AI Ethics: What Is It and How to Embed Trust in AI?
The next step of artificial intelligence (AI) development is machine and human interaction. The recent launch of OpenAI's ChatGPT, a large language model capable of dialogue of unprecedented accuracy, shows how fast AI is moving forward. The ability to take human input and permissions and adjust its actions based on them is becoming an integral part of AI technology. This is where the concept of ethics in artificial intelligence research begins, and this is the area I am focusing on for the rest of this article.
Previously, humans were solely responsible for educating computer algorithms. Instead of this process, we may soon see AI systems making these judgments instead of human beings. In the future, machines might be fully equipped with their own judgement system. At this point, things could turn for the worse if the system miscalculates or is flawed with any bias.
The world is currently experiencing a revolution in the field of artificial intelligence (AI). In fact, all Big Tech companies are working hard on launching the next step in AI. Companies such as Google, Open AI (Microsoft), Meta and Amazon have already started using AI for their own products. Quite often, these tools cause problems, damaging company reputations or worse. As a business leader or executive, you must also incorporate AI in your processes and ensure your data scientist or engineers team develops unbiased and transparent AI.
A fair algorithm does not bias against any single group. If your dataset does not have enough samples for a particular group, then the algorithm will be biased for such a group. On the other hand, transparency is about ensuring that people can actually understand how an algorithm has used the data and how it came to a conclusion.
AI Ethics: What Does It Mean, and How Can We Build Trust in AI?
There is no denying the power of artificial intelligence. It can help us find cures for diseases and predict natural disasters. But when it comes to ethics, AI has a major flaw: it is not inherently ethical.
Artificial intelligence has become a hot topic in recent years. The technology is used to solve problems in cybersecurity, robotics, customer service, healthcare, and many others. As AI becomes more prevalent in our daily lives, we must build trust in technology and understand its impact on society.
So, what exactly is AI ethics, and most importantly, how can we create a culture of trust in artificial intelligence?
AI ethics is the area where you look at the ethical, moral, and social implications of artificial intelligence (AI), including the consequences of implementing an algorithm. AI ethics are also known as machine ethics, computational ethics, or computational morality. It was part of my PhD research, and ever since I went down the rabbit hole of ethical AI, it has been an area of interest to me.
The term "artificial intelligence ethics" has been in use since the early days of AI research. It refers to the question of how an intelligent system should behave and what rights it should have. The phrase was coined by computer scientist Dr. Arthur Samuel in 1959 when he defined it as "a science which deals with making computers do things that would require intelligence if done by men."
Artificial intelligence ethics is a topic that has gained traction in the media recently. You hear about it every day, whether it is a story about self-driving cars or robots taking over our jobs or about the next generative AI spewing out misinformation. One of the biggest challenges facing us today is building trust in this technology and ensuring we can use AI ethically and responsibly. The notion of trust is important because it affects how people behave toward each other and towards technology. If you do not trust an AI system, you will not use it effectively or rely on its decisions.
The topic of trust in AI is broad, with many layers to it. One way to think about trust is whether an AI system will make decisions that benefit people or not. Another way is whether the system can be trusted to be fair when making these decisions.
In short, the main ethical consideration at this point is how we can build trust in artificial intelligence systems so that people feel safe using them. There are also questions about how humans should interact with machines as well as what types of capabilities should be given to robots or other forms of AI.
In the past few years, we have seen some of the most significant advances in AI — from self-driving cars and drones to voice assistants like Siri and Alexa. But as these technologies become more prevalent in our daily lives, there are also growing concerns about how they could impact society and human rights.
With that said, AI has also brought us many problems that need to be addressed urgently, such as:
- The issue of trust. How can we ensure that these systems are safe and reliable?
- The issue of fairness. How can we ensure that they treat everyone equally?
- The issue of transparency. How can we understand what these systems do?
Strategies for Building Trust in AI
Building trust in AI is a challenging task. This technology is still relatively new in the mainstream, and many misconceptions exist about what it can and cannot do. There are also concerns about how it will be used, especially by companies with little or no accountability to their customers or the public.
