AI or Not

E042 – AI or Not – Ryan Coffee and Pamela Isom

• Pamela Isom • Season 2 • Episode 42

Share episode

Welcome to "AI or Not," the podcast where we explore the intersection of digital transformation and real-world wisdom, hosted by the accomplished Pamela Isom. With over 25 years of experience guiding leaders in corporate, public, and private sectors, Pamela, the CEO and Founder of IsAdvice & Consulting LLC, is a veteran in successfully navigating the complex realms of artificial intelligence, innovation, cyber issues, governance, data management, and ethical decision-making.

What happens when a philosopher becomes a physicist and then stumbles into artificial intelligence? You get the fascinating perspective of Dr. Ryan Coffee, Senior Staff Scientist at SLAC National Accelerator Laboratory, who shares his unique journey and vision for our technological future.

Dr. Coffee takes us deep into the world of federated machine learning, explaining why preserving intellectual property—what he calls "secret sauce"—is crucial for innovation while still enabling collaboration across competitive boundaries. As someone working with an X-ray laser that generates a million frames per second, he's tackling data processing challenges that require AI systems capable of making decisions in microseconds, creating what he calls "autonomous science."

The conversation expands into energy systems of the future, where Coffee envisions interconnected microgrids with diverse power sources managed by intelligent systems. Rather than choosing one energy technology over another, he advocates for an ecosystem approach where each source—from nuclear fission to renewables—serves its unique purpose. His enthusiasm for modular nuclear reactors deployed near data centers reveals practical solutions for powering our AI-driven future.

Perhaps most striking is Coffee's timeline for these transformations. While many experts talk about technologies being decades away, he believes we're underestimating the pace of change: "If you think we're 10 years away, because humans are linear thinkers, it's really three." This accelerating innovation means we'll soon see technologies we can't even imagine today.

Coffee leaves us with a powerful analogy about the interface between AI and humanity, functioning like the corpus callosum connecting the brain's hemispheres. His call for psychologists, artists, and humanities experts to help shape this interface reminds us that creating our technological future requires not just engineering expertise but the full spectrum of human understanding. Listen now and discover why our autonomous future is arriving faster—and looking different—than you might expect.



[00:00] Pamela Isom: This podcast is for informational purposes only.

[00:27] Personal views and opinions expressed by our podcast guests are their own and not legal advice,

[00:35] neither health, tax, nor professional nor official statements by their organizations.

[00:42] Guest views may not be those of the host.

[00:51] Hello and welcome to AI or Not, a podcast where business leaders from around the globe share wisdom and insights that are needed now to address issues and guide success in your artificial intelligence and digital transformation journeys.

[01:07] I am Pamela Isom and I am your podcast host.

[01:12] I am so excited to have Dr. Ryan Coffee,

[01:16] senior staff scientist from the SLAC National Accelerator Laboratory, with us today.

[01:22] Ryan, congratulations on all of your success and thank you for being here. I think the listeners are going to be excited and anxious to hear from you. So welcome to AI Or Not.

[01:35] Ryan Coffee: Great. Thank you. It's really wonderful to be here, Pam. It's always, always fun talking with you.

[01:40] Pamela Isom: This is going to be exciting. So let's start with you talking about your story and your journey.

[01:47] So we have people that probably don't know you or if they do, may not be familiar with your journey. So tell me more about your background and your slack experience.

[01:57] Ryan Coffee: Oh, man, we can go all the way back to when I was a, you know,

[02:01] an aspirational coffee shop owner and philosophy major, or we could just start with when I fell in and stumbled into physics. Which one do you want to go for,

[02:09] the long story or the short, the condensed one?

[02:12] Pamela Isom: I think you can go with whatever you want.

[02:15] Ryan Coffee: All right. So, you know, all I would say is I started out in philosophy and that, you know, clued me into quantum mechanics may have been being misused. And that's what caused me to stumble into physics.

[02:27] And then physics led me into this. You had mentioned in one of our previous conversations about this atomic, molecular and optical physics,

[02:36] I wanted to figure out how quantum mechanics matters to the world. And so that's what drove me into lasers and molecules. And that brought me into slack to use their X ray laser to study molecules.

[02:47] And now that we have a machine that runs at superhuman scale, it drove me into AI for this case of we have to process this information as it's coming in before it hits disk.

[02:58] And so that's. It's been an evolution of taking options as they come and somehow being lucky enough to fall up.

[03:10] Not many people get that opportunity. So if you do enjoy it and make them and use it for good. And so that's what we've been trying to do when it comes to AI's use in science is how do we make sure this thing gets used Correctly and well for the sake of all of humanity's use of science and its output.

[03:31] Pamela Isom: So, AI artificial intelligence accelerators for streaming data interpretation and information. So let me, let me read this quote that you have, Yaddle's quote that says atomic,

[03:41] molecular and optical physics is a scientific home base that serves as proving ground for high speed sensor edge AI accelerators for streaming data interpretation and information extraction.

