Machines & Genes: The Future of AI in Biology with Dr. George Church
Resumo
TLDRThe video features a conversation with George Church, a leading geneticist who discusses his experience with the Personal Genome Project, ethical concerns in genomics, and the role of AI in advancing biology. Church shares insights from his journey in genetics, his views on the revival of woolly mammoths, and the importance of public access to genomic data. He also touches on the impact of machine learning in protein design and potential changes to healthcare costs. Ethical dilemmas surrounding AI applications in medicine and the future of genomic testing in personal relationships are explored, emphasizing a balance between technological advancement and societal responsibility.
Conclusões
- 🔬 George Church is a pioneer in genetics and synthetic biology.
- 📊 The Personal Genome Project emphasizes public access to genomic data.
- 🤖 AI has transformative potential in protein design and gene therapy.
- 🧬 Church advocates for understanding genetic risks in relationships.
- 🌍 Reviving woolly mammoths could aid climate change efforts.
- 🤝 Ethical concerns about privacy and data sharing are paramount.
- 💡 Machine learning can significantly reduce healthcare costs.
- 📈 Genomic testing is becoming more prevalent and accessible.
- 🎯 Transparency in medical data is essential for ethical research.
- 🌐 The intersection of technology and ethics remains critical for the future.
Linha do tempo
- 00:00:00 - 00:05:00
The speaker discusses their experience with Institutional Review Boards (IRBs) while working on an ethical and technological study, known as the Personal Genome Project. They express concerns about privacy and the inability to share vital health information with patients, advocating for a more open approach to medical records.
- 00:05:00 - 00:10:00
The introduction to George Church, a prominent scientist in genetics and synthetic biology. The hosts highlight his groundbreaking work in genomics and his ambitious projects, such as reviving the woolly mammoth and the Personal Genome Project.
- 00:10:00 - 00:15:00
George Church shares his early fascination with science and technology, recalling how he began building computers and programming in high school, which led him to pursue a career blending technology and biology.
- 00:15:00 - 00:20:00
Church discusses his motivations for wanting to revive the woolly mammoth, focusing on ecological benefits, such as carbon sequestration in the Arctic, rather than merely bringing back an extinct species.
- 00:20:00 - 00:25:00
He explains the technical challenges involved in engineering cold-resistant elephants, outlining the process of genome editing necessary to achieve this goal, referencing his previous successful work on pig genome editing for xenotransplantation.
- 00:25:00 - 00:30:00
The conversation shifts to Church's personal decision to make his own genomic data publicly available, which stemmed from a desire to explore the implications of genomics on health, privacy, and patient information access.
- 00:30:00 - 00:35:00
This public availability has had personal benefits for Church, prompting patient alerts about potential health issues and revealing risk factors that impacted his medical decisions.
- 00:35:00 - 00:40:00
An overview of Church's lab’s contributions to using artificial intelligence in biotech, emphasizing their work on protein engineering and enhancing CRISPR technologies by leveraging machine learning's capabilities.
- 00:40:00 - 00:49:49
Church reflects on the evolving landscape of artificial intelligence in biology, urging that the integration of machine learning is becoming crucial across all aspects of the field, while emphasizing the ongoing importance of human oversight and engagement.
Mapa mental
Vídeo de perguntas e respostas
What is the Personal Genome Project?
The Personal Genome Project is an initiative founded by George Church that aims to make genomic data publicly accessible while emphasizing ethics and participant consent.
How does AI impact protein design?
AI enables massive parallelization in protein design efforts, allowing researchers to achieve higher efficiency in mutagenesis and optimization compared to traditional methods.
What are the ethical concerns with genetic data privacy?
Many researchers, including Church, express skepticism about the feasibility of keeping genetic data private and advocate for informed public sharing of such information.
What is George Church’s view on bringing back the woolly mammoth?
Church argues that reviving the woolly mammoth could help restore Arctic ecosystems and combat climate change by enhancing carbon sequestration.
How does Church suggest approaching genomic testing in relationships?
He advocates for considering genomic testing prior to dating to avoid potential genetic risks in offspring, promoting a more humane approach to genetic matching.
Will machine learning reduce healthcare costs?
Yes, there is potential for machine learning to lower costs in diagnostics and drug development, as evidenced by recent achievements in gene therapies.
What role does AI play in the future of genomics?
AI is expected to contribute significantly to genomics by enhancing data analysis, improving gene therapy results, and enabling personalized medicine.
What is one of Church’s most controversial opinions?
He believes everyone should consider getting their genome sequenced, particularly those of reproductive age, to understand their genetic risks.
What companies is George Church involved with?
Church is involved with several companies including Nabla, Dino Therapeutics, and Manifold Bio, all focused on using machine learning for biotech applications.
What keeps George Church up at night regarding AI?
He concerns himself with the potential for discrimination through biased AI algorithms and the ethical implications of personalized weapons created using AI.
Ver mais resumos de vídeos
- 00:00:03and so we put in two pages on ethical
- 00:00:05legal social implications and how we
- 00:00:08were going to do the IRB and the process
- 00:00:10this is the first time I've actually
- 00:00:12done an IRB i' collaborated with people
- 00:00:14that had done irbs before this is first
- 00:00:15time I did it and I was just appalled at
- 00:00:18what some of the things that they were
- 00:00:19forcing you to do they they said like
- 00:00:22you have to promise that you'll keep it
- 00:00:24private and I said I don't know how to
- 00:00:27keep my credit card private much less
- 00:00:29somebody's medical records and in fact
- 00:00:31there had been numerous uh hacking
- 00:00:34events and people losing disc drives at
- 00:00:37various hospitals in the Boston area
- 00:00:38that will remain nameless and so I said
- 00:00:42I think the better Assurance would be
- 00:00:44that we get people who are okay with it
- 00:00:45being public that was one thing that I
- 00:00:48was horrified about the other one was
- 00:00:49they said you couldn't return
- 00:00:51information to the patients even if it
- 00:00:55could save their life and I felt that
- 00:00:57was crazy too uh and said said well but
- 00:01:00a consequence of their data being public
- 00:01:02is it also be available to them and so
- 00:01:05they can you know if we give them the
- 00:01:06right software they can draw their own
- 00:01:08conclusions and with proper Consulting
- 00:01:11with their Physicians and Specialists
- 00:01:13they they will make the right decision
- 00:01:15and so we framed it as a ethics study as
- 00:01:18well as a technology study and that
- 00:01:21became the personal Genome
- 00:01:24[Music]
- 00:01:25Project welcome to another episode of
- 00:01:27any Jam AI Grand rounds I'm your host
- 00:01:30Raj Mon and I'm here with my co-host
- 00:01:33Andy beam today we're delighted to bring
- 00:01:35you our conversation with George Church
- 00:01:37of Harvard George is a professor of
