How to approach AI more responsibly, according to a top AI ethicist

Creative has never been more important in verticals such as finance and health, due to the recent privacy changes. How to make data the backbone for your campaigns All Transform 2021 sessions are available for immediate download. You can watch the video now. Women in the AI field are leading the ethical discussion, making groundbreaking research, and inspiring the next generation. VentureBeat Women in AI Awards was created to highlight the importance of women's voices, work, experience, and shine a spotlight on some of these great leaders. This series features Friday interviews with the winners of this year's Transform 2021 awards. Last week's interview with the AI entrepreneur winner is available here. DataRobot, an enterprise AI platform, claims it creates 2.5 million models per day. Haniyeh Mahmoudian personally invests in making sure that they are ethically and responsibly constructed. VentureBeats Women in AI Responsibility and Ethics Award winner Mahmoudian literally wrote the code. Astrophysicist and data scientist, she became the company's first global AI ethicsist. She has also raised awareness about responsible AI in the wider community. She is also a speaker at panels such as the World Economic Forum and has driven change within her company. Coworker, I cannot emphasize how important her work was in pushing the thinking of our engineers to include ethics in our software. Ted Kwartler (DataRobot VP for trusted AI), was one of the people who nominated her. Mahmoudian's work has been a valuable resource during the crisis. Moderna used her research on risk level modeling for its COVID-19 forecasting. Eric Hargan was the deputy secretary of the U.S. Department of Health and Human Services at the time. Mahmoudian's work was crucial in ensuring that the simulation was fair and unbiased in its predictions. VentureBeat is proud to present this award to Mahmoudian for all she has done. We recently met up (virtually), to discuss her impact and AI regulation. Also, ethics as a buzzword. And her advice on how to deploy responsible AI. This interview was edited to be concise and clear. VentureBeat: What would you say about your approach to AI? What is your driving force? Haniyeh Mahmoudian says: It's all about learning new skills. AI is becoming more integral to our daily lives. It was fascinating to me to discover new uses cases and new ideas when I first started my career as a data scientist. It also gave me the insight that this field is extremely vast. There is a lot of potential there. However, you should be careful about certain areas. VentureBeat: The code that calculates statistical parity in DataRobots platform and provides natural language explanations to users was yours. These codes have been used by companies across a range of industries, including banking, insurance, tech, manufacturing, CPG, and CPG to eliminate bias and improve their models. What does this look and what is its importance? Mahmoudian says: I was very surprised to discover that it was difficult for non-technical people to understand the technical details of how the model behaves when I began my journey towards responsible AI. They must be able to communicate in a language they can understand. However, telling them that your model is biased does not solve the problem. The natural language component of it is useful in helping them to understand that their system may have some biases. Take a look at the data XYZ. Here's what we found. This applies to both the case and the general levels. There are many definitions of bias and fairness. It can be difficult to know which definition to use. We want to ensure you are using the right one. For example, hiring is a case where you would be more interested having a diverse workforce. Equal representation is what you are looking for. In a healthcare scenario, however, representation is not as important as ensuring that patients have access to the model. VentureBeat: VentureBeat is impressed by your efforts to reduce algorithmic bias in models. You have also briefed dozens Congressional offices about the issues and are dedicated to helping policymakers make AI regulations. What importance do you think regulation is in preventing AI-related harm? Mahmoudian, I believe regulations are very important. AI bias is a problem that companies are working to address, but there are still gray areas. It is not standardized and it is uncertain. These types of things require clarification. In the EU regulations, for example, they attempted to clarify what it meant to have a high risk use case. They also tried to outline the expectations (having confirmatory tests, auditing, and other such things). These are the types of clarifications that regulations can provide, which will help companies understand the process and reduce their risk. VentureBeat: Responsible AI and AI ethics are a hot topic these days. It is really, really important. Do you feel that it is becoming a buzzword or do you already fear it? How can we ensure that this work is not just a box or facade? Mahmoudian says: It is used in the industry as a buzzword. However, I believe it is being used as a marketing tool by companies. This is because it actually benefits them. One of their biggest fears is losing customers, as shown by the AI bias surveys. Their brand would be at risk if a headline was published about their company. They are aware of these types of issues. They are also aware that having an AI system and framework in place can help them avoid this kind of risk. They have my full support. They are considering it and are working towards it. It's possible to say it's a bit late, but it is never too late. It is a buzzword but it takes a lot of effort. VentureBeat: What is often overlooked in discussions about ethical and responsible AI technology? What is more important? Mahmoudian says that sometimes people will directly connect ethics to fairness and bias when they are talking about ethics. Sometimes it can be seen as one group pushing their ideas on others. This is why I believe we should remove bias from the ethics process. It's not about individual bias, it's about the entire process. It can be detrimental to your customers if you create a model that doesn't work well, and then your customers use it. This might be considered unethical by some. There are many ways to include ethics and responsibility in different aspects of the AI/machine learning pipeline. It is important that we have this conversation. This conversation is not limited to the final stage of the process. Responsible AI should be integrated throughout the entire pipeline. VentureBeat: What advice would you give to companies deploying AI technology? How can they approach it responsibly? Mahmoudian: Get to know your process and create a plan. Every industry and every company has its own criteria and project types. You should choose the dimensions and processes that are most relevant to your work. This will guide you through the entire process.


Post a comment

Your email address will not be published. Required fields are marked *