Changing the Timeline for Discoveries through Use of Artificial Intelligence (AI)

From Manual Research To Automation with AI

For some time, scientific advances relied solely on the insight of brilliant scientists and not Artificial Intelligence (AI). They then had to design and test their hypothesis manually through custom experiments. This was most often a painstakingly slow process, taking years from the initial ideas to the actual results.

Recently, higher levels of automation have allowed for a reduction in manual labor in fundamental research. For example, Next-Generation Sequencing (NGS) has allowed for genome sequencing in just a few days and at a lower cost, when the first human genome sequencing in 2003 had cost $3B and took a decade to achieve.

NGS Sequencer- Source: Illumina

Automation has allowed for the physical steps in research to be done at a massive scale and lower costs. But the intellectual effort, for example, analyzing the data coming out of genome sequencing, was still relying solely on the human brain and computer-assisted mathematical models. That is until AI became a new game changer.

Assisting The Human Brain

The more science advanced, the more complex the data became. Finding new material relies on a relatively simple process, like mixing together elements that had not been tested together before.

But advanced material sciences, for example in battery technology or semiconductors, require manipulation of components at the nanometre level, and sometimes at the individual atoms level. This makes the modelization and understanding of the process too complex for the human brain to fully grasp all the available data.

For example, finding the right material for a new battery design might include deciding which one to pick among 32 million potential inorganic materials. This was the task facing Vijay Murugesan, the group lead of the materials sciences at PNNL (Pacific Northwest National Laboratory).

Previously, educated guesses, rough computer models and manual testing would have been required to narrow down the possibilities and would have taken years. Instead, PNNL established a collaboration with Microsoft to leverage the tech giant's expertise in AI.

Teaching Multiple AIs Chemistry

Microsoft has been working for a long time on such applications, through its AI4Science program. For example, its MatterGen generative AI “enables broad property-guided materials design”.

Source: Microsoft

In collaboration with PNNL, Microsoft has specifically developed several different AIs to evaluate all the workable elements and their combinations.

It also has an AI dedicated to finding which materials are stable. And then an AI evaluating molecules based on their reactivity. And another AI judging the molecules' ability to conduct energy.

This combination of AI is not designed to find every possible solution to a given problem. Instead, it is looking at the massive haystack of possibilities (in the tens of millions) and trying to narrow it down to a few good candidates.

This approach took 32 million potential materials down to 500,000 candidates, and then to 800.

It fundamentally differs from a purely computer-based mathematical model approach, which will instead try to calculate with “brute force” the chemical properties of a material, often by simulating it atom by atom. Instead, the AI makes an educated “guess” of what should work by using analogies, pretty much the same way a human would do. Except the team of AIs could screen 32 million ideas in less than 80 hours.

Reducing months or years of work into a few days or weeks is a complete revolution in the pace of scientific progress.

If applied more broadly, this could completely change the pace of technological progress in human societies.

Saving Precious Resources

Once the list was narrowed down to just 800 candidates, researchers could then deploy the more standard method of High-Performance Computing (HPC). It allowed for the calculation of all the possible energy states of the 800 candidates. Then a mix of dedicated AI and HPC was used to simulate the movements of each atom and molecule inside each material.

By using HPC on only a few hundred candidates, this saved PNNL a lot of time and money, as this is a method very hungry for computing power. This reduced the candidate list to only 150.

From there, an evaluation of costs, availability, and other practical considerations reduced the initial 32 million candidates list to a mere 23. Interestingly, out of the 23, 5 were already known, demonstrating the pertinence of the method, as it “re-discovered” these 5 materials independently.

Previous research methodology would have relied heavily on expensive and slow HPC. The AI method instead reduced HPC to just 10% of the total computing time. The fact that the AI computation was cloud-based, instead of using precious & rare research institutes' supercomputer time, also helped make it more efficient.

The Next Steps

The collaboration between Microsoft and PNNL over battery material was just the beginning of a multi-year collaboration agreement. Ultimately, the idea would be to generate enough data that knowing with what new material to design a new battery would be as simple as just asking the AI system about it.

“The vision we are working toward is generative materials where I can ask for a list of new battery compounds with my desired attributes,” – Nathan Baker, Product Leader for Azure Quantum Elements.

It is also worth paying attention that the current method relies on classic computing. But Microsoft's AI programming and software are designed to be upgraded to using quantum computing as soon as this technology is mature enough.

This would give unprecedented computing power for topics like chemical and biological simulations, increasing the computing power by several orders of magnitude (x100-10,000). So we are still in the very early stage of AI working on designing new molecules and materials.

Fields Of Applications

The first experiment was focused on battery materials. But many other fields are likely to benefit from such advanced insight on chemistry, as well as applying AIs to other sciences, all already in Microsoft’s pipeline:

• Semiconductors.
• Quantum computing & cryptography.
• Visual and audio
• Molecular biology & genomics.
• Biotech & and pharmaceuticals in general.
• Renewables & climate change.
• Agriculture & Ecology.
• Economics.
• Social sciences.
• Theoretical mathematics.

Each of these fields has seen research struggling due to the sheer mass of data and complexity of the issues at hand. Similarly to the pilot project on batteries, each could benefit from insight from AIs.

AI-Driven Science Stocks

1. Microsoft

Microsoft has been at the center of the tech industry almost since its inception with its still-dominant operating system Windows. It is now also a leader in enterprise software (Office365, Teams, LinkedIn, Skype, GitHub), gaming (Xbox and multiple videogame studios acquisitions), and in cloud (Azure).

