AI in Life Sciences Panel Replay: Using AI to Improve Patient Outcomes and Redefine Health

Science Stops
9 followers
AI Generated·May contain errors
Video Summary
AI GeneratedThis panel discussion examines the integration of AI into healthcare, biotechnology, and life sciences. Experts analyze the transition toward computational medicine, the synergy between mechanistic modeling and machine learning, and the systemic cultural and educational shifts required to overcome barriers in drug discovery and clinical translation.
Welcome everybody to the AI plus Health and Life Sciences panel.
It's really an honor to be moderating this all-star team. Very, very quick introductions.
We've got Ray Winslow at my far left over there, Director of Life Sciences and Medical Research
at the Ruh Institute. Ray's an expert in computational medicine and has really built
his career on multi-scale modeling of the molecular basis of heart disease and arrhythmia
and predictive analytics in acute and home care settings. Next to him is Aileen Huang Saad.
She is the Director of Life Sciences and Engineering at the Ruh Institute and is really
involved in the programming and education of AI and life sciences. And Aileen, I just learned that
you were promoted or added to the Maine State Workforce Board, so congratulations on that.
That's really, really cool. Yeah. Mel Landon is next to Aileen over there. She is the Chief
Strategic Officer at Cyclica, which is a leader in AI-enabled technologies for drug discovery.
And Mel has held numerous other leading positions at Schrodinger and Academia and Farm and
the list goes on. Jared, next to her, is the Associate Dean for Programs and Graduate Affairs
at the College of Science and also the Director of the Battle Lab, I love the name,
Biopharmaceutical Technologies Lab. And then finally, last but not least,
Laurent, who's going to help me moderate, is the CEO, founder, co-founder and Chairman
of Parthenon Technologies. Welcome, Laurent. So, when we think about AI and health and life
sciences, I'm going to maybe make a slightly provocative statement that the stakes are really,
really high, right? Let's think about the good, the bad and the ugly for a second in healthcare
and life sciences. In the best case scenario in healthcare, a decision will lead to a good outcome
or it might not lead to the desired outcome, just no change at all. And the ugly, where a decision
in healthcare can lead to an undesirable, devastating consequence. So, the stakes are
really, really high when we talk about AI's role in this space, where AI can potentially
be transformative in improving a clinician's decision-making process to improve somebody's
welfare, but at the same time, AI's involvement in the decision can potentially lead to an awful
outcome. And this is a very, very different, new-to-us component here when we think about
mistakes compared to the mistakes that could be potentially made by a human alone. Like
Ricardo said earlier, right? To err is human, but I'm not sure that the technology, we're ready to
deal with the technology erring yet. Let's imagine some of the spaces like medical errors. The third
leading cause of death is medical errors after heart disease and cancer. I mean, this is a
huge space where AI can improve things, but are we there yet? So, let me begin with a few questions
to the panel, and I'll start with Ray. Ray, 66 years ago, AI was coined at a small summer
workshop in Dartmouth, and here we are 66 years later. Where are we now in all this time since
the term was coined initially, and where do you see the future of AI going? And I don't know,
Ray shared with us a slide that he wants to use. I don't know. We don't need it. It's not,
it's not. We don't need it. All right, there we go. Gene, I thought your comment was right on.
I don't think it was provocative. I think it was right on. The stakes are very high.
There are two things that I often say when I give a talk or when I'm speaking, and they tell
you where I think AI and life sciences and medicine is going. The first is medicine and health care
is increasingly becoming a computational discipline. I think it's absolutely a true
statement. The second thing I like to say is that algorithms will become a member of the health care
team. That may or may not be true. We don't know. It hasn't been tested yet. There are plenty of
papers and bodies of work out there showing the value of predictive algorithms and diagnostic
methods in medicine and health care, but precious few of them move to the translational stage
where their value is firmly and convincingly tested in a health care setting where we know
whether or not they improve the treatment of patients and whether or not they impact on the
cost of health care. I think that's one incredibly important direction for AI and life sciences and
medicine to go, which is to the translational step. There's a reason that you don't see
a great deal of translation being done because it's incredibly hard to do. It's a very demanding
thing to build algorithms into the informatics ecosystem of a health care provider that's not
ready to yet accept that. That's a huge step. I will say that one of the things...
