When Web3 meets AI: a passionate collision with sparks | GAIR 2023
Latest update time：2023-08-17
Reads：
"
The technological wave of change is always on the way.
"
Author丨Dai Congfei Li Yangxia
Editor丨Lin Juemin
At the beginning of 2022, the Web3 concept, which has been born for eight years, began to sweep the technology circle. But less than a year after its popularity, the craze gradually faded; "Web3" was replaced by the era of large models detonated by ChatGPT.
In the past six months, more than 100 large models have been produced in China, ranging from general large models to vertical large models, each with its own merits. In essence, large models allow people to see the dawn of general artificial intelligence.
Back on Web3, the craze faded and the naked swimmers had to taste the bitter pill. But this does not mean that Web3 is an illusory concept. The fact is, people are becoming more aware of it.
Even Sam Altman, the creator of ChatGPT and founder of OpenAI, is interested in joining Web3 entrepreneurship. He has an "ambitious" project - "Worldcoin".
When Web3 truly collides with large models, what will be the result?
The 7th GAIR Global Artificial Intelligence and Robotics Conference, co-sponsored by GAIR Research Institute, Leifeng.com, World Technology Press, and Kotler Consulting Group, was successfully held in Singapore on August 14. In response to the latest topics in the industry, on the afternoon of August 15, GAIR specially set up
the ultimate collision sub-forum of Web3 and AI, inviting many senior industry players in the industry to share their insights.
Guest speakers at the meeting include:
-
Professor at Nanyang Technological University, Liu Yang
-
James Liu, Director of Emerging Business and Innovation Alliance, Alibaba Cloud Singapore Region
-
BitDATA founder, Wang Kai
-
Founder of Project Twelve (P12), boyang
-
CEO of Abbots Technology, founder and CEO of Dolphin Browser, Yang Yongzhi
-
Foresight X Partner, Min
At the meeting, guests such as practitioners and investors in the Web3 and AI industries conducted in-depth discussions from multiple dimensions such as security, application prospects, and the relationship between the two, resulting in many wonderful collisions and sparks.
When talking about the relationship between AI and Web3, BitDATA founder Wang Kai compared the two to productivity and production relations. He said: "If AI is productivity, then Web3 is production relations. Unlike AI, Web3 is a more complex It is not only a change in technology, but also a change in production relations, just like the Internet technology has surpassed the technology itself. "
Come and take a look at more exciting content from the guests in the sub-forum:
01
Liu Yang, a professor at Nanyang Technological University: AI can help application security, but its own security is also very important
How does AI help application security? Professor Liu Yang of Nanyang Technological University has led his team to conduct exploratory research in this area for many years. After the emergence of ChatGPT, Professor Liu Yang and his team are also very concerned about its application and performance in the security field.
Liu Yang
Professor Liu Yang shared many of the team’s thoughts on the application of large AI models in security. Professor Liu Yang said that in 2019, the team used the neural network method to find loopholes more. The idea here is actually very simple: first collect a bunch of vulnerability data, train a model, and then let the model predict whether there is a vulnerability in an unknown function.
Professor Liu Yang also pointed out possible problems. Since the initial starting point of large models and artificial intelligence is from the NLP scene, when most people do neural networks, the input or scene is mainly text, not code. But text and code are very different. If you use text to learn code, it is difficult to learn its precise semantics. Following this idea, Professor Liu Yang and his team applied the code structure to the neural network, and the results were very good. However, during the implementation of industrialization, it was found that the effect was not very good.
The reason is that if you want to build this vulnerability scanning tool, the amount of vulnerability data will be very small. If you add up all the open source component vulnerabilities in the world, the gap may still be very large.
After ChatGPT appeared, the industry was thinking about whether it could become a new methodology to better find vulnerabilities. But it may not be that simple. From a semantic level, vulnerabilities are a very volatile thing. If you change one letter in a code, it will change from a safe contract to a vulnerable contract. However, ChatGPT itself is a probability model, and it is impossible to convert such a small code into a vulnerable contract. Text-level changes are analyzed to the corresponding semantic level.
Professor Liu Yang and his team have done a lot of work on how AI can actually find loopholes in this scenario: first, let ChatGPT determine what the contract does, and further determine whether there are key variables in it that may be manipulated by hackers, and put this GPT The code understanding ability is combined with MetaScan, the automated smart contract vulnerability scanning product of MetaTrust Labs, co-founder of Professor Liu Yang, to form a unique AI scanning engine in the industry. This method can efficiently detect logical vulnerabilities that are some of the industry's pain points; secondly, Finding vulnerabilities through dynamic methods originally requires a lot of manual intervention, mainly the generation of the driver and Oracle requires manual participation, but ChatGPT can replace people to write the driver. Interestingly, the results show that the results of ChatGPT4.0 are much better than those of ChatGPT3.5; thirdly, in-depth testing is automated. We originally thought that penetration testing is a relatively systematic behavior that requires human participation. Now, by understanding the context of each action and generated data through ChatGPT, the penetration testing process can be automated.