As we work to improve understanding and awareness of AI, it is not too late to start building trust in AI. Here are some strategies that can help us achieve this:
1. Be transparent about what you are doing with data and why
When people do not understand how something works, they worry about what might happen if they use it. For example, when people hear that an algorithm did something unexpected or unfair, they might assume (wrongly) that humans made those decisions. A good strategy for building trust is to explain how algorithms work so that people understand their limitations and potential biases — and know where they should be applied. Make sure you have policies governing how your team uses data to create ethical products that protect privacy while also providing value to users. In addition, be transparent to your customers and inform them when decisions are made by algorithms and when by humans.
2. Provide clear explanations for decisions made by AI systems
AI systems are making important decisions about people's lives. These decisions can greatly impact how people live, from the applications they can access to the treatment they receive. So it is important that AI systems give people explanations for their decisions.
AI systems have become more accurate and useful over time, but they still make mistakes. In some cases, these mistakes may be due to bias in the data used to train them. For example, an image recognition algorithm might incorrectly identify a photo of a black person as an ape because it was trained on photos of apes rather than photos of black people.
In other cases, it might be due to limitations in the algorithm itself or possible bugs in its implementation. In both cases, the best way to fix these errors is by providing clear explanations for why they made certain decisions, which humans can then evaluate, and the AI can be corrected if need be.
3. Make it easy for people to opt out of data collection and use
Data collection is a big part of the digital economy. It is how companies can offer personalised experiences and improve their services. But as we have learned from the Facebook Cambridge Analytica scandal, collecting data is not always safe or ethical.
If you are collecting data on your website, there are some important steps you can take to make sure you are doing it the right way:
- You should have an easy way for users to opt out of any data collection or use. This will include a link or button that they can click on to do so. It is important that this option is prominent — not buried in a maze of other options. It should just be one click away when they visit your site or app and easy enough for anyone who visits your site to find and use it without having to go hunting around for it first.
- Give people control over their data. When someone chooses to opt-out of data collection, do not automatically delete all their records from your database — instead, delete the ones that are not needed anymore (for example, if they have not logged in for six months). And give them access to their own personal data so they can understand what information about them has been collected and stored by your system.
4. Encourage people to engage with your company
People can be afraid of things that are unknown or unfamiliar. Even if the technology is designed to help them, using it may still be scary.
You can build trust in your AI by encouraging people to engage with it and interact with it. You can also help them understand how it works by using simple language and providing a human face for the people behind the technology.
People want to trust businesses, especially when they are investing money and time in them. By encouraging people to engage with your company's AI, they will feel more comfortable with their experience and become more loyal customers.
The key is engagement. People who can see and interact with an AI solution are more likely to trust it. And the more people engage with the AI, the better it gets because it learns from real-world situations.
People should be able to see how AI works and how it benefits them. This means more transparency — especially around privacy — and more opportunities for people to provide input on what they want from their AI solutions.
Why Does Society Need a Framework for Ethical AI?
The answer to this question is simple: Ethical AI is essential for our survival. We live in a world that is increasingly dominated by technology, which affects every aspect of our lives.
As we become more dependent on technology, we also become more vulnerable to its risks and side effects. If we do not find ways to mitigate these risks, we may be facing a crisis where machines as the dominant species on this planet replace human beings.
This crisis has already begun in some ways. Many people have lost their jobs due to automation or the computerisation of tasks that humans previously performed. While it is true that new employment opportunities are being created as well, this transition period can be difficult for both individuals and society at large.
Extensive research by leading scientists and engineers has shown that it is possible to create an artificial intelligence system that can learn and adapt to different types of problems. Such "intelligent" systems have become increasingly common in our lives: they drive our cars, deliver packages and provide medical advice. Their ability to adapt means they can solve complex problems better than humans – but only if we give them enough data about the world around us, which should involve teaching machines how we think about morality.
A fair algorithm does not bias against any single group. If your dataset does not have enough samples for a particular group, then the algorithm will be biased for such a group.
The algorithm can be tested to measure its impartiality level by comparing your algorithm's results with those of a non-biased algorithm on the same dataset. If the two algorithms give different results for any given sample, then there is a bias in your model that needs to be fixed. Then it will produce more accurate predictions for those groups than those that do not have enough data to train against (such as women or people of colour).
