The Evolution and Impact of AI: From Predictive Models to Autonomous Systems

What is Artificial Intelligence (AI)? The most common definition refers to the development of a computer system(s) that can perform tasks typically requiring human intelligence - including recognising speech, making decisions, and identifying patterns. It covers a wide scope.

Artificial Intelligence (AI), particularly Large Language Models (LLMs) for example - ChatGPT Gemini, Co-Pilot, and OpenAI, are powerful tools that can provide educated predictions. However, they lack human-like reasoning and critical thinking abilities. These systems often present results with confidence, which can be misleading if the information is incorrect or misused.

AI operates on statistical patterns and algorithms, using past data to predict and guide future outcomes. However, the reliability of these predictions is not guaranteed, as the data and models used can be flawed or unreliable. The goal is to find repetition, patterns, self-organisation, interconnectedness, and self-similarity amidst the chaos and to implement constant feedback loops.

There are many types of AI. Fundamentally, AI attempts to simulate human intelligence and decision-making in machines. The AI term was first officially recognised in 1956. By 1982, at least 66% of Fortune 1000 companies had one AI project under development. AI has been extensively used since 1982 to predict stock market movements or financial market trends.

A Personal Journey with AI

In my first full-time role over twenty years ago, I worked with statisticians, statistical patterns, algorithms, and models to deliver information to a wide UK audience. My background in mathematics, statistics, and computer science was crucial in understanding the work. Mathematics, statistics and data are fundamental to building AI predictive and prescriptive models.

The data provided had to be of sufficient quality to yield an acceptable output. Despite having limited data, the process methodology, statistical model generation, governance and quality assurance processes were well established. A decentralised approach was utilised for data ownership, IT infrastructure (owned by the department outside of core IT) and the raw data was transformed into a product for public consumption that delivered results.

The computational power did not exist to extrapolate historical performance and convert this into predictive and prescriptive individual-recommended actions for improved performance. The information delivered to the primary stakeholders allowed them to subjectively interpret the results and define their plan to take action to improve their results. Computing power has improved to the point where it is now possible to use AI techniques to predict future performance and generate prescriptive actions from the same data.

Case Study: Banking Industry - Financial Crime - Transaction Monitoring

AI has been actively used in the banking industry for many years. I have performed in-depth reviews of machine learning optimisations to detect potentially suspicious transactions that may indicate criminal activity. Transactions that are identified as potentially suspicious generate alerts which should be investigated to determine if they are truly suspicious and require reporting to the appropriate authorities.

It is possible to use historical alerts, and the determination of expert investigator's past evaluations of these alerts to reduce the number of false positives that need to be investigated in the future. These types of AI systems help investigators concentrate on the highest-risk transactions and reduce the overwhelming amount of white noise that would have been generated by a simple rule-based approach used in the past.

The models used require updating, customisation over time as criminals constantly evolve new tactics to evade detection. It has created new jobs, whilst enhancing the efficiency of the process used as the number of financial transactions continues to grow exponentially.

Case Study: Weather Forecasting

The UK’s national meteorological service, the MET Office, combines weather and climate science data with expert insights to help people make informed decisions. Over 200 automatic sensor stations across the UK collect data for this purpose. It took billions of pounds and over 30 years to achieve the same level of accuracy for a 4-day forecast as a 1-day forecast.

AI is fundamentally about predicting outcomes with a certain degree of confidence. However, these predictions can still be incorrect, leading to potentially disastrous results. For instance, five-day weather forecasts are about 90% accurate, while ten-day forecasts drop to around 50% accuracy.

The Role of Chaos in AI

Chaos theory plays a significant role in AI. A small change in initial conditions, such as a butterfly flapping its wings, can lead to large-scale events like tornadoes. Accurate prediction of such events would only be possible in a predetermined simulation.

Humans, with their free will, add another layer of complexity. They can be influenced, deceived, persuaded, or encouraged to take certain actions, but the outcome is ultimately their choice. This unpredictability is similar to the techniques used by psychologists and magicians.

The Power and Limitations of Mathematics

Mathematics has proven that there are problems that no human, computer, or AI can solve. For instance, Alan Turing proved in 1936 that Hilbert’s Decision Problem cannot be solved by any computer, no matter how powerful. This problem, posed by David Hilbert and Wilhelm Ackermann in 1928, sought a way to determine if any statement is universally valid and true for every possible input.