[04:01] Okay, I'm going to need you to explain,

[04:04] to disentangle it.

[04:06] Ryan Coffee: Yes, I have a tendency to go very long in the sentence and very technical detail instead of summaries. I'm not a summarizer.

[04:15] So let's see,

[04:16] my home base as I mentioned, is atomic molecular physics. And so I use open science and this sort of basic energy science to domain to work on the computing infrastructure that is needed for our real time processing of this scientific information that's coming in.

[04:33] And so that's what I mean by saying I'm using this as a proving ground or demonstration ground.

[04:39] Now our X ray laser is going to be running at a million frames per second. So if you're,

[04:44] if you follow data processing,

[04:47] man, that's way too many frames per second to handle.

[04:50] Pamela Isom: That's a lot of frames.

[04:51] Ryan Coffee: That's right. And so that's why I see this as something that is relevant for a broad swath of the future market as we move into this world of AI everywhere and autonomy everywhere.

[05:04] And when we go into that world, we need a few open science domain test cases and test benches. And so that's where I like to think about my science hobby.

[05:14] Please don't tell my employers that I see my science research as a hobby, but it kind of is,

[05:21] it's a hobby that I'm leveraging to prove out how you would do this technology for real time sub microsecond decision making in autonomous systems. Because we are moving into a regime of autonomous science where it's not people scratching their heads for a year, doing an experiment for six months and then scratching their heads for another year to figure out what happened.

[05:43] It's going to be in a tight closed loop. And that's what we call autonomous science, which is,

[05:49] I would argue here now it's not something of the future, it's actually of the present.

[05:54] And that's the thing that I use my brand of curiosity into quantum mechanics as a,

[06:00] as a showcase ground for what this kind of autonomous control system can look like.

[06:06] Now that's what I mean by, I use that in service of this broader application of AI and science.

[06:12] Pamela Isom: Okay, so what does a Typical day look like for you as a senior staff scientist?

[06:18] Ryan Coffee: Well, senior means you do a bunch of stuff that isn't just your research,

[06:23] basically.

[06:24] So I'm sure you experience this as well if you've been in a role or you've developed the community around you. And I've been at lcls. LCLS is Linac Coherent Light Source.

[06:34] That's the name of the X ray laser.

[06:36] And I've been here the whole time. So since it started, since actually before it started. And when you do that, you tend to accumulate connections across multiple domains of science and even industry.

[06:49] And so your day ends up looking a little chaotic in that you're advising students and postdocs. I tend to work with postdoctoral students. So these are folks who have their PhD already.

[07:01] In the rare case I will work with an undergraduate, which I have one who was really fantastic recently.

[07:07] And I warned him, I said, I don't know how to teach people. I only know how to work with people who already have their PhDs.

[07:14] Kind of facetiously I tell him that. So I am a particularly pushy mentor.

[07:19] So part of my day looks like driving people to do more than others might expect of them. But I think this is the time to do that, especially in science.

[07:28] So I would say it's probably about 30% mentorship. 30% trying to affect the strategy of how Slack sees its role in Department of Energy and how Department of Energy sees its role in this AI ecosystem.

[07:43] That is a national ecosystem. Right. It's not just for federal. It's also including our private sector partners. And so figuring out how to navigate that those waters and who to bring together to the same table, I think would be,

[07:57] is how I spend about a third of my time.

[08:00] Pamela Isom: Okay, so that's kind of a busy day. That must be why I have a hard time getting in touch with you.

[08:09] Ryan Coffee: Yeah, that's right. I mean, I didn't, I judiciously didn't tell you what my other third of my time is spent doing. I get it's probably less than a third, but I still like to code.

[08:18] I got into physics actually from philosophy because you could actually do experiments.

[08:23] And when you can actually do experiments,

[08:26] making measurements and making plots that actually show the effects that you're looking for is incredibly mentally gratifying to someone who came from a background where all we do is sit around and argue all the time.

[08:38] And so I think, you know, that's why I still reserve some of my time. 20 to 30% still code, still plot data, still make measurements and try to Understand what's going on here?

[08:49] What are the trends? Yeah.

[08:51] Pamela Isom: So good for you.

[08:52] So let me ask you this. So we have collaborated federated research, which has been fun.

[08:58] Federated data.

[09:00] Tell me more about federated machine learning.

[09:05] And what I'm really after is,

[09:07] first of all,

[09:10] I think that it is a great concept,

[09:16] but I'm not sure how real it's being applied and in particular because of security that we mentioned earlier,

[09:26] but also privacy. So tell us more about the real deal behind federated machine learning.

[09:36] Ryan Coffee: Sure. Well, I don't know if I know the real deal behind FedML,

[09:40] but what I would say is if you look at individual hyperscaler companies and the way they use AI to turn a profit, there's not really an incentive to federate.

[09:52] Their incentive is to really own or somehow keep in their wheelhouse as much proprietary information as they can.

[10:00] And so this idea of sharing across boundaries isn't really incentivized. If you look at large companies and how they make a profit.