More recently, it made good progress on AI. This includes some consumer-grade AI like the Bing Image Creator and its deepening partnership with OpenAI. It also includes more business-focused initiatives, like Copilot for Microsoft 365 and Microsoft Research. Copilot is now being deployed to retail and smaller companies as well.

Microsoft has acquired a reputation for being the enterprise-centered tech giant, compared to more consumer-focused companies like for example Apple or Facebook. With AI becoming increasingly important in business models, the preexisting presence of Microsoft in cloud and enterprise services should give it a head start in deploying AI at scale and in customer acquisitions.

The collaboration/quasi-ownership with AI development leaders like OpenAI will also cement Microsoft's position as an AI powerhouse.

2. NVIDIA

NVidia initially had a dominant position in the graphics card (GPU) market, mostly used for high-end gaming and 3D modelization. GPUs are able to run calculations in parallel and differ in that regard from processors (CPU).

The design of its hardware proved a very good fit for cryptocurrency mining (especially Bitcoin), creating a strong wave of growth for the company.

Now, it appears that it is equally powerful for training AIs, making Nvidia’s hardware the backbone of the AI revolution.

NVidia is now developing custom computing systems for different AI applications, from self-driving cars, to speech and conversational AIs, generative AIs, or cybersecurity.

It is likely that Nvidia is not done finding new use cases for its AIs hardware, as shown by Microsoft research with PNNL. For example, NVidia is now developing a whole range of solutions for drug discoveries, as well as AI-powered medical devices and AI-assisted medical imaging.

Source: NVidia

It is likely that in the very long run, competitors to NVidia might start to seriously challenge the company’s initial head start. But for the foreseeable future, considering the explosion in demand for AI-dedicated computing power, NVidia will stay the prime supplier of all the new AI-training data centers being built.

3. CrowdStrike

The more the world relies on AI and digitalization, the more connected cybersecurity will become important.

CrowdStrike was founded with a cloud-first approach to cybersecurity. The company's offer covers all categories of cybersecurity threats, and among its clients are 15 of the 20 largest US banks, 70 of the Fortune 100 companies, and 556 of the Global 2000.

CrowdStrike's growth is supported by a quickly expanding total addressable market (TAM), expected to grow 13% CAGR in the next 2 years. With additional offerings still in development, the company expects to expand its TAM from the current $76B to $158B by 2026.

Source: CrowdStrike

Another growth factor for CrowdStrike is the expansion of business with pre-existing clients. When a client starts with at least one cybersecurity module, it usually goes on and keeps integrating more modules, with 62% of clients using 5 or more modules and 23% using 7 or more modules.

This dynamic creates an environment that allows CrowdStrike to grow its margins when a relationship has developed for long enough, with an impressive total gross margin of 78% in 2023.

The transition to the cloud is still mostly ongoing for many large companies. This creates a large opportunity for a market leader like CrowdStrike to help them transition their cybersecurity strategy to the cloud as well.

The company should also see its international business grow, with still 3/4th of the Global 2000 companies yet to enter the CrowdStrike ecosystem.

The cloud-first approach of CrowdStrike has allowed it to take market share quickly and is now replicated by all the large cybersecurity companies. So, investors will want to pay attention to CrowdStrike's ability to retain its advantage despite mounting counter-attacks by the industry.

4. Adobe

At first glance, the emergence of generative AI, especially image generation, could be a threat to the owner of important graphic software like Photoshop, InDesign, After Effects, Lightroom, or Illustrator.

This would be forgetting that Adobe has been a trailblazer in the software industry, being one of the first to move to a cloud-based subscription model, when the industry practice was selling the latest software version for thousands of dollars.

The same is true for AI, with Adobe Firefly. This AI image generation tool is now being seamlessly integrated with Adobe's other software programs, allowing you to generate images from simple text, create a rich image from a simple 3D model, or even “generative fill”, expanding an existing image in a photorealistic fashion.

Source: Adobe

With AI lowering the skill barrier required for creating logos or, magazines, or for modifying images, the offer and the demand for original content will likely grow.

By embracing change and AI, Adobe will likely retain its position as a leading software package for all visual creative works, and even grow further its reach.

5. Upstart

Upstart is a lending marketplace powered by AI, launched much before AI became a center of tech conversation in 2023.

Upstart's process is mostly automated, with 87% of granted loans fully decided through automation.

The idea behind Upstart is that the existing credit score system is inefficient and outdated. With a lot more data available, it is possible to identify loan risks better and, as a result, provide cheaper loans to a large portion of the population.

This means Upstart's method can identify people with high FICO scores but who, in practice, have a high risk of defaulting on their loans. And reversely, people with low FICO scores are not that likely to default.

Source: Upstart

The Total Addressable Market is large, with a yearly $4T originating from personal, auto, home, and small business loans.

Due to rising interest rates and reduced demand for loans, Upstart has experienced some decline in revenues and losses in 2023, together with the rest of its industry.

This temporary loan volume setback has not slowed Upstart's expansion of its lending partner network, with 100 banks, up from 71 the year before and only 10 at the IPO in 2020, and 61 dealerships, from 39 in early 2023.

Investors in Upstart will need to hope that the growing network is a clear sign of the value of Upstart technology and of its potential to become a large originator for loans in the US market.

Continuous progress in AI calculation might also give it an edge in a very competitive loan rating market.