5月19日，由全球数字金融中心（杭州）、全联并购公会联合举办的“科技向善：强AI时代的变革浪潮”人工智能与数字金融研讨会在杭州顺利举办，来自国内外研究机构、高校、行业协会、金融媒体、相关从业机构等20余位专家学者通过线上或线下方式参会发言，围绕AI的技术、应用、道德伦理、影响等方面开展深入交流研讨。
在AI核心科技的进展方面，与会专家阐释了以ChatGPT为代表的AI核心科技进展。作为以人工智能技术驱动的自然语言处理工具，ChatGPT凭借其作为“转换器”的优势，超越了较为传统的卷积神经网络，优化了模型结构、训练效率等能力，并通过预训练及强化学习达到能够通过理解和学习人类语言进行对话交流，甚至完成文稿撰写等任务的效果。ChatGPT等一系列大模型的成功表明模型具备涌现能力，而只有模型和数据都跨过门槛，且经过充分的训练，才能出现规模效应，发挥模型的涌现能力，产生强大能力和海量知识。在这一过程中，较于模型构建，需给予数据质量更多的重视，避免过度参数化，加强数据应用，最终提升模型的整体效果。
在未来AI的应用方向和生态环境方面，与会专家指出，随着模型和数据的逐步扩大，未来AI技术将逐渐向多模态、专业化的方向发展，当下AI模型的发展将推动新一轮生成式AI应用的爆发。AI系统通过多轮对话的方式进行问题理解、问题检索、答案生成等过程，最终生成用户反馈，未来AI技术将对心理教育、学生管理、医疗服务、银行业务、文案工作及其它一系列领域产生重要的影响。而随着AI技术融入人类生活的程度不断加深，引导AI向善发展尤为关键。为达到AI向善的目的，首要因素就是保证输入的数据本身向善，需以内在核心的基本要素及原则为首要评估标准，实现AI技术与人类的共赢与和谐。
在AI的伦理问题方面，与会专家指出信息技术伦理问题产生的根源在于信息技术的摩尔定律远远快于社会文化体系的发展速度，在人工智能等新兴技术在造福人类的同时，应做到相应的伦理治理体系也能与技术进步一同进化。同时，由于人工智能与人类智能的界限变得模糊，以及其大量实施了使机器目标符合人类意图的人机价值对齐工程，加之其创新性和影响存在高度不确定性，为此，应推动顶层法律制度设计、技术伦理研究与宣传方面的工作，推动开源创新、构建合作研究和开源创新平台、发展一系列的测试工具，对其安全性和伦理问题进行量化测试，以构建完善的技术伦理治理体系。
在AI对金融行业的影响方面，与会专家指出发挥算力的作用可以提升金融行业的运行效率。发挥极致算力解决高频交易的瞬时计算问题，发挥超级算力支持大模型训练、金融复杂建模等系统工程。具体来看，一是对瞬时计算速度极敏感的场景，可以通过计算机体系架构级优化、预计算、软件硬件化、实时数据复杂指标增量计算等技术方式提升实时计算速度；二是对超大规模复杂计算的算力发挥领域，优化应用层原子计算效率，做到对数据结构、算法、编译器的原子计算优化。在调度层的算力分配优化方面，平衡好各类约束，在网络延迟、数据通信效能、节点处理能力、原子计算处理时间预算、任务颠簸做好算力调度优化。
在AI的投资热点方面，与会专家指出AI对相关行业起到降本增效的作用，加速大模型的商用落地，推动其实现商业变现，为行业发展创造更多可能。同时，AI检索信息的全局性会扩大其在投资领域的适用范围，从高频投资逐步向低频投资渗透，不断扩展其应用价值。
"Technology for Good: Wave of Change in Strong AI Era"
Artificial Intelligence and Digital Finance Seminar held in Hangzhou
On May 19th, the seminar on artificial
intelligence and digital finance "Technology for Good: Wave of Change in Strong AI Era" jointly organized by the GDFC and the
CMAA was successfully held in Hangzhou. More
than 20 experts and scholars from domestic and foreign research institutions,
universities, industry associations, financial media, and relevant institutions
attended the conference online or offline, and conducted in-depth exchanges and
discussions on the technology, application, morality and ethics, and impact of
AI.
