AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
Challenges in Training Generative AI Models
The chapter explores the complexities of training generative AI models, discussing issues such as bias in training data, lack of common sense, and unpredictability in responses. It emphasizes the crucial role of human intervention in ensuring the reliability and value of AI outputs, highlighting the risks of errors and misinformation without oversight.
On this episode of Looking Outside we explore the reality and risk behind the hype of AI, with Executive Director of the Institute for Experiential AI at Northeastern University, Founder of Open Insights, Data Scientist and AI expert, Usama Fayyad.
Usama has been in the field of AI for three decades and has lived through three AI hype peaks followed by three winters. In this conversation he contextualizes how the current infatuation with the ‘eloquence’ of data-driven AI stacks up.
Usama first points to an important problem with AI - not with the algorithm but with the data sets that inform the outputs. As the first person to ever hold a Chief Data Officer title, Usama stresses how critical it is to scrutinize the data sets that are feeding the algorithm, as these large data sets are really the breakthrough in this wave of AI, he says, not the machine learning advancements, and the data is filled with errors.
Jo and Usama discuss the onus of the user in not over relying on the AI for our thinking, as the risk here is equally in erroneous output as it is in missing the ‘true contribution’ behind the source material. Usama puts this plainly to say AI has the potential to speed up banal tasks but can, for other tasks, be completely inappropriate – particularly when these require critical thinking and finding what’s between the lines. The algorithm is auto completing answers based on information fed into it: that information may be incorrectly summarized, incompletely inputted, biased, misrepresented or just plain incorrect.
Usama says the user must be aware and in control, because at the moment most generative AI tools are like black boxes that hold things nobody understands. And when the AI gets it wrong, it’s up to us to catch the mistakes, otherwise a world of hurt in the form of misinformation, misrepresentation and perpetuation of bias lies ahead.
>>>
Jump to key points in the episode:
>>>
To look outside Usama works to catch his own bias – bias built on his personal experience, training, and business objective. He does this by taking a situation and trying to see its effect on someone's live he wouldn't normally consider. This forces him to consider interesting aspects he wouldn't otherwise like social and ethical impacts that may arise. He marries this with talking to people, specifically in asking questions he knows the answer to, seeking to understand why a different answer may be given.
>>>
Usama joined Northeastern University Khoury College of Computer Science as Professor of the Practice, and the Office of the Provost as the Inaugural Executive Director of the Institute for Experiential AI. He continues as Chairman of Open Insights, a company he founded as a technology and consulting firm in 2008 after leaving Yahoo! to enable enterprises to get value out of their data assets and optimize or create new business models based on the new evolving economy of interactions.
He was the first person to hold the Chief Data Officer title when Yahoo! acquired his second startup in 2004. At Yahoo! he built the Strategic Data Solutions group and founded Yahoo! Research Labs. He has held leadership roles at Microsoft and founded the Machine Learning Systems group at NASA’s Jet Propulsion Laboratory.
Usama has published over 100 technical articles on data mining, data science, AI/ML, and databases. He holds over 20 patents and is a Fellow of both the AAAI and the ACM. Usama earned his Ph.D. in Engineering in AI/Machine Learning from the University of Michigan, Ann Arbor. He is an active angel investor/advisor in many early-stage tech startups across the U.S., Europe and the Middle East.
>>>
Looking Outside is a podcast dedicated to exploring fresh perspectives of familiar business topics. The show is hosted by its creator, Joanna Lepore, consumer goods innovator and futurist at McDonald's. Find out more at looking-outside.com.
Connect with Jo and join the Looking Outside community on LinkedIn.
>>>
All views are that of the host and guests and don’t necessarily reflect those of their employers. Copyright 2024. Theme song by Azteca X.
Listen to all your favourite podcasts with AI-powered features
Listen to the best highlights from the podcasts you love and dive into the full episode
Hear something you like? Tap your headphones to save it with AI-generated key takeaways
Send highlights to Twitter, WhatsApp or export them to Notion, Readwise & more
Listen to all your favourite podcasts with AI-powered features
Listen to the best highlights from the podcasts you love and dive into the full episode