Data Science has the ability to influence people, learn new things, automate a task and literally change the world. With this power comes a responsibility to ensure the ethical design and used.
The integrity of Artificial Intelligence (AI) technologies is dependent on acquiring data, the integrity of the data and how the technology is eventually used. More specifically, the ability of AI technologies to perform ethically, and without bias, is dependent on quality data combined with well-designed algorithms that make honest, unbiased decisions.
Therefore, AI technologies can be far-reaching yet properly solve real problems in the world. This is a growing discussion in the AI technology space. Identifying and mitigating bias in AI is important to building trust between the companies that develop and use new tech and the population.
This blog discusses the main issues regarding data science, ethics, and bias in regards to its application to artificial intelligence technologies. The ethics conversation has three broad areas:
- Data Acquisition
- Data Integrity
- Use of Technology
Data acquisition is the gathering of the data used by AI technologies. How that data is acquired has raised some ethical discussions and its implications on artificial intelligence include:
- data ownership, and
In regards to privacy, some people might not want to share certain private or sensitive data about themselves. Do they know what data is being shared and do they have a choice to opt out? And if individual opt-out of having certain personal data acquired does that skew the data now. For example, some people might not want to share their age, race or sexual preference. Does that create a higher data pool skewed towards the more conventional population? This can subsequently create inaccurate and incorrect results when used in AI technologies. Further, certain groups of people, such as minorities, might be more prone to withhold data about themselves than other groups for fear, as shown by history, of usage against that population.
This overlaps and contrasts with the next major issues in regards to data acquisition. These issues include who owns the data once it’s acquired and how the data is acquired. Often times an individual does not even own the data about themselves given today’s legislation. Individuals should likely have more control over choosing what data about them is saved, and who owns it. Further, individuals should be able to delete data that is collected about them as per new laws in Europe (see GDPR (more info))
All of these extra controls afforded to the individual increase the ethical conduct of data acquisition and ownership. However, they increase the challenges of creating high integrity, accurate AI technologies.
Here are some real-world data acquisition dilemmas. Can a phone using Siri always monitor and log your conversations? Can Alexa always record what is being said around it? Should your Fitbit always record where you are and how you are doing? Once the data is acquired does the individual have the right to know and see that data? Do you have a right to delete your data? Can I use all the cameras in the city and watch your every move? Does an individual lose complete control of their personal data that is acquired by the technologies that they use and rely on in their day-to-day lives? All of these issues are being discussed now because the technology exists to make this happen.
All modern AI techniques use data as a foundation for building models and algorithms that make decisions. The ability of AI technologies to make accurate, unbiased and ethical decisions depends on the integrity of the data that is used by these technologies. The foundational data is what defines whether the model will work. With that said, certain biases can be reflected depending on how the data is acquired. For example, in a lot of medical studies, the subjects used in the past tended to be white (more info).
Therefore, models built around this data have built-in racial biases. This can be true of any data.
Another example, businesses that acquire and own individual data often tend to sell it. There are markets for this data and some data might be priced higher than others. Prices might vary based on demographics and socioeconomic biases. Therefore, making some data less or more accessible to builders of AI technologies. Even worse, large groups of the population might be completely missing from data that is available for sale or that is directly acquired by AI companies. To elaborate, certain groups might not even use or have access to the technologies or programs that acquire data. As a result, AI technologies that use this data will inherently have biases and potentially unethical decision-making abilities embedded into them. This creates an extra area of caution to those developing AI technologies in regards to preventing bias and maintaining integrity and ethics. So, understanding data bias and then working to compensate for it is an ongoing discussion.
Use of Technology
Technology is often used to influence the behavior of people or groups. This raises an ethical dilemma to the limits of how technology should be used. For example, Russia’s use of technology via Facebook to influence the election created a lot of controversies. The identification and targeting of pregnant women by Target to send coupons created privacy discussions. Technology is also being used to predict crimes. However, this could create biases based in the judicial system (more info).
Essentially, technology is being used in many cases to influence behavior or make a decision that affects people. Today, after someone searches for something on the internet, and they open Facebook or even AliExpress, an ad is shown, offering to sell them something related to previous searches. This should make us ask the question of whether or not this is ethically correct. Should this data be used to drive corporate bottom lines, or should it be used to ethically drive intelligence that solves problems that benefits the individuals in the world? There is a saying, perception is reality. When technology is used in ways to drive consumption or influence voting behaviors these behavioral influences, for better or for worse, often become reality. These are just ethical discussions going on about the use of new powerful, AI technologies.
We need to be cognizant of both obvious and indirect ethically questionable biases in both our data and intelligence technologies that use this data. Further, managers of AI technologies should understand the ethical usage issues and make well-informed decisions. To gain trust in your company with the general public maintaining integrity and honesty in the data and how it is used is valuable. Artificial intelligence can be used to create value to people and make your company more competitive.