[10:08] Pamela Isom: And energy companies. Right. I run into this with energy data all the time. Not the Department of Energy, but utility companies.

[10:15] Ryan Coffee: Sure, sure.

[10:16] There's a lot of, there's a lot of proprietary information there they don't want to share.

[10:20] Now the banks have it, right? Because the banks have been looking at how they make market decisions and make trading decisions. Whatever they're doing for their financial stuff, they realize they need to be able to share,

[10:32] but they're sharing with competitive collaborators.

[10:36] So they want to collaborate so they can make better decisions for their own bank.

[10:41] They don't want to share their secret sauce, but there is some information that they do want to share so they can make a better forecast of the market.

[10:48] And this is where I like the position of Department of Energy when it comes to AI strategy and how we're actually doing machine learning and using these models because we're a vendor agnostic player in this space and therefore we get to see the benefit.

[11:04] If you can link all of these independent private sector endeavors,

[11:09] also some of our internal endeavors,

[11:12] in a way that preserves that secret sauce for each player,

[11:16] but you still get to leverage everybody's coherence so that you do a better forecast of, say, the market,

[11:22] be it the energy market, be it the financial markets, if it's the banks, or for us, basically thinking about technology and inventions as a marketplace.

[11:33] And so if invention is a marketplace,

[11:36] then I would argue Department of Energy knows how to handle that because we are kind of an engine for invention in the nation.

[11:44] And as such, we're here to Spin off technology,

[11:48] right. We basically work on problems like Manhattan Project, like space. Well, space race was also NASA, but federally funded research has a tendency to do things.

[11:59] It's essentially a similar philosophy, as I was saying for myself, using amo science as a demonstration ground,

[12:06] Department of Energy will create particle accelerators, right, that smash particles very near the speed of light. They look at very basic energy science. They look at all kinds of research that is tickling the curiosity of astronomy and cosmology sort of everywhere.

[12:22] Why do we fund that?

[12:24] Well, partly because it's what's good for humanity intellectually, but mostly it spins off a lot of interesting technology that keeps our economy running and innovating and inventing.

[12:36] And so if we want to optimize that invention cycle and accelerate it,

[12:41] that's where federated machine learning can help. Or essentially federated computing for science, invention and market.

[12:49] And that's the thing that I think,

[12:51] if you don't have security,

[12:53] you keep it only in the domain of open science.

[12:56] And like I said, Department of Energy, I think, is similar to my philosophy, let's use open science, but let's do it to innovate so we can inspire a market and we can spin off technologies that will be national resources.

[13:10] And if you can't preserve the secret sauce, you won't ever work on those problems.

[13:15] And that's why I think federated machine learning is so important, because it helps you preserve the secret sauce for your own participation.

[13:23] It's this kind of idea of like sovereign data. There's data that you are happy to share with your collaborative competitors,

[13:31] but there's also data that you don't want to share because that's your secret sauce.

[13:35] And we.

[13:37] This is where I, I like to say that a third of my role is thinking about how do we set up the role of Slack in doe? And DOE in the national ecosystem is let's support this thing that helps people preserve their secret sauce.

[13:52] I don't want someone to avoid the DOE because they're afraid that their ideas are going to leak somehow.

[13:58] They shouldn't. Right.

[13:59] In particular, you know, because you were on the inside for a while.

[14:03] We worry about things that are absolutely secret,

[14:06] so. And need to stay that way. So if you want a trusted partner who can keep your secret sauce private,

[14:12] I would trust the Department of Energy.

[14:15] But we need to tell that story a little better, I think. And so that's what I'm trying to work on is letting that story of why federated ML gives you more than the sum of your parts.

[14:28] It doesn't compromise your secret sauce.

[14:31] Pamela Isom: Okay.

[14:31] Ryan Coffee: And so that's what I like about FedML. And why is it not like, I think,

[14:37] why doesn't everybody know that word and understand it? I think it's because you have success stories like Google or OpenAI or some of the hyperscalers that benefit significantly from people not quite thinking about preserving secret sauce anyhow.

[14:53] So I, I care about the secret sauce. I want, I don't want to take anyone's secret sauce. And I think most scientists actually feel that way when, when it comes to their own inventions.

[15:03] Pamela Isom: So from a security perspective, do you think that. What can we do for. From the security side?

[15:11] Ryan Coffee: From the security side, I'm really excited about some of the compute acceleration that's coming up. There's this thing called fully homomorphic encryption.

[15:20] It allows you to do computation on data without ever decrypting that data.

[15:25] And to me, that allows us to use this ecosystem in a dynamic computing ecosystem that the nation is building. Right. We just saw the announcements come out of Department of Energy for data centers on federal lands.

[15:38] We know that there's a support for dynamic energy to support such computing.

[15:43] But these are going to still be places of compute that are going to be shared by a broad swath of the community,

[15:51] not just, just open compute.

[15:54] And if it's not open compute, we have to secure it.

