Artificial intelligence (AI) has vastly improved the manner and capability in which we do things. Research shows more than half of companies globally, especially in emerging nations, have adopted AI in at least one function. It is used in all sectors, from optimising service operations through to recruitment, and its functionality is extending to the capture of biometric data, judicial systems and finance.
So, AI is making key decisions in people’s lives. But while it brings multiple advantages, the technology poses both risks and challenges. It is therefore imperative to consider the ethical implications which the use of AI poses.
AI is the technology which gives a machine or a digital computer the ability to execute tasks performed by humans, such as thinking and learning. However, it emanates from programs and algorithms generated by humans. These analyse, disseminate and study information effectively, whilst algorithms follow the programmed rule-set used in calculations and other problem-solving operations. Thus, some of the ethical issues surrounding the use of AI stem from the fact that AI is crafted by human programming.
But there are no global laws overseeing AI, its development and usage, leaving the ethical aspects of AI usage unclear.
Globally, there are discussions about the under-and overregulation of AI, both of which pose ethical concerns. Whilst there are several legal guidance aspects around consent and appropriately informing users, the legal interpretation and practical implementation of requirements, such as AI fairness, are still in their development stages. Essentially, there is no one-sizefits-all approach for assessing trustworthy AI principles.
Discrimination and bias
Artificial intelligence bias is regarded as the underlying prejudice upon which AI is created. It can lead to discrimination and other social impacts, heightening ethical concerns. Both human bias and resulting discrimination can be replicated when programs or algorithms are developed.
According to Harvard Law Review, when the integration of bias and discrimination is applied at a scale in sensitive application areas, the situation can worsen. For example, in some instances algorithms determine which prisoners should be granted parole, which sectors of the public should be afforded housing opportunities and so forth. In some instances, the output of AI can be regarded as replicating bias and discrimination which are already prevalent in society.
Since 2019, the Dutch government has been involved in the highly publicised scandal about the self-learning algorithm through which the tax authorities detect fraud in respect of those applying for childcare benefits. Families were penalised on a mere suspicion of fraud, based on an algorithm, and thousands were plunged into poverty. Some committed suicide whilst several children were incorrectly placed in foster care.
Privacy
Artificial intelligence can identify patterns unseen to the human eye, learn, and make predictions about individuals and groups, as well as create information that would not have otherwise existed. Inferring information in this manner challenges not only the definition of personal information but also raises privacy aspects as to whether it is acceptable to infer personal information about an individual who could have chosen not to disclose it.
Collecting data via AI also raises privacy issues, such as whether informed consent is freely given, whether the holder of information can opt out, whether the data collection can be limited and whether it can be deleted on request.
The ultimate question remains as to whether an individual would even be aware that their data had been collected, which would allow them to make a reasonably informed decision as to the next steps about its retention.
Even when data is publicly available, the use of such data can breach what is referred to as textual integrity. This fundamental principle in legal discussions of privacy requires that an individual’s information is not revealed outside the context in which it was originally produced.
In such instances, the traditional definition of privacy would not be technically or legally practical, nor particularly helpful as an effective way to protect the privacy of individuals. With the lack of general regulation about AI, privacy cases are dealt with in an individual manner based on country-specific regulation.
At the end of 2021, the Office of the Australian Information Commissioner found that Clearview AI had violated the Australian Privacy Act, based on the collection of images and biometric data without the consent of data holders. A joint investigation was concluded with Australia’s OAIC and the UK’s ICO, which confirmed its intention to impose a potential fine of more than £17 million. This was supported by three Canadian privacy authorities as well as France’s CNIL which ordered Clearview AI to stop processing and to delete the data.
Plagiarism
Recently, plagiarism concerns surrounding ChatGPT have been widely published in the media. The AI chatbot, released in November 2022, uses natural language to enable a normal conversation that is humanlike.
ChatGPT can provide responses to questions, content for essays and compose articles. Concerns emanating from ChatGPT relate to whether the information will be used in an unethical manner to cheat, plagiarise or misuse it in some other manner.
There are varying views about the experience AI can deliver to students, in particular. It can be used to personalise their learning experience and offer real-time feedback to both students and teachers. ChatGPT, in particular, can also help with grading, generate reports and deal with frequently asked questions, freeing up time for educators to focus on more creative and engaging teaching methods.
In the absence of a standard or regulation around the use of AI, and more specifically chatbots, educational facilities such as New Zealand’s eight universities are attempting to align their individual AI policies.
Some aspects for consideration of ethical AI practices can include:
- Program and algorithm developers questioning the consequences of the design at development stage;
- Regular and thorough audits to be conducted on the data collected from a new AI product, along with seeking responses from developers, society and others impacted by the product to detect and possibly deter bias; and
- Instead of creating additional legislation, enable regulatory sandboxes to be created to allow for innovation both in technology and regulation.
Vaheeni Naidoo is an executive working with ADLS’ committees.
0 Comments