This article was written by DataRobot and originally appeared on the DataRobot Blog here: https://www.datarobot.com/blog/deliver-ai-you-can-trust-with-datarobot-discover-transparent-ai/
Artificial intelligence, machine learning, and deep learning are powerful technologies that have created endless possibilities. They contribute to AI chatbots, have brought us self-driving cars, and help translate documents into human-like speech in dozens of languages. For many narrow applications, we have become accustomed to superhuman performance, such as in the case of AlphaGO beating the world’s best human player, or IBM’s DeepBlue beating Garry Kasparov. We also celebrated the recent win of AlphaFold being able to outperform previous efforts on protein structure prediction, a very exciting win for the medical research community.
But AI is a tool that can amplify both our best and worst decisions, and as such, we need to handle it with care. The world is full of examples of AI’s potential hazards, from self-driving cars that don’t recognize road hazards to natural-language processing algorithms that create startlingly real – and completely bogus – stories and articles. In the case of recruiting and building models from job applicants, our implicit biases can surface and become amplified by AI if we aren’t proactively working to identify and prevent this. We have seen high profile examples of this in the news where a resume model was sexist and gender classifiers performed poorly with Black women.
At DataRobot, we’re committed to AI you can trust, offering expertise and tools to test your systems across multiple dimensions of trust. That allows you to design AI that performs exceptionally, maintains operational excellence, and reflects your values. The goal is to strive to be proactive and catch problems early.
We build our approach to AI you can trust on three key principles:
- Performance. Our platform includes guardrails that ensure top performance and democratizes AI so that any business can take advantage of its value safely.
- Operations and Reliability. A model is only as reliable as the system that it is deployed on. DataRobot’s MLOps allows for ongoing monitoring of deployments, and Humble AI allows a model to identify when it is not confident in its predictions. It creates “triggers” that proactively prevent potentially faulty AI decisions. Feature drift is a real threat for model deployments; many build-it-yourself applications lack this type of support and protection. This is a must-have for anyone deploying models that matter.
- Ethics and Explainability. Our AI platform is not a black box. It includes dozens of tools that allow technical and non-technical users alike to explain and understand AI models with transparency so that you can ensure that your models align with your company’s values.
We see AI as a co-pilot for your decision-making, not a replacement for it. One analogy we like to use is that of medical imaging. Today’s sophisticated imaging devices – combined with AI – can be powerful diagnostic tools. But it’s important to have human expertise working in tandem with AI, both to validate or correct the AI’s predictions and to determine next steps. Before implementing AI, many processes in an organization are stale and haven’t changed for many years. With model insight and explainability, we find that subject matter experts can work in tandem with AI to improve and innovate on new feature ideas. AI enables process evolution and improvement and helps increase human creativity.
We’re also aware of the risks to privacy that AI can present. What if an insurance carrier was able to closely track a policy holder within a few feet of their actual location, and noticed that the holder’s car was parked near a bar at midnight? Could they use that information to increase premiums or even cancel a policy? Just as AI is rapidly evolving and changing, so are the new ethical dilemmas we must face and discuss. On our More Intelligent Tomorrow podcast, we invite thought leaders in AI Ethics, such as Thomas Chamorro-Premuzic and Fiona McEvoy who was recently listed as one of the top 100 most influential women in ethics.
We’re committed to building AI that helps an enterprise solve tough problems in ways that are ethical, fair, and trustworthy. Our AI tools are working in industries such as manufacturing, financial services, and retail, and many others, providing best-in-class accuracy and insight into potential anomalies. Our customers rely on us to enable their most important predictions where accuracy and reliability matter most. Drawing on the combined experience of our customers, partners, and employees, we have learned through practical experience the importance of proactive trust measures, and this experience helps to protect our customers from the unexpected model issues that arise.
To learn about DataRobot’s commitment to ethical AI, watch the replay of a live event we recently hosted on LinkedIn, featuring our VP of Trusted AI, Edward Kwartler as he talked about Trust, Ethics, and AI.
Also, be sure to visit our “More Intelligent Tomorrow” series of podcasts. These weekly conversations with AI innovators cover technology trends, best-practices for AI, career advice, and more.Finally, visit our AI You Can Trust page, which outlines our approach to AI and offers resources that help you understand why you’ll want to work with us.