- 00:01:39genetics at Harvard Medical School and
- 00:01:41professor of Health Sciences and
- 00:01:42Technology at Harvard and MIT he has
- 00:01:45many other titles and honors Andy I was
- 00:01:47really struck by both the depth and the
- 00:01:49breadth of cuttingedge science that
- 00:01:51George oversees in his laboratory and
- 00:01:53also by how effectively he must contact
- 00:01:55switch he's very well known for his
- 00:01:57pioneering work in genomics and
- 00:01:58synthetic biology his involvement with
- 00:02:00the Human Genome Project and the
- 00:02:02personal Genome Project his work on Gene
- 00:02:04editing and even Revival of the woolly
- 00:02:06mammoth just to name a few areas he also
- 00:02:09shared aspects about his career journey
- 00:02:11and how he approaches both academic and
- 00:02:13entrepreneurial efforts which were
- 00:02:15enlightening you know Raj George has to
- 00:02:17be one of the most fascinating
- 00:02:18individuals I've ever met I think when
- 00:02:20most of us decided that we wanted to be
- 00:02:22scientists we actually probably wanted
- 00:02:24to be someone like George he's always
- 00:02:26working on the most intriguing things
- 00:02:28and I'm constantly Amazed by the range
- 00:02:30of truly sci-fi projects that he has
- 00:02:31going on I really love that he sees his
- 00:02:34mission as making the future happen
- 00:02:35faster you know in that way he really
- 00:02:37reminds me of a scientist from a movie
- 00:02:38almost he's got this big beard this big
- 00:02:40personality and he does crazy things
- 00:02:42like try to bring back the woolly
- 00:02:44mammoth I always enjoy hearing about
- 00:02:46what George is up to and this
- 00:02:47conversation was no
- 00:02:50exception the njm AI Grand rounds
- 00:02:52podcast is sponsored by Microsoft and
- 00:02:55VII we thank them for their support and
- 00:02:58with that we bring you our conversation
- 00:03:00with George
- 00:03:01Church welcome to AI Grand RS George
- 00:03:04we're happy to have you here oh it's
- 00:03:06great to be here so George a question we
- 00:03:09ask all of our guests tell us about the
- 00:03:11training procedure for your own neural
- 00:03:13network how did you get interested in
- 00:03:15science and in AI in particular what
- 00:03:17data and experiences LED you to where
- 00:03:19you are today I've heard there is a
- 00:03:21framable Duke letter as part of this the
- 00:03:25story yeah so I got interested in
- 00:03:28science both both natural natur and
- 00:03:30unnatural my third father was a
- 00:03:32physician and so I was fascinated by the
- 00:03:34technology in his he would do house
- 00:03:36calls back then bit of uh history there
- 00:03:39and then I went to the World's Fair and
- 00:03:41saw computers and robots or animatronics
- 00:03:45and I decided that was definitely cool
- 00:03:49and I came back to Florida where they
- 00:03:51didn't have any science classes to be
- 00:03:53taken that I could find so I just
- 00:03:55started building my own computers
- 00:03:57because I was desperate not very good
- 00:03:59one and so it wasn't until 9th grade
- 00:04:01that I actually got access to a GE 635
- 00:04:04at Dartmouth been programming ever since
- 00:04:07basically that was actually a time
- 00:04:09sharing interactive one this is in 1968
- 00:04:12I'm in ninth grade then when I went to
- 00:04:15the college at Duke was kind of a step
- 00:04:17backwards forwards in terms of science I
- 00:04:19was doing crystallography but backwards
- 00:04:21in ter everything was Punch Cards into
- 00:04:23IBM 360 and then I got two major degrees
- 00:04:27in two years and then proceeded to flunk
- 00:04:29out of graduate school they gave me a
- 00:04:32nice letter hoping that would do well in
- 00:04:34some other field but I stuck to this
- 00:04:36field and my lab has been mostly mixture
- 00:04:39of Technology development and
- 00:04:41computational biology pretty much since
- 00:04:43the beginning in 1986 at AR medical
- 00:04:46school and naturally as the names kept
- 00:04:49changing as to whether it's neural Nets
- 00:04:51or deep learning or machine learning but
- 00:04:53we've been part of that Revolution at
- 00:04:56least as applied to things like Le
- 00:04:59biology and protein design in particular
- 00:05:02thanks that's super interesting and I
- 00:05:03actually didn't know that you had such
- 00:05:05an early interestes in computer science
- 00:05:06and I can see that writing code whether
- 00:05:08it's in bits or in DNA has sort of been
- 00:05:11with you since the beginning that you're
- 00:05:12still programming the software the
- 00:05:14substrate has just changed a little bit
- 00:05:16over time yeah yeah my first computers
- 00:05:18were a analog computer and then a a
- 00:05:23digital mechanical computer so it was
- 00:05:25only the third one that was was a real
- 00:05:28Von noyman digital
- 00:05:30interactive computer we're going to hop
- 00:05:32into your research in just a little bit
- 00:05:34but there's an icebreaker question that
- 00:05:35I think I'm obligated by law to ask you
- 00:05:38and it's about bringing back the woolly
- 00:05:39mammoth and so you've been on record on
- 00:05:42the record is saying that we should
- 00:05:43bring back the woolly mammoth and I'd
- 00:05:45love to hear the motivations for doing
- 00:05:47that and sort of how far are we away
- 00:05:49from being able to do that
- 00:05:51now well this is not going to fit into
- 00:05:54140 or 280 characters but I'll try my
- 00:05:57best so it's really not about bringing
- 00:06:00back an extinct species we're mainly
- 00:06:02interested in endangered species and
- 00:06:05restoring ecosystems and the ecosystem
- 00:06:08that seems like it needs the most help
- 00:06:10is the Arctic where there's a lot of
- 00:06:12carbon sequestered because of the cycles
- 00:06:14of freezing and then summer
- 00:06:17photosynthesis results in sometimes 500
- 00:06:20meters thick a top soil compared to say
- 00:06:22one meter in the lot of the
- 00:06:25rainforests and that's all melting and
- 00:06:28it's melting a lot of it in the form of
- 00:06:30methane which is 80 times worse than
- 00:06:32carbon dioxide and so we're hoping that
- 00:06:36making cold resistant elephants which
- 00:06:39would help save them from their
- 00:06:40endangered species categorization and
- 00:06:44would also help restore that when we're
- 00:06:46not saying this is the only answer or
- 00:06:48even this is a good answer it's just our
- 00:06:51fot at increasing sequestration and
- 00:06:54decreasing loss in the form of methane
- 00:06:57and we're doing that by genome
- 00:06:59engineering which we've already done in
- 00:07:02pigs for making transplants
- 00:07:05xenotransplants which are making their
- 00:07:07way through now 600 successful days in
- 00:07:11in non-human primates and some and now
- 00:07:14going into human clinical
- 00:07:16trials we can do dozens 42 edits in pigs
- 00:07:21genome dermaline we hope to be able to
- 00:07:24do the same thing in elephants and make
- 00:07:26them cold resistant maybe also resist to
- 00:07:29herpes viruses that are killing them if
- 00:07:33you have to do 42 edits for pigs what is
- 00:07:36the difference in scale to do something
- 00:07:38to like turn the elephant into a cold
- 00:07:40resistant woly
- 00:07:41Mammoth a cold resistant elephant that
- 00:07:44has genes that are that are resurrected
- 00:07:46from woly math we already resurrected
- 00:07:48two genes that seem to have the right
- 00:07:51properties when resurrected we don't
- 00:07:54know the exact number we're doing a
- 00:07:57computational analysis of dozens of of
- 00:07:59elephant and extinct
- 00:08:02relatives and focusing on genes that go
- 00:08:05to fixation that means that they're
- 00:08:08homozygous both their maternal and
- 00:08:10paternal lils are the same and both of
- 00:08:12those are different from the existing
