I really enjoyed this conversation with Regina. Here's the outline: 0:00 - Introduction 0:48 - Literature 5:02 - Science personalities and ideas 8:37 - Medice and computer science 11:49 - Breast cancer and facing mortality 17:58 - Machine learning - detection of and curing cancer 23:25 - Lack of medical datasets 26:54 - Data privacy, value, and future 40:52 - Open problems in application of AI in medicine 50:06 - Natural language processing 54:54 - Language understanding and deep learning 1:02:42 - Human-level intelligence 1:05:41 - Neuralink and augmenting human intelligence 1:09:06 - MIT Introduction to Machine Learning course 1:13:40 - Meaning of life
Favorite quote from this one: ""ideas on their own are not sufficient, and many times... it's the personalities and their devotion to their ideas... [that] changes the landscape". So true.
I love how honest and humble you are when you don`t understand something. Even though you are by far one of the smartest interviewers I have listened to, you don`t hesitate to show and press on when there is something that you are not familiar with (the term screening in this interview). In the process of your own learning, you are teaching millions of other people invaluable knowledge. I have been involved in cancer research for two years now and i thought i knew everything about screening, listening to you and this awesome interviewee made me realise that I dont know anything. Keep it up and if ever you come to Singapore let me know, we are heavily involved in AI innovations and research here. Would love to show you around.
i feel so sorry for you. In our family (father side) it also runs in the blood. breast cancer killed 2 aunties of mine, that is why I'm really searching for the updates on anything related to breast cancer from the machines that will be able to detect (coz what we have here in the Philippines is like an ancient tool that will press ur boobs-so hurtful) up to the cure🙏🙏🙏
Is it as simple as Hospitals and clinics begin "required" to ask you if you want to option your data into a "national consortium" of medical data that can be accessed by researchers and companies? Maybe we need that Federal law. The patient would have the option to say no. Call it CURE - Consortium of Unified Research Evidence
Make CURE part of the NEW HIPPA - Under this act, patients would OWN their data & have the option of updating a thumb-drive &/or having a file in the cloud that would be updated every time they see a doctor or other health professional or get a prescription filled. Patients would then (under CURE) have the option of uploading anonymized complete records to research institutions either from the cloud or from their thumb-drive or other device. Thanks to HIPPA & LAWSUITS, this would have to involve Federal legislation that required that hospitals, clinics, doctors etc. do this, & the legislation would have to specifically protect them from any lawsuits that might arise from complying with this new law.
I also don't care if people know what my x-ray MRI's etc. look like. There are other things I do not want in a database like my conversations with people, my location, daily routine, or political leanings.
"...Track the eye gaze to detect when you lost attention and instead of writing your paper, you dream about what you will do in the evening..." Omg, she caught me red handed :P
Wonderful interview and thanks to Regina for sharing hr story. AI/ML will play a huge role in the future of all healthcare, especially when the tsunami of Whole Genome Data hits. Precision medicine for all please. Whilst this is one piece of the puzzle there are several more if we're going to crack this. Regina touches on a general lack of coordination. Technology will help that although govt plays a bigger role. Govt can organize the frameworks, and fund projects only for those who collaborate. Govt should set drug policy which includes pointing the pharma industry in the right direction and set pricing rules However, are they set up to do that? Does business get in the way of real progress? It's also interesting to look at other models outside of the USA. The UK is well organized and now has a national genome project. Whilst the signs are good I would urge everyone to reflect that there have only been 4 cancer drugs developed for children in the last 30 years. Survival rates aren't understood and other health related complications aren't known. the facts are shocking. That is totally unacceptable considering how much money goes into research. Billions of dollars over 30 years in to the funnel with only 4 drugs released. Does research lead to drugs and treatment? Or does it lead to great ideas that are shelved by drug companies because they aren't multi million dollar opportunities? This needs more open discussion. Without a single payer healt system in the US it's way to fragmented. Check out the amazing work done by Curesearch in the kids cancer space. curesearch.org/Our-Strategy Don't let business get in the way of true precision medicine. This should be govt lead and the burden of cost not shouldered by individuals. Maybe the billionaire philanthropists will clud together and disrupt this system. Pricing needs to be cost plus rather than emotional i.e. pharma pricing plays on people's fears and emotions and therefore they charge based on that.
