Quick view: Overview of AI in business (USA)

Introduction

This Quick view is aimed at global in-house counsel, private practice lawyers, and risk and compliance professionals. It can be used in most jurisdictions and has a specific focus on the United States. It will assist in understanding how artificial intelligence (AI) is being used in business and the associated legal implications.

This Quick view covers:

Overview of AI
Business purposes for using AI
Evaluating systems and developing policies for proper use of AI

This Quick view can be used in conjunction with How-to guides: Understanding AI-driven risks, Understanding the risk of negligence claims when using AI and AI and smart contracts and Checklist: Steps to mitigate risks associated with AI use in business.

Section 1 – Overview of AI

AI has become an increasingly popular tool for organizations, with many business applications already developed and more in development. The adoption of this new technology also carries with it new legal risks and liabilities, which are, to some extent, still unclear. However, any issues likely to arise can be identified ahead of time and precautions against them can be put in place.

1.1 Definition

AI is usually defined as computer systems that can perform tasks that would normally require human intelligence. For example, see the Oxford Reference definition. For further information about AI terminology, see Quick view: Key AI terms.

1.2 Types of AI

According to the IBM Data and AI Team, there are three kinds of AI based on capabilities, as follows:

1.2.1 Weak AI/narrow AI

Weak AI, or artificial narrow intelligence (ANI) refers to AI systems that are designed for specific tasks or relatively narrow ranges of tasks, such as language translation or image recognition. Apple’s Siri, Amazon’s Alexa, and IBM Watson are examples of narrow AI. Narrow includes two subset types of AI which are categorized based on their functionalities:

Reactive machines AI

Reactive machines are AI systems that have no memory. They are task-specific, meaning that an input always delivers the same output. Reactive AI is generally reliable and is most useful in applications such as self-driving cars. Reactive machines cannot predict future outcomes unless they have been fed the appropriate information. The IBM Deep Blue, IBM’s chess-playing supercomputer AI that beat chess grandmaster Garry Kasparov in the late 1990s, is an example of reactive machine AI. The Netflix recommendation engine is another example of reactive machine AI.

Limited memory machines AI

A limited memory machine imitates the way the neurons in human brains work together, meaning that the machine gets smarter as it receives more data to train itself on. Limited memory AI is able to look into the past and can monitor specific objects or situations over time. These observations can then be programmed into the AI system so that its actions can be performed based on both past data and on present moment data. This data is not saved into the AI system’s memory as experience to learn from, and so the system cannot derive meaning from its own successes and failures, as a human mind might. The AI system improves over time as it is trained on more data.

1.2.2 General AI

Artificial general intelligence (AGI) is a theoretical form of AI that possesses the ability to understand, learn, and apply knowledge across a wide range of tasks similar to human intelligence.

Theory of the mind AI is a functional classification of AI that falls under the general AI category. This type of AI would understand the thoughts and emotions of other entities; in other words, it would allow the AI to simulate human-like relationships. To date, there are no real-world examples of theory of the mind AI.

1.2.3 Super AI

Artificial superintelligence (ASI) represents AI systems that surpass human intelligence and capabilities across all domains, potentially leading to advanced problem-solving and decision-making abilities beyond human comprehension.

Self-aware AI is a functional classification of AI that falls under Super AI. Self-aware AI would understand its own internal conditions and traits along with human emotions, thoughts, needs, and beliefs. Like theory of the mind AI, this form of AI remains theoretical and there are no real-world examples.

Section 2 – Business purposes for using AI

There are already a number of established business purposes for using AI and its use continues to grow. In fact, AI is currently being utilized across a wide range of business industries, such as healthcare, banking, postal services, manufacturing, retail, hospitality, technology, education, social media, and logistics and supply chain (University of San Diego – 10 examples of Artificial Intelligence in Business).

2.1 Expected gains and efficiencies

2.1.1 Eliminate need for human intervention

AI-augmented robots or machines can easily perform various tasks automatically, without the need for constant human intervention. AI technologies are presently being used to reduce human involvement in assembly, packaging, customer service, and HR, among other areas. The move to AI has reduced operational and employee costs substantially (eg, see the Okuma smart factory).

2.1.2 Instantaneous decision-making and monitoring

Organizations are turning to AI to gain insights into their data in order to make ‘data-driven decisions’ (see Harvard Business School Online, The advantages of data-driven decision-making). As they do so, organizations are finding they do indeed make better, more accurate, and faster decisions based solely on objective data, instead of decisions based on personal instincts or intuition tainted by personal biases and preferences (see Pew Research Center, The Future of Human Agency).

For example, AI systems can be employed to use information gathered by devices installed on factory equipment, to identify problems and predict when maintenance will be necessary. This information can help to prevent disruptive equipment breakdowns and to avoid costly maintenance work being performed because it is scheduled, rather than actually needed.

