What is generative AI – and how can government use it?

Photo Shutterstock

Artificial intelligence has the potential to impact almost every area of life. This second article looks at the technology behind generative AI, and how you can use it.

You’ve probably heard about generative AI over the last year or so since a free version of ChatGPT was made available to the general public.

Generative AI (GenAI) is at the forefront of today’s focus, not merely as an advancement of artificial intelligence but as transformative technology that has the power to reshape how businesses, governments, and humans innovate, operate, create and deliver value. 

GenAI is an AI method that learns from real-world data to generate new content – and this could be text, images, audio, code, video, or tabular data, with similar characteristics of the data it is trained on.

Three primary applications of generative AI include large language models (or LLMs), synthetic data, and digital twins. In this article – building upon the last article looking at other types of artificial intelligence – we will look at these different models and how they work.

Large language models

A large language model (LLM) analyzes huge amounts of text – millions or billions of words – to train itself to be able to know the relationships of words, and to then produce human-like text.

For example, how would you finish this sentence, “After he messed up, the boy was in the dog-” Did you say “dog treat” or “dog days of summer”? No, of course not. You said “doghouse” because that is a common phrase that you have heard over and over again. “He was in the doghouse” is commonly associated with making a mistake, messing up, acting badly.

The inventors of LLMs took the natural language processing concept of scanning each word in a sentence and translating it in a sequential process to instead reading an entire sentence at once, analysing all its parts rather than the individual words. This provides better context.

So, LLMs track and learn relationships in sequential text – or text in which the arrangement matters – to respond to prompts using similar text.

When we ask an LLM like ChatGPT or Copilot to write a sonnet about our favorite dog and his love of slippers, it seems like the system is being creative and generating completely new ideas. In reality, the system – using millions or billions of data – sequentially lines up the most probable next best word or groups of words. With unlimited computing power and tremendous speed, tasks that would take humans hours, weeks, or even years to perform can be done in seconds.

One of the most prevalent uses of LLMs is as an advanced search engine. Not only does it retrieve information from the far corners of the internet or an internal database, it often assembles the information in a relevant and consumable manner. But it is important to note that an LLM alone does not solve business tasks. The key is to integrate an LLM into a decisioning process. Combining an LLM with other computer or human systems – such as layered on top of other AI algorithms or part of a investigator’s inquiry process – accelerates value for an organization.

An excellent example of this is the use of NLP and a LLM to consume large volumes of public commentary. Hours of reading and synthetising public comments can be processed in a fraction of the time as NLP models ingest and organize commentary into groups and an LLM interprets the findings for easy consumption.

Synthetic data

Synthetic data is artificial data that accurately mimics real data. This on-demand, automated data is generated by algorithms or rules, as opposed to traditional data sets gathered from the real world.

Synthetic data reproduces the same statistical properties, probabilities, patterns, and characteristics of the real-world data set from which the synthetic data is trained, and has been found to be as much as 99% statistically valid.

Governments can use synthetic data for various purposes, including research, testing, and analysis, without violating privacy regulations or exposing sensitive information.

There are three primary reasons why governments will want to use synthetic data:

1. Synthetic data can supplement your data set when there is not enough real-world data. For example, you want to test the concept of making a road improvement, but you only have a few months of traffic data. Creating synthetic data – such as simulated traffic flows – allows you to test possible road improvements. In cases such as this, creating artificial data assists in training or testing models, running what-if scenarios, or identifying optimization.
2. Synthetic data can also be used to protect sensitive data. Synthetic data can mimic actual data sets without containing any personally identifiable information. So, you could create synthetic data to train and test a system that processes health records, student records, or tax information. It allows governments to harness the benefits of data-driven decision-making while respecting individuals’ privacy and data protection rights.
3. Synthetic data can be used to complete a data set when the real-world data is dirty or has gaps, thereby improving the usefulness of the data set.

Digital twins

A digital twin is a virtual model of a physical object or system from the real world. For example, a government might build a digital twin of a road network, a supply chain, or a financial system.

A digital twin can be used to make predictions about the real-world impacts, such as that from an accident on a highway, a supply shortage, or an economic disasters. What-if analysis can be used to virtually test the effects that certain decisions might have in the real world.

