Have we seen changes like this before? Yes we have.” 

Tracking AI 

But are these changes to stay, or is AI just a fad? 

To analyse this, Marshall looked at the Gartner Hype Cycle, calling upon his previous work at the company.

The Gartner Hype Cycle tracks different technologies’ journeys as they grow in popularity and use. 

Importantly, the complete Gartner Hype Cycle tracks successful innovations as it slopes up and down – eventually plateauing. 

Starting with the ‘Innovation Trigger’, technologies then rapidly increase in popularity to reach the ‘Peak of Inflated Expectations.’

For AI, a lot of these expectations could also be fear. For example, as AI developed from its early stages, the public was terrified of major job loss across industries, and a dystopian armageddon of machines taking over humans. 

While these expectations – or fear – dissipate, technologies, even the eventually successful ones, plummet to the ‘Trough of Disillusionment.’ This is where the nay-sayers, pessimists, and sceptics grow in number as applications get more complicated. 

In AI, discussions on plagiarism, the lack of a universal truth, and bias had many people feeling wary of the exponential growth of the tech.

Even now, the proliferation in development and use of AI has seen a number of notable tech leaders sign an open letter, imploring a pause in AI development until regulation can catch up.

The ‘Trough’ is also where less impactful technologies are most vulnerable to being thrown away to the ‘Skip of Insignificance’. But for the innovations that make it, this is where the real popularity and lasting growth occurs. 

Innovations – once they overcome these early issues – begin their ascent on the ‘Slope of Enlightenment’ where more developments occur. This is where the ‘hype’ levels out to the mainstream, and real development makes the tech fundamental. 

According to Marshall, AI is now at the ‘Plateau of Productivity’. While it continues to expand its reach, the technology’s place in certain areas has stabilised, and major tech companies are making AI projects a central pillar in their growth plans after massive layoffs. 

But despite its rapid expansion, it still seems to be early days. How do we know how to understand the significance of new technology? 

Marshall advises to be wary of listening to people with stakes in the game.

“Loud mouths with vested interests” are to be avoided, according to Marshall, showing a photograph of a Bitcoin investor. 

Equally important is understanding what the technology is for. “Half-baked technology pitches trying to control the future,” he urges the audience to steer away from, giving the Metaverse as an example of a technology likely heading straight for the skip. 

Listening to analysis is important. AI is resoundingly popular – ChatGPT, just one of many AI chatbots – is the fastest-growing new technology, ever, reaching 100m users in less time than phones, computers, and the internet. 

“That tells us that it’s triggering the belief that it is useful,” Marshall said. “These are tools that will help people do their jobs.” 

What to Worry About 

“AI is an accelerator; it may not be refined but it is powerful and very fast,” Marshall explained about generative AI. 

While some may like to focus on the ‘powerful’ and ‘fast’ aspects of AI, Marshall delved further into its issues. 

“These things do not understand what they are doing, there is no ‘intelligence.’ There is no world model, and there is no concept of truth behind it.”

These are some of the overarching concerns about generative AI as a whole. While ‘intelligence’ may be in the name, Marshall assures us there is no intelligence in the model. 

Marshall detailed some of the other overarching issues regarding generative AI, including bias, which can be introduced by the historic data used to train an AI model. 

“Incomplete training will let bias get in there,” Marshall said. AI models have been known to introduce bias based on the data it was trained on – Marshall gave the example of women’s CVs being thrown out by an Amazon AI recruitment bot. 

Marshall urges users to be careful with the data they feed AI models as this can perpetuate biases in the data. These biases are not generated by the AI, but existing biases can be exacerbated from a system continually regurgitating the same information.  

Another central issue in the AI expansion is plagiarism, which has entered mainstream debate especially in the case of intellectual property. Artists and writers have dealt with conundrums of how AI uses intellectual property to create ‘new’ pieces of work. 

Ultimately, it is up to the user, not the AI platform, to be careful of plagiarism, according to Marshall. But defining what plagiarism truly means in these cases only fosters further conundrums. 

“The problem is that as an artist, or a writer, or anything, we are all relying on hundreds of years of history – so how much plagiarism is going on in our own head, and is that any different than AI?” 

The concept of truly original thought generation makes the debate much less clear cut – AI ethics and AI philosophy is causing the public, and technologists, to ask much bigger questions than expected. 

This includes the very concept of truth. As mentioned by Marshall, generative AI has no concept of truth. 

Ask ChatGPT a question, and depending on the phrasing, it could generate completely fake, but completely real-sounding, scientific papers and references to back up a false claim. Tell ChatGPT that 2 + 2 = 5, and it might just believe you. 

The data inputted and derived from generative AI therefore cannot yet be fully trusted. 

Further, generative AI, in its current form, is not auditable. As it does not understand itself how it comes to its own conclusions, it is impossible to check, or repeat, these results. 

“The problem with generative AI is that it’s almost impossible to explain why it did what it did – the lack of auditability is a really big issue for generative AI,” Marshall said. 

Recommended

• Q1 VC Funding at Lowest Level Since Pre-Pandemic
• Scots Investors Archangels Make Senior Appointment
• CMA Blocks £55bn Microsoft-Activision Merger

“No matter how big the model it” – no matter how much data the model uses to generate results – “it still doesn’t know what it’s doing.” 

Fundamentally, Marshall explains, it is a powerful probabilistic approach. And the probability may change. And this probability can be unpredictable if it cannot be explained. 

Play Time is Just Beginning 

But while generative AI is plagued by all of these technological challenges and philosophical issues, Marshall was certainly not trying to scare us. 

He provided the audience with basic advice, such as how to protect data by not sharing it for no reason in public AI models, and on using secure models. 

Most importantly, he stressed the importance of education. 

“Education is always the best policy: explain to people why they shouldn’t do something rather than just telling them not to do it,” he urged. 

By educating people on the dangers of poorly monitored AI, or what AI can and cannot do for you and your business, we can better use the technology for good. By understanding how biases are introduced, how to secure and use data, and how to avoid plagiarism, we can responsibly use AI before we get to some of the more ethical concerns regarding its use. 

And the best way to learn is by doing, Marshall exclaimed. “Experiment, don’t ignore it. Use it, understand it; it’s really important,” he said. 

Setting clear targets, so that AI is being used in a fundamental manner for a defined purpose, is also essential. 

Most importantly, Marshall declared that AI is certainly not a fad. 

“This is not going away, this is fundamental: generative AI is fundamental,” he said. 

“So go out and play, folks.”