Why AI should also stand for All Included
Some time ago my attention was drawn to a blog on the World Economic Forum website, written by Nahia Ordu?a, Senior Manager in Analytics and Digital Integration at Vodafone. She pointed to the importance of diversity in AI-driven companies. A topic – I realised – that often remains underexposed. And that’s a pity, because the foundation we lay today will determine the AI applications of the future.
At Vodafone, inclusion is one of the strategic pillars. The telecom company states that they understand the risk that technology can create, but are working hard to build a digital future that works for everyone. It’s an attitude that all AI-driven companies should adopt. Why? Because we live in a diverse world and because it simply pays off.
The big market impact of AI
Today, innovative, digital technologies are transforming the world as we know, but AI has a changing power that goes beyond our imagination. From smart products, such as drones, to virtual agents, decision support, predictive analytics and even decision automation. A recent study of Fortune Business Insights shows we are still at the beginning, with prospects on a tenfold increase of the AI market size in the coming years: from USD 20.67 billion in 2018 to USD 202.57 billion by 2026. According to the same study, the growth is mainly driven by an increased adoption of cloud-based applications and services and an upswing in the connected device market. Besides, the big investments in 5G technology and demand for intelligent virtual assistants are boosting the AI market development.
However, as with many new technologies, there is a weak spot: the unconscious bias. Bias sneaks in via two ways. Firstly, AI is only as good as the data that it is trained to analyse, which can include pre-existing bias. Secondly, bias can be imported during the coding process by the IT engineers.
That’s where the importance of diversity in tech comes into play.
Biased AI algorithms
Nadia Ordu?a refers in her blog to the example of Amazon, that had high hopes for the AI-powered recruiting engine it developed in-house over several years. However, the project backfired. Amazon discovered that its machine learning models were introducing bias and favoring male candidates because it had been trained on a male-dominated pile of resumes submitted to the company over a 10-year period. The old paradigm ‘garbage in, garbage out’ applies here.
But also the gender diversity in tech – which I wrote about in one of my earlier blogs – has an impact on AI development. The World Economic Forum Global Gender Gap Report 2020 reveals that just 26% of Data & AI professionals are women. If we look at the big tech giants such as Google or Facebook, than just 10 to15% of the AI professionals are female (figures of 2018). That algorithms are largely developed by men often subtly reflects into the end product. A study from Microsoft Research and Boston University researchers, for example, found that word embeddings trained on Google News articles exhibited gender stereotypes. More specific: a nurse will be indicated as a ‘she’, while a doctor will be identified as a ‘he’.
Although, diversity goes beyond gender and also refers to race, culture,… MIT Media Lab, for instance, researched three major facial recognition systems, using a dataset from dermatologists. The study unveiled that white males are recognised in more than 99% of the cases, while the accuracy for black males and white females is – depending on the system – between 88% and 99%, and for darker females the chance to be recognised is less than 80% in all systems.
If technology developers are all similar in profile - think white male - then they are likely to unconsciously produce the same kinds of rationale and products. And when AI eventually gains control of greater functions in our everyday lives, we will be ruled by the biases inherent in its coding.
Diversity guarantees future success
Companies must be aware of and recognise that algorithms are not neutral, but created by humans with biases and beliefs, and make every effort to eliminate those biases. For example, by coaching their management to identify bias, but also by stimulating diverse teams, certainly in the tech development.
And if striving for a better place for everyone to live and to remove inequality is not enough to convince your organisation, then think about the following: diversity also pays off economically. People with different backgrounds have different experiences and can come up with different ideas. A McKinsey study even revealed the link with the profit of a company. They compared more than 1,000 companies worldwide and the ones in the bottom quartile for both gender and ethnic/cultural diversity were 29% less likely to achieve above-average profitability than all other companies in the data set.
Inclusion of highly diverse individuals, in the myriad ways in which diversity exists, can thus be a key differentiator among companies.
Will you use the All Included formula for your next AI project?