Limitations of AI

Like any technological advancement, AI comes with its own set of limitations. Often, we are more interested in the answer to our problem than in the validity of the approach used to find it. Both humans and AI systems have their limitations and imperfections. Humans can be influenced by emotions, biases, and fatigue and create fictitious answers. AI systems can make mistakes due to the information provided to them and may replicate human flaws in their outcomes and decision-making processes.

Understanding the Data

It is crucial to understand the data used by AI systems, including its source, the processes it undergoes, the models used to make decisions, and how it will be refined over time. Without this understanding, you run the risk of unforeseen consequences. The inputs and outputs need to be checked and determined if they are valid before use to minimise potential harm.

As leaders, it is our responsibility to ensure that these systems are guided, learned from, and corrected when necessary.

Regulation of AI

The use of AI has grown rapidly in the absence of regulation and with regulations it is forcing us to consider the ethical and potential harms that it causes. The framework for AI regulation will always be fragmented and presents challenges for global markets. China has a diverse fragmented approach, the EU focused on publishing a singular Artificial Intelligence Act - other countries will have their own regulations and these will change over time.

It is important to consider the legal, political, industry sector, societal and ethical factors to ensure that your use of AI is compliant within the law and will be accepted. Existing AI systems may need to be checked as regulations and public opinions evolve.

Evolution of Change

AI will continue to provide new jobs; and modify the processes we use to reach our decisions and solve our problems. It is a programmed assistant which in many cases is not yet capable of replacing a human.

Humans are essential to augment AI systems to provide the necessary oversight, reasoning and critical thinking. It has the power to be used to automate, accelerate and increase the accuracy of many decisions with massive investment and time. It requires many steps to be taken which require human intervention to source data, build, refine and model the AI that we use.

AI will continue to find its way into different sectors and drive increased productivity. The farming sector now includes autonomous tractors. It is being used across many sectors to inspect product quality autonomously. It has found its place in the ability to anticipate and predict potential failures in hardware through sensors and visual inspection minimising downtime and maintenance costs.

The best development of AI focuses on a business problem that can generate cost-effective strategic value and align with the goal you intend to solve. If a person can demonstratively and repetitively do a task without critical thinking or undetermined reasoning - the chances are high that autonomous AI could do it. It is often more economical to employ someone than to put in place a replacement AI system for many tasks. This may require other changes in regulation, law, insurance, and finance before it is permitted.

Fully autonomous cars are a prime example of where changes are required and discussion in these sectors is ongoing. The product may be ready before the safety, regulatory, legal and insurance frameworks are ready for all markets.

Consumers have learnt expectations from futuristic science fiction movies that show us how great AI could be. Marketing talks about AI products - the promise is exciting, but the reality of the delivered product is often a stark shock, expensive, disappointing and underwhelming when it arrives. To properly understand the product a deep dive into the capabilities, functionality, risks and limitations is required. It is a tool to help us automate, its definition is so broad that it has become overused and permutated into our daily lives.

Image AI systems which can understand and respond to the emotions and needs of individuals. It would analyse the tone of voice, choice of words and prior conversation history. This could potentially revolutionise existing customer service lines where we dread the automated voice recognition or script reading agents as our first point of contact. If AI can determine a person's emotional state and generate a tailored empathetic response to reach the desired outcome. The potential of AI may be more powerful than many human customer service agents achieve. It has the potential to improve customer understanding, interactions and satisfaction. This reality of achieving such advanced AI is complex and requires significant technological advancements.

It does not generate a wholesale replacement of jobs; it changes them due to the many limitations it possesses. What it does have the power to do is change the nature of how we work and what is required of us.

AI will continue to evolve and improve. When we reach a strong AI scenario that can critically think, reason and take full autonomous action without requiring human feedback the future may be very different. AI has the potential to supersede humans in the future.

AI is not yet fully autonomous and I have no doubt we will reach a point where many existing professions will become fully autonomous AI. AI has a broad definition and it is unlikely that anyone can be a full AI expert encompassing the full scope of AI.

This then raises the question - if AI can do everything what use is there for humans - why keep us at all?

I hope that we will all have the foresight to prevent this scenario.