I'm going to make a comment about the IEAI. I think an important aspect of the IEAI,
I'll call it savvy. By that I mean every time I talk to Usama, I learn a tidbit about
software engineering and the practical issues of technology development and data analytics,
often from the business perspective, but I really appreciate Usama's business perspective
on the challenges that you face when you bring this technology into application. I think having
that experience is going to be really valuable for us as we move forward with clinical translation.
I think the big challenge, the big opportunity for AI is translation, but there are a couple
other more fine-grained, detailed comments that I want to make. I use this term computational
medicine to refer to the process of building quantitative models of disease, constraining
them using data from patients, and then harnessing them to do good by patients.
I use the term modelling, not AI, for a reason. I think there are two broad approaches to
computational medicine out there. The first is what I would call mechanistic modelling.
Modelling that's driven by really detailed, deep insight into how biological mechanisms work
in health and disease. These kinds of models usually take the form of ordinary differential
equations, which are numerically simulated to do predictions, to make predictions.
The strength and weakness of this approach are that the strengths are they're foundationally
based on mechanism, constrained by fundamental physical principles about how biological and
chemical systems work. The weaknesses, they're based on those things, and that kind of information
and insight is really hard to get in biological systems. It's a very slow process building up
these models. The new emerging parallel track is to learn how to relate properties of biological
systems to phenotype, biological phenotype or clinical phenotype, through machine learning
methods. You learn a function f that takes you from things that you measure in biology to a
phenotype that you want to predict. It doesn't need to be based on mechanism, and so it's faster,
and it can be accurate, and it can be reliable, and you can gain insights into mechanism from
doing this. The point I'm building to is I think an opportunity is to bring these two methods
together, these two approaches together, to understand how to harness machine learning
under the constraints of the physical world and our knowledge of how real biological systems work,
to learn the mechanistic models more efficiently and more quickly from large data sets. So I think
that's an opportunity area. Second opportunity area is we have to understand how to do continual
learning from data. There's no AI algorithm out there that, if developed at one hospital or within
one health system, that can be transported and work as well as it did at that location,
at a different location, because the data distributions are different.
The algorithm may be the same, but the data distributions on which the algorithm operates
are different, and so we need to be able to continually learn the algorithm, even to apply it,
and continually learn from data as it accrues to make algorithms better, and this is a big deal.
This is a challenging thing. It's traditionally been called the, this is what the phrase the
learning healthcare system really means to me. That's a very popular phrase out there. It was
coined about 10 years ago, and to me it means really using machine learning to do continual
learning in the healthcare environment and improve the performance of systems all the way.
Final thing I'll say, and then I'll stop, is that I remember the first time I talked to Usama,
when I came to the room, and he used the phrase human in the loop, and it stuck in my mind because
I'd never heard it before, and I thought about our loop, and that's what's in the slide gene,
the loop that's up there. Oh great, data of different kinds coming to predictive algorithms,
machine learning methods, that make interpretations of the data, and so the first part of the loop
are the caregivers, the people who AI is making recommendations to, not decisions, recommendations,
and these caregivers can be all different kinds of caregivers. They could be the charge nurse,
they could be a technician, they could be a physician that's an attending and experienced
physician who's head of the ICU, and they use these recommendations as another piece of information
to feedback and helpfully improve the condition of the other group of people that are in the loop,
the patients. So there's multiple kinds of people in the loop, there's multiple kinds of caregivers,
and the caregivers who are acting on AI information are trying to use it in a way that benefits the
ultimate people in the loop, the patients, and so doing all of this is going to be an enormously
challenging task, and one that's right for this kind of organization. By that I mean Northeastern.
