On the afternoon of October 9, Beijing time, the Royal Swedish Academy of Sciences decided to award the 2024 Nobel Prize in Chemistry to three scientists. Among them, two Nobel laureates, Demis Hassabis and John M. Jumper, are from Google DeepMind. They used the AI model AlphaFold2 to pass the amino acid sequence with an accuracy of more than 90%. Predict 200 million protein structures known to mankind.
No coincidence. Earlier on the 8th, the 2024 Nobel Prize in Physics was also awarded to two scientists in the field of AI: John J. Hopfield, a professor at Princeton University in the United States, and Jeffrey Hinton, a professor at the University of Toronto in Canada. (Geoffrey E. Hinton), for their fundamental discoveries and inventions in realizing machine learning through artificial neural networks.
Obviously, this year has become the "year of AI" for the Nobel Prize, and both the physics prize and the chemistry prize were awarded to AI-related work. So, does this mean that AI can already replace the work of scientists? Why did this year's winner win the prize in chemistry rather than physiology or medicine? Regarding the AI bubble, especially the lower than expected industrial returns on AI software, how far is it from technology to application and generating positive profits?
In this regard, Titanium Media App held exclusive dialogues and exchanges with many scholars including Alex Zhavoronkov, founder and CEO of InSilico Medicine, and Professor Dou Dejing, chief scientist of Nortel Digital Intelligence.
How to interpret this year’s “Year of AI” for the Nobel Prize? In this regard, Zhang Hongjiang, founding chairman of the Beijing Zhiyuan Artificial Intelligence Research Institute and foreign academician of the American Academy of Engineering, said in a video shown to TMTpost App that AI actually plays a very important role in scientific research and physical research. This time the two The award is well deserved. "I think this is a very good recognition of the future potential of AI. I believe that future physics is also inseparable from AI."
"Hinton used RBM to do DNN self-supervised pre-training in 2006 and successfully trained a deep neural network. It can be said to be the forerunner of this round of AI revolution. The Hopfield network laid the foundation for RBM." Zhang Hongjiang said that the two people are actually very related to physics. Close connection. In addition, it is very important that the Nobel Prize given to the field of network machine learning is actually a recognition and expectation of the importance of AI or machine learning based on neural networks.
Matt Strassler, a theoretical physicist at Harvard University, said, "Hopfield and Hinton's research is interdisciplinary, integrating physics, mathematics, computer science and neuroscience. In this sense, it belongs to all of them field."
Professor Dou Dejing, chief scientist of Nortel Digital Intelligence, told TMTpost App that first of all, this year’s Chemistry Nobel Prize was awarded to DeepMind Hassabis and Qiaopu. Their “contribution to protein structure prediction” is indeed indispensable. Using AlphaFold’s high accuracy It has predicted the complex structure of proteins at low cost and at low cost, which had previously been time-consuming and laborious for biological scientists to obtain, and promoted changes in the biological research model. The Nobel Prize in Physics awarded to the field of AI represents the recognition of the contribution of AI by the entire scientific community. Since the birth of ChatGPT, AI has developed rapidly for two years and has been accelerating. Although it has not yet achieved great commercial realization, it has had a lot of impact on all walks of life, especially on the scientific community. This physics prize is awarded to Hopfield and Hinton in recognition of their fundamental discoveries and inventions that promoted the use of artificial neural networks for machine learning. The core of this award is the application of basic principles of physics to the field of AI neural networks.
However, Dou Dejing believes that “AI’s contribution to physics itself is not obvious enough.”
He introduced that one of the past contributions of AI to the physics community was in 2017, when astronomers used computer vision technology to help process humanity’s first black hole photo. Although the progress of large-scale model technology that has been hot in the past two years also relies on basic disciplines such as mathematics, statistics, informatics and physics, Hinton was surprised when the Nobel Prize in Physics was awarded to Hinton. He did not expect it to happen.
Dou Dejing emphasized to TMTpost App, “In summary, mathematics, statistics, physics, and informatics are the basis of computer science. These basic theories help the development of computer science and AI and are also the basis of AI. But in turn, AI It has not yet truly affected the basic principles of physics and helped the development of physics. In the future, as new substances and theories are constantly discovered, we expect that AI will interact more frequently with physics and other basic disciplines. In addition to physics and chemistry, Nobel Prizes in biomedicine may also recognize the contributions of AI scholars.”
Alex Zhavoronkov, founder and CEO of Insilicon Intelligence, told TMTpost App that AI has had a profound impact on science and technology and will change all aspects of human life.
"I think the Nobel committee recognized this and had to push the boundaries to recognize this profound change." Alex said that there are many unusual facts in this year's award. First, AI is mostly mathematics. John McCarthy, Alan Turing, Marvin Minsky, Allen Newell, HerBERT A. Simon ), Nathaniel Rochester, and Claude Shannon are mostly mathematicians and engineers. When the Nobel Prizes were first introduced, there was no computer science or artificial intelligence as a separate discipline. So for deep neural networks, they had to classify AI as physics, and it was expected that AlphaFold won the Nobel Prize.
In Alex’s view, the Nobel Prize will inspire more people, and the value of neural networks to the industry is huge.
“A lot of very simple tasks have been taken over by AI. Even at Insilico, we have replaced a lot of annotation, writing and even coding jobs with AI, and have had to reskill and upskill a lot of our employees who prepare the data. The economic benefits are not yet felt, but It does. And nothing is more profound than the impact of drug discovery. Since raising its first large round of funding in 2019, Insilico alone has successfully nominated 19 preclinical drug candidates, advancing nine projects into the clinic. , and passed a II phase trials. Typically, big pharma nominates 5-7 preclinical drug candidates per year and has more resources - with AI, one company has more capabilities in drug discovery than most in the world. Developed countries are stronger. Few new drugs have been discovered. Most countries have never nominated PCC. But thanks to AI and the power of China, you can really feel the impact of AI on this industry without spending decades training local scientists. Impact. But just like the Internet or social networks - there are very few winners, maybe 2-3,” Alex said.
