Understanding the issues using Artificial Intelligence (AI) in OHS

Artificial intelligence is an innovative and transformative technology that has been making waves across various industries & today is the big buzzword in business. But very few of us understand what AI is and means, including OHS practitioners who are starting to use AI in the OHS field.

Firstly, the Occupation Health & Safety field is a technical subject field that requires the application of the science (facts and knowledge), art (skills and ability) and management (processes and systems) by a competent OHS practitioner to affectively impact on the health and safety of people in a workplace.

Secondly, OHS practitioners are now starting to apply AI-based outcomes in the workplace, without understanding the how and where these requested AI outcomes are generated. Consequently, there could be OHS risks involved by not understanding AI functional operations and technology.

Artificial intelligence (AI) is not a new technology as it came into existence in the 1950s when the term “artificial intelligence” was first coined and has grown through several phases to achieve today’s status.

Phase I AI: The first phase of AI research was marked by the development of “ES” (expert systems) in the 1960s. These computer programmes use a set of rules to make decisions and solve problems in specific domains, such as medical diagnosis or financial analysis.

However, these systems had limited capabilities and could not learn from new data.

Phase II AI: The second phase of AI research, which started in the 1980s, focused on machine learning. Researchers developed algorithms to learn from data and improve their performance over time. One of the most important developments during this period was the introduction of neural networks, which were inspired by the structure and function of the human brain.

Phase III AI: The third phase of AI research, which started in the 1990s, was characterized by the development of more sophisticated machine learning techniques, such as support vector machines and decision trees. During this period, AI was used in real-world applications, such as speech recognition, computer vision, and natural language processing.

Phase IV AI: The fourth phase of AI research, which started in the 2010s, is marked by the rise of deep learning, which uses neural networks with many layers to learn complex data representations. Deep learning has led to significant advances in image and speech recognition, natural language processing, and robotics.

The present scenario: Today, AI is used in many applications, from self-driving cars and personalised healthcare to virtual assistants and smart homes. While early AI systems were limited in capabilities, machine learning and deep learning development have led to significant progress in recent years.

There are two main types of AI currently:
1. Narrow AI (ANI), Focused Intelligence for Specific Tasks:

Also known as weak AI, narrow AI is designed to carry out specific instructions-based tasks. These systems are trained to recognise patterns and make decisions based on the data provided to them. 

Some common examples of narrow AI in everyday life include:
i. Virtual assistants such as Siri and Alexa respond to voice commands and perform tasks like setting reminders, making calls, or playing music. (i.e. task-specific expertise)
ii. Image recognition software used in social media apps like Facebook and Instagram automatically tags people in photos. (i.e. task-specific expertise for image recognition and Natural Language Processing (NLP)
iii. Email providers use spam filters to filter out unwanted messages automatically.
iv. Navigation systems like Google Maps provide real-time traffic updates and route suggestions. (i.e. task-specific expertise real-world applications)
v. Navigation systems like Google Maps provide real-time traffic updates and route suggestions. (i.e. task-specific expertise real-world applications)

ANI is what is being used by OHS practitioners currently, but we must be aware that the data sets information used by the machine logic is frequently using overseas OHS information (e.g. the USA’s OSHA information), which may not be in accordance with or comply with our local legal requirements. So, OHS practitioners must assess and review the AI generated information they may use to ensure it is suitably applicable for local application scenarios and conditions.

2. General AI (AGI), The Quest for Human-Level Intelligence
Also known as strong AI, general AI is designed to perform any intellectual task that a human can do. AGI aims to replicate human-like intelligence and comprehension. Unlike Narrow AI, General AI possesses the ability to understand, learn, and apply knowledge across diverse domains. Achieving General AI remains an ongoing endeavour, but its implications are profound. These systems are capable of learning from experience and adapting to new situations. 

While general AI is still largely a theoretical concept, some potential examples of general AI in everyday life include:
i. Adaptive learning capabilities, enabling it to accumulate knowledge from various sources, reason logically, and apply that knowledge in novel situations, like:
Personalised medical diagnosis and treatment recommendations based on a patient’s medical history and genetic profile.
Automated customer service chatbots that can understand and respond to complex inquiries
Intelligent personal shopping assistants can recommend products based on a person’s preferences, budget, and style
Autonomous vehicles that can safely navigate complex environments and make real-time decisions based on changing road conditions

ii. Ethical Considerations: As General AI becomes more sophisticated, ethical considerations gain significance. Questions regarding AI’s decision-making processes, accountability, and potential consequences must be addressed to ensure responsible development and deployment. Recently, ethical issues are being raised in legal court cases over the unethical use of copyrighted material being used in the machine learning training data sets.

iii. Other AI types: Apart from the main narrow and general AI types, several other types of AI are used in different applications, like supervised learning, unsupervised learning, deep learning, reinforcement learning, transfer learning & cognitive learning, for example.

Describing What is Artificial Intelligence?
Artificial intelligence is generated from developed computer systems that are capable of performing tasks that are able to learn, make decisions, and take actions, such as decision-making, visual perception, and speech recognition – even when it encounters a situation it has never come across before – provided that a related and suitable machine learning training data set is available for the machine (computer) learning analysis.

At its core, Artificial Intelligence refers to the simulation of human intelligence in machines. It encompasses the development of intelligent systems capable of perceiving, reasoning, learning, and problem-solving, mimicking human cognitive abilities. It involves the development of complex algorithms and advanced computer power, through machine learning systems that can process and analyse vast amounts of data, recognize patterns, and learn from experience to make predictions or decisions, in relation to:

• High energy consumption and scalability issues
• Regulatory and legal challenges
• Security risks such as 51% attacks and smart contract vulnerabilities
• Lack of standardization and interoperability between different Blockchains
• Limited public awareness and adoption in some industries and regions.

How does AI work?
Currently, most AI relies on a process called Machine Learning to develop the complex algorithms that constitute their ability to act intelligently.
There are other areas of AI research – like robotics, computer vision (via images, videos, sensors), and Natural Language Processing (NLP) (i.e. AI developed complex algorithms understanding and applying the rules and syntax of language) – that also play a major role in many practical implementations of AI, but the underlying training and development still starts with Machine Learning and the sets of training data information available for it to process.

With machine learning, a computer programme is provided with a large training data set – the bigger, the better. (The question is, are there positive and negative data information risks included within such training data sets? This is important if AI is being used in the OHS field).

The OHS practitioner needs to understand the functional operation of AI that generates information outcomes related to the training learning data sets information it analysed (usually USA’s OSHA data set information) so the AI material generated will not include the specific legal requirement information, say from the OHS Act’s 2014 Construction Regulation if the AI request was construction works related. The OHS practitioner must review and ensure that all the AI generated material is local legislation compliant before implementing any AI generated material locally.

References
https://www.blockchain-council.org/ai/what-is-artificial-intelligence/
https://www.blockchain-council.org/ai/the-ultimate-guide-to-artificial-intelligence/ 

This series, by SHE & Risk Management Consultant Leighton Bennett of Benrisk Consulting, is written with occupational health and safety officers in mind. He can be contacted at +27 (0)83 325 4182, benrisk@mweb.co.za.