While Artificial Intelligence (AI) has only recently become a topic of mainstream conversation, the technology has been in use for quite some time. Let’s say for example you’re shopping on an eCommerce site and click on a product in which you have an interest. In most cases, the underlying algorithm will show you a “People searching for this item also considered—” prompt.
The system arrived at this decision by aggregating and analyzing data generated by other people who performed the same action. As benign as this might seem, it demonstrates how AI’s ability to amass, examine, and act on massive quantities of data can raise substantial privacy issues. Should your purchasing decisions be used to promote products to other people without your knowledge or permission? This is but one example of why addressing AI privacy concerns is of paramount importance.
Evolution of the Data Privacy Paradigm
Before this technology became commonplace, data privacy initiatives were largely focused on protecting your personal information. While this remains an important concern, the paradigm has expanded to include the recognition and governance of how that data is gathered, as well as employed and shared. This, in turn requires developers to think beyond what should be done to protect your privacy to also contemplate what has to be done.
Consider, for example, the way facial recognition systems are being employed at theme parks, stadiums and other entertainment venues to provide ticketless entry to patrons who paid online. Would providing that data to a law enforcement agency constitute illegal surveillance? Would it be a violation of a ticketholder’s civil liberties?
Fundamental AI Privacy Concerns
Misuse of Data and Consent
The quantities of data required to inform machine learning, predictive analytics and natural language processing are massive. However, gathering this data can potentially infringe upon the privacy of the individuals observed. Additionally, AI systems performing the analysis of personal data could unintentionally expose personal details. Thus, the challenge becomes how to amass and employ the data these systems need to function, without overstepping the boundaries of privacy.
Enhanced Surveillance
AI surveillance isn’t always confined to areas in which security is a primary concern. The technology is employed in shopping malls, on city streets and in other public areas in which we have the right to expect our privacy to be respected. Essentially, we are under constant scrutiny wherever this technology is deployed. Governments, corporations and hackers can use this information in ways that could violate our privacy.
Perpetuation of Ethnic Biases
AI algorithms are no more enlightened in this regard than the people who write them. When social biases are programmed into their coding, whether intentionally or inadvertently, social discrimination can result. Hiring decisions, lending decisions and credit scoring can all be affected by these biases. Meanwhile, the people who are affected often have no idea they’ve been judged in this fashion.
Addressing AI Privacy Concerns
Ultimately, addressing AI privacy concerns is an ethical consideration. Respecting the privacy and dignity of individuals should be fundamental to the design of any AI algorithm. It’s important to implement privacy by design as one of the fundamental underpinnings of any AI system. Toward this end, leading speech generation researchers here at Tomato.ai developed our AI models with these concerns in mind.
Data collection should be narrowly focused to that which is needed to serve the intended task, rather than broadly based. Data retention should be limited to that which is absolutely required too. Encryption strategies should also be key to the development of AI algorithms. Conducting privacy audits and impact assessments at regular intervals are of equal importance.
Transparency is also important. People should always be informed when AI is being employed and the privacy ramifications of that deployment should be clearly outlined. Developers must also be careful to ensure systems perform based on fairness and equality doctrines to avoid the potential for biases and discrimination.
