00:00I can't think of a better person who's uniquely placed to talk about the intersection between artificial intelligence and energy.
00:07That's a big theme here at ADIPEC this week.
00:10Great to see you.
00:11Yeah, good to see you, too.
00:12So just talk to us about the extent and the scale of the opportunity that you see in this space right now.
00:18Well, it's really fascinating.
00:2099% of the AI that our viewers here are familiar with, ChatGPT, Google Gemini, wonderful applications of AI.
00:27But that's AI for the digital world.
00:29Now we're talking here at ADIPEC about AI for the physical world.
00:33And that's the AI that's really going to impact the energy industry and all of the consumers of energy around the world.
00:39So what does that mean exactly?
00:40So let's talk about, for example, when we take oil out of the ground, and many of the companies that do that are right here at ADIPEC,
00:47what happens to that?
00:48How do we turn it into useful products, both fuels and products that we can use?
00:5280% of all the products that we use in our world come from the hydrocarbon cycle, come from these kind of companies.
00:58And so we need something called a catalyst.
01:00A catalyst is a chemical that helps us turn raw product into finished product.
01:05And to make catalysts, it's really difficult.
01:07On average, it takes about 5, 10, sometimes 15 years to make a new catalyst.
01:12Those are essential ingredients to making all the goods in our world, the cars, the chairs, the pens, everything you use in the world comes from using a catalyst.
01:21And now with our AI that we just announced last week with NVIDIA, this is an AI that is the best AI, the best model in the world to make a new catalyst for the energy industry.
01:31So what is it specifically that you're doing differently from your competition in this space?
01:36Great question.
01:37So instead of training on words, so large language models are trained on all the words of the Internet, and that's great if you want to write an essay, if you want to summarize a document, marketing, customer service, great applications for large language models.
01:51We're one of the leaders in the world of large quantitative models, meaning that we don't train on words.
01:56We train on numbers, equations, and molecules.
01:58And this is a very different training set and therefore a very different model.
02:02And what we've shown now with NVIDIA's help, one of our investors, is that we can now produce catalysts at much faster speeds and also speed up other areas of the energy industry.
02:13An example is when you look at the fluid dynamics, the flow of fluid inside of our reactor as it's going through the refinery, we have sped up that process as well.
02:22So AI for the physical world is now going to be the next revolution of AI.
02:26Yeah, and what sort of productivity or efficiency gains are we targeting here, right?
02:31Because one of the criticisms of AI is there's been so much spending on AI applications, but perhaps many companies are still not getting the return on investment.
02:38That's absolutely true.
02:39An MIT study just came out.
02:41Correct, 95%.
02:4290%, exactly.
02:43And that's because, again, large language models are really good for some things, but not for all things.
02:48I'll give you another example.
02:50We talked about the energy industry here at ADIPEC, but also let's talk about the banking industry.
02:54Banks, it turns out, deal with numbers and equations.
02:57Yeah.
02:57And large language models can be helpful in a bank, for example, to summarize some documents from the SEC or things like that.
03:04But the majority of what a bank does is numbers, and we cannot have any kind of hallucinations when it comes to the numbers in your account.
03:11Your account matters to a bank.
03:13It should, anyway.
03:14And so we have to make sure that it's exact and precise.
03:16That's the domain of large quantitative models, models based on numbers trained on equations of quant finance in that case, or in the case of energy,
03:25trained on the physical laws to make new chemicals for the world.
03:29So these are very different models.
03:31We don't have any hallucinations because we use pristine scientific data instead of the words of the Internet.
03:37That's a big fundamental change, and that speaks to the issue that the MIT study talked about.
03:42And ostensibly, you're also dealing with troves of data that go back decades, right?
03:46You're talking about Aramco, Adnock.
03:49These companies are sitting on decades and decades of data specific to the energy industry, specific to the type of data that's going to inform your models better.
03:57With that kind of data, you do not want to feed that into a language model.
04:01That is very different kind of data.
04:03And even when you talk about biopharma, biopharma is in the same kind of ballgame where if you look at Sanofi, look at Roche, they want to develop new drugs.
04:11That's not really the large language model to do that.
04:14That's, again, an AI for chemistry and biology.
04:17How do you see quantum computing coming into this in coming years?
04:20Well, quantum now is really starting to get in people's minds, and that's very exciting because quantum computers will add to the mix.
04:26A quantum computer fundamentally is different than the computers we have.
04:30It's not just faster.
04:31It's a different kind of computer altogether.
04:34The good news is there's about a dozen companies out there right now, both large companies like Google, IBM, Amazon, Microsoft, as well as a lot of startups such as CyQuantum and others.
04:44QEar is another one that are now rising up.
04:47And I think in the next five years, we're going to see significant, significant progress on the part of these quantum computers.
04:53Sandbox AQ, we write the software that runs on top of those quantum computers.
04:58And so it's an exciting coming together now, a convergence of the software algos and the hardware.
05:04But we may also want to talk about the cyber aspect as well to all this.
05:07Yeah, well, I mean, obviously, there are cybersecurity concerns.
05:10How do you think about creating a system that is bolstered and immune to those types of shocks?
05:17Jumaane, this is a fundamental issue, particularly here when we talk at ADIPEC about the energy industry.
05:23Energy is a critical national infrastructure for every country in the world.
05:27We cannot exist without energy.
05:30And so we must protect, from a cybersecurity perspective, the energy, the power generation, the electric grid itself, the usage of all this, the management of all this has to be protected.
05:40We know that there are state actors and other hackers that have penetrated into the electric grids in the United States.
05:47We know that's happened for sure, and probably happened in other places as well.
05:51AI is both a great tool for hackers, but also the tool we need to defend against these hackers as well.
05:57So AI is really a dream tool for anyone wanting to go into a certain area, for example, for phishing.
06:04They can write a beautiful email, Dear Jumana, please click on this wonderful link.
06:08And that's a great threat right now.
06:11And so energy companies and national companies as well need to really think about how do they defend themselves against this.
06:19And the best way to defend yourself against an AI attack is AI itself that can look at that attack, detect that attack, and stop it in its tracks.
06:27Four or five years from now, we also have to worry about the quantum attack.
06:31That's also going to be something four or five years from now.
06:33But right now, we've got to worry about the AI attacks.
Be the first to comment