Standing at the intersection of computer science, robotics, and electrical engineering, Elaheh Ahmadi’s passion for enhancing intelligence in robotics has led her into the world of artificial intelligence (AI). Ahmadi’s academic journey saw her delve into a range of disciplines, not limiting herself to one. This holistic approach, involving an intertwining of computer science and hardware creation, allowed her to gain substantial knowledge about an understudied area of AI – its reliability.

In the sphere of robotics, Ahmadi points out, there are more often than not serious deficiencies in terms of intelligence. This led her to the realization that the implementation of AI methods could be a potential solution for taking robotic applications from ‘dumb’ to smart. The perspective invites Ahmadi, during her undergraduate studies at MIT, into the realm of AI. From there, her fascination with artificial intelligence only grew.

However, the gigantic machinery involved in AI operations, which involves millions to billions of parameters, increases the opaqueness of the models. The inability to decipher when these models are reliable became the intended focal point for Elaheh during her master’s research. 

Now Ahmadi and her workforce at Themis AI, of which she is a co-founder, have set out to make these complex models more conceivable. They diagnose the particular causes of inefficiency in AI models, determining whether it is the model or the training data triggering its failure. The mission is to help corporate partners understand their models and to develop mechanisms to rectify the aberrations.

Differentiating from competitors, Themis AI offers tools not just for monitoring but for rectifying these anomalies. Other enterprises may help with identifying issues over time, but rectifying them in real time to save client companies from profound losses is where Themis AI takes the lead.

In addition to this, Ahmadi has a keen interest in what she refers to as ‘generative AI applications.’ The future of AI, as she envisions it, is in personalized applications. Rather than having colossal general models, she expects to see more emphasis on personalized experiences. In essence, an AI assistant retains information about an individual, enabling it to make informed decisions on their behalf.

Whether it entails designing a dietary plan based on one’s genetic makeup, assisting in creating workout regimes, or taking over labor-intensive tasks, the possibilities are endless. The only obstacle that hampers such initiatives, according to Ahmadi, is reliability. AI models are still too unpredictable, and failures can occur without any prior warning.

To counteract this uncertainty and to make AI more consistent and trustworthy, Ahmadi and her team are hard at work at Themis AI. The goal is not only to alert companies to potential failures in real-time but also to help them understand the reasons behind these failures. Their ambition is to remove the uncertainty surrounding AI’s reliability, ultimately making it a more dependable asset in various fields.