Internship Blog on Bias in Artificial Intelligence
Authored by Tenzin Yangzom and Rithik Sam. Supervised by Dr Chih Wei Teng.
The potential of AI revolutionising drug development is immense. Once considered far-off dreams, AI can enable transformative possibilities such as slashing development timelines and expenses by up to 70%. These efficiencies can only be gained by harnessing AI's ability to analyse large amounts of data and reveal intricate patterns, improving every step of the drug discovery process, from identifying targets to optimising clinical trials. Yet, amidst this transformative potential, there is a challenge: the issue of biases in AI systems. If not adequately addressed, biases in AI can result in unequal treatments that will affect healthcare delivery and potentially endanger lives.
We delve into the multifaceted impact of AI in biopharma, examining its transformative potential, current applications, and challenges. At the heart of AI's transformative power lies its capacity to analyse vast and diverse datasets with unprecedented speed and precision. In drug discovery, this translates to accelerated timelines, significantly reduced research and development costs1 and, most importantly, the identification of novel therapeutic targets previously hidden in the data deluge.
Through predictive modelling and simulation, AI offers researchers valuable insights into a drug's effectiveness, safety profile and potential side effects, which enables them to make more informed decisions and focus on the most promising candidates for further investigation2.
Traditionally, drug discovery has been hindered by lengthy timelines, high expenses and a high failure rate. AI technologies can speed up drug discovery processes up to ten times compared to traditional methods. For instance, AI-driven autonomous labs have demonstrated the ability to conduct experiments 100 times faster than conventional approaches3,4. Researchers at the University of Oxford, utilising generative AI and surrogate simulations, identified and tested 20 antimicrobial candidates in just 48 days—a significant improvement over traditional drug discovery timelines, which can take years3. Furthermore, AI has the potential to transform this pattern by streamlining procedures, optimising resource management and reducing risks associated with clinical trials.
Some examples include emerging companies such as Atomwise, which is transforming small molecule drug discovery by focusing on challenging targets and utilising its AtomNet platform for structure-based drug design. Similarly, companies such as Insilico Medicine and Exscientia are also making significant strides. By leveraging sophisticated algorithms and machine learning techniques, AI sifts through genetic data, molecular structures, and clinical trial outcomes to identify promising drug candidates and biomarkers, paving the way for more effective treatments for various diseases. Insilico Medicine utilises AI to discover and design drugs for idiopathic pulmonary fibrosis (INS018_055) and COVID-19, with INS018_055 being the first entirely AI-discovered and AI-designed drug to enter a phase 2 clinical trial5.
However, the rapid advancements in AI also bring to light critical issues of bias. A compelling example of AI perpetuating bias emerges in the underrepresentation of women in clinical trial data, particularly for cardiovascular drugs. A study by the National Institutes of Health found that AI systems trained on such biased datasets could perpetuate this gender bias, potentially leading to less effective treatments and higher risks of adverse effects for female patients6. This issue extends far beyond gender, affecting various demographics. Clinical testing often underrepresents specific populations based on race, ethnicity, gender, and age, undermining universal health coverage7. Consider paediatric versions of adult medicines for diseases like HIV/AIDS and antibiotics- these often face significant delays8. AI models trained on biased data may exacerbate these disparities, widening the lack of understanding of how these drugs affect children and other underrepresented groups and furthering healthcare inequalities.
The bias in AI is not just confined to the biotech sector. Bias in AI is a well-documented problem across various fields. For instance, in the criminal justice system, AI algorithms used for predicting recidivism have been shown to exhibit racial bias. A study by ProPublica found that the COMPAS algorithm, widely used in the United States, was biased against African Americans, wrongly flagging them as high risk at nearly twice the rate of white defendants9. Similarly, facial recognition technologies have been criticised for their higher error rates in identifying people of colour compared to white individuals10. In another prominent case, Amazon scrapped its AI recruiting tool because it exhibited bias against women, favouring male candidates for technical roles due to being trained on resumes submitted predominantly by men over ten years11.
It is crucial to implement de-biasing strategies within AI systems. One effective method is reinforcement learning, where AI models are trained on diverse datasets and continuously adjusted based on feedback to reduce bias12,13. This approach involves creating algorithms that identify and correct biased patterns in the data and learn to avoid these biases over time. By incorporating reinforcement learning, AI systems can become more equitable, ensuring their benefits are distributed more evenly across different populations.
The McKinsey Global Institute (MGI) predicts that AI has the potential to create economic value ranging from US$60 billion to US$110 billion for the pharmaceutical and medical product sectors14. This significant economic benefit stems from AI's ability to increase productivity, speed up the discovery of drug compounds, expedite development and approval processes and improve marketing strategies. Beyond cost reduction, the impact extends to transforming healthcare itself. Envision a future where AI does not hasten drug development but also results in tailored patient care and accelerates disease eradication like never before. While this vision is achievable, it requires significant consideration of ethical and practical hurdles.