[15:57] Up till now, I would say homomorphic encryption was computationally too expensive to execute on on a large scale.

[16:05] I'm hopeful that this emerging technology,

[16:09] and we're trying to demonstrate it actually at Slack, that this emerging technology will accelerate it by multiple orders of magnitude and it becomes computationally acceptable.

[16:19] If you do that, then you preserve secret sauce, yet maintain the benefit of models that have intelligence across multiple domains.

[16:28] Pamela Isom: Homomorphic encryption, yeah, it's like super mathy.

[16:31] Ryan Coffee: But I, I love it because it's essentially like you convert your language into something that is uninterpretable by anything mathematically uninterpretable. But if you ask a question, you get a response back.

[16:45] And as long as you use the same key that you used for your own question,

[16:50] decrypt the answer,

[16:51] but no one else can because they don't have your key, which is really cool. I love homomorphic encryption.

[16:57] Pamela Isom: I like the idea that that will help scale federated machine learning. Yeah, I like that.

[17:07] Ryan Coffee: Yeah, I think you need it because we worry a lot as a nation about IP leakage in actual property leakage.

[17:16] And these federated systems, if they're not like really hardened secure,

[17:22] they can leak intellectual property.

[17:25] I once saw A presentation at Supercomputing a couple years ago where someone was listening just to the exchanges of the weights for one of these models for voice and they were able to reconstruct, without ever having seen one of the members of the training set, they were able to reconstruct their voice sufficiently enough that they could fool voice recognition systems.

[17:46] Yeah, yeah, this is, that is secret sauce that someone could actually extract that I'm sure our private sector partners, especially the startup world,

[17:57] absolutely cannot afford.

[17:59] And so we, if we're going to support innovation, we need to support our startups and our innovative projects that are developed in the nation.

[18:08] And that means that we need pretty good secret sauce preservation.

[18:13] And that's why I'm, as you know. Right. This is why I've been a big proponent of proper, secure, federated machine learning for a long time.

[18:20] Pamela Isom: Yep. That's why we're working together. We're going to keep on doing it, not going to stop.

[18:24] That's right. Actually, I want to discuss some things so I am paying close attention to the AI action plan.

[18:32] Okay. There are two portions of it that I know is relevant. All of it is relevant to you,

[18:39] but there are two portions that I know are relevant to you that I want to talk about a little bit more.

[18:45] You may have discussed some of this, but let's just kind of go in there because I'm pretty sure the listeners are going to want to know like what Ryan think about the action plan.

[18:54] So let's talk about this one section that says prioritize the interconnection of reliable, dispatchable power sources as quickly as possible and embrace new energy generation sources at the ecological frontier.

[19:08] To me, when I see slack all in here, right. I'm from energy, so I know that this is us. Right. But what does that mean to you when you hear that?

[19:19] Ryan Coffee: So I get really excited about the whole ecosystem of energy production in the us.

[19:25] I live in Silicon Valley.

[19:26] There are so many electric vehicles it is really hard to look at. I still drive my gas powered manual transmission convertible because I love it and I love hitting the gas pedal and I can hear that engine.

[19:39] So maybe I'm biased because I like my manual.

[19:43] Pamela Isom: I think people say that you're set in your ways.

[19:45] Ryan Coffee: I am a little set in my ways.

[19:47] Potentially that may be true. However,

[19:50] I believe that every energy production has its niche and sometimes you want an electric vehicle and sometimes you want a gas powered vehicle.

[19:59] And so by that same token, the energy supply for our nation is the same.

[20:04] Every energy supply has its niche. I don't think we can cut off any one of them for the favor of a different one. I am a proud proponent of nuclear fission.

[20:14] I think it is a very viable and interesting and required solution for our future.

[20:22] And when you say deployable energy sources,

[20:25] I would argue that that is this idea that we're doing innovative modular reactors that are sometimes they're called micro reactors so that you can deploy them. They've already been used in the military quite a lot for deployed energy.

[20:40] And now I think that is another role to play in domestic energy production,

[20:47] where you have energy where it's needed and when it's needed, and not necessarily forever. Instead of building gigantic gigawatt scale nuclear facilities,

[20:57] you could build 100 megawatt scale facilities where a new data center crops up, for instance,

[21:04] and that helps us support AI that might be a more consistent load supplied by something like nuclear fission.

[21:12] And then the dynamic loads that are used by small cities could be compensated with other energy sources.

[21:20] And so that way we can keep our vision away from our population centers. You could imagine there's a whole ecosystem of highway transport that might make great use of electric vehicles,

[21:32] electric autonomous vehicles that can drive themselves up to one of these plants, charge, and then continue on with their freight delivery. Right. There's a whole future that you can easily see with these kind of deployed solutions.

[21:45] I actually have a research project on machine learning for nuclear fusion.

[21:51] But you'll notice I'm really excited about fission.