- 00:08:16elephants so basically you've got this
- 00:08:19branching in the philogenetic tree where
- 00:08:23it's gone to fixation where all of the
- 00:08:26mammoth genes are different from the
- 00:08:29existing ones and we can do that for
- 00:08:30multiple mammoths and multiple existing
- 00:08:33genomes both African and Asian elephants
- 00:08:37and we're focusing on those both the
- 00:08:38coding regions and the non-coding
- 00:08:41regions and if we did all of them it
- 00:08:44would be in the hundreds of thousands if
- 00:08:48we just do the ones that we think are
- 00:08:49involved in cold a priori then those
- 00:08:52could be in the dozens so it's somewhere
- 00:08:54in between and we have Technologies
- 00:08:57editing Technologies I think that are up
- 00:08:59up to the high end of the spectrum our
- 00:09:01record so far for repetitive elements is
- 00:09:03editing
- 00:09:0524,000 edits simultaneously in one cell
- 00:09:09that happened to be a human plur poent
- 00:09:11stem cell but it could have been any
- 00:09:12amalian genome as far as we know got it
- 00:09:17thanks so George I want to switch gears
- 00:09:20to your work in genomics you're a
- 00:09:22Pioneer in genome sequencing genetics
- 00:09:25synthetic biology many other areas and
- 00:09:28your work has been influential on many
- 00:09:30scientists myself included uh but I
- 00:09:32actually want to ask you about a
- 00:09:34different perspective which is your
- 00:09:35perspective as a patient you famously
- 00:09:38made your own genome sequence your lab
- 00:09:41values and medical records publicly
- 00:09:43available for anyone to download on the
- 00:09:45internet I want to ask how did you come
- 00:09:47to that decision and what uh you might
- 00:09:50have learned since you've made that
- 00:09:52information
- 00:09:53available yeah that's an interesting
- 00:09:55question so we uh applied for ni Center
- 00:09:59of excellence in genomic science and got
- 00:10:02it and in the process we were proposing
- 00:10:05to develop a new way of doing DNA
- 00:10:07sequencing now this is in 2003 and we
- 00:10:10were partly there anyway so that's why
- 00:10:12we proposed it it's now called nextg
- 00:10:15sequencing but back then it had other
- 00:10:17names and we proposed that in the
- 00:10:21process of the five-year Center of
- 00:10:23Excellence Grant we would get as far as
- 00:10:26doing a 1.7 million based pair bacterial
- 00:10:30genomes this is a very tiny bacterial
- 00:10:32genome not the smallest but very tiny
- 00:10:34with helicobactor pylori the positive
- 00:10:36agent of stomach cancers and
- 00:10:39ulcers and we had already contributed to
- 00:10:41the first time that was sequenced back
- 00:10:43in 199 4ish so in 2003 we proposed to do
- 00:10:48that again after 5 years of work as it
- 00:10:52turned out 8 months into it we had
- 00:10:54finished something three times bigger
- 00:10:56than that and by the end of the project
- 00:10:59we had finished five human genomes at
- 00:11:01six billion base pairs each so diploid
- 00:11:03high quality genomes now that was not
- 00:11:06completely unanticipated we thought we
- 00:11:08might get the human genomes even though
- 00:11:10we were only promising to do a really
- 00:11:12tiny bacterial gen thousand times
- 00:11:13smaller we thought we might get there
- 00:11:15and so we put in two pages on ethical
- 00:11:18legal social implications and how we
- 00:11:20were going to do the IRB and in process
- 00:11:23this is the first time I've actually
- 00:11:24done an IRB i' collaborated with people
- 00:11:26that have done irbs before this is the
- 00:11:28first time I did it and I was just
- 00:11:29appalled at some of the things that they
- 00:11:31were forcing you to do they they said
- 00:11:34like you have to promise that you'll
- 00:11:35keep it private and I said I don't know
- 00:11:38how to keep my credit card private much
- 00:11:41less somebody's medical records and in
- 00:11:43fact there had been numerous uh hacking
- 00:11:46events and people losing disc drives at
- 00:11:49various hospitals in the Boston area
- 00:11:51that will remain nameless and so I said
- 00:11:54I think the better Assurance would be
- 00:11:56that we get people who are okay with it
- 00:11:58being public that was one thing that was
- 00:12:00horrified by the other one was they said
- 00:12:02you couldn't return information to the
- 00:12:05patients even if it could save their
- 00:12:08life and I felt that was crazy too uh
- 00:12:11and I said well but a consequence of
- 00:12:13their data being public is it also be
- 00:12:15available to them and so they can you
- 00:12:18know if we give them the right software
- 00:12:19they can draw their own conclusions and
- 00:12:21with proper Consulting with their
- 00:12:24Physicians and Specialists they they
- 00:12:26will make the right decision and so we
- 00:12:28framed it a ethics study as well as a
- 00:12:31technology study and that became the
- 00:12:34personal Genome
- 00:12:35Project that's fascinating that history
- 00:12:38I'm curious so you also as part of this
- 00:12:41you made your own genome public and I
- 00:12:44think to this day you can go and
- 00:12:46download a copy of your genome as well
- 00:12:48as your medical records and just
- 00:12:50connecting the the dots you were one of
- 00:12:52the participant who was willing to have
- 00:12:54their genome released right you are a
- 00:12:56participant in your own that's correct
- 00:12:58the
- 00:12:59IRB felt that I mean it was an unusual
- 00:13:03proposal to the IRB because I was
- 00:13:05breaking at least two of their rules a
- 00:13:06few others that we don't need to go into
- 00:13:08but but you know transparently and
- 00:13:10politely and it took us about a year to
- 00:13:13negotiate it but that's not that long
- 00:13:15compared to other IRB approvals uh even
- 00:13:17non-controversial ones but part of that
- 00:13:20is they said well you know you should be
- 00:13:21willing to eat your own dog food uh you
- 00:13:23know you we want you to be the sole
- 00:13:26participant and I said well how about 10
- 00:13:29participants let's compromise and I'll
- 00:13:30be the first of the 10 and then we'll
- 00:13:33expand if nobody gets hurt we'll expand
- 00:13:35it from there and John hamka was
- 00:13:37actually number two Esther Dyson was
- 00:13:39number three all their names were known
- 00:13:42that was approved by the IRB as well and
- 00:13:44John was at the time just brand new CIO
- 00:13:47at Best Israel deaconness and also had
- 00:13:50crafted the I think it was Norway's
- 00:13:52government's medical informatics
- 00:13:54policies and so he was really perfect
- 00:13:57second if in fact there's a funny story
- 00:13:59about my posting so I was a patient at
- 00:14:02Beth Israel Deacon this as it turned out
- 00:14:04and before we had the IRB approval I had
- 00:14:06just posted my own medical records as a
- 00:14:08test and some patient had been looking I
- 00:14:12get browsing the internet and had run
- 00:14:14across my medical records and she
- 00:14:16freaked out she thought this is standard
- 00:14:18policy at pical deaconness was to
- 00:14:20release patient medical records on the
- 00:14:22internet and the her complaint worked it
- 00:14:25all the way up to the president and then
- 00:14:27back down to John halamka who said oh I
- 00:14:30know what this is this is George church
- 00:14:32and he contacted me and I said oh yeah
- 00:14:34yeah oh I didn't even think that some
- 00:14:36patient would find this I I put a big
- 00:14:38disclaimer on it I said this is not the
- 00:14:41Beth Israel policy this is just an
- 00:14:42experiment and then they thought that
- 00:14:44was okay and we went
- 00:14:46forward and so have you learned anything
- 00:14:49about your own genome since it's been uh
- 00:14:52publicly available or your own medical
- 00:14:53records since you've made these widely
