Russian Scientists Feud With DeepMind Over Quantum AI Research

Russian Scientists Feud With DeepMind Over Quantum AI Research Scientific advancement is the most spectacular and motivating thing there is. But what takes place when several scientific communities can’t seem to agree on the science?

In an intriguing research report last year, DeepMind, a London-based Alphabet research organization, said that it had overcome the enormous difficulty of “simulating matter on the quantum AI scale using AI.”

Now, over eight months later, a team of university researchers from South Korea and Russia may have found an issue with the first study that calls into question the paper’s whole conclusion. If the paper’s results are accurate, there might be significant repercussions for this cutting-edge study. In essence, we’re discussing the possibility of using artificial intelligence to find novel approaches to manipulating the constituent parts of matter.

Hope for the future quantum AI

The key concept is the ability to model quantum AI interactions. Our universe is composed of matter, which is composed of molecules, which are composed of atoms. The difficulty of simulating anything increases with the amount of abstraction. When you get to the quantum level, which is where atoms are, it is very hard to simulate how things might interact.

According to a blog post from DeepMind:

Quantum AI The simulation of electrons, the subatomic particles that control how atoms come together to form molecules and are also in charge of the flow of electricity in solids, is necessary to carry out this task on a computer.

Even after decades of work and a number of important improvements, it’s still hard to accurately simulate how electrons behave in quantum mechanics. The fundamental issue is that it is very difficult to forecast the likelihood that an electron will end up in a certain place. And as you add more, the complexity rises.

It was later realized that it was unnecessary to follow every electron individually by Pierre Hohenberg and Walter Kohn. The electron density, or likelihood that an electron will be present at any point, is all that is needed to precisely calculate all interactions. After demonstrating this, Kohn won the Chemistry Nobel Prize, establishing Density Functional Theory (DFT).

Unfortunately, DFT was only able to streamline the procedure so much. The “functional” part of the approach meant that all of the labour-intensive tasks had to be done by people. When DeepMind Quantum AI released a study in December headline “Pushing the boundaries of density functionals by addressing the fractional electron issue,” everything changed.

Quantum AI In this study, the DeepMind team says that the creation of a neural network has greatly improved the ways that quantum behaviour can be simulated: We develop functionals free from significant systematic mistakes by expressing the function as a neural network and including these precise features in the training data. This leads to a better representation of a large class of chemical interactions.

 

The academicians respond.

 

The original, official review procedure for Quantum AI DeepMind’s work was successful, and everything was OK. up until August 2022, when a group of eight academics from South Korea and Russia submitted a response that questioned the findings. According to a statement issued by the Skolkovo Institute for Science and Technology, DeepMind AI may not be able to generalize the behaviour of these kinds of systems based on the results shown, so more research is needed.

In other words, experts disagree with the methods used by DeepMind’s AI to achieve its results.

The academics who commented claim that DeepMind’s neural network was trained to remember the solutions to the particular challenges it would encounter during benchmarking, the procedure by which scientists decide if one strategy is superior to another.

The scientists write in their comments

Although Kirkpatrick et al.’s assessment of the importance of FC/FS systems in the training set may be accurate, it is not the sole reason for their findings. We think that the reason DM21 did better than DM21m on the BBB test dataset than DM21m was that the training and test datasets overlapped by accident. If this is true, it would mean that DeepMind didn’t really train a neural network to predict quantum physics.

 

The Quantum AI is back.

 

Quantum AI DeepMind responded right away. The company’s response, which came out the same day as the comment, was a quick and harsh reprimand. We don’t agree with their analysis, and we think the problems they raise are either untrue or have nothing to do with the paper’s main findings and the overall quality of DM21.

In its response, the team elaborates on this:

The DM21 Exc fluctuates across the whole range of distances taken into account in BBB and is not equal to the infinite separation limit, as shown in Fig. 1, A, and B, for H2+ and H2, demonstrating that DM21 is not remembering the data. For instance, the DM21 Exc is about 13 kcal/mol from the infinite limit in both H2+ and H2 at 6. (although in opposite directions).

Also, even though it’s outside the scope of this paper to explain the words used above, we can be sure that Quantum AI DeepMind was ready for that particular criticism. It remains to be seen whether it resolves the issue. The academic team hasn’t told us yet whether or not their concerns have been addressed in response to our questions. In the meantime, it’s feasible that this discussion’s effects will extend well beyond the scope of a single scientific report.

The areas of artificial intelligence and quantum physics are entwining more and more and are becoming more controlled by well-funded corporate research organizations. What happens when corporate interests are involved and there is a scientific standstill when opposing parties cannot come to a consensus on the usefulness of a certain technical technique using a scientific method?

 

Now what?

 

The inability to explain how Quantum AI models “crunch the numbers” to get their findings might be the root of the issue. Before producing a result, these systems may go through millions of possibilities. We need algorithmic shortcuts and AI to brute force mass-scale issues that would be too big for a person or machine to address head-on since it would be difficult to explain every step of the process.

We could eventually reach a point where we run out of tools to fully comprehend how AI systems operate as they expand. When this happens, we might be able to tell the difference between the technology used by a company and that which passes an outside peer review. Not that Quantum AI DeepMind’s article serves as an illustration of this. As stated in the news release by the academic team that made comments:

Quantum AI Other than the use of fractional-electron systems in the training set, DeepMind’s work is innovative in many ways. Their technique of imposing physical sense via training on the appropriate chemical potential and the notion of putting physical limitations through the training set into a neural network is anticipated to be extensively used in the future when creating neural network DFT functionals.

However, a daring, fresh, AI-driven technological paradigm is now in play. It’s perhaps time we begin thinking about what life would be like after peer review.