It seems like a really good argument for universal medicine. I would like to ask her if she tried to get data from Canadian sources! Does that make it easier to gather data I would be more willing to give up my data if I knew profit wasn't going to be made from it.
Agreed. It''s a huge problem when there is little access to large data sets. Countries like South Korea are making interesting discoveries because properly de-identified data from the entire population is in a national database that is made available to researchers. But you would think that a national opt-in system where people could offer their data for research could work, even in the US. Perhaps government funding could be tied to data collection efforts to incentivize hospitals.
Near the end Regina mentioned how the running metrics forced her to do something irrational. It is the reinforcement learning mechanism. Metrics have a lot of power! They are good carrots and sticks at the same time. Its use in school can sometimes drive students to do suboptimal learning tasks. Governments could use metrics to control certain behavior. For example companies should comply with X or Y metric, or else. These metrics could be good sometimes, but not always.
As a long term cancer survivor, I can attest to the massive gaps between AI & the health research machine in spite of billions in research funding. We need a new approach, a new way of thinking & advanced cognitive #AI to assist humans in reaching beyond our current dimensional limits.to cure cancer & other diseases. ML is only one component of the needed change. Far more powerful will be #cognitiveai.
It is not as easy as it seems though. For one blockchain doesn't have alternative to normal big data databases for ordered data (which support range queries). Then, encrypting that data is tricky too. One piece of data can be encrypted with a single key. This is a problem, because giving access to read some data in a blockchain context means giving a decryption key. But if you give access to that data to hospital A and to hospital B, then either of them can leak your decryption key and not only you wouldn't know who leaked it because it's the same key, but now anyone who got the leaked key can have access to your data. The third problem is selective sharing. There are some queries on data which are only possible on a private server. One such example is asking whether a database has a record which contains certain genes, without revealing anything else. In that example, this is done in order for a researcher then to contact whoever has those records and ask permission to access them. But you can imagine all kinds of statistical questions which are impossible to ask in a decentralized trustless context like a blockchain.
11min45 secs in.. I dont think she agrees with him but is 2 polite to interrupt but if she was his friend she wood n at length. Reductionism is a strong n foundational scientific view (clockwork/Meck) but just one view and we moved on to include it in the ingredients of the current batch of bread we r cookin up
![[FULL EP.13] เซียนหรั่ง พา ไททศมิตร ลงเรือฝึกสกิลหว่านแห | เฮ็ดอย่างเซียนหรั่ง | One Playground](http://i.ytimg.com/vi/EWFuf2Av0K0/mqdefault.jpg)
I really enjoyed this conversation with Regina. Here's the outline:
0:00 - Introduction
0:48 - Literature
5:02 - Science personalities and ideas
8:37 - Medice and computer science
11:49 - Breast cancer and facing mortality
17:58 - Machine learning - detection of and curing cancer
23:25 - Lack of medical datasets
26:54 - Data privacy, value, and future
40:52 - Open problems in application of AI in medicine
50:06 - Natural language processing
54:54 - Language understanding and deep learning
1:02:42 - Human-level intelligence
1:05:41 - Neuralink and augmenting human intelligence
1:09:06 - MIT Introduction to Machine Learning course
1:13:40 - Meaning of life
Are you on Apple Podcasts? I hate TH-cam but enjoy your content.
@@CraigTalbert Yes, here's the link: itunes.apple.com/us/podcast/artificial-intelligence/id1434243584
Please do a podcast with Prof Robert Langer if it is possible.. Please
Favorite quote from this one: ""ideas on their own are not sufficient, and many times... it's the personalities and their devotion to their ideas... [that] changes the landscape". So true.
Its kind of a sad quote tbh. I heard that and thought of all the difficulties I encounter as a researcher. All the politics and push-backs etc. :(
@@p5742j Maybe there is a role for AI to play in helping to identify and highlight great ideas.
