**Machine Learning Conference Leads Physicist to Ethical Awakening**
In 2017, Savannah Thais attended the NeurIPS machine-learning conference in Long Beach, California, hoping to learn about techniques she could use in her doctoral work on electron identification. However, what she encountered at the conference transformed her worldview.
During the conference, Thais listened to a talk by artificial intelligence researcher Kate Crawford, who discussed bias in machine-learning algorithms. Crawford highlighted a study by Joy Adowaa Buolamwini, which revealed that facial-recognition technology had picked up gender and racial biases from its dataset. Women of color were 32% more likely to be misclassified by the technology compared to White men.
This study became a landmark in the machine-learning world, shedding light on how seemingly objective algorithms can make errors based on incomplete datasets. For Thais, who had a background in physics, this was a pivotal moment that led her to pivot towards studying the ethical implications of artificial intelligence in science and society.
**Trusting Your Models**
Traditional computer models rely on human input to make decisions, while machine-learning algorithms learn parameters from the data they are given. These algorithms can come up with millions of parameters, each with its own “phase space,” which determines their usefulness for a specific task.
However, algorithms can exhibit bias based on the dataset they are trained on. Buolamwini’s thesis highlighted how a facial-recognition algorithm, trained on mostly photos of White men, struggled to differentiate between individuals of other ethnicities and genders.
This bias in facial-recognition technology can have far-reaching consequences, from denying access to services to misdiagnosing individuals in medical settings. Thais emphasizes the importance of understanding how algorithms work across different demographics as both a scientific and ethical question.
**Responsibilities and Opportunities**
Physicists, who have a history of ethical dilemmas such as the creation of nuclear weapons, are now facing similar challenges with artificial intelligence. Thais advocates for incorporating ethics into the training of physicists in machine learning to raise awareness of biases and ethical considerations.
By integrating ethics into their work, physicists can not only improve the use of machine learning in society but also enhance the science of machine learning itself. Thais suggests that physics data, with its controlled nature and quantifiable biases, provides a perfect platform for building unbiased models.
In conclusion, Thais stresses the importance of physicists engaging in ethical discussions surrounding machine learning and its implications. By incorporating ethics into their thinking, physicists can contribute to a more ethical and unbiased use of artificial intelligence in both science and society.