[21:54] Part of that is that I made a plot recently where I was very excited about energy production over since like 1750 or something, since recorded energy, global energy production. And you see these jumps periodically and we're coming up to another one and that that one is happening essentially now,

[22:13] sort of this decade is what this trend seems to show. We've been through three industrial revolutions so far. The first was animal power to steam.

[22:23] The next was internal combustion.

[22:25] The next was computing in the 1970s.

[22:28] And this trend shows that there's one that looks like it's centered around 2025.

[22:33] So I have a feeling that we're right in the middle of the fourth industrial revolution today.

[22:38] And because of that,

[22:41] every time there's one of these jumps, it's the previous energy technology that actually supplies that jump.

[22:48] But it only happens when the new technology is demonstrated.

[22:52] And so that's why I'm optimistic that we will demonstrate a nuclear fusion pilot plant within potentially this decade. I know that's a little aggressive, but I also am a firm believer in exponential growth of technology.

[23:07] And humans think linearly, so it's hard for us to predict that these things happen sooner than we expect.

[23:13] So I expect fusion very soon,

[23:16] But I don't think fusion is going to give us the energy we need.

[23:19] I actually think fission is going to.

[23:21] And so that's why I'm enthusiastic about having a diverse portfolio of energy production,

[23:27] including these modular reactors. And I'm pretty excited to see what the next five years will unfold.

[23:34] I think we're going to see a very big difference what our life looks like in 2030 and what our life looks like today in 2025.

[23:42] Pamela Isom: Now it also talks about financial incentives with the goal of grid stability.

[23:49] So coming up and developing more financial incentives. But then it also says create a strategic blueprint for navigating the complex energy landscape of the 21st century, which is basically all connected.

[24:03] Right. That's all saying the same thing. It just needs a strategic blueprint. And I think part of that is because there needs to be some understanding of what some of these sources are.

[24:14] Because I know one misnomer is that when we talk about like microgrids, for instance,

[24:21] micro grids are typically thought of as renewable energy sources, when it really could be all types, which is what they have here, dispatchable power sources.

[24:32] And then they name nuclear fission and fusion.

[24:36] Ryan Coffee: Yep.

[24:36] Pamela Isom: But when you. Geothermal goes to the renewable side. But yeah, it's kind of like a misnomer that when you think about microgrids, you think that they're predominantly renewable sources because they could be other dispatchable power sources like traditional energy.

[24:53] Is that correct?

[24:54] Ryan Coffee: Yeah, And I think that's. To me, that's the thing that is similar to how we were saying we want science and private sector to be more than the sum of its parts by using federated machine learning.

[25:07] The same goes for our energy sector. These microgrids, if they're interconnected,

[25:12] you can imagine they will be more than the sum of their parts. If something starts to fail, they can draw a little. If something is overproducing, they can push a little.

[25:20] And so this idea that we have an intelligently interconnected ecosystem of microgrids, I think is not crazy. And I would imagine,

[25:30] and this is something where I like to look at human society as just another representation of a biological system, a population system.

[25:38] We're more resilient when we have more diverse micro grids. It makes us resilient as a nation.

[25:44] And so the thing that is extremely fragile in my mind is if you have one solution that is spread to all, I would rather have lots of solutions all working together because it is that sort of ecosystem of varied sources is how nature solves her problems.

[26:01] She uses lots of species to fill an ecosystem, not just one. If you have one, boy, are you going to suffer a virus or an infection,

[26:08] right? So that idea that you have a living, breathing energy ecosystem, I think is what this idea of critical control systems is going to help us with.

[26:20] It's too complex for a human to handle.

[26:24] But we're not in this alone anymore, right? We have,

[26:28] we're starting to see the growth of trusted,

[26:31] essentially AI partners that will help us act our goals and our actions at superhuman speed.

[26:40] And I think that's something that is arising now and needed for this ecosystem of microgrids. I think, you know, as this happens throughout human history,

[26:51] you always have the technology that gets invented right at the time that humanity needs it. And then you have a new energy source that's invented right at the time that humanity invented that technology to help with something.

[27:02] And so it all happens together.

[27:04] You could call it magic, you could call it deity, or what have you. I am formally an agnostic because I'm a philosopher, but I do believe there is something out there that manages to put humanity in the right place at the right time.

[27:18] And I think that's what's happening now.

[27:20] Pamela Isom: You said that to me before in other conversations. I know you mean it.

[27:25] You mentioned in a previous conversation about the American semiconductor manufacturing.

[27:32] Can you elaborate your perspectives on that?

[27:35] Ryan Coffee: Well, this is a little bit where I'm excited about what will happen to the semiconductor fabrication that will be different than just doing N plus one.

[27:48] So right now, you know, obviously everybody knows that there's a sort of global competition on how do you make the next generation of semiconductor and where are we going to do it and who's going to do that sort of innovation.

[28:00] And this is where I think the convergence of what we're doing currently in AI and accelerated real time control systems is going to help us do manufacturing very differently, significantly differently.