Recently, Meta launched an artificial intelligence model called Galactica. It says it was trained on a dataset containing over 100 billion words of text to summarise a large amount of content easily. This included books, papers, textbooks, scientific websites, and other reference materials. Most language models that model the characteristics of a given language are trained using text found on the internet. According to the company, the difference with Galactica is that it also used text from scientific papers uploaded to the website PapersWithCode, a Meta-affiliated website.
The designers emphasised their efforts on specialised scientific information, like citations, equations, and chemical structures. They also included detailed working-out steps for solving problems in the sciences, meaning a revolution for the academic world. However, within hours of its launch, Twitter users posted fake and racist results generated by the new Meta bot.
One user discovered that Galactica made up information about a Stanford University researcher's software that could determine someone's sexual orientation by analysing his or her Facebook profile. Another was able to get the bot to make up a fake study about the benefits of eating crushed glass.
For this and many other reasons, the company took it down two days after launching the Galactica demo.
The Accuracy of the Algorithms
The most common way to test whether an algorithm is fair or not is by using what is called "lack-of-fit testing." The idea behind lack-of-fit testing is that if there were no biases in an existing data set (meaning all the records within one specific category were treated equally and the dataset has been analysed for biases that were accounted for). A well-organised database is like a puzzle: the pieces should fit together neatly and show no gaps or overlaps.
In the earlier example, both men and women were assigned gender roles based on their birth sex rather than their actual preferences. If every role had been filled before moving on to something else, we would not see gaps in between categories—but instead, what we see here is something that does not add up one way or another.
They should also be able to explain how you can change its behaviour if necessary. For example: "If you click here, we will update this part of our algorithm."
As we have seen so far, the potential of artificial intelligence (AI) is immense: it can be used to improve healthcare, help businesses and governments make better decisions, and enable new products and services. But AI has also raised concerns about its potential to cause harm and create societal bias.
To address these issues, a shared ethical framework for AI will help us design better technology that benefits people rather than harms them.
For example, we could use AI to help doctors make more accurate diagnoses by sifting through medical data and identifying patterns in their patients' symptoms. Doctors already rely on algorithms for this purpose - but there are concerns that these algorithms can be biased against particular groups of people because they were only trained on data from those groups.
A Framework for Ethical AI
A framework for ethical AI could help us identify these biases and ensure that our programs are not discriminating against certain groups or causing harm in other ways.
Brown University is one of several institutions that have created ethical AI programs and initiatives. Sydney Skybetter, a senior lecturer in theatre arts and performance studies at Brown University, is leading an innovative new course, Choreorobotics 0101—an interdisciplinary program that merges choreography with robotics.
The course allows dancers, engineers, and computer scientists to work together on an unusual project: choreographing dance routines for robots. The goal of the course is to give these students — most of whom will go on to careers in the tech industry — the opportunity to engage in discussions about the purpose of robotics and AI technology and how they can be used to "minimise harm and make a positive impact on society."
Brown University is also home to the Humanity Centered Robotics Initiative (HCRI), a group of faculty members, students, staff, and faculty who are advancing robot technology to address societal problems. Its projects include creating "moral norms" for AI systems to learn to act safely and beneficially within human communities.
Emory University in Atlanta has done a lot of research to apply ethics to artificial intelligence. In early 2022, Emory launched an initiative that was groundbreaking at the time and is still considered one of the most rigorous efforts in its field.
The Humanity Initiative is a campus-wide project that seeks to create a community of people interested in applying this technology beyond the field of science.
I think exploring the ethical boundaries of AI is essential, and I am glad to see universities weighing in on this topic. We must consider AI's ramifications now rather than waiting until it is too late to do anything about it. Hopefully, these university initiatives will foster a healthy dialogue about the issue.
The Role of Explainable AI
Explainable artificial intelligence (XAI) is a relatively new term that refers to the ability of machines to explain how they make decisions. This is important in a world where we increasingly rely on AI systems to make decisions in areas as diverse as law enforcement, finance, and healthcare.
In the past, many AI systems have been designed so that they cannot be interrogated or understood, which means there is no way for humans to know exactly why they made a particular decision or judgement. As a result, many people feel uncomfortable with allowing such machines to make important decisions on their behalf. XAI aims to address this by making AI systems more transparent so that users can understand how they work and what influences their thinking process.
Why Does Explainable AI Need to Happen?
Artificial intelligence research is often associated with a machine that can think. But what if we want to interrogate or understand the thinking process of AI systems?