2.1.3 Transactions processed immediately

AI can be used to monitor financial transactions in real-time. This will also help to detect fraud or suspicious activity. When a user is notified immediately about a fraudulent transaction, they can control the damage before it worsens, therefore protecting both consumers and businesses.

2.1.4 Improved Resource Management

AI systems can optimize resource allocation by making rapid projections of demand and adjusting supply chains accordingly. This leads to reduced waste and more efficient use of materials and energy.

2.1.5 Advanced Risk Assessment

AI can analyze vast amounts of data to identify potential risks in areas like finance, healthcare, and cybersecurity. By providing early warnings, organizations can take preventive measures to mitigate these risks effectively.

2.1.6 Streamlined Operations

AI technologies can automate repetitive tasks and workflows, allowing employees to focus on more strategic activities. This increases productivity and can lead to innovation and growth within the organization.

2.2 Potential applications

There are many different ways in which the efficiencies from AI can be applied in practice. Some of the most common uses are set out below.

2.2.1 Fulfilling and placing orders for the purchase or sale of goods

Sales and purchase order processing is a vital aspect of many businesses. However, it is a time-consuming operation that involves entering information into systems manually, checking for errors, and verifying orders. This process uses up a significant amount of time and resources, and often leads to mistakes and inefficiencies. To address these challenges, businesses are relying on AI to automate their sales and purchase order processing. AI can be used to automate data entry, order verification, and error detection. This saves time and reduces human error. Additionally, AI systems often have a predictive functionality that will assist in optimizing inventory levels based on data regarding previous purchases.

2.2.2 Insurance underwriting

Using AI can make the insurance underwriting process more efficient and more accurate. Predictive analytics – defined as the use of data to make predictions about future events or trends – can be used to develop a better understanding of risk. AI is also a powerful tool that helps to address the persistent problem of bias in underwriting. AI can also allow insurers to gain deeper insight into their customer base than is possible with traditional underwriting methods, by rapidly and accurately analyzing data-driven sets.

See How-to guide: Risks and liabilities of AI algorithmic bias.

2.2.3 Credit scoring

Traditional credit scoring models are unable to account for the complex modern financial ecosystem. Many critics say that traditional credit reporting is error-ridden, biased, and exclusionary.

AI can rapidly classify and organize the financial data utilized in credit scoring, reducing the reliance on manual processes and the risk of human error.

Lenders and credit bureaus can build AI models that uncover financial patterns from historical data. These patterns may then be applied to new data to predict future behavior. Unlike the traditional rule-based decision-making process of credit scoring, AI can continually learn and adapt to new patterns, improving accuracy and efficiency.

Lenders and researchers can also use AI to analyze traditional credit models and identify variables that are having a disproportionate impact on certain groups of borrowers. AI models can then be trained to account for known biases, therefore improving objectivity and fairness in lending decisions.

2.2.4 Market research

AI market research tools use AI technology to develop efficient and effective research solutions, such as automated survey creation, automated data cleaning, and automated report generation. These automated tools separate themselves from traditional research methods, and offer a user-friendly way to speed up research project timelines without sacrificing quality or substance.

2.2.5 Employment screening

AI technologies are increasingly being used to streamline employment candidate assessment and selection and to enhance the efficiency and effectiveness of the hiring process. AI is used for:

resume and application screening: AI-powered algorithms analyze resumes and job applications to identify relevant information, such as skills, qualifications, and experience. This saves time by allowing an employer to automatically shortlist candidates based on specific criteria;
video interviews: AI-powered video interviewing platforms utilize natural language processing and facial recognition to assess candidates’ verbal and non-verbal communication skills, their facial expressions and body language, and overall fit for the role;
personality and behavioral assessments: AI-based assessment tools can analyze candidate responses to psychometric questionnaires and provide insights into their personality traits, work styles, and cultural fit; and
social media screening: employers frequently use a review of social media profiles to gather information about a candidate not available through the usual components of the application process. AI algorithms can be used to scan candidates’ social media profiles and gather information about their interests, behavior, and professional reputation. This helps employers to gauge a candidate’s suitability for a role and to assess the candidate's online presence.

AI offers employers numerous benefits in employment screening, but it is important to ensure fairness, transparency, and, most importantly, compliance with privacy regulations and anti-discrimination laws. Using AI will not necessarily eliminate bias, since bias may be present in the training of the system. For example, Amazon found that its automated recruitment system for new hires was biased against women. The system judged an applicant’s qualifications by comparing qualifications with those of past candidates. Since prior candidates were overwhelmingly male because the pool of candidates had STEM backgrounds, an area in which women have historically been underrepresented, the algorithm assumed that male candidates were preferred, and so rejected applications from women. (See MIT Technology Review, Amazon ditched AI recruitment software because it was biased against women.)