[28:15] And if this is conventional manufacturing, it's fairly obvious we have a convergence of robotics together with autonomous learning systems, computational learning systems that's going to completely revolutionize what the factory floor looks like.

[28:28] In fact, I think it's going to make a factory floor look like it's making, making custom items one at a time, except with the superhuman speed of an unbelievable scaling factory floor right now,

[28:41] map that to the semiconductor.

[28:44] Now maybe we start making custom chips in a way that is somehow highly scalable, as scalable as making billions and billions of all of the same chips.

[28:58] So now you have a very dynamic semiconductor fabrication process that, you know,

[29:04] could do something very differently than it was done before.

[29:08] And that's what I'm enthusiastic about in the U.S.

[29:11] i think that we're on the verge of making a very big step change in how we do fabrication and that that step change will jump far enough ahead of where we are today that it'll be unrecognizable.

[29:26] You'll look at this fabrication and you'll say, I don't even see a conventional fabrication plant in that design.

[29:32] And so I think by definition we can't say what that looks like because we can't imagine it because we're still stuck in Henry Ford time.

[29:42] But I think we're leaving Henry Ford behind.

[29:44] I think we're moving into a world where we're going to reinvent industry.

[29:50] And I think that happens in semiconductor and it's happening now. And we're,

[29:54] we see quantum computers coming up, we see photonic computation coming up, we see all kinds of interesting novel analog electronics and bio mimicking compute systems. All of this stuff is going to be a little bit different, or I would argue a lot bit different than our conventional silicon solution.

[30:16] And so that's why I'm excited about microelectronics. And you know, yes, Slack is playing a role. I'm a part of a group where we're looking at what kinds of compute happens for these real time autonomous control systems and what would make sense to not do in silicon anymore.

[30:32] Let's do it with the photons that are directly being detected and keep it photonic. For instance,

[30:37] right. If we're reading analog, let's build in memristors, let's design bio inspired memory preserving resistors that actually act similar to neurons in your brain and do that directly where you're doing your signal acquisition.

[30:53] And so this is something that I think is going to make computing and the microelectronics ecosystem very interested. Again, in five years,

[31:03] like nothing is 10 years out and 20 years out we think it is, we talk like it's 10 to 20 years out.

[31:09] No, I think we're going to invent much faster than we thought we were.

[31:13] Pamela Isom: So I've always been impressed with Slack because first of all, because you all were so supportive of me and still remain supportive of me. So I appreciate that. But it's because of things that you just described, right?

[31:25] So taking things to the next level, the American,

[31:28] the semiconductor manufacturing, using different,

[31:32] I mean, advancing the fabrication process.

[31:36] I wasn't,

[31:38] you know, I wouldn't think I just didn't think that would be something that Slack would be pursuing. But it makes sense.

[31:45] Ryan Coffee: It grows from a project that came out of Paul McIntyre and Angelo de Groni. Angelo is head of our detectors group.

[31:54] And the reason that SLAC has a role to play here is that we have historically for decades designed the detectors for high energy physics experiments.

[32:05] And again, like we said at the beginning of this, this talk,

[32:08] we use those as demonstration grounds for technology that you spin out for other reasons.

[32:14] And that technology is how do you make a pixel that can interpret the information it's measuring immediately on that pixel before you ever send a signal off of the detector.

[32:26] And so we've been doing, you know, Slack has been working on the intelligent trigger systems for experiments at Slack, starting and then at cern we're doing satellite work, we're doing detectors for Fermi looking for neutrinos, we're doing all kinds of stuff where we have to analyze the signals as they come in.

[32:48] The X ray laser is another example of where we're doing that. We've leveraged the tradition of SSLAC to use ASICS, application specific integrated circuits, conventional silicon. But we have material scientists in collaboration with Stanford and Slack that are working on photonics,

[33:05] that are working on analog electronics, that are working on neuromorphic electronics.

[33:09] And so this is the thing that we're putting intelligence into the pixels of the cameras.

[33:15] And that's what nature did. Your retina has a convolution filter in the,

[33:20] I think they're retinal ganglial cells and it measures your convolution neural network starts there, starts in your retina.

[33:30] And so nature did that too. She does her compute inside the pixel and we should, should follow her example.

[33:38] And so that's what Slack has been working on for a long time. And that's why I think we have something to say in this microelectronics world.

[33:45] Because we're doing that, we're trying to put intelligence into that pixel.

[33:50] Pamela Isom: That's good, that's great to hear.

[33:53] So now let's talk about society and the AI transformation. Since you've kind of, you've touched on that several times.

[34:00] So how are we as a scientific community and as a society succeeding or struggling when it comes to the AI transformation journey?

[34:12] Ryan Coffee: That's a tough question because it's kind of loaded.

[34:16] You added the struggling part. And I, you know, so far I've been optimistic about most things we've talked about. This one needs a real change I think because like our discussion about federated machine Learning people don't want to share with their competitors and so they want to preserve their secret sauce.