Digital twins use a combination of different data as inputs such as historical, real time data, synthetic data, and system feedback loop data as inputs. These inputs can be processed in batch or in real time.

While a digital twin of a physical environment might be the most obvious, digital twins can be used to test the impacts within a policy framework. For example, a government can use a digital twin to test the implications of tax changes before making them.

The rise of AI: how you can use it

Many people ask why AI is taking off now if it has been in existence since the 1950s. It comes down to the maturity of three elements: computing power, data, and analytic models. For anyone who ever did programming, you recall how long it took to churn a program that used a lot of data. In the 1980s, it was common to start a program running before leaving work with the hope that it would run without errors overnight. Today, computers’ power to churn data is so great that those same programs run in seconds.

The computers have a whole lot more to churn now as we have lots of data – especially in the public sector. All of this data is the fuel that AI needs to produce results. And, finally, we have sophisticated analytic models that emulate tasks previously performed by humans.

Even with advancements in machine learning, deep learning, and generative AI, there is still a clear distinction between what humans and machines do well. Humans use common sense, intuition, creativity, empathy, and versatility. Machines take on tasks that humans could perform if we had all of the time in the world and didn’t get fatigued: processing large data sets – not only consuming, but learning from massive amount of data, performing complex calculations, and automating tasks which can be performed without human intervention or assistance.

While the advancement in Generative AI suggests that machines will be able to take over the world, let’s remember how machines generate their output. In large language models, when given a prompt, machines grab the most probable next word, line of code, or image. Some marvel at the LLM’s creativity. But the creativity occurred when humans put information into the dataset. A machine doesn’t understand what it means to have a broken heart but will finish the sentence “When he left me, he broke my…” with “heart” because millions of people have written that. A large language model is operating without common sense and true intuition. Because this approach is effectively guessing, LLMs only assemble and present what humans have previously rendered.

Many people are wondering if AI is going to take away their jobs. Of course, we can’t predict the future, but what we are seeing is that AI is being deployed to complement the work of humans. Humans make sense of patterns and trends surfaced from large data sets. Humans interpret results from complex calculations. Humans assess the impacts of simulated scenarios. Human handle exceptional cases apart from those that can be automated, and so on.

In short, we expect machines to take on the work that is too cumbersome, too time-consuming – too boring – so that humans can optimize their use of time to improve outcomes.

AI technology is very powerful and will become even more so. But as with any great power, there is great responsibility. The advancements that we are seeing in AI technology have far-reaching effects and implications. Therefore, we must have a trustworthy and ethical approach as we set our strategy and guardrails for our use of AI and generative AI. We must place human-centricity, citizen interests, and doing the right thing first.

Consider these six principles as you adopt any AI or GenAI solution.

• Human centricity – remember that the AI solution is developed and run in order to promote wellbeing of people.
• Inclusivity – ensure that a system is created by and for people from a variety of backgrounds, with diverse perspectives and experiences.
• Accountability – be proactive in identifying and stemming adverse impacts.
• Transparency – provide a “clear box” rather than a “black box” – in other words – be open about how the AI system works, why it produces certain results, and what data it uses.
• Robustness (also referred to as stable AI or resilient AI) – implement an AI system that is able to function effectively even when operating in unexpected or changing environments.
• Privacy and security – protect the identities of the people who are subjects of the AI system.

These six principles can be difficult to navigate when working with AI and generative AI. It is critically important for humans to qualify the input, the prompts, and the output when AI and Generative AI is used. By keeping the “human in the loop,” governments will be able to take advantage of the strengths of AI while limiting the risks. Rather than diminishing the role of the human in government work, it will change and elevate the work that humans do.

But, with the potential of AI to take on so much work, it is understandable that people are concerned about AI taking over their jobs. Is AI going to take your job? It is more likely that someone who knows how to use AI could take your job.

This is the second of a two part series of articles looking at how generative AI works, to ensure public servants understand how the systems work, and how they can deploy them. Read the first article here: An artificial intelligence primer – from machine learning to computer vision.

To learn more about how AI can benefit government organisations please contact Jennifer Robinson, Global Public Sector Strategic Advisor, SAS [email protected] or visit the website sas.com/public-sector.