Thank you, Ray. You know, one thing that you said that I think is a nice segue
to my next question is about building the mechanistic model, a mechanism model of a system,
but as we know, the body is a complex of multiple systems, and building mechanistic models of
multiple organs and systems in the body and then tying them together is no
easy lift. So my next question to Laurent is, can we build AI ML models that are agnostic to
disease or even generalizable across multiple systems in the human? Thanks, Gene. So let me
address this question, but first I also want to acknowledge in my own ways why we're here,
why I'm here, and so Osama, I mean, I think you're doing an amazing thing. I think the
university, yourself, all the supporters doing this around life sciences and more broadly across
so many sectors is absolutely the right time and the way to go, and I'm going to seg into
providing a perspective on the question, and mostly from a perspective of let's call it the
biotech pharma slash life sciences lens here. Five years ago, AI was not mainstream in biotech.
I still remember three years ago, people like rolling their eyes, you know, and I think in
some ways it's because we haven't done a very good job at defining what is AI, how can it help,
and I think it's because there are a few good case studies and there's a lot of fluff, and so
having an institute like the IEAI that can actually put a little bit of a shape and be so
cross-disciplinary as I think is an amazing thing. So I'm just really grateful to be part of the
discussion and part of the story, and I think, you know, the future is ours, let's put it this way.
But just to go back more specifically to the question, so the answer I think, like often,
right, it's a little bit of both. So, you know, I was trying to structure a little bit
how to describe one way to think about the question, and so in the field of biotech pharma,
as it applies to life sciences, there's probably three verticals. One is around the technologies
that are enabling. The second one is around our insights into disease biology, and then the third
one is around the iterative processes, basically the back office, the loops, the ability to reduce
cycle times, the ability to reduce the cost curve of everything we do. And so going back to the
first one, you know, it was amazing. I was super excited, you know, what was this, maybe 12, 18
months ago, I can't remember, you know, alpha fold, right? We're going to use AI to predict
the folding of proteins. Really cool. Problem is, you know, there are 65,000 structures already
uploaded in the protein database. That doesn't really mean that we'll be able to bend the curve
of healthcare just because now we can go really fast in terms of predicting the structures of
disease-causing proteins. However, it is a little piece in the long value chain that ultimately
will lead to us better understanding disease. So that piece, the technology, I think is somewhat
disease agnostic, but it does require loops of teeing up the hypothesis, testing it, and then
showing that you're on the right track. So I think, at least in the life sciences, you know, we talk
about loops for AI and the IAI. I think it's already wired into the everyday workflow. That's
kind of how the mindset is, right, of most practitioners, is that you use the data to
generate hypotheses, and then those hypotheses are tested, and then you make sort of some improvements,
or you've refuted it, or you go on to the next hypothesis. So technology-wise, I think there's
an opportunity. So one example was alpha fold protein structure prediction. I think that's
okay. The other one is on the design of medicines, right? There it gets a little bit tricky because,
at least from my side, I don't think I buy yet the headlines that we've discovered novel medicines
using AI. I just don't quite get it. I don't think we have enough data to actually have this
end-to-end AI type of solution already sort of off the shelf. So I think the technology side
on some fronts is probably more ready than others in terms of readiness for AI sort of output.
The second one is disease insights. So that one is fascinating, right, because ultimately that
speaks directly to your question, Gene. And I think there we're dealing with really sort of the core
of this, let's call it multi-dimensional, multi-parametric data sets, right, big data. And that's
really what life sciences, biotech by extension, pharma is all about, right? If ultimately you want
to bend the curve of quality of life or of life expectancy, you got to understand diseases in a way
that's actionable. And I think that's where you probably cannot be agnostic to the disease. I think
the data sets that you generate are so unique to the specific disease, the specific pathophysiology
that you're interested in, it's just right now it's not upon us. And I'm having a hard time even
within the next decade or two thinking of big algorithms trained by big data sets that are
disease agnostic. So I think that second pillar, I think that's where it's not agnostic. The third
one, this iterative processes part where because of our ability to trial and to repeat quickly, I
think that's where actually we can probably think of disease agnostic based approaches. So it's a
bit of a mixed bag. And I think from at least my perspective, we might get there sometime in terms
of this disease agnostic AI platform, but at this point we're not there just from a blanket statement.