Shen Qi, a permanent teaching associate professor at the School of Chemistry and Chemical Engineering at Shanghai Jiao Tong University, said that with the emergence of AI, the accuracy and efficiency of protein prediction have been improved unprecedentedly, solving major scientific problems that have troubled chemists for many years and becoming a popular choice for the majority of chemists. A powerful tool in the hands of scientific researchers, this award is well deserved.
In fact, since the Nobel Prize was first awarded in 1901, the Nobel Prize has often emphasized the impact of research on society and rewarded practical inventions rather than just pure science. This year's awards are not unusual in this regard, as they are sometimes awarded to very outstanding engineering projects. These include the laser and PCR fields.
It is understood that the 2024 Nobel Prize in Physics and Chemistry will equally share the total single prize of 11 million Swedish kronor (approximately RMB 7.4446 million).
Although this year's Nobel Prize has been announced, there has been controversy on the topic of "whether the generative AI craze has formed a bubble."
According to the Gartner technology cycle, AI has passed the peak of overexpectation and will enter the trough of disillusionment. The report predicts that by 2025, 30% of current AI projects will be abandoned after proof of concept. At the same time, many AI projects will fail due to poor data quality, insufficient risk controls, unclear business value, or rising costs.
Gartner points out that implementing generative AI projects can cost millions of dollars and incur significant ongoing costs. For example, launching a new generation of AI virtual assistants may cost $5 million to $6.5 million, with an annual recurring budget expenditure of $8,000 to $11,000 per user.
In this regard, Alex told TMTpost Media App that in the short term, AI is like many other technology bubbles. It (generative AI) is a bubble. Many low quality companies received funding and even some lower level university professors received funding for new startups and are now struggling to create a product or revenue.
Dou Dejing told TMTpost App, “ We believe that current AI has not yet been able to help companies achieve economic profitability. Although some software companies now use copilot to automatically program, which can save programmers’ time and some costs, it is not yet possible. Use AI to completely replace programmers. In addition, AI’s current returns are lower than expected because the operating costs of the large model industry are too high. It takes several months and thousands of cards to train a model, even if there is a profit model. , it will also take a very long time to pay back the cost.”
In Dou Dejing's view, it was just like the emergence of search engines back then, which gave everyone better access to information. But at that time, he was also thinking about a monetization return model. Later, he relied on advertising to start a profit model. At present, there is no profit model similar to advertising in the AI field. It is currently unclear whether OpenAI will be able to achieve profitability by advertising on its platform in the future. After all, the daily user activity of large model companies is far lower than that of search engines such as Google, and a profit model like advertising is needed.
However, from capital to companies themselves, the market is changing, and companies in the large model field are accelerating the implementation of applications and working hard to gain revenue.
According to Dou Dejing, as an AI-native state-owned enterprise, Nortel Digital Intelligence solves how to effectively utilize existing multiple computing resources in the current competitive landscape of computing resources to enhance the core competitiveness of the AI industry while reducing enterprise usage. The threshold of AI computing power and help the development of AI industry.
Specifically, Nortel Digital Intelligence is using the core technology of Hunyuan Adaptation to more efficiently use domestic chips to process different types of data, while ensuring the security of data and the performance of models, and promoting the realization of domestic chips from "usable" Change to "easy to use". At the same time, the widespread application of AI requires innovation not only in the technology itself, but also in processes, systems and organizations; in addition, Nortel Digital is building an AI The production line of the times promotes the development of infrastructure. In addition to the computing power layer cooperating with domestic chips, the model layer provides universal support for mainstream base models and open source models. The data layer creates a trusted data space and creates a vertical for sensitive industries. Class model matrix; secondly, each Spark Intelligent Computing is also equipped with exhibitions, roadshow spaces, laboratories, and open and closed seminars, etc., to accelerate industrial development, make AI available and accelerate the arrival of the AI era.
Alex said that currently in the field of AI, only a few startups can achieve scale and industrial capabilities - OpenAI is doing a good job in inference, Insilico is doing a good job in drug discovery, but companies like Google, Microsoft , Amazon , and Meta Big companies hold all the keys to mainstream industrial applications. On the very positive side, in terms of drug development, we see Insilico driving the first Phase II clinical study of a drug entirely generated by AI to be completed, "I'm proud that this was completed in China, and if we're lucky If so, it may be the world’s first approved AI drug.”
According to the financial report information submitted by Yingsi Intelligent to the Hong Kong Stock Exchange in June this year, Yingsi Intelligent's revenue in 2021 and 2022 will be US$4.713 million and US$30.147 million respectively, mainly from medical research and development services. In the future, Insilico will expand the combination of AI, real estate and healthcare, and Alex pointed out that the company is working with some leading real estate companies.
According to industry research data, global pharmaceutical R&D expenditures increased from US$165.2 billion to US$217.9 billion from 2017 to 2021, with a compound growth rate of 7.9% during the period. It is expected that the scale of expenditure will increase from US$242.1 billion to US$313 billion from 2022 to 2026, with a compound growth rate of 6.9% during the period.
"I think 50% of the success is due to very powerful generative AI, and 50% is due to China's high-quality talent, capabilities and work ethics. I think the next biggest wave of productivity we will see in AI will be In China." Alex said.