AI and machine learning dominates our conversations. Modular unpacks how to approach the complexities of AI.

6 min read

Updated: 15 Jun 2024



The debate around artificial intelligence has reached something close to fever-pitch, and AI itself has become a truly vast and rapidly evolving field of innovation.

Here at Modular our approach to digital technology has always been pragmatic and evidence-based; we believe in dealing with complex issues in tech with a cool head. Artificial intelligence may be a technology that’s destined to transform the world completely, just as the birth of the internet itself did a generation ago. If that’s true, then it’s particularly important to approach it with care, attention, and precision.

Artificial Intelligence is in reality best viewed not as a single form of technology, but as three separate technologies: ANI, AGI, and ASI. Two of these remain — for now — largely confined to science fiction. The AI everyone is talking about right now, the dominating our feeds and minds is Artificial Narrow Intelligence, or ANI. Also, otherwise known as Machine Learning (yup, not quite so catchy, is it?).

Artificial Narrow Intelligence (ANI/machine learning)

ANI describes the field of computer science that is broadly termed ‘machine learning’. ANI technology will typically consist of a learning algorithm, which is a piece of computer code that can train itself using a certain dataset and make predictions based on what it learns.

Such algorithms can complete specific tasks without any human assistance or interaction. Easily recognisable examples of ANI technology currently deployed are image recognition systems, such as those that operate the electronic gates now used for passport verification in many international airports, and language auto-translate software applications.

In the world of digital and marketing, AI is applied to content creation, image creation, software development and product names you no doubt will of heard of include ChatGPT, Jasper, and Adobe Firefly. Marketermilk has pulled together a list of the top AI tools along with a good overview of how to employ them within your business.

As the most basic form of artificial intelligence, ANI systems are often deployed to perform relatively simple, repetitive tasks that would previously have been done by human workers. The tasks that ANI can be extremely good at don’t involve the more complex executive, analytical, and cognitive functions we associate with higher levels of human intelligence.

Artificial General Intelligence (AGI)

Artificial General Intelligence, or AGI, is a much more sophisticated and potentially far-reaching technology than the basic machine learning systems outlined above.

Whilst it remains currently a hypothetical phenomenon, AGI (which is sometimes also called ‘Strong AI’ by technologists) describes a form of artificial intelligence that has the capacity not only to learn things for itself, but also to understand and apply that knowledge to a diverse set of tasks and actions.

In theory, an AGI could learn and perform any cognitive task that a human being can currently perform. It could fly a fighter jet in a direct combat situation, defend or prosecute a complex criminal case in a high court, or design an end-to-end supply chain logistics system for a large multinational company. It would be as clever as any human being, and at the same time it would be able to do things much faster than any human being.

The creation of such an AGI is one of the goals of the world’s leading artificial intelligence research companies such as OpenAI, DeepMind, and Anthropic.

Interestingly, significant debate exists about whether or not AGI might present a threat to humanity.

Leading AI development firm OpenAI regards AGI technology as a potentially existential risk to humans which must be carefully controlled, whereas others believe the existence of a fully functioning AGI remains so unlikely that it presents very little risk.

Time, as ever, will tell.

Artificial Super Intelligence (ASI)

A final form of artificial intelligence has been defined which would exist and function at a level above the scope even of the AGI discussed above.

Just like an AGI, it remains at present completely hypothetical.

A so-called Artificial Super Intelligence, or ASI, would possess a level of intelligence far beyond the sharpest and most talented human minds.

The pioneering chess supercomputer Deep Blue, which beat several of the world’s best chess players in the 1990s, first pointed out the potential capacity of computing power versus the highest level of human intelligence. Its 1997 victory against Russian grandmaster Gary Kasparov is regarded, in fact, as one of the key landmarks in the history of artificial intelligence.

Researchers have pointed out that if intelligent systems rapidly become super-intelligent, they may take unforeseen actions that evade human oversight. In this context, a self-improving Artificial Super Intelligence could become so powerful that humans would very quickly lose control of it, and it would likely operate independently as a so-called ‘Intelligent Agent’.

Unlike the chess supercomputer Deep Blue, or even a human-friendly AGI, this most advanced kind of AI could probably not be controlled or switched off by humans.

The fictional battle between astronaut David Bowman and the supercomputer antagonist HAL 9000 in Stanley Kubrick’s film 2001: Space Odyssey shows some of the potential problems of artificial super intelligence with remarkable foresight.

Returning from the speculative to the practical, there are a host of machine-learning tools currently available that can be applied by businesses and organisations to help improve customer experience, optimise sales, and gather important data.

Rather than worrying about robots taking over the world, it’s perhaps worth asking exactly how artificial intelligence as it currently stands might help businesses and organisations right now.

The robots are still learning

The machines are still learning.

Despite the considerable opportunities that artificial intelligence offers, it would be a mistake to assume it is a seamless, perfectly functional technology without caveats. As with any fast-evolving field, the commercial use of even the most basic machine learning systems can throw up problems.

These include so-called ‘hallucinations’, a specific problem in large language model processing, in which the model produces content that is either nonsensical or unfaithful to the provided source content.

Other issues include problems with data processing and server capacity, since the deployment of machine learning can use up enormous amounts of processing power. If the server for the website or app using artificial intelligence isn’t powerful enough to handle the high level of processing required, systems can simply shut down as a result.

Another issue directly linked to this is the potential increase in a company’s digital carbon footprint that the use of this technology creates; ever-bigger servers and more powerful processing requires more electricity, so there’s a direct increase in energy costs as a result of using artificial intelligence in any digital system.

Another intriguing area surrounding the use of machine learning is the issue of data starvation.

In computer science, this term defines a situation in concurrent computing where a process is denied the resources it needs to operate. In machine learning systems, sometimes it may be the case that an algorithm actually ‘runs out’ of meaningful data to access for its output to be successful.

This may be particularly true if it is being asked to do something for which little data exists, or possibly if it is being used in an area that’s outside the one for which it was purposely designed.

In the same way an aerodynamic time-trial bike doesn’t work on a downhill mountain bike trail despite the fact that both are clearly bicycles, a machine learning algorithm designed for one task won’t necessarily be very good at another one, even if the other task is very similar to the original one.

Algorithms can learn to do things by themselves, but only within the technical parameters that their developers have set for them.

Looking at the product options on the market, the prospect of bringing AI into your business can be daunting, combined with the mounting FOMO pressure to be an early adopter or left behind, it is not surprising many of us are a little overwhelmed.

The Robots are here to help

In summary, machine learning systems are generally better at assisting humans with certain tasks than actually doing those tasks themselves. Right now, artificial intelligence is arguably more effective in the real world in an advisory capacity than an executive capacity.

Machine learning algorithms can fly planes and play chess better than human grand masters. But they can’t do everything.

An algorithm will probably never be able to perform tasks that require emotional sensitivity, creativity, or critical thought as well as a human being.

These powerful new technologies have the potential to improve customer or user engagement, streamline supply chains, supercharge data collection, and directly increase sales.

At the same time, if they’re not implemented in the right way, they can create far more problems than they solve.

The most important thing of all before launching a plan to integrate artificial intelligence into your business is to talk to an expert about exactly what you need to achieve, and how machine learning systems might help you get there.

Emma Millington


Smart thinking

Balancing technological advances with human needs is the greatest challenge we face today.

In each of our evidence based case studies and insights, we explore unique audiences and how you best meet their needs whilst at the same time embrace cutting edge technologies.