- 00:14:56available uh oh yeah sure there been
- 00:14:58various advantages there was one case
- 00:15:02where I was in giving a lecture in
- 00:15:05Seattle and somebody in the front row
- 00:15:09said while we were waiting for everybody
- 00:15:11to settle in you know there was kind of
- 00:15:13a
- 00:15:14pause and he said you know you should
- 00:15:16get your blood work checked out because
- 00:15:19according to your public medical records
- 00:15:20you're on Statin and there's no evidence
- 00:15:23that you've checked on your cholesterol
- 00:15:25or on the possible negative consequences
- 00:15:28for muscle damage and I said you know
- 00:15:31you're right I haven't been checked and
- 00:15:33it has been a long time and so I went
- 00:15:35off and I checked and actually there was
- 00:15:38no lowering of cholesterol and there was
- 00:15:39muscle damage so or potentially the
- 00:15:42biomarkers for that and so we changed
- 00:15:45the Statin and the dose and and
- 00:15:47monitored
- 00:15:48the that probably you know gave me extra
- 00:15:5110 years of life for not using drugs
- 00:15:54right and but you know I learned that I
- 00:15:57had risk factors for Peno 1 for Alpha 1
- 00:16:00antirion which put me at risk for a
- 00:16:02whole variety of respiratory problems
- 00:16:05and so I've been cautious about living
- 00:16:08in downtown Beijing and Los Angeles and
- 00:16:11things like that or or hanging around
- 00:16:14with Co too
- 00:16:15much great so we're gonna switch gears
- 00:16:18just a little bit again and talk about
- 00:16:21some of the work that your lab has been
- 00:16:22doing in artificial intelligence and
- 00:16:24deep learning for biotech applications
- 00:16:26so I've you know I tried to pick one
- 00:16:29paper to focus in on but you've done so
- 00:16:31much work on things like protein
- 00:16:33engineering designing aavs for gene
- 00:16:35therapy improving the efficiency of
- 00:16:37crisper so I was hoping you could pick
- 00:16:39one from the list of the many papers
- 00:16:42that you've written in this area and
- 00:16:44help us understand what machine learning
- 00:16:47is bringing to the table so I think
- 00:16:49there's often a lot of confusion about
- 00:16:52when and how and why to use machine
- 00:16:54learning but I was hoping you could
- 00:16:55really help us zero in on the types of
- 00:16:58new question questions that machine
- 00:16:59learning has let your lab ask and
- 00:17:02answer yeah so the one that I would pick
- 00:17:05or the two the first one was the one
- 00:17:08that I would pick is is a nature biotech
- 00:17:10paper we published in year 2000 which is
- 00:17:13Bryant at all Eric kelik uh was uh
- 00:17:17postto on my lab who was a senior author
- 00:17:18on that paper and it was titled
- 00:17:20massively parallel deep diversification
- 00:17:23of aav CD proteins by Machine
- 00:17:26learning and we have been working on
- 00:17:28machine learning for protein design for
- 00:17:31quite a while at that point but that was
- 00:17:34different in it it Illustrated your
- 00:17:36quest your point was when is it time to
- 00:17:38do machine learning when is it not
- 00:17:40typically there's one prerequisite for
- 00:17:43machine learning you have a lot of data
- 00:17:45we had just published some paper that we
- 00:17:48refer to as low-end machine learning but
- 00:17:51it isn't entirely lowend and there's
- 00:17:53some additional background information
- 00:17:55about proteins in general but lowend for
- 00:17:57your particular experiment but but
- 00:18:00anyway that's prerequisite you need to
- 00:18:01have a lot of data typically then
- 00:18:03there's the question of when is it
- 00:18:04better so in this particular case I
- 00:18:06think we showed that it's a lot better
- 00:18:09because we did a comparison of naive or
- 00:18:13random semi- random uh models for
- 00:18:16mutagenizing a stretch uh key stretch of
- 00:18:2028 amino acids that's important to the
- 00:18:23aav capsid for gene therapy delivery and
- 00:18:27we asked you know how many amino acids
- 00:18:29can be changed simultaneously and the
- 00:18:30reason you might want to change a lot of
- 00:18:32them simultaneously is that's where the
- 00:18:35immune system interacts with it and if
- 00:18:37you want to use the gene therapy more
- 00:18:38than once and you want it to be or even
- 00:18:41once you want to make sure the immune
- 00:18:43system doesn't attack your precious
- 00:18:45therapeutic also if you want to Target a
- 00:18:47new tissue you might have to radically
- 00:18:49change the surface of the virus so it
- 00:18:51will Target tissue a and not B so those
- 00:18:54are our
- 00:18:55objectives and we in the naive Eve model
- 00:18:59it was very hard to get more than four
- 00:19:01changes out of 28 and this was very
- 00:19:04consistent with all the work we've done
- 00:19:05in many other proteins before four out
- 00:19:08of 20 was actually a pretty good day in
- 00:19:10one round of diversification and
- 00:19:13selection but using logistic regression
- 00:19:16or various concurrent neural Nets or
- 00:19:19various neural Nets uh we could get up
- 00:19:22to we could get about 90% at 25 out of
- 00:19:2628 about just doing this from memory 70
- 00:19:2980% at 26 out and even 20% of this big
- 00:19:34Library focused on 28 out of 28 so it
- 00:19:37wasn't like one solution it was lots of
- 00:19:38solutions while at four out of 28 with
- 00:19:42the naive model you were getting close
- 00:19:43to zero per. so that to me is a dramatic
- 00:19:47validation that that this is an
- 00:19:49improvement over the the naive models
- 00:19:52yeah so so this all started with uh an
- 00:19:54undergraduate in my lab Harvard
- 00:19:55undergrad Ethan alley and surge biswash
- 00:19:59and others Pierce Ogden who had
- 00:20:01developed unir which was a language
- 00:20:04model so there kind of two broad
- 00:20:07categories in protein structure and
- 00:20:10design one is focusing on the structure
- 00:20:12and that got a lot of attention with
- 00:20:14Alpha fold from Deep Mind Google but it
- 00:20:17just predicts the three-dimensional
- 00:20:19structure of the protein just I should
- 00:20:21say that's a 50-year Holy Grail but from
- 00:20:23a protein design standpoint knowing the
- 00:20:25threedimensional structure is just
- 00:20:28starting point and not necessarily the
- 00:20:30right direction so for example I'm just
- 00:20:32going to end on this example that the
- 00:20:34the just contrasting the language model
- 00:20:36which is essentially linear with the 3D
- 00:20:39structure is if you if you have a
- 00:20:41searing protease which is named because
- 00:20:44sering at the Active side is critical
- 00:20:46and you change that sering to an alanine
- 00:20:48so now it's completely dead but it has
- 00:20:50exactly the same threedimensional
- 00:20:52structure now there's the version of
- 00:20:55this Alpha fold which is Alpha fold
- 00:20:57multimer where you can ask whether two
- 00:20:59proteins can stick together that might
- 00:21:02save you in certain circumstances but
- 00:21:04basically point is you can kill a
- 00:21:05protein without changing its
- 00:21:06threedimensional structure but not in
- 00:21:08the language model it's two it it
- 00:21:10catches us stuff right right so that I
- 00:21:14think dovet Tales nicely into a
- 00:21:15follow-up question that I'd like to ask
- 00:21:17is you've seen many Revolutions in
- 00:21:20biology I think it's fair to say um
- 00:21:22genomic sequencing synthetic biology
- 00:21:25could you help us understand um how AI
- 00:21:28is going to make an impact in biology
- 00:21:30over the next 10 years and the sort of
- 00:21:32flip side of that is are there areas of
- 00:21:35biology are there corners of biology
- 00:21:37that you think are immune from sort of
- 00:21:39the AI um incursion that we're seeing