Yeah, so I mean we may not need it at every stage, right? Because at some point it'll lose the
precision medicine, personalized medicine component if it's disease agnostic. But maybe we need it for
the discovery, the earlier stages. Yeah, absolutely Gene. And I think there are experts here on the
panel that will provide a lot more insights and sort of better context to that comment. But I
completely agree with you. So kind of answering then your comment from another perspective, if we
break down sort of again into three categories, how do we get to real world impact of our
understanding of life sciences? So let's talk about the preclinical space. So I agree with you in that
preclinical space. So before you get into clinical trials and you start to really test your hypotheses
with therapeutics or otherwise, I think that's probably where you're more likely to get
agnostic type of impact in that preclinical space. As soon as you get into the clinical space,
yeah, in this era of precision medicine. And I think to your point, maybe we are beginning
that journey even though it's mainstream, but we're only going to do more of it. We're just
scratching the surface. Absolutely. And then the third one, which is real world
evidence of the impact of all of our knowledge and manipulations in clinical trials and in the labs,
I think that's also where I just don't see it being agnostic. You know, we're going to have
to put on the lenses that's fit for purpose for that particular condition that we want to address.
Jared, let's move beyond drug discovery. What are the opportunities faces in other
sectors of life sciences? Yeah, thanks, Gene. And that's exactly what I was thinking about
when you were talking around. So that's fantastic. And kudos to Osama, a fantastic opportunity here
in the backdrop of Northeastern. And I say that on purpose, because, you know, if I look around
the room, I can probably count, I don't know everyone, but I know a lot of you. I can count
on my hand how many of you I would consider life scientists or health scientists, right?
So there's an opportunity to really leverage the infrastructure and all of the institutes
and technologies at the university to really let's do AI together, right? And using AI as
a tool, because generally speaking, AI is more accepted in the biotech space, but still,
I'm a life scientist. AI is a black box to me. I'll turn to my friend Mel, who I've known for
a long time to help me with that, right? Or Laurent, right? Generally speaking, we don't have a clue,
right? So there's a huge opportunity and challenge there, which I think is exciting.
And to that extent, and to your question, Gene, really an opportunity, I think, to think outside
of the box. Drug discovery is crucial to humanity, but I think in the AI space, it's a low-hanging
fruit, which you should go after. But what other fruit are out there that are just as impactful
that one should pursue, for example? And I'm not saying something super complicated or
up in the clouds, right? Like a lot of my time is spent thinking about biotechnology, which I can
probably guarantee if I pass this microphone around, I would get 100 different answers on
what biotechnology is, but essentially using cells to create drugs, right? Or a product, in our
case drugs, or in my case drugs. So we do that using lots of upstream and downstream processes
that in big pharma are really locked down, right? But if you start thinking about the new technologies,
the new medical technologies coming to market, cell and gene therapies, for example, we've probably
all heard about CAR-T cells. Those are going to be manufactured next to or adjacent to a hospital,
right? There's a tremendous opportunity for AI to play a role in ensuring those products are made
at a high quality so that when the cells are taken from Osama this morning and 15 hours later he's
getting his manipulated cells back, they're what they're supposed to be, and that AI has ensured
the quality of that product is instilled throughout the process, right? So that from the first cell
that's grown, nothing's happened to it that's changed it, and if things are changing, there's
ways that information can be fed back to the system and thought about how to manipulate the
system to change processes so the product's proper or appropriate, right? And that's where I spend a
lot of my time, and actually the AI and ML space, which is not my background, but thinking about how
we can apply those technologies in the manufacturing space, for example, or more broadly speaking,
biotechnology, which is an application-based science.
