I, for one, welcome our new [ insert existential threat ] overlords.
Whilst H.G. Wells vision of a future ruled by insect overlords hasn’t come to pass (yet) we are still facing exponential growth in one of the most fascinating technologies developed in recent times. AI and Artificial General Intelligence, has jumped to the forefront of recent technological discourse. Some, like Alan Turing, Elon Musk, Sam Altman, Yoshua Bengio and other CEOs and tech leaders have all campaigned for a cautionary approach to AI and AGI development.
Unfortunately for them, through source code leaks, and the inventiveness of the human mind, these AI agents are now in development worldwide. Whilst the idea of an AGI might seem far away, the exponential growth of AI and LLMs (Large Language Models) feels like crossing the Rubicon in terms of information delivery.
This might be no bad thing for humanity.
Freedom Technology or Nomophobia Enabler?
The upshot is simply a question of time, but that the time will come when the machines will hold the real supremacy over the world and its inhabitants is what no person of a truly philosophic mind can for a moment question.
Samuel Butler – Darwin Among The Machines
No one can deny the positive effects of technological advance on our standard of life in the developed world. It has enabled us to build immense cities, provide food and energy for billions, and networked humanity to a level of interconnectedness that wasn’t imaginable 100 years ago.
This same technology has also changed how we function on a social level. We talk to each other more, but connect in person, less. A Penn State survey showed that 77% of respondents thought we were too dependent on technology, especially as 75% of children in the West spend more time indoors engrossed in video games and online worlds, than outdoors.
More staggering is the rise in addiction to mobile phones. So much so that it has been called “possibly the biggest non-drug addiction of the 21st century”. The term ‘nomophobia’, an irrational fear of, or anxiety caused by, not having a working mobile phone, is behaviorally similar to drug addiction.
If the internet, or any other modern technological substructure of society failed today, it would be catastrophic. We are already wholly dependent on the networks that we have built in order to sustain our modern way of life. The argument, in this case, is that we are already under the supremacy of machines in one way or another. We’re just using it wrong.
Put down the phone (speak slowly for the robot please)
Most current technology requires constant interaction with a device we carry with us for most of our life. Primitive digital assistants in the form of static devices have grown increasingly popular as a way to access information and order our lives. Home devices such as Amazon’s Alexa use Natural Language Processing, automatic speech recognition and weak AI to perform basic tasks, such as controlling connected devices, or basic information requests.
We already know our brains love efficiency. The rise of AI agents with voice recognition and the ability to venture further than a set of limited instructions offers the opportunity for true interaction with software that removes a lot of the friction in our daily lives. Already, AI agents are in development that would allow you to request an itinerary for a weekend away, including flights, restaurant and hotel bookings based on your preferences. This entirely removes the chore of trawling through review websites, flight planners, checking schedules, transportation and the minutia of organisation.
Tailored LLMs (Language Learning Models) and AI agents will soon be able to run complex project management tasks within companies. Delegating tasks to individual workers it knows will be more efficient or suited, adjusting timelines and deliverables in real time, controlling supply chains based on predictions of demand or supply. All whilst remembering every employees favourite color, their birthday and interacting with customers and clients on a one to one basis.
A true personal digital assistant is on the horizon.
It just hasn’t been invented yet (your job)
Luddites destroyed technology and machinery because they feared that it would make their lives worse and destroy their livelihoods. Indeed, since the beginning of the industrial revolution there has been an outcry from every career path and craftsman that saw the demise of their jobs through technological advances. This cannot be viewed in isolation. Where machines replace humans, in time, there is economic growth and rising employment.
The Edelman Technology Trust Barometer has revealed that around half of people around the world feel that technology is moving too fast and are leading to changes that are not good for them personally.
There can be economy only where there is efficiency.
Benjamin Disraeli
Machines have been taking over the most laborious and routine manual jobs simply because they are more efficient and readily serve as a substitute for human labor. All jobs lost in disciplines and fields out-paced by technological development are recovered by a large margin in other fields.
The rise in development of AI and LLM software is giving birth to a new generation of jobs. A Prompt Engineer is someone who optimises and designs prompts for use within AI software. A front end user may achieve basic results using a Chat LLM input but when designing software for a specific task (such as reviewing document archives, or creating an AI assistant) an improved user experience will be optimised through prompt engineering.
Transformers (LLMs in disguise)
The rate of physical technological development is exponential up to a point. Moore’s Law is a well known observation that the number of transistors in an Integrated Circuit doubles about every 2 years. Recently, however, it has been realised that we may soon come to the later end of the exponential curve when it comes to IC due to the physical constraints of material science.
AI development is showing that same exponential curve of slowly at first, then all of a sudden. With linear progress from the early years (1960-2010), exponential progress (2010-2020) and now exhibiting signs of compounding exponential growth from 2020.
With an LLM, the key to efficiency is the Transformer Architecture. This is the fundamental building block of Large Language Models used in AI applications such as ChatGPT. With the primary driver of AI efficiency being software based instead of hardware based we are seeing an increase in efficiency far beyond previous technological progress. LLM transformer size has been increasing at around 240x every 2 years.
Don’t trust, verify
So where’s the downside in this AI powered, hyper-efficient future where we use LLMs to streamline our personal and business lives all whilst creating new economic opportunities and advancements?
At the most basic level, all of our interactions with AI agents and LLMs are about condensing and delivering information to the end user. It has already been seen that the developers of LLMs are using their own inherent biases to curate and steer the information delivery process to end users. This could be relative to social, political or scientific issues. For example, on release, ChatGPT would respond to questions and provide relatively unbiased results, but in the subsequent development of the LLM it has been increasingly steered towards a pro-environmental, left-libertarian orientation.
The control of information in our modern world is vital to a healthy discussion on almost every single topic. If the controllers of an LLM comply with a Health Organisation’s legal demands that specific treatment advice or information is broadcast to the exclusion of ‘misinformation’ then this is what the end user will see. If that information is subsequently proven to be misleading, or false, there are serious repercussions for the end user.
LLMs use a combination of data, mathematics and rules set by the designer to respond to the users input. When the designer of that system creates a bias, then this is what the users will see.
Conclusion
There is no way to create an AI or LLM without some inherent bias. Bias is a relative concept, and something that one person considers neutral might be considered biased by someone else.
We will also find that due to recursive programming and compounding exponential efficiencies that AI development will be rapid and frightening for many, whilst ushering in incredible technological advances for many others.
In the short term, what we do need however, is clarity on who is determining the real life human feedback process that determines the information output. An LLM in the hands of an oppressive or censorial government would be dystopian indeed.