- 00:21:44today yeah so uh I kind of have a policy
- 00:21:48not to say anything's impossible because
- 00:21:51it's one way you can show how foolish
- 00:21:53you are I mean I'm not questioning
- 00:21:55whether I'm foolish it's hard to prove a
- 00:21:56negative yeah and and it's easy to get
- 00:21:59embarrassed two years later when
- 00:22:01somebody shows that it does work so and
- 00:22:03in fact almost everything that I've
- 00:22:05worked on every project I've worked on
- 00:22:07in my 68 years has been 50 years as a
- 00:22:10scientist one person or another has said
- 00:22:13it's impossible sometimes a couple years
- 00:22:15after I published peer- reviewed paper
- 00:22:17on it they still say it's impossible but
- 00:22:20anyway so I I can't think of anything
- 00:22:22where you know the laws of physics or
- 00:22:24Computing would prevent it from making a
- 00:22:27contribution ify may I qualify an area
- 00:22:30of biology that you think has low
- 00:22:32probability of being impacted or or is
- 00:22:35just too hard for AI over the next 10
- 00:22:38years so I've relaxed the condition a
- 00:22:39little bit yeah I still think it's like
- 00:22:44I think they're all sufficiently high
- 00:22:46probability nonzero uh that they that
- 00:22:50they're worth considering I think things
- 00:22:52where it's really hard to get data but
- 00:22:55on the other hand it's not clear any
- 00:22:57kind of intell elligence artificial or
- 00:22:59otherwise is going to solve some of
- 00:23:01those problems so yeah I think there's a
- 00:23:04a good room for it it doesn't mean that
- 00:23:07human beings aren't going to play a
- 00:23:09gigantic role in nudging it programming
- 00:23:13it interpreting it interfacing with
- 00:23:16people who don't need to know all the
- 00:23:20the Gory uh computational details so
- 00:23:24yeah I I think it's going to affect
- 00:23:25everything really and I and I think the
- 00:23:27same thing's true for synthetic biology
- 00:23:28synthetic biology is going to affect
- 00:23:31everything including making computers
- 00:23:34that might be better than current
- 00:23:36computers they'll probably be hybrid
- 00:23:38computers of various sorts uh in in a
- 00:23:40certain sense machine learning is based
- 00:23:43on is inspired at least by um natural
- 00:23:47neuronal computers and I think the
- 00:23:50hybrids will have a good shot at it if
- 00:23:53it was anyone else I would say that was
- 00:23:54a hedge but when George Church says
- 00:23:56Never Say Never I believe it
- 00:24:00100% thank
- 00:24:02you so George I want to ask you about a
- 00:24:05field where you think it won't
- 00:24:07work no I I I've just heard illuminary
- 00:24:10of the field uh advise me against uh ask
- 00:24:14you first
- 00:24:15yeah anyway so George I want to ask you
- 00:24:19about commercialization so you've been
- 00:24:21involved in many companies I think a few
- 00:24:23that are relevant to protein engineering
- 00:24:26and AI including manifold bio nabla Dino
- 00:24:30Therapeutics could you tell us about
- 00:24:32what these companies do and maybe how
- 00:24:34their missions relate to one
- 00:24:36another right so the three that you
- 00:24:39mentioned nabla Dino manifold and two
- 00:24:43more patch and shape Therapeutics all
- 00:24:45are using machine learning for protein
- 00:24:48or or nucleic acid design they're wildly
- 00:24:52different so nabl is focusing on
- 00:24:54antibodies which are one of the key
- 00:24:56therapeutic categor and have diagnostic
- 00:24:59uses as well Dino is on delivery we were
- 00:25:02just talking about aav manifold is very
- 00:25:06interesting and they're developing ways
- 00:25:09to make protein binding pairs but that's
- 00:25:12not the endgame then those can be used
- 00:25:15for testing multiple protein
- 00:25:17Therapeutics simultaneously in one let's
- 00:25:20say expensive test animal for
- 00:25:24preclinical trials let's say a non-human
- 00:25:26primate so you can do thousands of
- 00:25:29simultaneous measurements of pharmac
- 00:25:32kinetics and Dynamics and tissue
- 00:25:34targeting and so forth all at once with
- 00:25:36protein Therapeutics which don't
- 00:25:38normally have a nucleic acid barcode
- 00:25:41that you can follow so this is a protein
- 00:25:43barcode and then shape is working on RNA
- 00:25:47therapies tras and uh ADR and also
- 00:25:51delivery and then patches on CIS
- 00:25:53regulatory elements DNA and RNA and
- 00:25:56there could probably be about 10 more of
- 00:25:57them that would not overlap one another
- 00:26:00it's it's a very important subset of
- 00:26:03things that you can do with machine
- 00:26:05learning so you know as you think about
- 00:26:07those companies that are making use of
- 00:26:09machine learning and artificial
- 00:26:11intelligence and the others that you've
- 00:26:13been involved in I'm curious have you
- 00:26:16identified maybe a set of let's say one
- 00:26:19or two key questions that allow you to
- 00:26:22decide whether an idea is right as a
- 00:26:24commercial entity or whether it maybe
- 00:26:26belongs in
- 00:26:28in Academia for more development before
- 00:26:30it it moves into into a commercial
- 00:26:33entity yeah this is a really tough call
- 00:26:36that that every postto that wants to
- 00:26:40start a company and the pi that maybe
- 00:26:44wants to start it with them or wants to
- 00:26:46jump from Academia into industry and I
- 00:26:49say it's not so bad to be number two to
- 00:26:52be the second one in the field even if
- 00:26:54you thought of it first and it's kind of
- 00:26:56irritates You That Somebody body jumped
- 00:26:58in there with your idea you just want to
- 00:27:00make sure you got enough intellectual
- 00:27:02property that you have freedom to
- 00:27:03operate and it doesn't really matter who
- 00:27:05gets funded first
- 00:27:07necessarily and very often the second
- 00:27:10one does a better job of it either from
- 00:27:13a business side or science engineering
- 00:27:16side so that's part of the
- 00:27:18decision the part where you know that
- 00:27:20it's ready is either you're getting a
- 00:27:22lot of feedback from your peers they
- 00:27:24like it they want it one case you know
- 00:27:27when we were developing DNA synthesis
- 00:27:31the very high throughput DNA synthesis
- 00:27:32like 10,000 times previous throughput
- 00:27:36and suddenly we had a lot of friends
- 00:27:38that wanted to Buddy and and collaborate
- 00:27:41on making big DNA constructs cheaply and
- 00:27:44we said oh this is going to be a
- 00:27:45tremendous academic distraction we
- 00:27:48really have to spin this out just for
- 00:27:49our own sanity so we did that same thing
- 00:27:52with with chrisper that was clearly
- 00:27:54going to be so popular I mean we we
- 00:27:56announced it in January 2013 and by
- 00:27:59March there were like 10,000 users who
- 00:28:02distributed it through adene which is
- 00:28:04nearly free and that would have been a
- 00:28:07real hassle to have done that licking
- 00:28:10stamps in our
- 00:28:11office so those kind of things that tell
- 00:28:14you that you're ready don't have to rush
- 00:28:17you can the the more mature it is the
- 00:28:20the slower you will get deluded out
- 00:28:22where if you go there too early you may
- 00:28:25think it's magic money it's easy than a
- 00:28:27grant but pretty soon it'll be taken
- 00:28:29away from you and and it may not go in
- 00:28:31the direction you want it to go once you
- 00:28:33lose control over it yeah I I like the
- 00:28:36emphasis on it's not so bad to be second
- 00:28:38uh there's a saying that I'm reminded of
- 00:28:40that the early bird gets the worm but
- 00:28:42the second mouse gets the cheese and
- 00:28:44that sometimes being second is
- 00:28:45strategically good all right that's good
- 00:28:48second mouse is a cheese yeah yeah and I
- 00:28:50think there a little bit the second worm