So I want to take a pause here and allow folks from the audience to ask questions before I go
to my next question. I was hoping there would be some sort of interactive dialogue.
No, so while we wait, Mel, you know, as I was preparing for you guys, and some of this is
outside of my research area of expertise, let's say, and I was doing some homework,
there are 10 to the 60th power small molecule compounds out there. And to say that it's big
data, which is a word that's been recurring all day today, is an understatement. So you in industry
and the pharma and drug discovery space and having a role, a leadership role, in the strategic
development of your organization, what do you see as some of the main barriers for using AI
in drug discovery? Yeah, I think, first of all, just to go back, Gene, to your earlier comments
about, you know, the potential impact in both directions from a human health perspective,
there's also a huge economic impact, which we realize. So, you know, raise of hands, how many
of you have gotten a prescription at the pharmacy and thought, Jesus, that's a lot of money to
spend on a drug? All of us. Okay. Well, what you may not realize is that over 90% of drug discovery
programs prosecuted in the pharmaceutical and biotech industry, starting with, you know,
disease biology and going through and trying to make a drug, over 90%, probably through 95,
97% of them fail. Most, two-thirds of them fail actually before they even get into clinical
development, before they even enter a patient for the first time. And so if you account for all the
failure, it costs on average $2 billion to bring a drug to the market. Okay. And so how that
translates to pricing is that when you are paying for your prescription drug, you are paying for all
the failures that have incurred, right? So I think one of the hopes of really this industry is
to bring that attrition rate down. It also takes, you know, typically over a decade to bring a new
drug to patient, right? And then you talk about, you know, you have typically patent life, about
20 years on a drug. So by the time a drug has entered the market, you have about 10 years to
or less to recover your costs. So I think there's huge economic reasons for wanting, you know, I
think the promise of AI and other enabling technologies is to drive down, you know,
attrition rate and cost, right? And failure. So going back to your question, Gene, I think
one of the challenges, someone recently referred to drug discovery, and I think this is true of a
lot of life science research, and I apologize if I'm offending any life scientists in the audience,
I started my career as a bench scientist before transitioning to a computational scientist.
It is what people, a friend of mine referred to as very artisanal in some ways, right? So you're,
you know, you generate data, you develop a hypothesis, you generate data, you perform
experiments. And then typically what happens is, you know, if you're generating a fair amount of
data, you go to a, you know, to a computational scientist and say, hey, I got all this data,
what can I do with it? Help me derive insight from this data. And really, in order to derive
maximal value and impact, and this goes to, I think, a comment that Gregory made earlier in
one of the discussions, is to incorporate this design thinking into your engineering mindset.
So how are you actually designing experiments to, and thinking about, and also how are you
scaling experiments and generating the right kind of data? Because we know biological data is
extremely messy, it's very heterogeneous, it's a lot of noise, there's a lot of bias, there's a lot
of, you know, areas where you have a lot of data, and then there's a lot of areas of low data.
So how do you actually, from beginning, design experiments so that you get the most value
out of an AI algorithm? And going to Ray's earlier point, how do you actually maximize learning?
So I think there's part of, there's a cultural challenge there, because as scientists, we don't
ever really want to be wrong, right? So what happens is, is that you typically will use an
algorithm, or at least, you know, this is I've seen in my 20 years working in this industry, you know,
to, you will, you know, a human will look at a prediction, say, okay, I'm predicting that these
things are going to work, therefore, I'm going to test these things, and only these things. Well,
what about the other things that you that, you know, an algorithm may be predicted wouldn't work,
right? We don't test those, because we don't want to potentially test something that could be wrong.
And in fact, up until recently, synthetic chemists and pharma were rewarded based on how many
compounds they made that were active, right? So they were incentivized not to make something that
they think wouldn't work. So from a learning perspective, from, you know, an AI perspective,
you know, we're never going to really train algorithms to perform well and be robust,
if we don't design experiments to maximize this learning, appreciating that we're going to have,
you know, success and failure. So I think that there's a cultural sort of shift, if you will.