For further information about carrying out social media checks on candidates, see How-to guide: How to investigate the social media activity of prospective employees (USA). For further information about AI and bias, see How-to guide: Risks and liabilities of AI algorithmic bias.

2.2.6 Automate manufacturing processes

The integration of AI into manufacturing processes has led to the increased automation of various tasks. This has resulted in increased productivity, efficiency, and accuracy. AI-powered systems have the capacity to analyze vast amounts of data in real-time as it is generated, to identify patterns, and to make intelligent decisions that optimize production workflows. Machine learning algorithms can predict equipment maintenance needs, reducing downtime and improving overall equipment effectiveness.

Robotic process automation (RPA), combined with AI, allows repetitive and mundane tasks to be automated. This automation frees up humans and allows them to focus on more complex activities. AI also enables manufacturers to implement predictive quality control, identifying potential defects or deviations from standards early on, leading to improved product quality and reduced waste. AI-driven automation has the potential to revolutionize manufacturing, driving cost savings, faster time-to-market, and improved customer satisfaction.

An example of AI automating manufacturing processes is the use of robotic arms equipped with machine learning algorithms that direct them to assemble products in a factory. These robotic arms can be trained to perform intricate tasks, such as picking and placing components, welding, or quality control inspections. These tasks are performed with high precision and efficiency, ultimately reducing human labor and increasing productivity in the manufacturing process.

2.2.7 Knowledge management

Knowledge management is the process of capturing, organizing, and leveraging an organization’s knowledge for better decision-making, innovation, and efficiency. AI plays a crucial role in knowledge management. It allows the automation of the process of knowledge acquisition, storage, and retrieval AI-powered systems have the ability to analyze vast amounts of data virtually instantaneously, to extract relevant information, and to generate insights that support decision-making. Existing knowledge management will become a more powerful and effective tool for an organization

Section 3 – Evaluating systems and developing policies for proper use of AI

3.1 Performance metrics

To evaluate the efficacy of AI systems and processes, performance metrics must be established.

3.1.1 Standards for evaluating systems and processes

The following standards can be helpful to evaluate an AI system:

accuracy metrics, to measure how well a system performs its intended tasks and how regularly it produces correct outputs;
precision and recall metrics, to assess the system’s ability to make a correct identification of relevant information, and its capacity to minimize false positives and negatives;
response time, to evaluate the speed at which the system provides outputs or completes its tasks;
scalability, to measure the system’s ability to handle increased workloads without compromising performance;
robustness, to assess how well the system protects against errors, noise, or changes in input; and
user satisfaction to monitor user experience and gauge the acceptance of the system.

3.1.2 Ethical considerations

When evaluating the performance of AI systems, the ethical implications must be evaluated alongside standard performance metrics. These implications may include:

fairness –whether the system treats individuals fairly, without discrimination;
transparency – is the system explainable, and are its decision-making processes understandable; and
accountability – to measure the system’s responsibility for its actions and the ability to attribute errors or biases.

3.2 Usability

Assessing the performance of AI systems also considers usability metrics. These metrics evaluate the ease of use and the efficiency of a system, and the user satisfaction when interacting with systems. Metrics such as user interface intuitiveness, user engagement, cognitive load, and task completion time give real-life insights into the usability of systems.

Transparency metrics play a crucial role in assessing usability. Transparency refers to the ability to understand and interpret the decision-making processes of AI systems. Transparency is measured by metrics such as explainability, interpretability, and comprehensibility. It is important for users to have a clear understanding of how a system reaches its decisions and the underlying logic behind those decisions.

3.3 Limitations

3.3.1 What is the system designed to do?

Context is important. The performance metrics of an AI system should be assessed alongside the system’s intended purpose and limitations. For example, a system designed for document classification makes metrics like accuracy, precision, and recall relevant. However, it is also important to acknowledge the limitations of an AI system, such as potential biases, error rates, or situations where the system may not perform optimally. By considering the system’s design and limitations, organizations can effectively measure and improve the performance of AI systems in a meaningful manner.

3.3.2 What can it not do?

Performance metrics should account for the system’s boundaries and constraints. Understanding the system’s limitations helps set realistic expectations and avoid potential pitfalls.

For further information about mitigating the risks of using AI, see How-to guide: Understanding AI-driven risks and Checklist: Steps to mitigate risks associated with AI use in business.

Additional resources

US Department of Justice, Civil Rights Division, Algorithms, Artificial Intelligence, and Disability Discrimination in Hiring
US Chamber of Commerce, How AI is Leveling the Marketing Playing Field Between SMBs and Big Business
International Journal of Information Management, Opinion Paper: ‘So what if ChatGPT wrote it?’ Multidisciplinary perspectives on opportunities, challenges and implications of generative conversational AI for research, practice and policy, August 2023

Filed under

Topics

Artificial intelligence