- 00:28:52gets to live yeah right yeah yeah right
- 00:28:56um so so George just as a follow on to
- 00:28:58that having overseen many companies from
- 00:29:02genomics and from sequencing genomic
- 00:29:04Technologies and now increasingly with
- 00:29:07AI is there a difference you see in the
- 00:29:10decision- making of when to start a new
- 00:29:12company when genomics in the context of
- 00:29:14your involvement in companies there
- 00:29:16versus now with the more AI focused
- 00:29:18companies or the the principles more or
- 00:29:20less the
- 00:29:21same well there's some very significant
- 00:29:23differences well first of all when I
- 00:29:25started back in the'80s
- 00:29:28I was basically just the way I was
- 00:29:30deciding was if a investor or a
- 00:29:33colleague came forward and said hey we'd
- 00:29:35like to collaborate with you on a
- 00:29:36company and I'd say okay it was fairly
- 00:29:39reactive but recently it's been mostly
- 00:29:42postdoc driven and during that time we
- 00:29:44transitioned from Mostly analytic to
- 00:29:47mostly synthetic and the problem with a
- 00:29:49lot of the DNA sequencing scenarios was
- 00:29:53you had to convince people to get
- 00:29:56sequenced well with the synthetic
- 00:29:58biology which is basic you know in this
- 00:30:01case is basically pharmacology people
- 00:30:03are already consuming drugs and you're
- 00:30:05just making better and better drugs for
- 00:30:07diseases for which there were no drugs
- 00:30:09and it's just a it's an easier pipeline
- 00:30:11while I would say that our first
- 00:30:13sequencing Innovation was in 1984 and
- 00:30:16here we are in almost
- 00:30:182024 and we still don't have you know 40
- 00:30:21years later we still don't really have
- 00:30:23consensus enough that healthc care
- 00:30:26providers think that it's a good thing
- 00:30:28to give everybody their whole genome
- 00:30:30sequence or act on it in any way and
- 00:30:33most people don't do it on their own so
- 00:30:35there's a disconnect I think it's the 1%
- 00:30:39dilemma it's the seat belts smoking
- 00:30:43global warming and getting your genome
- 00:30:45sequenced they're all like hey I got a
- 00:30:4899% chance of doing okay if I went to
- 00:30:51Las Vegas with those odds I'd be fine
- 00:30:54but it's different for public health and
- 00:30:56it takes special effort to get people to
- 00:30:58stop smoking and to wear seat belts and
- 00:31:00so forth so I think that's what's going
- 00:31:03on here and and no no government agency
- 00:31:05has stepped forward to do what they did
- 00:31:08for seat belts and smoking which was a
- 00:31:10whole whole series of experiments like
- 00:31:12just passing a law to buckle your seat
- 00:31:14belt wasn't enough getting them present
- 00:31:16in every car was not enough they had to
- 00:31:18actually mandate a circuit that would
- 00:31:21close once you buckled it on top of your
- 00:31:24belly Not underneath your you know that
- 00:31:27hasn't happened yet for for even though
- 00:31:30the the carrier status alone plus adult
- 00:31:33onset diseases could save us a trillion
- 00:31:35dollars a year and a lot of pain and
- 00:31:38suffering it just hasn't happened yet
- 00:31:40but with synthetic biology totally
- 00:31:42different thing there's the orphan Drug
- 00:31:44Act that makes it very profitable to go
- 00:31:47after rare things and then there's lots
- 00:31:49of common diseases as well that can be
- 00:31:51treated and my favorite treatment is
- 00:31:53gene therapy for reasons that we could
- 00:31:55go into if you if you want to R shall we
- 00:31:58go to the lightning
- 00:31:59round sure so George we'd like to do a
- 00:32:02quick lightning round if that's all
- 00:32:04right with
- 00:32:05you we're going to ask a series of
- 00:32:07questions and the rules of the game are
- 00:32:09one to two sentence responses uh to each
- 00:32:12question yes or no is is also great does
- 00:32:15that sound all right okay I'm ready and
- 00:32:18some of these will be highly entropic
- 00:32:20questions in the the goal is to to learn
- 00:32:22more about George church and how he
- 00:32:23thinks about the world but you have to
- 00:32:25be brief um so uh the first question is
- 00:32:29kind of a Turing test for biology is the
- 00:32:31way that I think about it and an
- 00:32:33appropriate response to this could be
- 00:32:35that's a dumb question but will AI
- 00:32:38understand biology in any meaningful
- 00:32:40sense where understand is in air quotes
- 00:32:43here I think in a way it already does
- 00:32:47highly Advanced biotechnologists talk to
- 00:32:50each other essentially in biotechnology
- 00:32:52which is not really natural
- 00:32:55language and computers
- 00:32:57I gave an example for protein
- 00:33:00design all right George what's your
- 00:33:03favorite piece of
- 00:33:05music favorite piece of
- 00:33:09music gee you know I I kind of like
- 00:33:12Talking Heads oh nice you know this is
- 00:33:16not my beautiful house I don't know
- 00:33:19given a little more time I could with a
- 00:33:21few others yeah that's the point of this
- 00:33:23though is to inject some entropy so um
- 00:33:27uh you've had a storied career in
- 00:33:29science um what one thing has changed
- 00:33:31the most either from a technology
- 00:33:33standpoint from a society standpoint
- 00:33:35from a political standpoint in science
- 00:33:37over the course of your
- 00:33:39career well I would say all of the above
- 00:33:43in the Genome Project because it the NIH
- 00:33:46was entirely hypothesis driven now
- 00:33:49you've got one Institute that is
- 00:33:51Discovery driven and another one that's
- 00:33:53engineering driven and that coincided
- 00:33:56roughly with a two-fold increase in the
- 00:33:58NIH budget while we were starting the
- 00:34:00Genome Project I can't say it was cause
- 00:34:02and effect but it was it was a nice
- 00:34:05coincidence okay do biologists need to
- 00:34:08understand machine learning to
- 00:34:10contribute to machine learning
- 00:34:13projects do citizens have to understand
- 00:34:16GPS and atomic clocks in order to find
- 00:34:19directions on Google Maps all right
- 00:34:22we'll accept it one sentence we accept
- 00:34:25it as we know the the price of things in
- 00:34:28healthcare tends to be sticky so the
- 00:34:30question is will machine learning
- 00:34:32ultimately reduce costs for Diagnostics
- 00:34:34and
- 00:34:36drugs yes is the short answer uh but I
- 00:34:40don't and I give an example where it's
- 00:34:41not sticky defending on how you define
- 00:34:43things so gene therapy was $2.8 million
- 00:34:47a dose until we got to coid 19 and the
- 00:34:52top five vaccines were all formulated as
- 00:34:54gene therapy some as low as $2 a day do
- 00:34:57so 2.8 to two so give me you know that's
- 00:35:01not very sticky okay our pre-prints and
- 00:35:05net scientific
- 00:35:06good
- 00:35:09preprints I think net is the key word
- 00:35:11yeah know they're good and bad pretty
- 00:35:13high levels of good and bad but yeah I
- 00:35:16think they're that positive okay final
- 00:35:18question of the lightning round if you
- 00:35:20could have dinner with one person Dead
- 00:35:22or Alive who would George Church have
- 00:35:24dinner
- 00:35:25with uh oh jeez I think probably Netty
- 00:35:29Stevens you've probably never heard of
- 00:35:31her but she was on the little Google
- 00:35:33logo but I knew her before that in
- 00:35:35around 1910 1915 she found the
- 00:35:37chromosome theory of inheritance along
- 00:35:41with Morgan but we're separate from
- 00:35:43Oregon about the same time all right all
- 00:35:47right so we're gonna move to the final
- 00:35:49segment of the episode we're GNA talk
- 00:35:50about some big picture things I think
- 00:35:52it's fair to say we've already touched
- 00:35:53on some big picture topics so far but