The above methods are all based on direct calls to ChatGPT. It has disadvantages such as the need to pay and the inability to control it. Professor Liu Yang also shared a more in-depth method of using ChatGPT. In this regard, Liu Yang talked about two scenarios: First, the positioning and explanation of the vulnerability. The focus is still on applying the understanding of vulnerability semantics, integrating vulnerability semantics with neural networks in depth, collecting vulnerability-related paths, and then doing some ranking to put the relevant codes and paths of potential vulnerabilities at a higher priority and recommend them; 2. It is to do so-called fine tuning based on the large model, and to achieve a deeper integration with the large model, so that the large model can be transformed from a model with only common sense into a vertical field that understands application security, code vulnerabilities and patches. Large model; the third is to add a feedback mechanism to make the repair effect truly implementable.
Professor Liu Yang finally shared an interesting latest research result: using security to try to attack AI, and found that ChatGPT has various methods to attack AI, such as making a large model spit out things he doesn't want to say. In this case, humans may have countless ways to get the large model to express what they want. What's even more exaggerated is that we can not only make him spit out some content he doesn't want to spit out, but we can even control it, like traditional malicious code, to make the big model do things it shouldn't do, and even change some of its behaviors. .
Therefore, AI can not only empower security, but the security of AI itself has also become a very important matter. If AI itself is not safe, then everything else will be empty talk. Liu Yang said that this is something that AI must consider in the future.
02
James Liu, Director of Emerging Business and Innovation Alliance of Alibaba Cloud Singapore Region: Changes and Unchanges in the Web3 Era
Subsequently, James Liu, Director of Emerging Business and Innovation Alliance of Alibaba Cloud Singapore Region, conducted an in-depth discussion on the changes and trends in the Web3 and AI era. He highlighted the 4+1 elements needed to achieve Web3 success, including data, digital assets, DAO governance and business logic, in addition to the need to add AI elements. He called on us to return to our roots and think about what remains unchanged while paying attention to the rapid changes. For example, computing, storage, network and security in the infrastructure are constant elements.
As the largest cloud service provider in the Asia-Pacific region, Alibaba Cloud provides users with these infrastructures and the corresponding business ecosystem, laying the foundation for Web3 and AI innovation.
James liu
He also shared the industry map of Web3.0, compared Web2.0 and Web3.0, and introduced innovation and layout from the physical layer, IaaS layer, PaaS layer and SaaS layer.
James also shared application cases of Web3 and AI. One of them is the cross-chain Cloudverse solution jointly released by Alibaba Cloud, Avalanche and MUADAO. In addition, he mentioned the fully AI-driven e-KYC financial compliance solution, which has been applied in the Web3 industry.
When elaborating on Alibaba Cloud's innovation layout, James mentioned that Alibaba Cloud has established an innovation accelerator in Singapore to assist the development of multiple fields, covering emerging fields such as Web 3.0, AI, sustainable development, and medical care. He believes that innovation requires ecological support. Whether it is Web3.0 or AI, a sound ecosystem is needed to promote joint innovation.
Finally, he welcomed everyone to join the Alibaba Cloud ecosystem and jointly achieve innovation and common growth through the ecosystem.
03
BitDATA founder Wang Kai: In the era of artificial intelligence, the effectiveness and challenges of anti-money laundering
As a senior practitioner in the digital currency industry, Wang Kai shared the application of AI in anti-money laundering at digital currency exchanges.
BitDATA Exchange (BitEx) founded by Wang Kai
is a Singapore-based cryptocurrency exchange that focuses on security and promotes efficient and high trading volume with advanced technical support.
Wang Kai
After briefly reviewing the development process of digital currency, Wang Kai gave his own judgment on the future development trend of digital currency. He said that the first is around financial derivatives; the second is centralized and decentralized exchanges; the third is transactions based on the chain; and the fourth is towards compliance and licensed operations. Under this premise, with the circulation and application of large amounts of funds in the market, anti-money laundering has become an important issue.
So, what are the applications of AI in the field of anti-money laundering?
Wang Kai gave the following answers: First, use big data to analyze ledger data, transaction records, customer information, market dynamics and other data on the chain to better identify risks; second, based on machine learning, the product (BitDATA) By continuously learning and analyzing data to change the recognition model, abnormal trading behaviors can also be discovered and predicted. In addition, through machine learning, we can discover some strongly related transactions, and through further correlation analysis of big data, we can discover some hidden relationships between these transactions. Third, through long-term accumulated transaction data and machine learning models, trend analysis can be conducted to determine future risk trends.
Through the AI model, BitDATA can continuously learn and optimize to improve the accuracy of subsequent identification. Of course, the entire technology is still inseparable from human intervention. Wang Kai said that this process is an automation + manual collaboration, and a large number of automated decision-making and processing processes reduce the cost of manual review and improve compliance.
