Artificial Intelligence without bias: Why it is needed and how we can ensure a fairer future
Organisations working on developing unbiased data-driven AI solutions need to do so without any link to race, gender, or any other prejudice of humankind. Some have suggested removing the labels that make the algorithm biased in the first place. There are significant benefits of artificial intelligence. In parallel, balancing it with a holistic approach to remove bias helps AI to achieve its true potential, says Apratim Purakayastha, CTO, Skillsoft
By Apratim Purakayastha
Artificial Intelligence (AI) is now here to stay as part of our lives. Numerous organisations across the globe have already taken AI maturity to the next level, and in just a few years, we will witness this number jump even higher. Large firms are working on revamping their business strategy amidst the pandemic with AI at its core. Sectors like manufacturing and agriculture are developing solutions for predictive analysis and sectors like transportation and consumer goods are developing deeper customer insight. The Indian government is making strides in incorporating AI for intelligence, cybersecurity, and the digital economy.
With such growth, there has also been a growing concern about AI bias. There have been some striking consequences in the real world with reported cases of AI biases. AI-based hiring software used by a multinational company was choosing men over women for various roles. In another example, AI used in courts had denied bail to people just by looking at their pictures. In yet another example, racial bias has shown up in AI-based healthcare treatment systems.
So, what is AI bias? In simple terms, AI bias is an incongruity in the output of machine learning algorithms due to the prejudiced assumptions made during the algorithm development process and training data. AI systems learn to make decisions based on training data, which is the culprit here. It is because this data can include biased human decisions or reflect social inequities. These biases may arise at all stages of AI-based decision-making, such as when information is collected, when algorithms turn data into decision-making capacity to results. Besides algorithms and data, researchers and developers developing these AI systems are also responsible for AI bias. To best understand AI bias, think about the development of a child. If a child grows up in a community which let’s say believes that humans never landed on the moon, and all community members re-inforce that in all conversations, pictures, and other material—the child inherits an inherent bias toward believing that humans never landed on the moon.
So, what needs to be done? Well, the idea is to build an AI solution that is holistic in nature and eliminates discrepancies or flaws related to data bias or algorithmic bias. The reason to create such a solution is that AI has a significant potential to transform processes. The current shortfall of AI professionals in India presents a grim outlook as there are thousands of vacant AI positions. This number is only going to increase. Organisations need to focus on upskilling, reskilling their workforce to deal with the demand and be future fit.
Therefore, it becomes imperative to move beyond the conventional AI algorithms, and at the same time, requires companies to upskill and reskill their employees rapidly to ensure bias-free integration of AI. Along with technical understanding of AI, a holistic curriculum must include keen awareness of bias in business and Diversity and Inclusion knowledge. To implement your AI initiative, here are some dos and don’ts that you should consider:
1. Align AI projects with business outcomes: Your AI project might address a significant need gap. However, until its results are quantifiable, it will become hard for you to prioritise it. So, always align your AI projects with tangible business outcomes and KPIs.
2. Create an AI dilo: You must ensure that your AI operates in a silo, at least until you train the algorithm perfectly. It also needs you to deploy AI in an area that you have a good grasp of. Doing so will help you detect and eliminate any errors that might surface initially.
3. Balanced learning approach: An AI curriculum must also included key areas of leadership development including a strong emphasis on Diversity and Inclusion, or a DEI curriculum. Without this exposure, employees are not equipped to spot and address bias in AI algorithms and results.
4. Plan your in-house AI talent: Have a proper framework in place to build your in-house AI talent along with succession planning. Sourcing AI talent by mostly internal growth and education has the added benefit of inheriting nuances of the company culture and focus on diversity, that may not be immediately available in external hires. In case of supplementing via external hires a DEI curriculum along with AI is an essential element of onboarding.
5. Track results: You must constantly track your AI projects’ performance, and it is also a good idea to link it with industry benchmarks or relevant offerings in the market.
There is no denying that for AI to reach its full potential, it must be free of bias. Only then can we have solutions delivering the correct output. This might take several years of development with much of the onus on its quality of input data. Organisations working on developing unbiased data-driven AI solutions need to do so without any link to race, gender, or any other prejudice of humankind. Some have suggested removing the labels that make the algorithm biased in the first place. There are significant benefits of artificial intelligence. In parallel, balancing it with a holistic approach to remove bias helps AI to achieve its true potential.