And then the other thing, I think this goes to Laurent's earlier point is that this is a very
complicated process. And if you're only using, let's say, you know, scalable technology at,
you know, points A and B, but you have A, B, C, D, E, F, G steps, and those steps are still done
in this sort of artisanal, low throughput manner, then you can only move as fast as your rate
limiting step. So really, you have to think about at each point, how do you, again, design experiments
to maximize learning to scale, and so that every piece of the process is moving at the same pace.
Get off my soapbox now. There you go. I'll get on your soapbox with you though,
Mel, right, because I think as a trained scientist who generates data, right,
I'm doing exactly that. I'm looking at one piece of data, the information that I'm wanting. I'm
looking at a whole cell perspective, ignoring 99.9% of what's happening in cells to find out
the one piece of information I want, missing God knows what in the process.
And what do you do with the rest of that data? I store it on a hard drive and shove it and let
it collect dust. Yeah, so I mean, in my industry, in drug discovery, the challenge of AI in general
algorithms, and there's, I think, at this point now, 300 companies in biotech claiming to do AI
enabled drug discovery, they're all training on data that's largely positive data. They're not
training on negative data, because that data doesn't exist, right? There is no journal for
negative results and negative data, right? And so that represents a huge challenge, frankly.
And in that space, all of that data that is collecting cobwebs on a shelf somewhere,
is it usable for future ML applications? Or do we have to go back to the drawing board
and collect the data in a completely different way? It could be, but it's how do you access it?
And that is another challenge, is data access, which we could talk about for a whole other hour
and I will not get in that. Gene, I saved my data in the hopes that that can be done one day.
Seven years, at least seven years. So one of the recurring themes has been that none of this can
be done in a vacuum. I mean, this has been a pervasive theme all day, that we need experts
in machine learning to have domain expertise and people with domain expertise to be able to have
some computational training. And so I want to move on to Eileen, who has been really involved in
many different types of program development initiatives in STEM. And I know that that's
what you're also doing here at the Ruh Institute. So what does it take to create this new
interdisciplinary type of workforce? Great, thanks. First of all, I want to thank you for
allowing me to join this great panel, because it's a real honor to sit with these people that
are really advancing the field. And I think being able to add the lens of how do we address it from
an education perspective is something that is something I'm very passionate about. And I loved
hearing all of our deans talk about how they're interdisciplinary and how we're trying to push
the envelope on these things. And I think AI and healthcare is one thing that is particularly
valuable in the world today, because it shows how complex and interdisciplinary the world has become.
And in order to address those things, we as higher education institutions have to do
better at making sure we're preparing our students to address these issues. And
as discussed earlier, as people were saying, we need to have a closer connection with industry.
We have to understand the problems that need to be solved. The other thing we need to understand
is that it's our responsibility to make sure that our students understand how the context
of which the problems that they're trying to solve and the implication of those things, right? So
I'm a biomedical engineer by trade, so I think about things, everything in terms of a technology
developer. And as a technology developer that is both in the healthcare space as well as in
biotechnology and engineering, it's really a complex educational system that you have to
put these students through, because you're trying to make sure that they understand all
the technical depth of each of these disciplines that they're impacting. But that being said,
it's not enough just to make sure that they understand the technology, because they have
to understand the impact on their stakeholders. And the stakeholders run wide and deep in these,
because there's the user, there's the implementer. In healthcare, there's the physicians who may or
may not even be exposed to what AI is or don't have a trust for it. You have patients that are
taking the resulting products. So you really need to think about all of those pieces. And I think
Northeastern has positioned itself where we have an opportunity to really think about what this
interdisciplinary education looks like, because traditional higher education really is something
like a mosaic, right? Where we're teaching each of these students individual tiles on a mosaic,
where you're expecting them to understand the color and the depth of each tile on that
print. And you're not helping them understand what that entire print looks like. You're expecting
them to be able to weave that together. And there are fundamental pieces that thread all of the
disciplines together that I think that the students need to understand, two of which are
user-centered design and ethics, right? So how do we make sure that this is part of their
educational system? Because traditionally, what you'll say is, oh, go take an ethics class,
right? Or go just take a user-centered design. Or we'll take one freshman experience that exposes
you to everything, and then we'll just expect you to go and pick up all these pieces later along
the way and just hope it works out. I think with Osama's mission and the idea of having a learning
hospital and things like that are an opportunity to really say, how do we scaffold a learning
experience in the context in which they're solving a problem so that they can take the
critical pieces along that journey and really help advance them? We know that when they understand
the context that they're deep embedded in the problems, their motivation will increase,
the learning increases. There's a huge impact there as well. We also potentially could talk
about how do we just shake up the entire system of experiential learning and say it should be
team-taught classes rather than inter-single classes being taught at them once at a time.
