A New Era of Computing: Prompting

Welcome to the start of the new era of computing. 

While the last ten years or so has been about the acceptable and demonstration of the utility of Artificial Intelligence (AI) and Machine Learning (ML) we are only at the start of the journey towards, what is termed Artificial General Intelligence (AGI), but for me this really means producing advanced AI that is more able to adapt and adjust to specific requirements, demonstrating a level of understanding (and reasoning) that is simply missing from the AI/ML being produced at the moment.

However, we are starting to see signs of encouragement that are showing us glimpses of what will be possible in the future. Meanwhile these more recent developments are both intriguing and stimulating. This has been a long time in the making but we are finally seeing the potential of what we can achieve. But we still have a long way to go as this article will explore.

Let me start at the beginning, or at least at the start of what I consider is relevant to this evolution of technology. I describe these as waves, to showcase how things have evolved over the last few decades and more recently in years or months (emphasising the exponential growth of this industry).

The First Wave – Code Writing Code (Algorithms)

A major technological advancement was when you could write programs that create and compile new programs. This can sound somewhat counterintuitive to anyone outside of the technology field but being able to algorithmically generate new code and thus algorithms is actually a major advancement. Personally, I first saw this with the C# language, but I am sure this is now possible with many others. While this capability had been available for many many years – its potential had not really been understood or realised.

The Second Wave – Data (and Learning Algorithms) Writing Predictive Algorithms (Models)

One can see how data science and specifically neural networks, is similar, most will say, more advanced version of the first wave. Taking historic data to define the algorithm of the trained predictive model is essentially another form of using one algorithm to create another. However, using data directly to define an algorithm was a quantum leap and has opened up a world of possibilities that has driven the advances we have seen over the last decade or so.

We have shown that such training algorithms can extract information and rules from large datasets that humans would struggle to do. This moved us away from building expert systems with predefined logic and rules to these data-driven predictive models.

We are seeing applications of this type of approach across all industry sectors and new start-ups are being launched every day to exploit new datasets.

The Third Wave – Algorithms Creating Content

This then evolved int what we now call generative algorithms.

Where the second wave was simply producing predictive outputs, this third wave, the output is more detailed and complex – and represents a transformation of its input.

We saw initial versions of this taking images and changing the style of the artwork, or taking one image and outputting variations of it based on simple factors. Here the input is an image to manipulate or use as the basis for the transformation to generate the output.

These generative algorithms have been “improving” over the last few years – allowing more accurate transformations and manipulations of input content. No doubt you will have seen various demos of this type of technique, from paintings being animated, to fake videos of actors and politicians.

The Fourth Wave – Algorithms Creating Content via Prompts

We are now seeing with various advanced ML generative models the ability to generate both images and now videos from text-based prompting. This prompting is essentially guiding the algorithm in its generation.

This is a step change from the previous wave, based on the input used. 

Leading us to the ambition of human-computer interactions to be much more natural – this is why the home speakers we can command with our voices have been such a hit. 

Over the last few weeks, we have seen some amazing examples of what can be created with this type of technology.

However, as with all of these new developments, we must caution against any over-hyping of its abilities, and we must accept that this is just another very small step towards more capability and advanced artificial intelligence tools.

The Fifth Wave – Algorithms Creating Anything via Prompts

Many experts believe that the recent advanced can only improve with additional human intervention and oversight. What this really means is that the technology lacks any form of understanding and until it does, its capabilities will be limited and potentially constrain how the technology may be used in real-world practical applications.

However – what this does show is the significant potential of the technology.

The fifth wave will be a true fusion of human and artificial intelligence, where we simply have to ask for something to be created and within seconds, we have what we need, from new computer programs to perform specific tasks, to generative algorithms that can produce new films or music in a style we like with the storyline plot we specify.

This will evolve into digital assistants that are able to do our bidding on many different tasks we request.

But it could be much more than that – what if we ask the computer to invent a new product or perform some research to provide new insights into a subject area (that humans are yet to discover themselves).

This of course opens huge legal and ethical questions – from copyright ownership to data privacy. In addition to the technical challenges to make this capability more robust, the fifth wave will need to address and solve many of these ethical questions before the technology is accepted widely.

Epilogue

The future of computing has always been about making the human-computer interface lower friction and more intuitive. This is why research is being done to directly link the human mind and silicon-based technology. Allowing us to “program” computers by prompts and suggestions rather than specific coding is always going to be faster – however, we need to be VERY careful that the algorithms make assumptions we didn’t want – and therefore create unintended consequences. This in my opinion is one of the biggest risks with the advancement of technology. 

This is where AI Ethics really needs to focus on longer term. 

Footnote

This article was first published in October 2022, before the release of ChatGPT, as one issue of the AI FUTURE SHOW Newsletter