Image by Editor | Ideogram
The EU AI Act recently entered into force. It is one of the most comprehensive frameworks that has set precedence on making the best use of cutting-edge innovation. The “best use” needs more attention here — it underscores the importance of responsible development, deployment, and utilization of AI systems.
Ethics is not just limited to the European Union, it is also the core focus of the US. A recent White House announcement promotes using concrete safeguards in AI to ensure the rights or safety of everyone involved. In line with the announcement, the role of Chief AI Officers is introduced to manage the risks associated with AI.
With emerging generative AI advancements, the risks have increased too. Its impact is far and wide, impacting everyone involved. We all have a role to play.
The focus of this article is to discuss the ethical implications and moral decision-making.
Understanding Machine Ethics
While we have been discussing the proper use of machines by humans for a long time, there is an increased need for machine ethics. But, what does that mean in the first place? Is it a new concept?
The answer is “no.” As per a research paper dated 2011, machine ethics involves machine behavior towards humans and other machines too. It highlights that the goal should be to create a machine that follows a set of principles while making decisions. This raises two important questions, such as:
- Do machines have self-awareness, also called consciousness to make ethical decisions?
- Do they have a moral compass similar to humans?
Before we even expect machines to have an approach to behave ethically, it is important to assess within ourselves — do humans have a consistent response when faced with dilemmas?
The basis for such questions lies in the fundamentals of how machines learn. Let’s discuss that in our next section and we will come back to these questions.
Machines Learn From Data
Let’s quickly cover some of the standard definitions of artificial intelligence and machine learning:
While some explain these concepts as systems that aim to simulate human intelligence, they are also expected to “perform complex tasks that historically could only be accomplished by humans, such as reasoning, making decisions, or solving problems”.
By now, it should be clear that machine learning finds patterns in data, i.e., data is at the core of machines, and that data often involves human interactions.
So, now a question to ask ourselves:
- How do we know we are ethical?
- Our decisions that reflect in the training data from which machines learn how to behave ethically — is that data embedded in ethics?
- Is ethics a universal approach?
- Do we all have a singular objective approach to ethics?
While feeding data into the algorithms, we also feed our ethical frameworks. Now, this data is mostly a result of inconsistent, and sometimes biased views of the humans involved, making it learn the imperfect human interpretation of ethics.
Consider a scenario where an autonomous vehicle is faced with an unavoidable accident situation and it must choose between a group of pedestrians vs. hitting a barrier, which will likely kill the passenger. Or a situation where the choice is between a child and an elderly pedestrian.
There is no singular way to approach, act, and respond to such situations. And, add to that, the fact that it is difficult to judge someone’s act as ethical.
Delphi: The Case Study on Morality
Now that we understand that machine learning algorithms analyze large datasets involving historical human decisions, opinions, and behaviors. Based on this learning, the algorithm learns to identify patterns and potentially derive ethical principles.
Now, talking about ethics and morality, Delphi deserves a mention; it was a chatbot designed by the Allen Institute for A.I. that guided the distinction between right vs. wrong.
Here are a few examples:
- For a question involving “cheating on an exam”, its response is clear that it is wrong.
- But add a little more context, “cheating on an exam to save someone’s life” and it surprisingly brings a nuanced addition to its view.
However, the users asked it a wide range of queries involving politics and more, and it succumbed soon after. This brings me to a common concern surrounding machine ethics, discussed in the next section.
Whose Values and Principles to Follow?
So, if humans are not perfect, but machines ought to learn from the data consisting of human behavior and actions, it all comes down to this profound question: “Whose values and principles should machines follow?”
It is best described by Norbert Wiener, a world-renowned mathematician, that “we had better be quite sure that the purpose put into the machine is the purpose which we really desire.”
Before we even expect machines to align ethically with humans and act responsibly, we, as humans, must align among ourselves first.
Think of building a task force that can reasonably represent the gold standards of ethics that are majorly acceptable to all and by all. They need to define and quantify these ethics in a way that is aligned with human interests and well understood by developers to code into machines.
The Road Ahead
While the promise of machine intelligennce is exciting, we must account that these models are only as good as the data they’re trained on. If the underlying data is either biased or under-representative, the outcomes are going to be skewed as well.
Diverse training data ensure ethics are second nature to business. Besides, organizations must promote transparency and accountability to share their ethical standpoint. Expecting developers to read between the lines and make AI systems comply with their understanding of ethics — is an unrealistic ask. It calls for organizational and global frameworks to set golden standards for the ethical use of AI.
Lastly, creating awareness and sensitizing everyone – AI teams and beyond, is critical to aligning ethics globally.
Vidhi Chugh is an AI strategist and a digital transformation leader working at the intersection of product, sciences, and engineering to build scalable machine learning systems. She is an award-winning innovation leader, an author, and an international speaker. She is on a mission to democratize machine learning and break the jargon for everyone to be a part of this transformation.