[34:37]  Ryan Coffee: Science has another problem in academic research and it's that our we're incentivized to be the masters of our own domain and to be a shining single solution hero.

[34:54] Yes, it's very hero mentality.

[34:57] And if you think about even doe's own examples of when we've had these success stories, right? And you know, whatever you may think about the Manhattan Project for what it was trying to do, it was a success story.

[35:10] All of these success stories are when people set aside their egos,

[35:15] they let go of this hero mentality. They may still have some of the ego and fight for their heroism, but they at least come together across expertise and work as a team.

[35:27] And that's something that I think is imminently needed right now. But it's hard for our academic incentive structure to support that because the funding silos fund a single domain.

[35:40] And to get your funding for that research, you have to be concentrated on that local domain.

[35:47] If you're in computing, you have to concentrate on computer science.

[35:51] If you're in engineering, you're concentrating on something that is not touch science as much, but it's focusing on optimization of existing, right?

[36:01] And so what we're talking about here is a way to break those silos.

[36:06] And if we can break those silos and come together interdisciplinary,

[36:11] then I think you can invent new things. You invent this new economy that's coming up. And that's the thing that AI can help us with. It can point to cross silo collaborations.

[36:23] That would be high potential for innovation.

[36:27] This is the thing that I think is probably interesting to our neighbors across the street at Slack.

[36:31] As you know, Slack is on Sandhill Road and our cross the street neighbors care about innovation and what might be a market potential for such.

[36:42] And so this is why I think, you know, leveraging this convergence of multiple domains to really accelerate it. You gotta kind of break out of this hero mentality of I need to be famous in my local domain and start looking at your technology that you're working on and how it might apply to people in completely different domains.

[37:04] Because I think when you do that, you start to see that AI has the power to pull us together and converge these threads, these initially independent threads together.

[37:14] And innovation happens when you take two unrelated domains and bring them together. That's innovation.

[37:20] And so this is what our current incentive structure makes challenging.

[37:26] And so I would be very excited to see this time of change, which as we all know we're going through some changes right now and this could be one heck of an opportunity to make this an innovation economy instead of a siloed economy.

[37:43] And so I'm very optimistic about that as well, though it is currently a challenge.

[37:49] Pamela Isom: Okay,

[37:50] well, I appreciate that. So, so let me ask you this.

[37:55] I'm to the place where I'm going to ask about the final words of wisdom or a call to action. But I was just sitting here thinking as you were explaining a few things there.

[38:04] And when it comes to the AI economy that you were describing,

[38:12] do you think that it's even possible with everything that you're talking about?

[38:19] So we need a better incentive structure,

[38:22] but can we do it?

[38:24] Ryan Coffee: Oh, you're saying, can we make the change that's needed and find this new way of approaching science and technology?

[38:36] I think we can and that we will. I'm actually,

[38:39] I'm pretty optimistic that we will do this.

[38:42] And I think to do it, it requires partnership between our federally funded agencies in our private sector and Washington.

[38:53] And so I think that we have a lot of incentive to work together now and that,

[39:00] you know, the U.S. although we may be reluctant to move when we're comfortable in a way of being,

[39:08] when we're really pushed, we will move quickly.

[39:10] And so I think we're being pushed right now.

[39:13] Pamela Isom: Global partnerships as well.

[39:16] Ryan Coffee: Yes, I think global partnerships, again,

[39:19] to do that, right, we have to solve the secret sauce problem.

[39:22] But I do think we have what we need to solve that secret sauce problem. At least the puzzle pieces are on the table.

[39:30] We haven't put them together yet.

[39:32] And I think the puzzle pieces currently are sitting in separate boxes and we need to take them out of those separate boxes and put them all on the table together and start working together to put the puzzle together and to make the picture.

[39:46] And that's the thing that I think we are at the verge of right now.

[39:50] And I'm,

[39:52] I think we're seeing the writing on the wall that the incentive structures are going to change very quickly so that we can actually start working together on this big puzzle.

[40:01] And I think the pieces will come together and you know, as far as what does it mean for society,

[40:06] I think we're going to see autonomous science,

[40:10] autonomous technology,

[40:12] autonomous investment.

[40:14] I think all of these things are going to surprise us with unimaginable technology.

[40:20] We are not going to be,

[40:22] we are no longer going to be good at predicting the future.

[40:26] In principle, you could say you look back at the old sci fi movies from the 60s, 70s and 80s and you see Technologies that look familiar to today.

[40:35] I would say we're not going to be able to do that this coming decade.

[40:39] We can't imagine what the next technology is going to look like in our science fiction anymore. That's what I think will happen when we solve this.

[40:47] Bring the puzzle pieces together.

[40:50] Pamela Isom: So do you think what you just described is going to make AGI seem more real, or what's your take? Is there a connection?

[40:59] Ryan Coffee: Yeah, I do. I think that. But right now,

[41:04] AGI looks like what we saw in the.

[41:08] In,

[41:10] for instance, Stanley Kubrick's 2001 Space Odyssey. Right. AGI looks like what we expected,

[41:16] a computer system that spoke to us as if it were human.