Incorporating AI in biopharmaceuticals offers promise for revolutionising drug discovery and development by providing unmatched speed and efficiency. The potential of AI to revolutionise drug discovery is not just promising; It's inspiring. It opens up a world of possibilities and opportunities that can transform the field of biopharmaceuticals that could lead to a healthcare revolution. Nonetheless, biases within AI systems could hinder progress. Overcoming these biases is essential to ensure that AI-driven advancements make effective healthcare solutions accessible to all population segments. Developing AI systems that are impartial and trained on diverse datasets is critical. Any missteps in this realm could introduce challenges that undermine the progress we strive for.
Reference:
1. Starmind. (2024, April 30). How leading pharma companies use AI to reduce the cost of R&D. https://www.starmind.ai/blog/how-pharma-companies-use-ai-to-reduce-the-cost-of-rd#:~:text=Drug%20discovery%20and%20development,spent%20and%20overall%20R%26D%20spend.
2. Hafke, T. (2023, December 16). AI in Biopharma: Use Cases and Key Considerations. AlphaSense. https://www.alpha-sense.com/blog/trends/ai-in-biopharma/
3. Biopharma Thought Leaders: How AI is accelerating and transforming drug discovery. (n.d.). Nature. https://www.nature.com/articles/d43747-023-00029-9
4. Hopper, L. (2023, April 27). How AI might speed up the discovery of new drugs. Physorg. https://phys.org/news/2023-04-ai-discovery-drugs.html
5. Shah-Neville, W. (2024, April 19). Five AI drug discovery companies you should know about. Labiotech.eu. https://www.labiotech.eu/best-biotech/ai-drug-discovery-companies/
6. EUGenMed, Group, C. C. S., Regitz-Zagrosek, V., Oertelt-Prigione, S., Prescott, E., Franconi, F., Gerdts, E., Foryst-Ludwig, A., Maas, A. H., Kautzky-Willer, A., Knappe-Wegner, D., Kintscher, U., Ladwig, K. H., Schenck-Gustafsson, K., & Stangl, V. (2015). Gender in cardiovascular diseases: impact on clinical manifestations, management, and outcomes. European Heart Journal, 37(1), 24–34. https://doi.org/10.1093/eurheartj/ehv598
7. Woo, M. (2019). An AI boost for clinical trials. Nature, 573(7775), S100–S102. https://doi.org/10.1038/d41586-019-02871-3
8. Penazzato, M., Lewis, L., Watkins, M., Prabhu, V., Pascual, F., Auton, M., Kreft, W., Morin, S., Vicari, M., Lee, J., Jamieson, D., & Siberry, G. K. (2018). Shortening the decade‐long gap between adult and paediatric drug formulations: a new framework based on the HIV experience in low‐ and middle‐income countries. Journal of the International AIDS Society, 21(S1). https://doi.org/10.1002/jia2.25049
9. Angwin, J., Larson, J., Kirchner, L., Matt, S., & Propublica. (2023, December 20). Machine bias. ProPublica. https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing
10. Najibi, A. (2020, October 26). Racial discrimination in face recognition technology - Science in the news. Science in the News. https://sitn.hms.harvard.edu/flash/2020/racial-discrimination-in-face-recognition-technology/#:~:text=Face%20recognition%20algorithms%20boast%20high,and%2018%2D30%20years%20old.
11. Dustin, J. (2018, November 11). Insight - Amazon scraps secret AI recruiting tool that showed bias against women. Reuters. https://www.reuters.com/article/world/insight-amazon-scraps-secret-ai-recruiting-tool-that-showed-bias-against-women-idUSKCN1MK0AG/
12. Stokel-Walker, C. (2024, February 20). ChatGPT replicates gender bias in recommendation letters. Scientific American. https://www.scientificamerican.com/article/chatgpt-replicates-gender-bias-in-recommendation-letters/
13. What is Reinforcement Learning? - Reinforcement Learning Explained - AWS. (n.d.). Amazon Web Services, Inc. https://aws.amazon.com/what-is/reinforcement-learning/
14. Viswa, C. A., Bleys, J., Leydon, E., Shah, B., & Zurkiya, D. (2024). Generative AI in the pharmaceutical industry: Moving from hype to reality. In McKinsey & Company. https://www.mckinsey.com/industries/life-sciences/our-insights/generative-ai-in-the-pharmaceutical-industry-moving-from-hype-to-reality