The issue is that AI systems can become so complex due to all the layers of neural networks — which are algorithms inspired by the way neurons work — that they cannot be interrogated or understood. You cannot ask a neural network what it is doing and expect an answer.
A neural network is a set of nodes that are connected together by edges with weights associated with them. These nodes represent neurons in your brain, which fire off electrical signals when certain conditions are met. The edges represent synapses between neurons in your brain. Each synapse has a weight that determines how much of an effect firing one neuron has on another. These weights are updated over time as we learn more about the world around us and change our behaviour accordingly (i.e., when we get rewarded for doing something right).
As you can see, neural networks are made up of many different layers, each of which does something different. In some cases, the final result is a classification (the computer identifies an object as a dog or not), but often the output is just another layer of data to be processed by another neural network. The result can be hard to interpret because multiple layers of decisions may exist before you get to the final decision.
Neural networks can also produce results in ways that are difficult to understand because they do not always follow the rules or patterns we would expect from humans. We might expect one input number to produce one output number, but it turns out this is not always true for neural networks either because they can be trained on lots of examples where this is not true and then use those examples as training data when making new predictions in the future.
In short, we are creating machines that learn independently, but we do not know why they make certain decisions or what they are thinking about.
AI systems have been used in many different domains, such as health care, finance, and transport. For example, an autonomous vehicle might need to decide between two possible routes on its way home from work: one through traffic lights and another through an empty parking lot. It would be impossible for an engineer to guess how such a system would choose its route — even if he knew all the rules that govern its behaviour — because it could depend on thousands of factors such as road markings, traffic signs, and weather conditions.
The ethical dilemma arises because AI systems cannot be trusted unless they are explainable. For instance, if an AI can detect skin cancer for medical purposes, it is important that the patient knows how the system arrived at its conclusion. Similarly, if an AI is used to determine whether someone should be granted a loan, the lender needs to understand how the system came up with that decision.
But explainable AI is more than just transparency; it is also about accountability and responsibility. If there are errors in an AI's decision-making process, you need to know what went wrong so you can fix it. And suppose you are using an AI for decisions that could have serious consequences, such as granting a loan or approving medical treatment. In that case, you need to know how confident you can be in its output before making it operational.
Other Ethical Challenges
In addition, this AI revolution has also led to new ethical challenges.
How can we ensure that AI technologies are developed responsibly? How should we ensure that privacy and human rights are protected? And how do we ensure that AI systems treat everyone equally?
Again, the answer lies in developing an ethical framework for AI. This framework would establish a common set of principles and best practices for the design, development, deployment, and regulation of AI systems. Such a framework could help us navigate complex moral dilemmas such as autonomous weapons (AKA killer robots), which can identify targets without human intervention and decide how or whether to use lethal force. It could also help us address issues such as bias in algorithms, which can lead them to discriminate against certain groups, such as minorities or women.
Consider the example of an autonomous vehicle that can decide whether or not to hit pedestrians. If the car hits a pedestrian, it will save its passengers at the cost of killing one person. If the car does not hit a pedestrian, it will protect itself but end up killing two people instead.
In this scenario, human morality would tell us that we should choose the option that results in saving two people over one person (i.e., not hitting pedestrians, which is what we want from our autonomous cars). However, if we ask an AI system to solve this problem without telling it any other information about morality or ethics, it might choose to kill two people instead of one.
This is known as a trolley problem — when moral dilemmas are presented in actions rather than outcomes — and it illustrates how difficult it can be for AI systems to make ethical decisions on their own without some framework for guidance.
How to Start Developing a Framework for Ethical AI Use by Businesses and Leaders?
AI is a tool that can be used to solve problems, but it has its limitations. For example, it cannot solve problems that require judgement, values, or empathy.
AI systems are designed by humans and built on data from their past actions. These systems make decisions based on historical data and learn from their experiences with those data sets. This means that AI systems are limited by the biases of their creators and users.
Human bias can be hard to detect when we do not know how our own brains work or how they make decisions. We may not even realise that we have prejudices until someone points them out to us — and then we still might not be able to change them quickly enough or completely enough to avoid discrimination in our own behavior.
As a result of these biases, many people fear that AI will add new types of bias into a society that would otherwise not exist if humans were making all the decisions themselves — especially if those decisions are made by machines programmed by humans who have their own biases baked in at an early stage of development.
