How does bias find its way into AI?
Bias in AI manifests in two ways. The first kind is a result of cognitive biases – unconscious errors in everyday thinking which directly influence individualsβ judgements and decisions – and unfortunately, all of us are guilting of having them.
Psychologists have identified and defined at least 180 human biases. Amongst these are gender bias, racial prejudice, and age discrimination – all the pesky traits that society has been trying to eradicate for decades.
These biases prevail as a result of our brains attempting to simplify the immense amount of information we receive about the world around us every second.
These accidentally end up into a machine learning algorithm by designers who unknowingly place them into a computerβs decision-making model.
Other times, biases appear in AI because there is a lack of complete data. When an adequate collection of data sets are not included during programming stages, that results brought up by AI wonβt be representative of the real world.
The wide majority of people working in the field agree that the more people there are involved in developing AI technologies, the better. Where unconscious bias goes unnoticed is when a limited number of unchecked people are responsible for programming.
Simply grouping a diverse team of AI developers and a team of ethicists wonβt be enough to ensure AI is without bias, though. In debates on how to optimise the scope of AI, programmers are being asked to think bigger.
How can we eradicate out bias?
Experts agree that widespread participation and data input from the public is the only way AI can get close to being sufficiently representative of the global population.
However, a lack of public understanding of AI, how it works, and how algorithms collect our information deters a large majority of people from wanting to get involved.
Resolving this would require educating the public on how their data is collected, stored, and used. Stronger legislation to ensure the safety and regulation of conduct with AI is currently being written up, which will likely bolster the publicβs trust in such technologies.
All in all, AI is still a novelty. In terms of regulation, principles, and guidelines, not much currently exists.
Legislation is only now catching up with these advances in technology and hopefully with that in place, the public will feel more comfortable with participating in building AI data sets. Without this, AI might never be able to deal with its problem of bias.
0 comments