[41:20] And so that is a not surprising computational outcome.

[41:25] Now,

[41:26] what would be surprising to me is whatever it would mean to have the power of that AGI that is sitting in my pocket,

[41:35] in my cell phone, or it's probably not a cell phone anymore. I don't know what it's going to be, an earbud or some such.

[41:43] And it will be autonomously connected to a robotic system that is not just a single robot that follows me around every day, but instead is the autonomous building that I enter.

[41:56] And then it moves with me as I move from building to building.

[42:00] And it's no longer located with a single physical robotic device.

[42:05] It travels with me.

[42:07] There's a lot of weird stuff that we will be surprised to find, I think.

[42:13] And it won't look like just what AGI looks like today.

[42:18] It will be an autonomous everything in a coherent way, focused around moving people.

[42:25] Right. People who move around in the world. We're not going to be more stationary. We'll be less.

[42:30] Because I think the notion of a computer that you sit at is gone, will be gone.

[42:35] Pamela Isom: How far away are we from that, do you think?

[42:37] Ryan Coffee: Again, I say five years.

[42:39] Pamela Isom: Okay.

[42:41] Ryan Coffee: I, you know, if.

[42:42] If you think we're 10 years away because humans are linear thinkers, it's really three. If you think we're 20 years away, it's really like 10 to 12.

[42:52] And so I think that's why I, you know, I'm excited about,

[42:55] you know, the 2000 and twenties.

[42:57] I'm not going to patiently await the 2000 and thirties.

[43:01] I think this is happening very quickly, and I'm excited to be alive for it. Right. I'm glad my kids are growing up watching this change happen as they're entering the world of the.

[43:12] Of adulthood.

[43:14] Pamela Isom: Well, because they're going to be a part of it, so that's good. All right, so I would like to Know if you have final words of wisdom or a call to action for those working at the intersection of science, policy and innovation,

[43:31] or for those listeners at large. Take your pick.

[43:35] Ryan Coffee: Ooh, both.

[43:37] Pamela Isom: Okay, go for it.

[43:39] Ryan Coffee: All right.

[43:40] Final words of wisdom are not so much words, but a question.

[43:44] I was in a meeting recently.

[43:46] We were talking about AI systems, digital systems and what they're good at and humans,

[43:53] human operators of accelerators, human operators of computer systems, what have you and humans, what are humans good at?

[44:02] And a woman at Slack, Jane, she's an operator and accelerator director.

[44:06] She said what she was curious about is when you interface the two and you have effectively, like if your right brain is the digital system, the computational AI system, and the left brain is the human system, them,

[44:20] what's this corpus callosum between them,

[44:22] the thing that enables the exchange back and forth between the right and the left brain?

[44:28] If you have that and you put some energy into that,

[44:32] what would you not foresee will be the future?

[44:36] And so think of that as a third thing. We always think about AI and humans in opposition,

[44:42] but this interface between the two is itself a system.

[44:47] And you know, if you imagine what happens sometimes they're. I think it's epilepsy, that they will sever the corpus callosum in the brain and it changes the personality of the human significantly, rather significantly.

[44:59] And so what is it that that corpus callosum between the AI and the human is going to help us create?

[45:08] And when that question came up, I raised my hand and asked how many psychologists there were in the room. Room.

[45:15] And surprise, surprise, there were none.

[45:17] And so what I'm really excited about and I want people to think about is how can your social sciences and how can your humanity's intelligence come to play as a full blown member of this ecosystem?

[45:32] I don't want a computer scientist to be the one who's helping me interpret AI.

[45:37] I want a psychologist.

[45:39] I want someone who's an artist, who's a musician, who's into literature, who studies history, who.

[45:45] Those are the people I want in these rooms.

[45:48] And so that's what I would say to the listeners at large and the policymakers.

[45:52] This is not a technology world.

[45:55] It is a cohabitated world between technology and humanity. And so our humanities have every bit as equal a role to play in this as our technology.

[46:06] So that's what I would kind of leave people with, I guess.

[46:08] Pamela Isom: That's amazing. That was a really, really good analogy. I never thought about that sinner.

[46:14] I never thought about that centerpiece as you described it. But what an opportunity for those that may feel like that AI isn't for them.

[46:23] Ryan Coffee: Yeah.

[46:24] Technology may not be for you, but whatever we're inventing as a, as a society, it is, it is absolutely needs every aspect of what it means to be human,

[46:34] to be a part of it, central part of it.

[46:38] Pamela Isom: Okay.

[46:39] Ryan,

[46:40] I want to say thank you for joining me on the show. Thank you.

[46:44] Today,

[46:45] of course. Very good discussion.

[46:47] I'm sure it's going to be valuable to others, and I'm so glad you were able to make time for me. And that would put the coding down for a minute.

[46:55] Ryan Coffee: Perfect. This was. This was more fun than Python, that's for sure. Sam.