- 00:35:56we're going to try and broad the
- 00:35:57aperture just a little bit
- 00:35:58further we've talked a lot about
- 00:36:01biotechnology and your work in the area
- 00:36:03I want to come back a little bit to a
- 00:36:06clinical focus and given sort of what
- 00:36:08you see happening either in Diagnostics
- 00:36:11or gene therapy what Medical Specialties
- 00:36:14do you think are most likely to be
- 00:36:15changed and impacted by
- 00:36:18AI Medical Specialties hopefully
- 00:36:21genetics interpreting The genome is
- 00:36:23increasingly engaging polygenic risk
- 00:36:26score
- 00:36:27and I think that probably could be done
- 00:36:30better um and then that could have
- 00:36:32impact on almost every field of medicine
- 00:36:35the other thing is age related diseases
- 00:36:38I think there there's an opportunity of
- 00:36:39having multiple genes involved in gene
- 00:36:44therapy and possibly even personalized
- 00:36:47or personalized medicine in general but
- 00:36:49especially related to aging because
- 00:36:52aging affects every disease basically
- 00:36:54almost every form of human morbidity
- 00:36:57immortality is impacted so I think those
- 00:36:59are a cluster of three things that
- 00:37:02interact with one another genetics aging
- 00:37:05and machine learning so thinking about
- 00:37:08our listeners who are clinicians and in
- 00:37:10particular early career clinicians Med
- 00:37:13students residents what do you think
- 00:37:15those folks should know about AI to help
- 00:37:18them prepare for a career in
- 00:37:21medicine I think that they have a fairly
- 00:37:24high level view of it it's it's like we
- 00:37:26no longer most of us don't program in
- 00:37:30zeros and ones we program with high
- 00:37:32level languages like python or or maybe
- 00:37:35even HTML or Excel or something so it'll
- 00:37:38be like that hopefully it'll be very
- 00:37:40easy to interface with as is the case
- 00:37:43for most really awesome software but
- 00:37:45they will have to know it and they might
- 00:37:47not have to memorize as much when I was
- 00:37:50a boy you know he had to memorize all
- 00:37:52the biochemical Pathways and all the
- 00:37:54pathologies and hopefully it'll be like
- 00:37:57how do you look for it how do you
- 00:37:58interface with the machine learning and
- 00:38:01the big databases so you can't know it
- 00:38:05all anymore but how do you know where to
- 00:38:08look for the answer there's not going to
- 00:38:11be a KB cycle of AI for for Physicians
- 00:38:14to memorize hopefully not on the
- 00:38:17contrary it's going to mean fewer people
- 00:38:19learning crb cycle excellent and they'll
- 00:38:23learn instead oh uh isocitrate dehydra
- 00:38:27enase is very impactful on certain gomas
- 00:38:31right and so it's one of the most
- 00:38:33treatable of the of the otherwise nasty
- 00:38:36category of cancer that hits the the
- 00:38:39brain yeah I'm married to a clinician
- 00:38:41and I think that the Mandate that I've
- 00:38:44been given is that if I develop anything
- 00:38:46in the AI space that makes her job more
- 00:38:48difficult if I have a new KB cycle that
- 00:38:50she has to memorize then that thing is
- 00:38:51not going to get very far clinically so
- 00:38:54I yeah yeah there's there
- 00:38:57in this day and age there's really no
- 00:39:00excuse for poor user interface in in
- 00:39:02Computing or another checkbox that you
- 00:39:05have to click or something like that
- 00:39:07exactly that's true there there's plenty
- 00:39:10of bat softare nevertheless but yeah but
- 00:39:12there's no excuse for it yeah so I'm
- 00:39:15glad that I get to ask this question
- 00:39:17because I'm sure that we'll get a great
- 00:39:19answer what is your most controversial
- 00:39:23opinion my most controversial opinions
- 00:39:26were
- 00:39:27placed on me not from me okay so like
- 00:39:30things having to do with Advocate I
- 00:39:32don't preced this but I don't Advocate
- 00:39:35but that people felt that I was
- 00:39:38advocating cloning human
- 00:39:41neanderthals so uh a controversial
- 00:39:44opinion that is yours not attributed to
- 00:39:46you one that's mine is probably that
- 00:39:50everybody should seriously consider
- 00:39:51getting their genome sequence in
- 00:39:53particular if they're of reproductive
- 00:39:56age age you know let's say 16 and up
- 00:40:00especially for men that keep going that
- 00:40:04they should know their carrier status uh
- 00:40:06and that could influence who they date
- 00:40:08or various other things I think the idea
- 00:40:10of dating app that is aware of your
- 00:40:14carrier status is the most Humane place
- 00:40:16to do it but the controversy is they
- 00:40:19either think that that's Eugenics which
- 00:40:21it isn't or it's controversial because
- 00:40:25yeah don't want to der
- 00:40:27romanticize something by being so
- 00:40:29technical but anyway I think that's a
- 00:40:32huge missed opportunity it's more Humane
- 00:40:34because I'm glad we're not in the
- 00:40:36lightning round it's more Humane because
- 00:40:39if you do it after you're pregnant then
- 00:40:43you have the tough decision about
- 00:40:45termination of pregnancy which is tough
- 00:40:49for essentially everybody where you're
- 00:40:51pro-choice or pro- lifee and if you do
- 00:40:53it after you're married then you got the
- 00:40:55tough decision are you going to have
- 00:40:57children with this person which means
- 00:40:59you're going to do invitro fertilization
- 00:41:01which is no walk in the park and so
- 00:41:03that's bad news but if you do it before
- 00:41:05you've even met the person then it means
- 00:41:07that out of a thousand people you could
- 00:41:09date you're going to date you know 990
- 00:41:12of them and at most and so you're going
- 00:41:16to eliminate a few and and there's no
- 00:41:17false positive problems at that point
- 00:41:20there's definitely a false positive
- 00:41:21problem if you're doing IVF or or
- 00:41:25termination but or or worse yet you know
- 00:41:28doing a surgery to remove organs that
- 00:41:31might be at risk for cancer but there's
- 00:41:33essentially no false positive problem
- 00:41:34when you're rejecting 3% of the
- 00:41:38potential suitors and at the risk of
- 00:41:40misattributing another quote to you I
- 00:41:42think I've heard you talk about this
- 00:41:44before and it would go something like
- 00:41:45there's a dating app like you said and
- 00:41:47silently behind the scenes you're
- 00:41:49getting screened out from people who
- 00:41:51have the same carrier status as you so
- 00:41:53that you're never matched with someone
- 00:41:55who would you so that two recessive
- 00:41:57geneses would would come together right
- 00:41:59so you You' kind of not even know that
- 00:42:01it was going on behind the scenes you
- 00:42:03would just be matched with people that
- 00:42:05you would not have this problem with
- 00:42:07right that's accurate and and that's
- 00:42:09Humane in addition to all the things I
- 00:42:11mentioned that's Humane because another
- 00:42:14awkward time is you've decided to marry
- 00:42:16somebody and then you get the score and
- 00:42:18then you decide not to uh this is or or
- 00:42:22for that matter you get people find out
- 00:42:24you know you're all ready to get married
- 00:42:25and then marriage is off um and then
- 00:42:28everybody knows that both of you are
- 00:42:30carriers and so in a certain sense maybe
- 00:42:32in less accepting parts of society you
- 00:42:36both get branded as why should anybody
- 00:42:38date them and in fact right everybody
- 00:42:40you know 97% of people should date them
- 00:42:42just not the 3% that are mismatched and
- 00:42:45so by avoiding anybody knowing that your
- 00:42:48carrier
- 00:42:49status I think it's the most Humane
- 00:42:52thing so both more Humane and no scarl
- 00:42:55letter no Stigma