@@jeff_holmes Inspiring thought, why not!
Brilliant conversation as usual Lex. Thank you for making these.
I love how honest and humble you are when you don`t understand something. Even though you are by far one of the smartest interviewers I have listened to, you don`t hesitate to show and press on when there is something that you are not familiar with (the term screening in this interview). In the process of your own learning, you are teaching millions of other people invaluable knowledge. I have been involved in cancer research for two years now and i thought i knew everything about screening, listening to you and this awesome interviewee made me realise that I dont know anything. Keep it up and if ever you come to Singapore let me know, we are heavily involved in AI innovations and research here. Would love to show you around.
Lost my Mom & Aunt to breast cancer. It is cool to hear where the current research is at. Thanks for sharing
Lost two favourite aunties to breast cancer. Really appreciated this conversation.
Thanks so much.
i feel so sorry for you. In our family (father side) it also runs in the blood. breast cancer killed 2 aunties of mine, that is why I'm really searching for the updates on anything related to breast cancer from the machines that will be able to detect (coz what we have here in the Philippines is like an ancient tool that will press ur boobs-so hurtful) up to the cure🙏🙏🙏
...idea is great...persuasion takes lots of devotion....very well said...great conversation...
I don't know why such an amazing conversation is watched relatively so little. It was very inspiring. Thank you Regina and ofc Lex.
This is incredible. Thank you for sharing all this amazing content. ❤
This is great, she is great, and your questions rock. Thank you
Very humbling. Great discussion!
This is a very enjoyable and interesting podcast, inspiring as well.thx a lot
Thank you, Lex. Regina was fantastic.
I waiting your video like fresh air. Thanks Lex for your great work.
great conversation once again.
10:52 lex ready to risk it all !
That was fun! Thanks for letting this humble brilliant scientist speak so eloquently at length
Thank you!
Regina, you are such warm and nice person! :)
Brilliant conversation 🎉
Thank You
Is it as simple as Hospitals and clinics begin "required" to ask you if you want to option your data into a "national consortium" of medical data that can be accessed by researchers and companies? Maybe we need that Federal law. The patient would have the option to say no. Call it CURE - Consortium of Unified Research Evidence
Make CURE part of the NEW HIPPA - Under this act, patients would OWN their data & have the option of updating a thumb-drive &/or having a file in the cloud that would be updated every time they see a doctor or other health professional or get a prescription filled. Patients would then (under CURE) have the option of uploading anonymized complete records to research institutions either from the cloud or from their thumb-drive or other device.
Thanks to HIPPA & LAWSUITS, this would have to involve Federal legislation that required that hospitals, clinics, doctors etc. do this, & the legislation would have to specifically protect them from any lawsuits that might arise from complying with this new law.
Needs to just be public. I'd never support giving certain companies who have access to the data set an advantage over anyone else.
I also don't care if people know what my x-ray MRI's etc. look like. There are other things I do not want in a database like my conversations with people, my location, daily routine, or political leanings.
incredible!
Yes. I want This.
Great conversation. Could you please do one with Andrea Morello on Quantum computing
Hi Lex, excellent interviews!
Would it be possible to interview Richard Socher in the near future?
45:54 reminded me of zizek as they both have similar accents and so on and so on.
יפה ממש יפה such a beautiful spirited woman with so much directly applied intelligence
Is there a way of expressing language processing in an image, musical, or dance form?
"...Track the eye gaze to detect when you lost attention and instead of writing your paper, you dream about what you will do in the evening..."
Omg, she caught me red handed :P
Wonderful interview and thanks to Regina for sharing hr story. AI/ML will play a huge role in the future of all healthcare, especially when the tsunami of Whole Genome Data hits. Precision medicine for all please. Whilst this is one piece of the puzzle there are several more if we're going to crack this. Regina touches on a general lack of coordination. Technology will help that although govt plays a bigger role. Govt can organize the frameworks, and fund projects only for those who collaborate. Govt should set drug policy which includes pointing the pharma industry in the right direction and set pricing rules However, are they set up to do that? Does business get in the way of real progress?