So I'm really excited about where we can go, particularly with the support of all of the
deans today talking about how we need to be interdisciplinary. So not only in research,
but I hope we see that in education as well. So basically, Eileen, you're saying you want
Osama to make sure you and I speak the same language? Absolutely.
So, Eileen, so is it easier to train, you know, a computational engineer some life science or
health science skills or the other way around? And don't say both.
I've heard that since I was training to be a biomedical engineer years ago, right? I mean,
well, a couple years ago. But, you know, I can't tell you what the truth is, but I will tell you
most of the industry will say I would rather have a well-trained engineer that I can teach
the life sciences than vice versa. So you can take that however it is, but that's what I hear.
Thank you. If I could comment on that. Yeah. I mean, I actually kind of find that
my experience has been sort of the opposite. And going back, I think, earlier today
when the panel and Guy was speaking to the need to, you know, there's a dearth of data scientists,
and that is true in the talent pool. And yet at the same time, we're talking about how important
it is to actually understand what the real problems are. What I find is that if you have
scientists who people who understand what the real problems are and who have maybe, you know,
have some battle wounds, that, you know, as tools get better, as they become easier to use,
if they have that intellectual, you know, curiosity to learn how to integrate those
tools, that that actually results often in more impactful application than bringing in, you know,
someone, you know, who's maybe a hardcore data scientist, but, you know, has no real
appreciation. And that's not to say, so, you know, in my company, we hire a lot of these
hardcore data scientists and software engineers, and the ones who do really well are the ones who
are very, I guess, scientifically curious, right, and interested, right? So, I can see it working
both ways, but I have seen, you know, especially as tools, I think, as, you know, I don't know that
AI maybe at some point will become a commodity, but, you know, as tools become more accessible
and easier to use, actually having people out, you know, in the real world who can just learn
how to apply these tools to their problems will, in some ways, be more impactful.
Okay, I sort of agree, but I want to make one. So, I guess, to my point, I think it's what we
would be doing a disservice to our students if we don't teach them how to go deep in an area,
right? So, they need to be able to go deep in an area, but they also need to be able to
work across disciplines, right, which is why it's so important, I think, we have this experiential
learning so that they can think about problem solving, and part of the reason why I say we
should have team-taught classes is so they have an opportunity to see how a problem is solved
from different lenses, right? And so, that, I think, causes this inquisitive nature that you're
referring to, but it's really, I think, one of the dangers I have with tools just being available to
people without being inquisitive about the nature of what that tool is telling you, that can be very
dangerous. So, the students have to still be able to have an intuition of where the results should
or should not go because we can make some very poor decisions. I agree with that, and education
is key, right? You have to educate, you know, users on what, you know, again, what is the domain
of applicability of these tools, what are the potential, you know, repercussions, you know,
what can these things do well, what can't they? Otherwise, I agree with you, it's extremely
dangerous, yeah. Can I make a quick comment? Eileen and I are from the same graduate program, so it's
not surprising we see eye to eye on this kind of, and I think an additional point is that
in the program we came out of, it was largely engineers who went into that PhD program,
and they were learning about biology. So, they came in well-trained in engineering,
continued training in engineering in depth, and trained in biology. And so, the fundamental
premise of the program was we're going to train you as deeply in engineering as any engineer out
there, and we're going to train you in the biology of the problem you're working on as deeply as any
biologist out there. I don't think there are any shortcuts. I really believe that the people who
are going to be powerful and make advances really span both worlds in a really meaningful way.
