The rise of artificial intelligence is already making an impact in industries as diverse as transport, healthcare and manufacturing. However, tech leaders agree that AI is just starting to develop and it will soon be able to do much more. At this relatively early stage of growth, experts like Tariq Khokhar, Head of Data for Science and Health at the Wellcome Trust, are working to ensure that AI is as fair, equal, and diverse as possible.
In his talk at Tech Show London, Tariq points to the Human Genome Project and the resulting fall in the cost of sequencing one human genome. This cost has seen a staggering decrease from around $100 million to around $100, making this as one of the many exciting developments happening at the intersection between AI and science.
“That’s the trend that exceeds Moore’s Law, which we’re seeing across biology. There has never been more data or more insight we can get into the sort of building blocks of life. But we need to make sure AI is working well in science and health. There is a lot of data and it needs to be thought about carefully. It’s not just about quantity of data, it’s really about quality,” said Tariq.
> Apply to be a speaker at Tech Show London 2024
The Importance of Diverse Data and Talent for AI
Many of the datasets used to train AI tools are believed to include biases that make these solutions perform ineffectively.
Tariq highlighted the work done by the Algorithmic Justice League to combat data bias. He also pointed to research by Massachusetts Institute of Technology and Stanford University that found when guessing a persons gender from their face, AI had an error rate of 0.8% for lighter-skinned men, compared to a 35% error rate for darker-skinned women.
“This is just one example of many biases that you get in AI systems. Of course, the root source of this is a lack of diversity in the training data. If you’re not training these models on appropriately diverse data, this is the kind of stuff that happens. It’s not all about minimising harms. If you are not thinking about the diversity of data, you are also going to miss potential discoveries,” added Tariq.
Beyond just including a range of diverse people in data that is used, different types of data also need to be included. For the future of AI, the leaders tasked with making decisions about AI should also be skilled in data and come from diverse backgrounds.
How can AI Reach its Full Potential?
When the right data sets are achieved, more still needs to be done if AI is to reach its full potential. Tariq suggested that professionals need to ask whether the AI systems actually have a real benefit. He referenced meditation apps used on smartphones that have surged in popularity due to the increasing importance of mental health.
“But right now, there is almost no regulation or evaluation of whether these things work or not. What is the correct dose of headspace to alleviate depression and do these apps actually make a difference,” Tariq asked.
In his concluding remarks, Tariq made two key recommendations. First, diverse data and talent is critical for the success of AI in science and health. Second, developers and users must consider whether the AI systems are useful and helping people’s lives.
“If we can put all those things together, I think the future for AI in science and health is really bright,” said Tariq.
