Home Artificial Intelligence (Un)Objective Machines: A Have a look at Historical Bias in Machine Learning What’s a Bias? Historical Bias Why Should We Care? What Can We Do About It? Conclusion Enthusiastic about Learning More? References:

(Un)Objective Machines: A Have a look at Historical Bias in Machine Learning What’s a Bias? Historical Bias Why Should We Care? What Can We Do About It? Conclusion Enthusiastic about Learning More? References:

0
(Un)Objective Machines: A Have a look at Historical Bias in Machine Learning
What’s a Bias?
Historical Bias
Why Should We Care?
What Can We Do About It?
Conclusion
Enthusiastic about Learning More?
References:

A deep dive into biases in machine learning, with a deal with historical (or social) biases.

Towards Data Science

Humans are biased. To anyone who has needed to take care of bigoted individuals, unfair bosses, or oppressive systems — in other words, all of us — this is not any surprise. We should always thus welcome machine learning models which may help us to make more objective decisions, especially in crucial fields like healthcare, policing, or employment, where prejudiced humans could make life-changing judgements which severely affect the lives of others… right? Well, no. Although we could be forgiven for considering that machine learning models are objective and rational, biases might be in-built into models in a myraid of the way. On this blog post, we will likely be specializing in historical biases in machine learning (ML).

In our day by day lives, after we invoke bias, we regularly mean “judgement based on preconceived notions or prejudices, versus the impartial evaluation of facts”. Statisticians also use “bias” to explain just about anything which can result in a scientific disparity between the ‘true’ parameters and what’s estimated by the model.

ML models suffer from statistical biases since statistics play a giant role in how they work. Nevertheless, these models are also designed by humans, and use data generated by humans for training, making them vulnerable to learning and perpetuating human biases. Thus, perhaps counterintuitively, ML models are arguably more prone to biases than humans, not less.

Experts disagree on the precise variety of algorithmic biases, but there are a minimum of 7 potential sources of harmful bias (Suresh & Guttag, 2021), each generated at a distinct point in the information evaluation pipeline:

  1. Historical bias, which arises from the world, in the information generation phase;
  2. Representation bias, which comes about after we take samples of information from the world;
  3. Measurement bias, where the metrics we use or the information we collect may not reflect what we actually wish to measure;
  4. Aggregation bias, where we apply the identical approach to our whole data set, though there are subsets which must be treated in another way;
  5. Learning bias, where the ways we’ve defined our models cause systematic errors;
  6. Evaluation bias, where we ‘grade’ our models’ performances on data which doesn’t actually reflect the population we would like to make use of the models on, and eventually;
  7. Deployment bias, where the model will not be utilized in the way in which the developers intended for it for use.
Light trail symbolising data streams
Photo by Hunter Harritt on Unsplash

While all of those are essential biases, which any budding data scientist should consider, today I will likely be specializing in historical bias, which occurs at the primary stage of the pipeline.

Psst! Enthusiastic about learning more about other forms of biases? Watch this beneficial video:

Unlike the opposite forms of biases, historical bias doesn’t originate from ML processes, but from our world. Our world has historically been, and still is peppered with prejudices, so even when the information we use to coach our models perfectly reflects the world we live in, our data might capture these discriminatory patterns. That is where historical bias arises. Historical bias can also manifest in instances where our world has made strides towards equality, but our data doesn’t adequately capture these changes, reflecting past inequalities as a substitute.

Most societies have anti-discrimination laws, which aim to guard the rights of vulnerable groups in society, who’ve been historically oppressed. If we usually are not careful, previous acts of discrimination could be learned and perpetuated by our ML models resulting from historical bias. With the rising prevalence of ML models in practically every area of our lives, from the mundane to the life-changing, this poses a very insidious threat — historically biased ML models have the potential to perpetuate inequality on a never-before-seen scale. Data scientist and mathematician Cathy O’Neil calls such models ‘weapons of math destruction’ or WMDs for brief — models whose workings are a mystery, generate harmful outcomes which victims cannot dispute, and which frequently penalise the poor and oppressed in our society, while benefiting those that are already well off (O’Neil, 2017).

Photo by engin akyurt on Unsplash

Such WMDs are already impacting vulnerable groups worldwide. Although we might think that Amazon, which profits from recommending us items we’ve never heard of, yet suddenly desperately want, would have mastered machine learning, it was found that an algorithm they used to scan CVs had learned a gender bias, resulting from the historically low number of ladies in tech. Perhaps more chillingly, predictive policing tools have also been shown to have racial biases, as have algorithms utilized in healthcare, and even the courtroom. The mass proliferation of such tools obviously has great impacts, particularly since they might function a solution to entrench the already deep-rooted inequalities in our society. I might argue that these WMDs are a far greater hindrance in our collective efforts to stamp out inequality in comparison with biased humans, for 2 essential reasons:

Firstly, it is tough to get insight into why ML models make sure predictions. Deep learning appears to be the buzzword of the season, with complicated neural networks taking the world by storm. While these models are exciting since they’ve the potential to model very complex phenomena which humans cannot understand, they’re considered black-box models, since their workings are sometimes opaque, even to their creators. Without concerted efforts to check for historical (and other) biases, it’s difficult to inform in the event that they are inadvertently discriminating against protected groups.

