Artificial intelligence can support a more equitable world – if we let it
Automation and artificial intelligence increasingly support decisions impacting people’s lives. It’s time to build systems that reflect our world’s diversity.
The use of artificial intelligence (AI) has exploded as digital solutions become more popular, whether in business or everyday life.
As AI and automation become more powerful, we’re seeing them applied to decisions that get made about people. The technologies are essentially learning from history and enable more efficient decisions at a much greater scale.
But this widespread adoption has its risks.
The potential for bias is creeping into automated decisions.
During the pandemic two years ago, statistical algorithms were used to determine grades for A-level and GCSE exams in the UK. While the resulting grade distribution appeared fair on the surface, a closer look showed catastrophic individual outcomes. Students from private schools and affluent areas were favored at the expense of high achievers from disadvantaged backgrounds. Overall, nearly 40% of students received grades lower than first assessed by teachers.
The event raised questions about how algorithms operating at scale work and how they evaluate people’s abilities. How do AI-supported decisions affect the future of the individual?
We see similar examples of disproportionate bias from automated decisions in recruitment, in loan decisions, insurance, justice and elsewhere.
Biases arise because AI programs are human creations.
There are significant – and often underestimated – risks associated with AI.
As someone who has worked with data for several years, I understand that when AI systems are trained on data that isn’t diverse enough, they may favor one group over another. This may lead to unfair hiring practices, pay inequality, erroneous medical diagnoses, and other negative consequences.
Bias can show up in AI through incorrect and skewed datasets, bad design and access limitations. It can affect every demographic, including different genders and racial groups.
We have two years to make AI a force for good.
By 2025, 90% of all new enterprise applications will include embedded AI functionality, Google’s 2023 Data and AI Trends Report forecasts.
As businesses embrace environment, social and governance (ESG) targets, we can use AI to create a fairer and more equitable society.
Addressing bias in AI won’t be fixed by better coding. It needs a comprehensive approach.
Good governance: Addressing bias in AI requires oversight, accountability and governance around how the system is built, and how training data is collected and used. We need to look beyond technical questions to consider ethical issues around design, development and implementation. Datasets, for example, must conform to diversity requirements.
Continuous monitoring and testing: Once the system is in place, constant observation is essential, particularly for self-learning programs. Metrics need to be put in place to ensure system performance, conformity thresholds and remediation strategies. Similarly, models must be tested regularly, not just before deployment, but also after release.
Diversity of teams: It is essential for companies using AI models to build diverse teams with the skillsets to address technical, human and ethical considerations of the system’s operation. With diversity comes a variety in perspectives, experience and ethical considerations, leading to stronger and more equitable products.
AI has proved its business value. Let’s use it to support humanity.
Automated and AI-infused systems are here to stay. They have proven their ability to promote agility, enhance productivity and efficiency, and spur innovation – while generating new business value.
I believe AI will prove massively beneficial to organizations and to people by enabling once-impossible outcomes. At the same time, we have an unprecedented opportunity to mitigate human biases with algorithms supporting fair and equitable outcomes for everyone, regardless of gender, race or ability.
Without diversity safeguards, AI will always reflect human prejudices. Let’s address gender and other biases in AI today to create inclusive and equitable systems that truly reflect the diversity of our world.
techUK is marching forward to close the tech gender gap in 2023. Throughout March, coinciding with International Women’s Day (IWD 2023) on 8 March, we are exploring how we embrace equitable workplaces. The UN’s theme for IWD 2023 focuses on Digital for All or DigitALL, and we are proud to support this.
For more information, please visit our Women in Tech hub.
Skills, Talent and Diversity updates
Sign-up to get the latest updates and opportunities from our Skills, Talent and Diversity programme.