#ronbrachman

Machines like Us: TOWARD AI WITH COMMON SENSE

Soundcloud link: https://soundcloud.com/bridging-the-gaps/machines-like-us-toward-ai-with-common-sense-with-professor-ronald-brachman?si=ea1cef08fb4c4c61bd9a90670a789eb8&utm_source=clipboard&utm_medium=text&utm_campaign=social_sharing

The discussion starts about how the guest Prof. Ronal Brachman sees the growth of AI but in a narrow way, in the sense performing singular tasks and not in a general way. The discussion went on about how AI can play chess and checkers. Because the computation power in those days, the tasks were aimed at doing single tasks, those were the starting point of systems that we have now to name a few, self-driving cars, personal assistance and so on. And the conversation started to dig deep on computer vision, deep learning and the data available to train these complex algorithms on important tasks.

Move from general AI approach – ultimate goal is to develop self-controlling AI and to develop human level intelligence. Interesting example given by the guest was sending robot to mars (mars rover). With the current robots that can’t be re-tasked when found something interesting like a metal in mars. More discussion on autonomous AI, where there is a level of knowledge and reasoning for the actions that carried out by the AI. Example about autonomous cars on making decisions on the go was discussed.

Failures on AI: Alexa gave a challenge to 10-year-old to plugging a phone partially and insert a penny between the probes, which suggest the lack of common sense or failure of AI. And another example was discussed on AI bot for suicide prevention, where the bot encouraged the person to go ahead with the suicide thought.

More insight on common sense in humans, where we remember or forget things, but comes from experience. And common sense on machines comes from the same, as an example when a cake was hidden in a washing machine for surprising someone, the AI should be intelligent enough not to turn it on, which was a great example!

The conversation went on about how in 30 to 40 years experts say that we will have fully generalized AI, our guest's thought is that it might be not possible but will take more time to get there. The discussion ended on McCarthy's programs with common sense, the ability to take responsibility and decisions outside the guidance, the advice taker.

Things that I learnt from the podcast:

Developing AI that is more autonomous and logical. Can make own decisions that don't cause any harm around.

Reinforcement learning on AI, training on the fly and experience where to take responsibility of their actions

Dangers on AI making mistakes that humans can’t predict, which could be destructible in a massive way. Which taught me as a data scientist that I should be more careful on the systems that I will be developing.

About the Guest: Professor Ronald Brachman (http://www.brachman.org/mybackground.html)

Author of Machines like us – towards AI with common sense

Pioneer in the field of Artificial Intelligence for more than 40 years. he served as President of Association for the Advancement of Artificial Intelligence from 2003 – 2006.  Before that he was the Secretary-Treasurer of International Joint Conferences on Artificial Intelligence, Inc for 9 years. He is currently co-editing the Synthesis lecture series on AI and ML. He was also a member and Treasurer of the Board of Directors of the Computing Research Association. He was an advisor for start-up called Segovia from 2015 – 2019.

He was the Chief Scientist of Yahoo and the Head of Yahoo Labs from 2012 - 1016.  Prior to that, starting in September of 2005, I was the Associate Head of the Labs and essentially its Chief Operating Officer.  before joining Yahoo in 2005, he was the Director of the Information Processing Technology Office at DARPA, the Defense Advanced Research Projects Agency.  The Personalized Assistant that Learns program that he created with help from Zach Lemnios led directly to the technology that Apple now offers in the form of Siri on its iPhone products.