Just to maybe add to the discussion, so I completely agree with the cross-disciplinary
training, you know, what comes first? I think we can debate this, but I'll just maybe offer
a perspective to try to add two aspects here in terms of training. One is defining the right
problem. I mean, that is such a critical limitation, at least in the companies I've been,
the industry I've been. You know, people who are able to capture what is the problem statement,
they tend to have a better grasp of the multiple pieces that are required in order to at least give
yourself a shot at the solution or solutions. And so often it's a bit of a handicap when we as a
company, as an organization, as a team, we're not able to actually put our finger on what is it
exactly that is the problem here. So what it's worth, you know, if there is some, I think there's
a class here, but you know, if somehow, you know, students, postdocs, you know, rising stars,
professors, et cetera, could actually be thrown into an environment where we are all asked to put
our finger on the problem statement, that could really help us in terms of integrating the pieces.
The other one is, you know, frankly, I can just see it on the job description, you know, has to
play well with others. So everything we do in our sector, it's all about what you don't know and what
others can bring to the table. And so that, you know, it's a good match, because at least in the
life sciences, and in particular, in the subsector of biotech pharma, there's no one person that's
going to carry that, you know, flag all the way across the finish line, right? And so yeah, you
need people that lead, manage, et cetera. But frankly, that's okay. I mean, what's more interesting is
how do you bring the pieces together? And that's a behavioral trait that, you know, frankly, how do
you select for that, right? And I know it's not a university conversation, per se, for today's panel,
but you know, the way we interview, right, we don't do a very good job interviewing for those behavioral
traits that are so critical once you hit the ground running in a work environment. So if there's,
you know, some way to kind of reverse engineer the coursework, that actually could allow us to
A, do a better job at problem statement, and the second one is a better job at teamwork. I mean,
that would be, I think, a big win, IEI and even beyond. Well, you just gave a commercial on why we
should teach engineering design to all disciplines, because that's exactly what we do, is start with
the problem statement and work on teamwork. And this is part of the reason why I think about
how we can change the curriculum of higher education, of starting with a problem to be
solved, right? And have a continuum of one problem that the students can be working on throughout
their career, and have them motivated to seek out these other opportunities, these disciplinary
classes that can actually advance their problem statements. But we can sit, and I can talk about
this all day long, so we, you know. I think just one comment, Laura, I mean, I agree with you 100%
about teamwork, and I'm just going to say, you know, not to be critical of academic institutions,
but the way that we train graduate students runs antithetical in a lot of ways to this,
because we train them, we're like, you have to be the master of your thing. You have to demonstrate,
you know, that you are the best at doing this, right? And you have to be first author on these
papers, right? And so forth. So there's so much, and if you look at how PIs, you know, get grants,
it's about, you know, their individual research a lot of times. I mean, there's, you know,
cross-disciplinary grants, and so forth. But I think what I've seen when I, you know,
hire scientists fresh out of either a graduate program or a postdoc, is like, they have not,
they don't have a concept of working in a team, because that's just not something that they've,
you know, frankly, been encouraged to, or incentivized. Except at Northeastern.
What's that? Except at Northeastern. Except at Northeastern. So I think as an institution,
if you, if Northeastern can train their scientists in particular, to embrace that,
that is huge. Because, you know, that is typically, it is not about people's technical
abilities. I mean, in this city, there's an incredible amount of talent and expertise.
But it's really, in this field, it is a team sport. It's 100% a team sport. And so how do
you instill that at the graduate level? Instill, you know, reward the individual, but also reward,
you know, them playing with others. I think that's a great place to end.
Are we out of time? Do we have time for questions? All right. Well, let's say thank you.
