Artificial intelligence (AI) is rapidly being adopted across industries, perhaps none bigger than the tech space. With tech and IT firms utilising cutting-edge AI tools and algorithms to streamline operations within HR, recruitment, customer service, accounting, and more, it’s no wonder why this sector-wide disruption is causing concern among business leaders and experts. There are reasons to suggest that AI empowers women in technology, and for all intents and purposes, when deployed methodically, this is true.
However, one of the most worryingly overlooked aspects of AI’s continued evolution is its propensity to perpetuate gender biases. If not properly monitored and adjusted, AI-generated content and algorithms can allow misinformed and harmful biases to spread, which is something that STEM organisations want to avoid, given its history as a male-dominated industry.
In a recent study conducted by UCL researchers, an AI algorithm used to predict liver disease in blood tests was largely unsuccessful at detecting findings in women. According to other recent reviews, there have been incidences of AI adopting gender bias from its human users, often grouping words with outdated gender stereotypes (e.g. ‘doctor’ with ‘man’ and ‘nurse’ with ‘woman’, etc.)
It begs the questions of how and why AI bias has been allowed to occur to such a degree, and what can be done to create more inclusive, ethical AI practices. This short guide explores why women in tech are rightfully expressing concern about this evidence and what hiring tech businesses can do to prevent this from escalating.
AI systems may appear neutral, but they fundamentally reflect the biases of their human creators.This includes gender bias, which can creep into AI in several ways:
AI is trained on large datasets of human-generated text, images, audio, and more. If these datasets are inherently imbalanced or contain biased depictions, the AI may amplify those biases. There is rarely a filter to recognise biased data, instead the focus is on the granular numbers and metrics to dispense findings and results.
Despite possessing some inherent security risks, AI systems are built to be as accurate as possible when making predictions or categorisations. However, prioritising its accuracy means that fairness and inclusion often get overlooked in development. For instance, algorithms trained to scan job applications and CVs in bulk may accurately predict that male applicants are more qualified.
The tech sector’s gender diversity gap is reflected in AI development teams. Most developers are men who often spearhead AI algorithm creation, meaning that there are huge gaps in perspective that can lead to biased, AI-led systems. If more women were represented in teams to produce and test these systems, it would be instrumental in spotting biases early.
Women already face numerous barriers in tech careers, from pervasive pay gaps and hiring imbalances to hostile company culture. Biased AI systems only threaten to stunt progress towards equality even further.
With regard to female recruitment in tech roles, AI can help hiring organisations and agencies screen and shortlist swathes of job applications and CVs at the early stages. In fact, most UK jobseekers had already noticed an AI presence when applying, according to a recent Beamery study. However, biased algorithms could unknowingly reinforce stereotypes, screen out women as less qualified or underestimate their performance and promotional potential.
Ethical human oversight still remains crucial, regardless of AI efficiency benefits. Reinforcing the status quo based on what algorithms recommend could continue to keep women underrepresented in tech management or leadership roles, not to mention dissuade more women from attempting to pursue careers in the field. Access to STEM education, networking, and mentoring programmes, among others, would therefore be less open to women due to increased fear of being turned down for tech opportunities, thanks to perpetuating biases.
As AI chatbots and tools explode in use with each passing day, there are continued concerns about how they portray underlying gender issues without consideration. Tools trained on non-diverse datasets and biased prompts risk spreading misinformation, fake news, skewed perspectives, and oftentimes dangerous discourse that could continue to impede the sector from progressing and becoming more inclusive.
Despite the existence of bias, the good news is that businesses and organisations in the tech sector can take proactive steps to improve gender fairness. This doesn’t just apply to AI systems that they use themselves, but also for greater inclusivity across the industry.
Companies should closely analyse the datasets that their AI tools use, ensuring that they represent a diverse and inclusive sample. This is especially important when using integrated AI chatbots as these can be easily tricked into performing harmful tasks. Make sure that any datasets that exhibit biased labels or language are sanitised and, if necessary, excluded from any tools deployed.
Having more women and other underrepresented groups directly involved in building, testing, and monitoring AI systems will help reduce blind spots and minimise biases. Diversity brings fresh perspectives and viewpoints which can be instrumental in helping catch any loopholes or weaknesses before, during, and after development.
Companies should, therefore, invest in STEM education and training, and present more opportunities for women to build a robust and inclusive talent pipeline. A recent study found that only 23% of people are receiving any form of workplace training, which needs to change if the gender pay gap and digital skills shortages are to be closed.
Once an AI system or tool is deployed, continuously monitor it for signs of biased decisions or recommendations. Track the outcomes and results across user demographics for evidence of any skewed text or imbalances in data. As you begin to find new areas to automate and optimise within your organisation with the help of AI, audit new tools and systems regularly to see if improvements can be made.
As more data is aggregated and dispensed, you may find yourself in need of more enterprise-level tools, but don’t be tempted by costs and benefits alone. Keep gender inclusivity at the forefront of your decisions when upgrading your setup.
Fundamentally, any AI tool needs human users to be at the epicentre of its use and effectiveness. We, as tech professionals, need to override AI decisions when necessary, and actively oversee its output, especially in the early stages of development and deployment. Any AI-made decisions that appear biased, unfair, or flawed must be given appropriate consideration by company teams, particularly women.
Additionally, teams should be equipped with the right feedback channels and permissions to extensively test and report on issues. Training all staff on ethical AI use will continue to raise awareness of biases from the subtle to the overt. The whole company, particularly executives, should understand these risks and support the integration of unbiased and inclusive AI. In turn, this will promote a workplace culture that values diversity and questions how to make AI fairer and more representative.
While AI can harbour biases, tech companies can still make a difference. Tech leaders, by example, can demonstrate thoughtfulness when harnessing AI for the right reasons, designing, testing, and deploying inclusive technology. The presence of women in tech roles still remains low, occupying 26% of the whole workforce and even as low as 5% of leadership roles, but companies can work diligently to change this with considerate use of AI.
This is by no means an easy task and work must start now. Tech leaders should relish the opportunity to audit and correct bias in AI systems as a means to set themselves apart from the crowd and contribute to creating a fairer and more inclusive sector. The recommendations above provide a blueprint for building more ethical AI systems that don’t impede diversity, but remains up to tech companies to recognise the gravity of these issues and act decisively to create meaningful change.