- Chelsea Tang
BEYOND Expo 2024: AI Industry Working Towards Multimodal, Human-Like Models
Artificial intelligence (AI) technology is advancing at an unprecedented rate, so much so that capabilities once reserved for science fiction are now within the realms of possibility.
At BEYOND Expo 2024, held from 22 to 25 May in the Venetian Macao, Mr Xu Bing, Co-Founder of AI software provider SenseTime, and Mr Zhang Wen, Founder and Chairman of semiconductor developer Biren Technologies, spoke about where the AI industry is heading and what possibilities they are exploring.
The two first commented on the release of GPT-4, OpenAI’s latest and most advanced large language model, praising it for how seemingly human-like it is.
“You can even interact with this AI, which has emotions and realistic interaction capabilities, generating a lot of emotional interactions,” commented Mr Xu in Mandarin Chinese. “So, I think this aspect will bring a huge breakthrough in social interactions.”
“The barrier between AI and humans, which has historically been about whether it feels real or not, will create an innovative opportunity in the field of social interactions.”
Mr Zhang agreed, noting how smooth and fast the model’s voice interaction capabilities now feel. “I remember in GPT-3.5 and 4.0, the delay in voice interactions was about three to four seconds, but in 4.0, it is only about 0.3 seconds, so the delay is very minimal. He also thinks the new model boasts a significant improvement in engineering, which decreases overall cost and boosts efficiency of the whole end-to-end process.
Both executives agree that human-like AI with realistic interactions and expressions is a goal that the AI industry will strive to achieve.
The industry will also be exploring multimodality because of the wealth of information that is transmitted today through videos. The catch is that such a capability requires large-scale computational power. That’s why the work of chip manufacturers is also crucial in the advancement of AI capabilities.
Of course, the capabilities of AI are being advanced to open new use cases and help humans become more productive. Many, if not all, industries in the world will benefit, though there are specific ones that the technology is perhaps best suited for.
“I believe that the two industries most actively using AI for consumer-oriented innovation are the internet industry and the smart hardware industry,” explains Mr Xu.
“In the internet industry, we see many apps integrating AI capabilities, such as chatbots, AI customer service, and even AI boyfriends or girlfriends for social experiences. Almost all internet companies are actively exploring and adding AI capabilities to their existing user base to provide better user experiences, improve user retention, and increase user monetization.”
“In the smart hardware industry, major players like Apple, Samsung, and various domestic Android smartphone manufacturers are integrating AI capabilities into their latest smartphones and computers,” he added. “They are using large-scale multimodal models for localised processing.”
“For example, Microsoft recently demonstrated the AI capabilities of its computers, which can recall and search through the user's historical interactions, providing an enhanced user experience.”
Meanwhile, there are also emerging smart devices like AR glasses and wearables with cameras or voice capabilities. By embedding AI models into these devices, they can enhance human capabilities and unlock numerous innovative opportunities.
AI is also being integrated to improve productivity in traditional industries like finance and manufacturing.
As popular a tool as AI is today, Mr Zhang said there is still a gap between the current scale of users and their future potential. “For example, Chat-GPT, despite its popularity, has approximately one hundred million daily active users, which is far from the scale of apps like WeChat, Facebook, and other national-level apps. However, I believe that with the increasing practicality of AI, the gap will narrow,” he said.
“For instance, AI applications in meetings are still challenging, and the practicality is not yet sufficient. But as the concept of computational power continues to evolve, we will overcome these challenges and achieve significant impacts on human life.”
Once those challenges are overcome, that’s when national-level usage of AI apps and services will emerge.
During Mr Xu and Mr Zhang’s discussion, it was also emphasised that for AI capabilities and application to see success, how fundamental models evolve must be understood. That way, developers can predict when a model can reach new capabilities and use them to solve existing problems. Such a transition could happen within the next decade, possibly even in the next three to five years.
Another factor for success, in both the rollout of AI to consumers and the use of AI assistants in business, is reducing costs on the edge side, including training at the edge. Tech giants like Microsoft are already focusing on this aspect of the technology. If AI on personal computers can be realised, it would greatly promote the development of AI assistants and expand usage.
As for AI eventually achieving Artificial General Intelligence (AGI), a type of AI that surpasses human capabilities, both executives believe it will eventually happen. They even went as far as to say that AGI is becoming clearer over time. A year ago, the majority did not believe AGI could be achieved, but now, there are already predictions as to when the technology might emerge, ranging from 2027 to 2030 to 2040. AGI is also now something that everyone is waiting for, according to Mr Xu and Mr Zhang.
The two likened it to Moore’s Law, which initially drew doubts. They note that as computing power and AI capabilities improve, the closer the phenomenon becomes a reality.
Two Chinese AI executives said at BEYOND Expo 2024 that human-like AI with realistic interactions and expressions is a goal that the AI industry will strive to achieve.
The industry will also be exploring multimodality because of the wealth of information that is transmitted today through videos.
Both also believe that AGI, a type of AI that surpasses human capabilities will eventually happen.