A world of intelligent agents
The SBL syllabus considers artificial intelligence, machine learning and robotics. This article looks at how developments in these areas are being utilised and considers the ethical issues associated with artificial intelligence.
Artificial intelligence (AI) is intelligence demonstrated by machines. One definition of artificial intelligence relates computers developing cognitive functions such as learning and problem solving, but other definitions focus on the perceptions that the AI systems gain from the environment and the actions they take to fulfil their objectives. AI has also been defined in terms of systems actively making decisions rather than just responding mechanically, showing intelligence by use of data analysis and recognition of what is relevant to solve problems, and having the ability to learn and adapt as they process information and make decisions. Practical definitions of AI focus on the tasks that can be undertaken using AI.
Algorithms are a vital element of AI. Algorithms can be defined as processes or sequences of steps to be used in problem-solving operations or for accomplishing tasks.
The scope of AI is being extended as technology advances. However, at the same time, simpler abilities, such as optical character recognition, are no longer strictly considered to be AI, as these abilities are now generally inherent in computers.
Uses of AI
There are many ways in which AI is being used in retail, most often to enhance the customer experience:
- Creation of environment – AI helps retail businesses to learn from each interaction with customers and improve the digital and physical environment they offer, adapting in-store displays in line with customer profiles and evolving the digital displays offered to customers.
- Changed relationship with customer – AI enables a continuously-developing relationship with customers to be established, so that what is offered to customers on their home page reflects developments in their shopping behaviour. AI can also help facilitate two-way conversation with customers, answering questions and using the information to build customer profiles. The data available about customers may not just take the form of purchase and query records, but also facial or audio responses that reflect what customers are thinking.
- Guiding purchasing decisions – Automated assistants can build on the information stored about customers to help them make appropriate purchasing decisions.
- Systems development – AI can build on customer information, using it along with competitor and market information, to develop supply chain planning, and pricing and promotional strategies.
- Product development – the data obtained about customers, not just purchasing but also feedback and sentiments expressed, can be used to develop future product and service designs that fulfil current customer requirements and anticipate future customer needs.
Many other sectors are making increasing use of AI:
- Medicine and healthcare – AI can analyse huge quantities of patient data in order to help prevent illness, by identifying people who are particularly at risk and prompting intervention to reduce the risk of disease developing. Medical diagnosis AI technology can use information about a person’s history and genetic make-up to personalise medical treatment.
- Transportation – AI-based features are now part of many cars, including automatic parking systems. More advanced AI is used in self-driving cars. AI can also assist in route planning, avoiding areas of heavy traffic or poor road conditions.
- Agriculture – algorithms designed for crop and soil monitoring can be used to track the health of crops. AI can also consider climate and other environmental conditions and, as a result, predict when crops will ripen, assisting planning.
- Cybersecurity – AI can help defend against hacking by analysing data to detect anomalies and allow focus on areas of greatest threat. It can also help by sorting data into high and low risk information.
- Human resources – AI can be used to distinguish candidates by how qualified they are for a particular job. Job matching platforms provide prediction for hirers of how successful candidates will be in particular roles, whilst allowing candidates to create profiles that facilitate their being matched with the right opportunities.
- Financial investment and trading – AI is used by large financial institutions to support their investment practices. AI systems use algorithms in high-speed trading, making decisions many times quicker than humans can. Some investment portfolios are managed purely by AI.
Machine learning involves using algorithms to gain experience through using data (known as training data) and harnessing this knowledge to establish relationships or learn how to do particular tasks without being programmed to do so. At a more advanced level, computer systems may be more effective in developing the algorithms needed to complete the tasks than humans would be.
The relationship between artificial intelligence and machine learning is not straightforward. Some see machine learning as wholly a part of AI, others believe that only some forms of more advanced machine learning can be called AI.
A key output in machine learning is prediction, where an algorithm has gained experience on a historical data-set and used the experience to make a prediction, based on new data, which is a prompt for action. Machine learning can also provide guidance on actions to take to achieve objectives.
One classification of machine learning is into three types:
- Supervised learning – this is based on training data and human feedback. It involves establishing the relationships between specified inputs and outputs. Regression analysis is a form of supervised learning. Supervised learning has been used in the fields of bioinformatics (for mapping biological processes) and developing speech recognition by computers. It also has applications in personalised marketing of products or services, based on information held in databases.
- Unsupervised learning – this involves establishing patterns between, or structures in, the specified inputs and therefore being able to classify the input data. Again there are marketing applications in being able to link customers who have similar buying histories with attributes that they have in common, for example age.
- Reinforcement learning – this is a learning process involving the maximisation of rewards for the actions taken. It can be most useful in situations where the input training data is limited or the output (target) cannot be clearly specified. This means that, in order to learn, the system has to gain experience through interaction with the environment. If it turns out that the system has not maximised its rewards, it learns from this. It is therefore possible that actions may need to be taken that are sub-optimal in the short-term, in order to gain the knowledge required to achieve longer-term maximisation of rewards. Reinforcement learning has been important in areas including investment portfolio management and robot control, as well as increasing the standard of play of machines programmed to play strategy games such as chess and backgammon.
There are other ways in which machine learning is classified. Self-learning is where there are no rewards for actions taken and no human guidance on what is optimal. The self-learning algorithm computes decisions about what to do and also takes into account ‘emotions’ (which it is programmed with initially) that respond to the consequences of the actions that it decides. As a result, it develops behaviour that will best achieve its objectives whilst also fulfilling its ‘emotional needs’.
