EMERGING TECHNOLOGY: Artificial intelligence

“I'M SORRY, Dave. I'm afraid I can't do that,” is a famous line from the classic movie, 2001: A Space Odyssey that possibly comes to mind for many of us when we ponder artificial intelligence (AI) and ask what could it become in the future? 
Just what is AI? Where did it come from? How does it work? What are the benefits and what could it become?

Development of AI

When we consider the history of AI, most of us will likely have some awareness of the advances of computing during the past 80 years, with this progress helping enable modern AI development. What may be less widely known, however, is that the concept of AI was considered much earlier and has been at the centre of philosophical discussions for thousands of years. These discussions considered human consciousness, thought and learning along with the notion that machines may be developed to learn and integrate with human life.  

Beyond these early philosophical ideas, many machines have been considered, designed and built to operate independently of humans, with this level of automation arguably an early form of AI. Furthermore, the 18th Century work by Thomas Bayes on probability inference, which is an iterative process refining probability, is a foundation for modern machine learning.

Moving further along, it was in the post-war environment of the 1950s where modern AI started to take its current shape, largely thanks to the significant work of Alan Turing and John McCarthy. Turing developed the ‘Turing Test’, which is a test of a machine's ability to exhibit intelligent behaviour equivalent to that of a human, evaluated through a series of questions. However, it is McCarthy that is generally considered to be the father of AI, with McCarthy coining the term ‘artificial intelligence’ as part of a research proposal in the 1950s, with the term defined as ‘the science and engineering of making intelligent machines’. 

From this early work, AI development has attracted global interest and significant research and development focus from academia as well as from government and the private sector, with the Defense Advanced Research Projects Agency (DARPA) and IBM being just two examples of government and industry investing heavily in AI in comparatively modern times.

AI today 

So just what is AI defined as now? A consolidated general definition of AI could be computer systems and computer controlled robots that simulate human thinking to learn and undertake tasks associated with intelligent beings. This definition of AI is often applied to the development of systems featuring human intelligence characteristics such as the ability to reason, discover meaning, generalise or learn from past experience (yes, these are all human characteristics!). 

There are two other key terms associated with AI, those being machine learning and large language model (LLM).  

LLMs

Machine learning is often incorrectly interchanged with AI, however, it is more appropriate to consider machine learning as a branch of AI. Machine learning is the ability of computers to learn without being directly instructed or programmed. LLMs are deep learning models built on understanding and analysing large amounts of data. As I’m sure most readers will be aware, commonly known LLMs include ChatGPT, Microsoft Copilot and Meta AI, with the LLMs pretrained on vast amounts of data, including copious amounts of words. 

It is reasonable to conclude that AI will at least drive changes to almost all human roles and tasks.

It is this pretraining and exposure to large datasets that set the capability for LLMs and more importantly, allowing the LLMs to learn the patterns and rules of language. This enables the LLM to understand meaning from text including phrases and the relationship between words. With this style of learning, it could be said that LLM learning is actually similar to the way humans learn to communicate through exposure to language. Again, similar to humans and a case of practice makes perfect, the more data and reading, the better the understanding that the LLM has. It is for the LLM’s ability to grasp natural language that it is considered superior to machine learning.  

Through this learning and access to data, the LLM can answer questions, translate between different languages, summarise and even write content, which of course is a significant disruptor when compared with traditional human centric methods. Looking further ahead, it is conceivable that newer and more capable LLMs will be significantly better in providing genuine insight, with a corresponding reduction in the need for human prompt and intervention. This will likely not only continue to change the way that we search for information, but more importantly, enable much greater automation using this information. While I have only discussed text as learning data above, the same logic applies to where the learning data is images and sounds. 

What's next for AI?

All around us we are seeing many benefits of AI, including automation of routine and complex tasks and generating predictions to improve or speed up processes. While there is seemingly no limit to where AI may apply in the future, some industries that are expected to change significantly include healthcare, retail, finance, law, HR, logistics, transport and entertainment.   

Unsurprisingly, there remain many predictions and differing views on what this means for humans and jobs. While it may not be clear yet on what roles will continue to be in demand and what roles may fade away, it is reasonable to conclude that AI will at least drive changes to almost all human roles and tasks. Furthermore, and similar to the broader benefits that other technological advancements have brought to humankind, there is strong opinion suggesting that the onset of AI will actually create far more jobs than it makes redundant.  

Only time will tell whether the future holds a scenario where AI has HAL disobeying Dave’s orders based on HAL’s own learning, processing and... consciousness.

This article appeared in the August | September edition of DCN Magazine