Why AI cannot be explained | Rickard Brüel-Gabrielsson | TEDxBoston

We derive a fundamental uncertainty principle of meaning that links precision to complexity. Demonstrating that power and precision within AI always comes at the cost of explainability. Thus, great advances like ChatGPT or Dall-E must come with close to trillion parameters. Indeed, explainability has never been a viable path forward for human-like intelligence, but how did we build trust without it? We use the tool of evolution: the test of time. Giving AI the chance to earn our trust through the test of time. Our great challenge lies in applying rigorous tests of time to this new AI as we are advancing almost a billion times faster than evolution did in creating human intelligence. We're reaching a pivotal moment. In fact, this could be our greatest test of time yet.
MIT researcher specializing in foundation models and generative AI, previously at Stanford. Serial entrepreneur and co-founder of Unbox AI, a leading company innovating in foundation models and generative AI for behavior-driven businesses. This talk was given at a TEDx event using the TED conference format but independently organized by a local community. Learn more at www.ted.com/tedx

[@ChituOkoli]

It sounds like he is equating explainability to trust. While trust is important, there is a lot more to XAI than just trust. A huge part of XAI is actionability--receiving guidance for what we might change in the data to get the outcomes that we want. While that might require some trust in the model, that's not the main point--we don't need the test of time to increase that kind of explainability. It is relevant right now. So, AI CAN be explained from that perspective.

[@KellieAguado]

I'd like to know if anything has changed from the last time he gave this talk or maybe what are his predictions for XAI in 2024.

[@sorjef]

Thank you for the talk. With the questions discussed it gives an opportunity to ponder and build some greatmental models around the concept. Also recommed this author's Future of AI set of lectures!

[@Red_Blue_Green]

Thank you so much Artem!! :D

[@sajeeshsaji4623]

Wow it's empty 🤣🤣🤣

[@sajeeshsaji4623]

I meant the comment box