It's also interesting to look at other models outside of the USA. The UK is well organized and now has a national genome project.
Whilst the signs are good I would urge everyone to reflect that there have only been 4 cancer drugs developed for children in the last 30 years. Survival rates aren't understood and other health related complications aren't known. the facts are shocking.
That is totally unacceptable considering how much money goes into research. Billions of dollars over 30 years in to the funnel with only 4 drugs released. Does research lead to drugs and treatment? Or does it lead to great ideas that are shelved by drug companies because they aren't multi million dollar opportunities? This needs more open discussion. Without a single payer healt system in the US it's way to fragmented. Check out the amazing work done by Curesearch in the kids cancer space.
curesearch.org/Our-Strategy
Don't let business get in the way of true precision medicine. This should be govt lead and the burden of cost not shouldered by individuals. Maybe the billionaire philanthropists will clud together and disrupt this system. Pricing needs to be cost plus rather than emotional i.e. pharma pricing plays on people's fears and emotions and therefore they charge based on that.
It seems like a really good argument for universal medicine. I would like to ask her if she tried to get data from Canadian sources! Does that make it easier to gather data I would be more willing to give up my data if I knew profit wasn't going to be made from it.
Agreed. It''s a huge problem when there is little access to large data sets. Countries like South Korea are making interesting discoveries because properly de-identified data from the entire population is in a national database that is made available to researchers. But you would think that a national opt-in system where people could offer their data for research could work, even in the US. Perhaps government funding could be tied to data collection efforts to incentivize hospitals.
Professor Regina Barzilay, congratulation on winning the Squirrel AI Award for Artificial Intelligence for the Benefit of Humanity
"There's not many people I'd like to speak to for an hour"
I'd have felt pretty awkward if someone I was interviewing for an hour said that.
is her machine learning course available in MIT OCW? or is it a separate course?
@Khaled Kholy Thank You!
Near the end Regina mentioned how the running metrics forced her to do something irrational. It is the reinforcement learning mechanism. Metrics have a lot of power! They are good carrots and sticks at the same time. Its use in school can sometimes drive students to do suboptimal learning tasks. Governments could use metrics to control certain behavior. For example companies should comply with X or Y metric, or else. These metrics could be good sometimes, but not always.
As a long term cancer survivor, I can attest to the massive gaps between AI & the health research machine in spite of billions in research funding. We need a new approach, a new way of thinking & advanced cognitive #AI to assist humans in reaching beyond our current dimensional limits.to cure cancer & other diseases. ML is only one component of the needed change. Far more powerful will be #cognitiveai.
It's a science poem.
Who gives thumbs down on a cancer pod cast? Especially a breast cancer survivor giving details on how to improve the system?
some morons for sure
we need a global blockchain distributed database for Health records
It is not as easy as it seems though. For one blockchain doesn't have alternative to normal big data databases for ordered data (which support range queries).
Then, encrypting that data is tricky too. One piece of data can be encrypted with a single key. This is a problem, because giving access to read some data in a blockchain context means giving a decryption key. But if you give access to that data to hospital A and to hospital B, then either of them can leak your decryption key and not only you wouldn't know who leaked it because it's the same key, but now anyone who got the leaked key can have access to your data.
The third problem is selective sharing. There are some queries on data which are only possible on a private server. One such example is asking whether a database has a record which contains certain genes, without revealing anything else. In that example, this is done in order for a researcher then to contact whoever has those records and ask permission to access them. But you can imagine all kinds of statistical questions which are impossible to ask in a decentralized trustless context like a blockchain.
Devotion of the person
Limited time. What is important
1.7 million new cases every year
600k related deaths every year
TH-cam should ML the comments and group in categories/tabs, so I can easily skip all the grateful comments. There is a like button people!
11min45 secs in.. I dont think she agrees with him but is 2 polite to interrupt but if she was his friend she wood n at length. Reductionism is a strong n foundational scientific view (clockwork/Meck) but just one view and we moved on to include it in the ingredients of the current batch of bread we r cookin up