Find out how AI works and what this technology could bring to our lives and the field of homeland security
This is the first article of a multi-part series about AI and how it can empower Home Team operations.
Over the past few years, companies from all kinds of industries have been jumping onto the artificial intelligence (AI) bandwagon with the aim of improving productivity and efficiency. From retail businesses using AI to provide a highly tailored customer experience to factories using this technology to handle tedious and repetitive tasks, the AI movement has been quickly gaining traction, even in the homeland security space. We launched our HTxAI movement on 1 June 2024 with the aim of creating an AI-first HTX to empower an AI-enabled Home Team.
But what exactly is AI? We break it down for you in this explainer.
What is AI?
AI is a way of simulating human intelligence in computers and machines—allowing machines to understand natural language, reason, learn from past experiences, and make predictions.
What are the different types of AI?
AI can be divided into two main categories. The first is Artificial Narrow Intelligence (ANI), also referred to as Weak AI, which is the current state of AI technology. It can perform specific tasks and operate within a limited context and is a catch-all category that includes machine learning, neural networks, and generative AI (GenAI), the last of which has made AI unprecedentedly accessible.
ANIs include reactive and limited memory machines. Reactive machines are the more basic of the two. These AI systems carry out specific tasks but cannot store memory or rely on past experiences to aid their real-time decision-making. In other words, a pre-programmed algorithm guides their decisions.
Limited memory machines, however, can store previous data and predictions and this helps them learn from past experiences and improve their decision-making capabilities.
The other type of AI is Artificial General Intelligence (AGI), or Strong AI, which refers to AI with human-like intelligence and abilities. AGI does not exist at the moment.
How do AIs work?
Deep learning AI models have a ‘brain’ that is made up of neural networks.
AI systems contain many AI models that use algorithms to recognise patterns or trends in data. Think of algorithms as a set of rules or the logic that is applied to datasets to achieve a specific purpose, like summarising text.
But how do you ‘train’ a model? Computer scientists use many techniques such as machine learning (ML) and deep learning (DL), a subset of ML.
In ML, computer scientists create algorithms for the model to follow. The scientists then feed the models a large amount of curated data. While the algorithms are mostly accurate, scientists must still intervene and make corrections if they spot inaccuracies.
Unlike ML models, DL models can learn on their own. They have a ‘brain’, or a series of algorithms called neural networks. Neural networks receive information or raw datasets via an input layer, allowing that information to enter hidden layers of the ‘brain’. These hidden layers function like neurons, processing information and weighing potential outcomes before deciding.
If the decision is incorrect, the ‘brain’ will learn from its past mistakes and augment its decision-making capabilities until it makes the right decision. This intelligence is what helps DL models perform higher-order tasks, like adapting to complex patterns in large data sets.
How can AI help you?
An AI-generated image.
In the workplace, AI can automate many tedious and mundane tasks, like sending out mass emails or transcribing meeting minutes.
AI assistants, like Microsoft Copilot, can also boost your productivity by helping you brainstorm, draft document outlines, and even gather information from various documents on your computer.
GenAI, like ChatGPT and Midjourney, can also help you unleash your creativity. With GenAI, you can turn the concepts in your head into riveting stories or beautiful works of art. They do this through chat interfaces that allow us to input text (and more) that gets processed by large-language models (LLMs) and large multi-modal models (LMMs).
How does AI augment homeland security?
In the realm of homeland security, AI can bolster the threat assessment capabilities of security officers. For example, sophisticated AI analytics can identify hidden contraband in luggage and immediately alert immigration officers to this threat. This helps safeguard our borders and prevent the smuggling of illegal goods.
The predictive capabilities of AI also make them excellent tools for identifying potential threats. For example, AIs trained to analyse vast amounts of cybercrime data excel at detecting cyber threats lurking online. Armed with this information, our police officers can swiftly take down malicious websites.
More impressively, AI can turn the tides against death itself. In disaster scenarios, AI-powered robots can survey the area before human responders arrive, relaying critical information that helps rescue officers tailor their life-saving efforts.
The evolution of AI through the decades
- 1950: Alan Turing developed the Turing Test to test for machine intelligence. In a test, humans were made to distinguish between computer and human-generated text.
- 1955: John McCarthy created the term ‘artificial intelligence’ at the world’s first AI conference.
- 1966: Joseph Weizenbaum created Eliza, a chatbot that convinced users it had emotions. That same year, the Standford Research Institute developed Shakey, a mobile AI robot that is the precursor to modern self-driving cars and drones.
- Late 1970s-Early 1990s: Known as the AI Winter, AI systems were underperforming vis-à-vis researchers’ expectations.
- 1997: IBM's Deep Blue AI beat Garry Kasparov, then world chess champion, in a chess match.
- 2012: Geoffrey Hinton, Ilya Sutskever, and Alex Krizhevsky used a convoluted neural network algorithm to win the ImageNet challenge, a competition for image recognition algorithms. This sparked further research in deep learning.
- 2015: Baidu's Minwa supercomputer identified and categorised images more accurately than the average human due to its convolutional neural networks.
- 2020: OpenAI released the large language model (LLM) GPT-3. This was a major development in AI as it was trained on a whopping 175 billion parameters (learnable components within the model), as opposed to GPT-2’s 1.5 billion parameters.
- 2021: OpenAI released DALL-E, a software that can generate images from textual input.
- 2022: OpenAI released ChatGPT, a chatbot modelled on the GPT-3.5 LLM. ChatGPT’s enhanced natural language processing abilities helped it interact with people more realistically than other chatbots.