Before discussing
the ethical issues that are created by AI ethics, let’s first discuss what
actually is meant by AI Ethics and who are the stakeholders.
In a layman
language, artificial intelligence is the process of incorporating intelligence into
the machines with an objective of simulating human intelligence and decision
making. Initially, it looked very nice and helpful but with the passage of time, it results in many risks to human safety also. So, we can say that with the advancement of AI, ethical issues will also come into being.
Ethical issues
are basically a set of rules, guidelines and principles which should be
considered by the stakeholders while developing or using artificial
intelligence. All the organizations working on AI should make “AI ethics
policies” based on the guidelines provided and train their manpower to develop
systems according to these policies. This not only minimizes the risks involved
with the development of AI but will also work towards attaining its actual
objective of improving human life.
There are many
fields in which people need to collaborate to develop ethical principles for
Responsible AI. Researchers and professors in Government agencies, academics, international
entities like UNO, many NGOs and private companies can depute their
representatives to discuss and develop AI Ethics. These all can be considered
as the stakeholders for AI Ethics. It is their fundamental responsibility that
they must examine how artificial intelligent machines and humans can coexist
with harmony.
There are mainly
five key principles that need to be considered for developing a strong AI
ethics Policies. These are Transparency, Accountability, Impartiality,
Reliability, Security and Privacy.
Transparency:- AI needs to be transparent in terms of its algorithms and
decisions. A common man should know how an algorithm works and why a decision
has been made by AI. For example, a person is denied a loan by a bank’s online
system. Now the system should be so transparent that the person should know why
the algorithm has denied the loan and what he can do to get it sanctioned in
future.
Accountability: As the algorithms are run by AI, the
question arises as to who should be held responsible and accountable in case of
wrong decisions? Well, the duty lies with people and teams developing AI
systems. They should ensure that the algorithms are developed properly. It
should be their responsibility to monitor that high-quality data is fed into
the system. In case of any ambiguity, they should be held responsible.
Impartiality: AI should not be biased and all human beings should be
treated equally by AI. Unbiased and high-quality data should be used to train
AI systems so that their decisions must not be biased at any developmental
stages.
Reliability: The AI systems should be reliable. The results should match
with the outcomes for which the system is basically designed. This is very
useful when we use AI, especially in healthcare and financial services.
Security and Privacy: Sensitive data should be the
topmost concern while developing AI systems. There should be clear security
policies to deal with data security and privacy.
AI
algorithms are being used by many healthcare providers to help them make
judgments about patient care, including which patients need special
attention. Researchers Obermeyer et al. find evidence of racial
prejudice in algorithms. It is shown that black patients are identified with
lower risks by the algorithm, despite being sicker than white patients. As
a result, white patients get chosen for additional care because they have
higher risk scores.
Case study 2: Banking and finance
Apple's AI algorithm, "Apple card," has a prejudice against women. The interest rates and credit limitations it offered to different genders varied dramatically. Men were being granted larger credit limits than women. It would be difficult for a bank to examine and determine the source of this bias with conventional "black box" AI systems.
Case Study 3: Hiring Process
Amazon made an effort to use its AI hiring and recruiting
tool. It enables companies to choose the top 5 resumes from thousands of
submissions. Every company desires this setup.
However, it was discovered in 2015 that the system's evaluation of applicants
for technical jobs, such as software development, has a bias against women.
The data utilized to train the system was the cause of this bias. It was
trained using applications from the previous ten years, so it could recognize
the format of resumes submitted. The majority of those resumes were from men,
which suggested that there was a male-dominated IT sector.
From the above case studies, we can conclude that Ethical AI hinges on high-quality data. Ultimately, it's all about data to bring AI ethics into practice. Biased and poor-quality data will result in poor outcomes. In a report published by Forbes, AI is patched up by a gang of unethical people thus making it a biggest challenge to achieve the expected levels of security, reliability and accountability in ethical AI. Ethical issues in any system can only be recognized if data and algorithms are understood thoroughly. Explainable AI is the core component to understand the data and algorithm. Explainable AI transforms a black box model of Ai into white box model to achieve transparency in AI systems. A unified approach is needed to develop across the entire lifespan of an AI system so as to save us from its harmful implications.
Being
one of the stakeholders in AI ethics, we at
CU Online follow all AI Ethics in terms of research and publications.
Workshops are conducted to make all the faculty and students aware about using
AI while adhering to AI ethics. Thus, we strive to fulfil our social
responsibilities while progressing towards an AI-driven world.
Author:- Shuchi Sharma - Assistant Professor
Computer Applications - (CDOE)