Secondly, the dimensions of injury which could be done by a historically biased model is, for my part, unprecedented and neglected. Since humans need to rest, and wish time to process information effectively, the damage a single prejudiced person might do is proscribed. Nevertheless, only one biased ML model can pass hundreds of discriminatory judgements in a matter of minutes, without resting. Dangerously, many also consider that machines are more objective than humans, resulting in reduced oversight over potentially rogue models. This is very concerning to me, since with the large success of huge language models like ChatGPT, an increasing number of persons are developing an interest in implementing ML models into their workflows, potentially automating the rise of WMDs in our society, with devastating consequences.

While the impacts of biased models could be scary, this doesn’t mean that we’ve to desert ML models entirely. Artificial Intelligence (AI) ethics is a growing field, and researchers and activists alike are working towards solutions to eliminate, or a minimum of reduce the biases in models. Notably, there was a recent push for FAT or FATE AI — fair, accountable, transparent and ethical AI, which could assist in the detection and correction of biases (amongst other ethical issues). While it will not be a comprehensive list, I’ll provide a transient overview of some ways to mitigate historical biases in models, which can hopefully enable you to on your personal data science journey.

Statistical Solutions

Because the problem arises from disproportionate outcomes in the actual world’s data, why not fix it by making our collected data more proportional? That is one statistical approach of coping with historical bias, suggested by Suresh, H., & Guttag, J. (2021). Put simply, it comprises collecting more data from some groups and fewer from others (systematic over- or under- sampling), leading to a more balanced distribution of outcomes in our training dataset.

Model-based Solutions

In step with the goals of FATE AI, interpretability might be built into models, making their decision-making processes more transparent. Interpretability allows data scientists to see why models make the choices they do, providing opportunities to identify and mitigate potential instances of historical biases of their models. In the actual world, this also signifies that victims of machine-based discrimination can challenge decisions made by previously inscrutable models, and hopefully cause them to be reconsidered. It will hopefully increase trust in our models.

More technically, algorithms and models to deal with biases in ML models are also being developed. Adversarial debiasing is one interesting solution. Such models essentially consist of two parts: a predictor, which goals to predict an consequence, like hireability, and an adversary, which tries to predict protected attributes based on the anticipated outcomes. Like boxers in a hoop, these two components go backwards and forwards, fighting to perform higher than the opposite, and when the adversary can not detect protected attributes based on the anticipated outcomes, the model is taken into account to have been debiased. Such models have performed quite well in comparison with models which haven’t been debiased, showing that we want not compromise on performance while prioritising fairness. Algorithms have also been developed to cut back bias in ML models, while retaining good performances.

Human-based Solutions

Lastly, and maybe most crucially, it’s critical to keep in mind that while our machines are doing the work for us, we are their creators. Data science starts and ends with us — humans who’re aware of historical biases, resolve to prioritise fairness, and take steps to mitigate the results of historical biases. We should always not cede power to our creations, and may remain within the loop in any respect stages of information evaluation. To this end, I would love so as to add my voice to the chorus calling for the creation of transnational third party organisations to audit ML processes, and to implement best practices. While it is not any silver bullet, it’s a very good solution to check if our ML models are fair and unbiased, and to concretise our commitment to the cause. On an organisational level, I’m also heartened by the calls for increased diversity in data science and ML teams, as I consider that this may help to discover and proper existing blind spots in our data evaluation processes. It is usually obligatory for business leaders to pay attention to the bounds of AI, and to make use of it correctly, as a substitute of abusing it within the name of productivity or profit.

As data scientists, we must always also take responsibility for our models, and remember the ability they wield. As much as historical biases arise from the actual world, I consider that ML tools even have the potential to assist us correct present injustices. For instance, while up to now, racist or sexist recruiters might filter out capable applicants due to their prejudices before handing the candidate list to the hiring manager, a good ML model may give you the option to efficiently find capable candidates, disregarding their protected attributes, which could result in worthwhile opportunities being provided to previously ignored applicants. In fact, this will not be a straightforward task, and is itself fraught with ethical questions. Nevertheless, if our tools can indeed shape the world we live in, why not make them reflect the world we would like to live in, not only the world because it is?