A survey conducted by Pew Research in 2020 found that 42% of people worldwide are concerned about AI's impact on jobs and society. A great way to tackle this concern could be to consider hiring an ethics officer in different fields in the near future.
There is no doubt that artificial intelligence will play a bigger role in the business world in the coming years. For these reasons, leaders from all fields need to develop an ethical framework for AI that goes beyond simply putting an AI system into place and hoping for the best.
Businesses need to develop a framework for AI ethics, but it is not easy. There are many considerations, including what is acceptable and what is not.
Here are several steps you can take to begin developing a framework for your organisation's AI ethics:
Define what you mean by "ethical AI"
AI is a broad term that covers many different technologies and applications. For example, some "AI" is simply software that uses machine learning algorithms to make predictions or perform specific tasks. Other "AI" may include robots or other physical devices interacting with humans. It's important for business leaders to clearly define what they mean by "ethical AI" before they start developing their ethical framework.
Clarify your values and principles
Values are general principles about what's essential for an organisation, while principles serve as guidelines for acting according to those values. For example, a value might be "innovation," while a principle might be "do not use innovation as an excuse not to listen to your customers." Values drive ethical decision-making because they provide direction on what's most important in a situation (for example, innovation vs. customer needs). Principles help guide ethical decisions because they outline how values should be translated into action (for example, innovate responsibly).
Understand how people use AI technology today
One way is by observing how people use technology daily — what they buy, what they watch, what they search for online, etc. This can give you insights into how organisations use technology and where there's demand for new products or services that rely on AI. It can also help identify potential downsides of using AI too much — for example, if employees are spending too much time using their devices at work instead of working as efficiently as possible or if customers feel stressed out because they spend too much time looking at their phones while they are with friends or family members.
Know what people want from AI tech
Understanding who your customers are and what they expect from you is important before integrating any new technology into your business strategy. For example, if your customers are older adults who do not trust technology, then developing an ethical framework for AI will be different than if your customers are younger adults who embrace new technologies quickly. You also need to know what they want from AI tech — do they want it to improve their lives or make them more efficient?
Knowing this information will help you set realistic goals for the ethical framework you develop.
Set clear rules for your organisation about how you want people to use AI tech
This can be as simple as creating a checklist of best practices for using AI technology that employees could refer to when making decisions about applying it in their jobs. For example, suppose someone at your company is considering using an application that uses facial recognition technology. In that case, there might be specific parameters regarding how it should be used, such as whether employees can use it in public places without first asking permission from passersby.
Create a list of questions that will help you assess whether or not using certain applications is ethical or not. For example, if someone wants to use facial recognition software to track attendance at meetings, they might ask themselves if this would violate anyone's privacy rights or if it would cause any harm.
Work with your employees and stakeholders to improve the framework
A great first step is gathering data and feedback from your employees and stakeholders about how they feel about AI and their thoughts on its ethical implications. This could be done through surveys, focus groups, or even casually talking with them during company events or meetings. Use this feedback to improve your understanding of how your employees feel about the subject, allowing you to develop an ethical framework that works for everyone involved.
Create clear policies around AI use
Once you have gathered data from your employees, it's time to create clear policies around AI use within your organisation. These policies should be clear and easy to understand by all employees, so there are no misunderstandings about what is expected when using AI solutions at work. Ensure these policies are reviewed regularly so they do not become outdated or irrelevant over time.
In an ideal world, all businesses would be ethical by design. But in the real world, there are many situations where it is unclear what the right thing is to do. When faced with these scenarios, business leaders must set clear rules on how people should act so that everyone in the company knows what's expected of them and can make decisions based on those guidelines.
This is where ethics comes into play. Ethics are a system of moral principles — such as honesty, fairness, and respect — that help guide your decision-making process. For example, if you are trying to figure out whether you should use an AI product that may harm your customers' privacy, ethics would help you decide whether you should use it or not.
AI ethics and its benefits
The technology industry is moving rapidly, and businesses need to keep up with the latest trends. But to build a future where humans and machines can work together in meaningful ways, the fundamental values of trust, responsibility, fairness, transparency, and accountability must be embedded in AI systems from the beginning.
