Women must demand a seat at the table to shape the development of Artificial Intelligence, which promises to be the biggest game-changing technology in 25 years. 

 How AI models are taught and the information used to develop their models will determine whether it produces real-world outcomes that benefit all sections of society rather than skewing outcomes towards a select few. 

Just as small children are shown by their parents how to function in the world and how to react to certain situations, large language models will develop in the mould of their makers with outcomes based on the information used to inform their learning. 

This makes inclusion and diversity paramount when it comes to seats around the table of AI’s development. 

If AI large-language models are developing using skewed information, then AI risks being unable to function effectively in the real world. 

Given the IT industry remains a male dominated domain, this should be a clarion call to women to enter the industry and to help shape the future. 

While the risk of inherent bias being passed to AI models is a concern, I am cautiously optimistic about AI’s potential.  

In health alone it promises to revolutionise detection and treatment of skin cancer and breast cancer by quickly processing huge tracts of information, speeding diagnosis, and increasing the chances of recovery. 

In professions such as law and medicine, for example, it promises to lift productivity by freeing skilled workers from highly repetitive tasks and increasing their opportunity to perform higher-value tasks. 

But this depends on minimising human cognitive biases, which have hindered the effectiveness of earlier iterations of AI systems. Diversity and inclusion are the key antidote to this risk.  

Data sets that feed into large language models for machine learning, and AI need to be truly reflective of the communities and subjects they are meant to represent.  

Just like in a business, the more views that are around the table from diverse perspectives, the better the chance of getting a balance and innovative outcome. 

There are examples in the past where there have been negative real-world outcomes because the inputs to models have fallen short in the diversity stakes. 

One basic example from decades ago was that women were less likely to be told to go to the emergency room straight away if they were having issues that might be a heart attack. 

This was because the protocols were based on data that was mainly collected from men, which may have suggested that common women’s symptoms, such as pain in the left arm and back, may be due to depression rather than heart attack.  

The problem was men and women show different symptoms of heart attack. Common men’s symptoms such as sweating, chest pain or tightness, heartburn, nausea, and vomiting were typically regarded as symptoms of heart attack. 

Women’s symptoms such as jaw pain, shortness of breath, chest discomfort or tightness, backache or extreme fatigue and nausea may not have been regarded as symptoms of immediate heart attack under this now discredited set of protocols.  

Another example can be found in skin cancer detection.  

Early iterations of an AI called CNN, used to detect skin cancer using data from the International Skin Imaging Collaboration, were found to be less likely to identify cancer lesions in patients of colour or misdiagnose non-existent skin cancers in patients of colour.  

This was because its algorithm was trained on a dataset primary from fair-skinned populations in the US, Australia, and Europe.  

If an algorithm is not trained with diverse images, then basically the algorithm doesn’t work. 

So, diversity matters in the development of AI, just as increased gender diversity matters to the IT industry. 

When I first started in the industry 25 years ago, I was typically the only women in the room. People would be afraid to knock on the door and talk to the ‘IT Gods’. 

As more women have entered IT over the years, IT teams have become a lot more user-focused both in terms of technical support and software development.  

Part of that is because technology has come a long way. There is more memory on computers.  

But a lot of that has also come down to the inclusion of women who are often more focused on the end-user experience and understanding the diversity of how this application is going to be used by people other than your typical young white male IT worker.  

Stakeholder management and change management has matured in leaps and bounds over the past 10 to 15 years and some of that is because women usually hold those roles. This is a critical IT capability to ensure that new IT services are properly rolled out to a user group.  

However, many women remain reticent to speak up in male dominated meetings and brainstorming sessions. Going forward, senior women must support their junior cohorts to speak up and have their opinions considered in the mix.  

Nowhere is this more important than in the development of AI.