Whether you might be a budding data scientist, a machine learning engineer, or simply someone who’s occupied with using ML tools, I hope this blog post has shed some light on the ways historical biases can amplify and automate inequality, with disastrous impacts. Though ML models and other AI tools have made our lives loads easier, and have gotten inseparable from modern living, we must keep in mind that they usually are not infallible, and that thorough oversight is required to make sure that that our tools stay helpful, and never harmful.

Listed below are some resources I discovered useful in learning more about biases and ethics in machine learning:

Videos

Books

  • Weapons of Math Destruction by Cathy O’Neil (highly advisable!)
  • Invisible Women: Data Bias in a World Designed for Men by Caroline Criado-Perez
  • Atlas of AI by Kate Crawford
  • AI Ethics by Mark Coeckelbergh
  • Data Feminism by Catherine D’Ignazio and Lauren F. Klein

Papers

AI Now Institute. (2024, January 10). Ai now 2017 report. https://ainowinstitute.org/publication/ai-now-2017-report-2

Belenguer, L. (2022). AI Bias: Exploring discriminatory algorithmic decision-making models and the appliance of possible machine-centric solutions adapted from the pharmaceutical industry. AI and Ethics, 2(4), 771–787. https://doi.org/10.1007/s43681-022-00138-8

Bolukbasi, T., Chang, K.-W., Zou, J., Saligrama, V., & Kalai, A. (2016, July 21). Man is to computer programmer as woman is to homemaker? Debiasing word embeddings. arXiv.org. https://doi.org/10.48550/arXiv.1607.06520

Chakraborty, J., Majumder, S., & Menzies, T. (2021). Bias in machine learning software: Why? how? what to do? Proceedings of the twenty ninth ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering. https://doi.org/10.1145/3468264.3468537

Gutbezahl, J. (2017, June 13). 5 forms of statistical biases to avoid in your analyses. Business Insights Blog. https://online.hbs.edu/blog/post/types-of-statistical-bias

Heaven, W. D. (2023a, June 21). Predictive policing algorithms are racist. they must be dismantled. MIT Technology Review. https://www.technologyreview.com/2020/07/17/1005396/predictive-policing-algorithms-racist-dismantled-machine-learning-bias-criminal-justice/

Heaven, W. D. (2023b, June 21). Predictive policing remains to be racist-whatever data it uses. MIT Technology Review. https://www.technologyreview.com/2021/02/05/1017560/predictive-policing-racist-algorithmic-bias-data-crime-predpol/#:~:text=It%27s%20no%20secret%20that%20predictive,lessen%20bias%20has%20little%20effect.

Hellström, T., Dignum, V., & Bensch, S. (2020, September 20). Bias in machine learning — what’s it good for?. arXiv.org. https://arxiv.org/abs/2004.00686

Historical bias in AI systems. The Australian Human Rights Commission. (2020, November 24). https://humanrights.gov.au/about/news/media-releases/historical-bias-ai-systems#:~:text=Historical%20bias%20arises%20when%20the,by%20women%20was%20even%20worse.

Memarian, B., & Doleck, T. (2023). Fairness, accountability, transparency, and ethics (fate) in Artificial Intelligence (AI) and Higher Education: A scientific review. Computers and Education: Artificial Intelligence, 5, 100152. https://doi.org/10.1016/j.caeai.2023.100152

Obermeyer, Z., Powers, B., Vogeli, C., & Mullainathan, S. (2019). Dissecting racial bias in an algorithm used to administer the health of populations. Science, 366(6464), 447–453. https://doi.org/10.1126/science.aax2342

O’Neil, C. (2017). Weapons of math destruction: How big data increases inequality and threatens democracy. Penguin Random House.

Roselli, D., Matthews, J., & Talagala, N. (2019). Managing bias in AI. Companion Proceedings of The 2019 World Wide Web Conference. https://doi.org/10.1145/3308560.3317590

Suresh, H., & Guttag, J. (2021). A framework for understanding sources of harm throughout the machine learning life cycle. Equity and Access in Algorithms, Mechanisms, and Optimization. https://doi.org/10.1145/3465416.3483305

van Giffen, B., Herhausen, D., & Fahse, T. (2022). Overcoming the pitfalls and perils of algorithms: A classification of machine learning biases and mitigation methods. Journal of Business Research, 144, 93–106. https://doi.org/10.1016/j.jbusres.2022.01.076

Zhang, B. H., Lemoine, B., & Mitchell, M. (2018). Mitigating unwanted biases with adversarial learning. Proceedings of the 2018 AAAI/ACM Conference on AI, Ethics, and Society. https://doi.org/10.1145/3278721.3278779

LEAVE A REPLY

Please enter your comment!
Please enter your name here