- 00:42:56like none of none of those issues yeah
- 00:42:59right so I think that's the time to do
- 00:43:00it or we could destigmatize everything
- 00:43:03but that's I think that's harder it's
- 00:43:05hard to say which is hard but it could
- 00:43:06be hard yeah George are there any
- 00:43:08examples of a I guess I think you called
- 00:43:10a dating app are there any examples of
- 00:43:12that in place or is this is just an idea
- 00:43:15there are not exactly that not an app
- 00:43:18and it's not quite but it's like that
- 00:43:20which is dorya sharim which was started
- 00:43:23by a rabbi I think mid 80s so it's been
- 00:43:26around for a while because he had I
- 00:43:29think four of his children had taacs
- 00:43:31which is a serious disease kills kids
- 00:43:34painfully at age
- 00:43:364ish and he just decided that there
- 00:43:39should at least be the option among
- 00:43:42his congregation is anybody that that
- 00:43:45could have similar afflictions and it
- 00:43:48scaled up to I think eight or nine genes
- 00:43:50typically that are enriched in the
- 00:43:52ashkanazi population but in a certain
- 00:43:55sense we're all at risk for those eight
- 00:43:57or nine genes and about a thousand more
- 00:44:00and it's not clear why it hasn't spread
- 00:44:02it's been very successful in in the
- 00:44:04populations that use it where it's
- 00:44:07lowered the risk of such births by at
- 00:44:10least a factor of 10 and why it hasn't
- 00:44:12spread to other populations is I don't
- 00:44:14think it's because one population knows
- 00:44:17more or less science than than the other
- 00:44:19ones it's something
- 00:44:21else it's not that one population has
- 00:44:24necessarily more genetic diseases than
- 00:44:25other it is true that some inbred
- 00:44:28populations have slightly higher but
- 00:44:30that's not point is we're all at at
- 00:44:32least a 3% risk
- 00:44:34right so our our final question for you
- 00:44:37George and you can take this in multiple
- 00:44:39possible ways it's up to you what
- 00:44:42applications of AI to biology keep you
- 00:44:45up at
- 00:44:47night oh yeah well first of all I'm
- 00:44:51genetically narcoleptic and so nothing
- 00:44:53keeps me up at night it's about 30
- 00:44:55seconds is the median time to falling
- 00:44:58asleep but what keeps me up during the
- 00:45:00day is anything involving discrimination
- 00:45:04so artificial intelligence could more so
- 00:45:08it's an interesting question when we
- 00:45:10worry about discrimination are we
- 00:45:11worried about it being too inaccurate no
- 00:45:15words we're we're stereotyping an entire
- 00:45:17people or category of people that maybe
- 00:45:21have a a priori a low probability of
- 00:45:24living up to the stereotyp type or are
- 00:45:26we worried it's too accurate is it are
- 00:45:28we worried that it's not accurate enough
- 00:45:30or it's too accurate and I think it's
- 00:45:32Case by case but any case it could have
- 00:45:35enough imp premature of accuracy that it
- 00:45:38would be used but still inaccurate
- 00:45:41enough that it could be abuse so that's
- 00:45:42one scenario the other scenario is they
- 00:45:45could use it you know make personalized
- 00:45:47weapons you know once once the
- 00:45:48Terminators come then are human failures
- 00:45:53could be in a certain sense public to
- 00:45:55the machine
- 00:45:56even though they're not public in other
- 00:45:58words I can't tell it but they could
- 00:46:01figure it
- 00:46:02out I guess a follow-up question to that
- 00:46:05is I know like you're working Gene
- 00:46:07editing you spend a lot of time thinking
- 00:46:08about how when this technology becomes
- 00:46:11democratized and you can buy like a a
- 00:46:13reagent kit for $10 like how do we what
- 00:46:16are a set of ethics and what are a set
- 00:46:18of protocols that we can use in a world
- 00:46:19like that is that at all similar to how
- 00:46:22you think about what's happening with AI
- 00:46:24either generally or a biology because a
- 00:46:27model that cost $10 million to create
- 00:46:29today will you know cost $10 to create
- 00:46:32five years from now so how do we think
- 00:46:34about these powerful technologies that
- 00:46:37are also being democratized at like a
- 00:46:39very very quick Pace how can we sort of
- 00:46:42balance safety and progress in that kind
- 00:46:45of
- 00:46:47world well we've demonstrated that we
- 00:46:50that there's no such thing as a slippery
- 00:46:52slope that is to say there are
- 00:46:55documented cases where we were able to
- 00:46:58keep ourselves off the slippery slope
- 00:46:59and other cases where we are were not at
- 00:47:01least not for the whole population
- 00:47:04there's always some percentage of the
- 00:47:05population that falls into the Trap so
- 00:47:07for example speed limits there is no
- 00:47:10magic point where suddenly it becomes
- 00:47:13unsafe but people tend to stay pretty
- 00:47:16close to the speed limits yeah I guess
- 00:47:18the question was about either
- 00:47:21professional societal Norms so I know
- 00:47:23that there's been a lot of this in gene
- 00:47:25editing where there are groups that meet
- 00:47:27to discuss safety and come up with
- 00:47:30regulation is there any of that lessons
- 00:47:33from that community that transport to AI
- 00:47:38generally or AI in
- 00:47:41biology right so you don't need special
- 00:47:44groups to monitor Gene editing uh you
- 00:47:48have the FDA and the FDA is very
- 00:47:50effective at keeping us bringing things
- 00:47:52out that are safe and effective that
- 00:47:55also applies to medical devices so it is
- 00:47:57possible the AI would fall in that
- 00:48:00category but it's also possible you can
- 00:48:02evade the category by making something
- 00:48:04that's not
- 00:48:05recognizable for example dating apps
- 00:48:07that don't seem to be reg regulated by
- 00:48:09the FDA even though they could have a
- 00:48:11trillion dollar impact on medicine in a
- 00:48:14certain sense they're not medicine but
- 00:48:16nevertheless so even though people try
- 00:48:20to make regulations on top of
- 00:48:21regulations so they wanted to have a
- 00:48:23moratorium on Gene editing on top of the
- 00:48:26FDA which has a moratorium on all new
- 00:48:28drugs I thought that was a little crazy
- 00:48:31redundant and it didn't really
- 00:48:34happen but with AI if it does slip
- 00:48:37between the cracks then there should be
- 00:48:38some kind of safety now our track record
- 00:48:41for that not so great in Computing uh if
- 00:48:44you look at the internet there was very
- 00:48:46little of the foresight that existed
- 00:48:48very little of it made it in in time so
- 00:48:51there's wide open doors for hacking for
- 00:48:54computer viruses for identity theft for
- 00:48:58abuse of children and pornography and so
- 00:49:02forth so we didn't do such a great job
- 00:49:04there and I hope we do a better job with
- 00:49:06AI partly because of fantastic
- 00:49:10educational media by which I mean the
- 00:49:15Terminator all right uh that maybe we'll
- 00:49:18edit it so that we don't end on a dower
- 00:49:20note like that but I think that yeah
- 00:49:22that would be
- 00:49:23better ending on The Terminator yeah but
- 00:49:26I can't remember the last time I've had
- 00:49:27a conversation where we discussed
- 00:49:29resuscitating Willie mammoths and dating
- 00:49:31apps within the same hour so it's it's
- 00:49:33been it's been a really special
- 00:49:35conversation thank you so much George
- 00:49:37for being on AI Grand rounds it's been a
- 00:49:38pleasure thank you it was great I I look
- 00:49:41forward to hearing it yeah great
- 00:49:45[Music]
- 00:49:47thanks
- George Church
- Genetics
- AI in Biology
- Personal Genome Project
- Ethical Implications
- Genomic Data
- Woolly Mammoths
- Machine Learning
- Protein Design
- Healthcare Costs