Generally speaking, artificial intelligence technology has great effects and benefits in the field of anti-money laundering, especially in improving detection accuracy, reducing operating costs and increasing customer trust.
Despite this, Wang Kai also admitted that AI is still under development and faces many problems and limitations: for example, the effectiveness of the AI model depends on data quality, and data quality will directly affect the result deviation. For another example, in terms of interpretability, deep learning models often lack transparency, making some decisions difficult to explain.
At the end of the speech, Wang Kai envisioned the huge potential of artificial intelligence in the field of anti-money laundering in digital currency exchanges.
04
Roundtable Forum: When AI and Web3 meet, what changes will happen in the future?
After the speech, a roundtable forum on AI and Web3 brought the atmosphere of the entire venue to a climax. Hosted by Boyang, founder of Project Twelve (P12), three guests, Wang Kai, founder of BitDATA Exchange, Yang Yongzhi, CEO of Abbots Technology and founder and CEO of Dolphin Browser, and Min, partner of Foresight X, discussed the connection between AI and Web3. Topics such as how to better leverage the value of the two were discussed heatedly.
From left to right are Boyang, Yang Yongzhi, Wang Kai, Min
Boyang first said that the technology wave of AI and Web3 is very important and is the focus of great attention of practitioners. And then raised the first question, what is the relationship between Web3 and AI, and whether there are potential opportunities to combine the two.
In this regard, Wang Kai shared a very interesting point of view in his eyes, that is, from the perspective of productivity, production tools, and production materials, AI can be regarded as a production tool, while Web3 is a relatively complex thing, which not only includes technology Change is also a change in production relations, just like the Internet has gone far beyond technology itself. Under such changes, Web3 may affect many aspects, and new technologies, such as AI, can also help Web3 develop better.
Foresight X partner Min started thinking from the perspective of interaction between the two parties. In her opinion, Web3's role in AI is more about infrastructure. All the features of Privacy, Trustless, Decentralization and Token Economy of Web3 can be used in the three most important elements of AI infrastructure - data, model and computing power. Modify accordingly. In contrast, AI has brought many changes to various application fields of Web3, such as on-chain data analysis tools developed based on large models such as GPT and LLaMA, DID (decentralized identity), NFT, blockchain game NPC, etc.
However, Yang Yongzhi is cautious about the impact of AI on Web3. He believes that the application of AI in Web3 and Web2 is no different. The essence is the same and will not create another huge opportunity for change. If there can really be a huge change, it may take at least 5 years.
At the forum, Boyang expressed his approval for the views shared by the guests. He also said that the application of AI in anti-money laundering, security auditing and other fields is also very interesting. Later, Boyang expressed his vision for the future development of Web3 and AI: How will cutting-edge technologies such as Web3 and AI add to the process of transforming the industry in the future?
Wang Kai said that AI itself is a technical tool, and the most important value is reflected in its application. How to combine it with new business models to achieve innovation and realize commercial value is the key. If Web3 wants to develop, it must lower the technical threshold, otherwise it will hinder development. He also gave a specific example. For example, the currently very popular AIGC and ChatGPT require people to feed Internet data. So can these things be combined with Web3 to allow more people with professional knowledge to share? ChatGPT can learn these Expertise generates better value. Is it a new application to combine this kind of contribution with token economic rewards?
Yang Yongzhi believes that the Metaverse game may be an explosive opportunity for the close integration of Web3 and AI. He explained that many games nowadays, such as NPCs, are essentially AI and have nothing to do with the Metaverse. The biggest difference between the Metaverse and games is co-construction, sharing and governance. At present, many Rangers essentially add VR or AR to expand the boundaries of the game. This kind of metaverse cannot be achieved. Only a metaverse that is built on the blockchain and truly allows the community to build, share and govern is the real metaverse. Of course, if you want to do a good job in the Metaverse, you cannot do without AI.
Min gave a corresponding perspective from the perspective of an incubator. She believes that the scarcest factor at present is actually people. At present, I see that in many teams the people in the two fields do not overlap: the people in AI are on the left, the people in blockchain are on the right, and there are very few people who are trained in both fields. However, with the rise of large models, the industry is increasingly accepting AI and Web3. In the future, we may see teams with stronger comprehensive capabilities produce something different.
With the free talk about the future, the Web3 sub-forum came to a successful conclusion.
05
Conclusion
On the afternoon of August 15, with the successful conclusion of the Web3 sub-forum, GAIR 2023 also came to an end. Looking back at this GAIR, from GPT to Web3, the conference discussed many hot topics.
Eliminating the false and retaining the true, there are still a group of real pursuers on the Web3 track. As a cutting-edge scientific technology, whether it is a large model or Web3, its development and improvement cannot be achieved overnight.
Now, although the craze of Web3 has subsided, the real doers have never simply chased the trend, but moved forward steadily and found new life in less visited places.
We look forward to these doers bringing us more surprises in the future.
//
Recent popular articles
京公网安备 11010802033920号