Deep learning is an increasingly significant field of machine learning. It is based on artificial neural networks (networks based on the biological networks of animal brains). Deep learning takes a non-linear approach to analysing unstructured data. Applications of deep learning have included consumer recommendation apps, drug design and medical image analysis. Deep learning has also been used in developing language translation applications.
Robotics is concerned with the development of machines (robots) that reproduce human actions and can be used instead of humans. All robots have certain basic features in common – mechanical construction, electrical components and computer programming code to guide their actions. One distinction is between robots that require human intervention to operate and robots that can operate autonomously. Humanoid-like robots are often seen in science fiction films, but robots can take other forms as well.
Uses of robotics
Broadly, robots are most useful where accuracy and repeatability are particularly important, whereas humans are best employed in jobs involving creativity and flexibility. Robots are often used in jobs which are undesirable for humans to do, for example because the jobs are dangerous (such as radioactive testing or bomb disposal). They are also used in environments where humans cannot function, for example underwater or in very high heat.
Reduced operating cost has been an important reason why robots are being used in manufacturing for repetitive and monotonous tasks, though their ability to perform tasks precisely is also important. Robots are also seen to be more reliable than humans and do not tire.
Welding is a good example of a task well-suited to robots in manufacturing. Robot welders can achieve higher productivity, lower usage of materials and higher yield (by avoiding mistakes) than human welders. Advances in robotics mean that designs can be used that require welds that cannot be made by a human welder. Welding also produces safety risks to humans such as flashing, splatter and exposure to fumes that do not affect robots.
The machine vision that robots have has also helped in developing the use of robots for quality control. Robots use AI to detect whether components are not fulfilling specifications or have other defects. These can then be removed from the production process early on, avoiding delays later or the components being part of finished goods.
Robots have also been used in the medical field. In operations, surgeons can control the robotic arms that use surgical instruments on patients. This can be done by using telemanipulators to control the arms directly, or by computer control, which can mean that the surgeon does not have to be present for surgery. The advantages of robotic surgery are precision and smaller incisions, resulting in less blood loss and pain than traditional methods.
Current areas of development of robotics include home robotics, for example robotic vacuum cleaners, to help with everyday tasks. How robots process and respond to their environment, which is an important area generally for robotics, is particularly significant in the home environment. There are also developments in robot learning, where algorithms work through self-guided exploration and interaction with humans, with the objective that the robots will acquire new skills.
Concerns about the use of robots include initial investment, the training costs needed for human employees to work successfully with robots, and health and safety concerns relating to humans interacting with large, powerful, robots.
The World Economic Forum has addressed a number of ethical issues relating to AI.
Some of the issues addressed also relate to risk and control systems:
- Mistakes – whilst computers may not make many sorts of mistakes that humans would make, equally they may make mistakes of their own, for example seeing patterns in randomly generated items.
- Bias – AI is created by humans and may therefore reflect the biases of its creators, or inadvertent bias may be programmed into the criteria used to make decision.
- Security – the increasing abilities of AI place a greater obligation on those who use it to have effective cybersecurity to protect it from being used for illicit purposes.
- Unintended adverse consequences – an AI system may produce the solution that it was intended to produce, but the solution may have adverse consequences which the system ignores because it was not programmed to think about them.
- Ultimate control – this is the longer-term issues of whether computers will one day be so intelligent that they will be able to counter any attempts to control them and/or they will start making attempts to control humanity.
Another aspect of the debate about AI ethics is the economic and social consequences of using AI:
- Unemployment – replacing of humans by AI or robots may result in unemployment and little prospect of re-employment, certainly for a similar sort of job, if AI’s use is widespread. This raises wider questions of those made redundant having to find different roles and society needing to be geared to support them.
- Distribution of income – the question of dealing with unemployment is related to how the benefits of AI should be distributed. There is the possibility that rewards will be concentrated in the hands of the owners of businesses that use AI effectively, businesses perhaps where there are few salary-earning workers.
Other, wider, ethical issues have also been raised:
- Influencing behaviour – interaction between machines and humans carries the possibility of human behaviour being influenced. AI can be created that focuses on building relationships with humans. Video games, for example, contain algorithms designed to capture the attention, but the consequence of this may be addiction.
- Humane treatment of AI – the reward functions of AI are becoming increasingly sophisticated, so the issue has been raised of whether giving negative rewards to a system can be said to cause it suffering and whether this is an ethical concern.
The World Economic Framework suggests that businesses using AI need to give consideration to developing an ethical framework for its use. This may include:
- Definition of AI ethics – this should be consistent with the organisation’s general ethical framework in areas such as respect for the law and commitment to society, but with a focus on AI areas such as data protection principles.
- Building AI ethics into product development – this involves consideration of areas such as limiting the use of data, protecting the data that is used and adherence to privacy.
- Obtaining customer/stakeholder feedback – this is particularly important during the development process.
- Continuing awareness of bias – as systems develop, being alert for signs and taking effective measures to deal with any bias that has developed.
- Transparency – this area of ethical policy is particularly important to reassure customers, with explanations being given of what data is being used, how it’s being used, and for what purpose.
This article has given a general introduction into how AI and other developments are being used and discussed the ethical issues that businesses may need to consider. Although these subjects are in the syllabus, in the SBL exam exhibits will clearly explain the business context in which AI could be, or is being, used. Answers will (as for other areas for the syllabus) need to focus on what is important for the organisation to score well.
Written by a member of the Strategic Business Leader examining team