Systems created with ethical principles built in will be more likely to display positive behaviour toward humans without being forced into it by human intervention or programming; these are known as autonomous moral agents. For example, suppose you are building an autonomous car with no driver behind its wheel (either fully self-driving or just partially so). In that case, you need some mechanism to prevent it from killing pedestrians while they are crossing the street—or doing anything else unethical. This type of system would have never gotten off the ground had there not been thorough testing beforehand.
Latest advances in the field of AI ethics
AI ethics is growing rapidly, with new advances being made every day. Here is a list of some of the most notable recent developments:
The 2022 AI Index Report
The AI Index is a global standard for measuring and tracking the development of artificial intelligence, providing transparency into its deployment and use worldwide. It is created every year by the Stanford Institute for Human-Centered Artificial Intelligence (HAI).
In its fifth edition, the 2022 AI Index analyses the rapid rate of advancement in research, development, technical performance, and ethics; economy and education; policy and governance — all to prepare businesses for what is ahead.
This edition includes data from a broad range of academic, private, and non-profit organisations and more self-collected data and original analysis than ever before.
The European Union Efforts to Ensure Ethics in AI
In June, the European Union (EU) passed AI Act (AIA) to establish the world's first comprehensive regulatory scheme for artificial intelligence, but it will have a global impact.
Some EU policymakers believe it is critical for the AIA to set a worldwide standard, so much so that some refer to an international race for AI regulation.
This framing makes it clear that AI regulation is worth pursuing its own sake and that being at the forefront of such efforts will give the EU a major boost in global influence.
While some components of the AIA will have important effects on global markets, Europe alone cannot set a comprehensive new international standard for artificial intelligence.
The University of Florida supports ethical artificial intelligence
The University of Florida (UF) is part of a new global agreement with seven other universities committed to developing human-centred approaches to artificial intelligence that will impact people everywhere.
As part of the Global University Summit at the University of Notre Dame, Joseph Glover, UF provost and senior vice president for academic affairs, signed "The Rome Call" on October 27—the first international treaty that addresses artificial intelligence as an emerging technology with implications in many sectors. The event also served as a platform to address various issues around technological advancements such as AI.
The conference was attended by 36 universities from around the world and held in Notre Dame, Indiana.
The signing signifies a commitment to the principles of the Rome Call for AI Ethics: that emerging technologies should serve people and be ethically grounded.
UF has joined a network of universities that will share best practices and educational content and meet regularly to update each other on innovative ideas.
The University of Navarra in Spain, the Catholic University of Croatia, SWPS University in Poland, and Schiller International University are among the schools joining UVA as signatories.
Microsoft's Latest Updates on Its Own AI Ethical Framework
In June, Microsoft announced plans to open source its internal ethics review process for its AI research projects, allowing other companies and researchers to benefit from their experience in this area.
A team of researchers, engineers, and policy experts spent the past year working on developing a new version of Microsoft's Responsible AI Standard. The new version of their Standard builds on earlier efforts, including last fall's launch of an internal AI standard and recent research. It also reflects important lessons learned from their own product experiences.
According to Microsoft, there is a growing international debate about creating principled and actionable norms for the development and deployment of artificial intelligence.
The company has benefited from this discussion and will continue contributing to it. Industry, academia, civil society—all sectors of society have something unique to offer when it comes t learning about the latest innovation.
These updates prove that we can address these challenges only by giving researchers, practitioners, and officials tools that support greater collaboration.
There is not just the possibility but almost certainty that AI will significantly impact society and business. We will see new types of intelligent machines with many different applications and use cases. We must establish ethical standards and values for these applications of AI to ensure that they are useful and trustworthy. We must do so today.
AI is an evolving field, but the key to its success lies in the ethical framework we design. If we fail in this regard, it will be difficult for us to build trust in AI. However, many promising developments are happening now that can help us ensure that our algorithms are fair and transparent.
Commonly, there's a belief that artificial intelligence will advance to the point of creating machines that are smarter than humans. While this time is far off, it presents the opportunity to discuss AI governance now while introducing ethical principles into the technology in an updated manner. If we stand idly by and do not take action now, we risk losing control over our creations. By developing strong ethics guidelines early on in AI development, we can ensure the technology will better benefit society and not harm it.
Cover image: Created with Stable Diffusion