In terms of the progress of core technology
development of AI, experts illustrated the progress of core technology
development of AI represented by ChatGPT. As a natural language processing tool
driven by artificial intelligence technology, ChatGPT, with its advantages as a
transformer, surpasses the traditional Convolutional Neural Network, optimizes
model structure and training efficiency, etc. Through pretraining and reinforceed
learning, ChatGPT manages to conduct dialogue and write articles after it
succeeded in communicating with human. The success of a series of large models
such as ChatGPT indicates that models have the ability to emerge, whereas only
when the model and data both cross the threshold and undergo sufficient
training can scale effects occur, which unleashes the models’ emergence ability
to generate strong capabilities and massive knowledge. In this process, more
attention needs to be paid to data quality compared with model construction, and
users ought to avoid excessive parameterization while strengthen data
application to ultimately improve the overall effectiveness of models.When it comes to the application direction
and ecological environment of AI in the future, experts pointed out that with
the gradual expansion of models and data, AI technology will gradually develop
towards multimodality and specialization in the future. The development of current
AI models will promote a new round of generative AI applications. Through
multiple rounds of dialogue, AI systems generate feedbacks by following the
process of problem comprehension, problem retrieve, and answer generation. In
the future, AI technology will have significant impacts on areas such as
psychological education, student management, medical services, banking,
copywriting, etc. Meanwhile, as AI technology continues to deepen its integration
into human life, guiding AI towards a right direction becomes particularly
crucial. In order to achieve the goal of AI being virtuous, the primary factor
is to ensure that the input data itself is ethical. We need to prioritize fundamental
core elements and principles as an assessment standard to achieve a harmonious
relationship between AI and humanity.In terms of the ethical issues of AI,
experts pointed out that the root of the ethical issues of information
technology is that Moore's Law of information technology is much faster than
the development of social and cultural systems. As emerging technologies such
as artificial intelligence benefit mankind, the corresponding ethical
governance system should evolve together with technological progress. At the
same time, since the boundary between artificial intelligence and human
intelligence has become blurred, and it carried out a large number of
human-machine value alignment engineering to make machine goals conform to
human intentions, plus its innovation and influence are highly uncertain, we
should promote top-level legal system design, technology ethics research and publicity,
promote open source innovation, build cooperative research and open source
innovation platforms and develop a series of testing tools to conduct
quantitative testing of its security and ethical issues so that we can build a
relatively perfect technology ethics governance system.In terms of the impact of AI on the
financial industry, experts pointed out that making use of the effect of
computing power can improve the operational efficiency of the financial
industry. We could exert extreme computing power to solve instantaneous
computing problems of high-frequency trading, and we can also exert super computing
power to support system engineering such as large model training and financial
complex modeling. Specifically, first, the scenes that are extremely sensitive
to instantaneous calculation speed. They can be improved by computer architecture
level optimization, precomputation, software to hardware, and incremental
calculation of complex indicators of real-time data to improve real-time
computing speed. The second is the field of computing power for super-scale
complex computing, it can optimize the atomic computing efficiency of the
application layer, and optimize the atomic computing of data structures,
algorithms and compilers. In terms of computing power allocation optimization
at the scheduling layer, all kinds of constraints should be balanced, and
computing power scheduling optimization should be done in terms of network
delay, data communication efficiency, node processing capacity, atomic
computing processing time budget and task turbulence.In respect of AI investment hotspots,
experts pointed out that AI plays a role in reducing costs and increasing
efficiency in related industries, since it accelerates the commercial
implementation of large models, promotes their commercial realization, and
creates more opportunities for industry development. At the same time, the comprehensive nature
of AI information retrieval will expand its scope of application in investment,
gradually penetrate from high-frequency investment to low-frequency investment,
and continuously expand its application value.
全联并购公会是2004年经全国工商联批准成立、2012年经民政部登记注册的非营利性民间行业协会。总部位于北京，行政主管单位为全国工商联。
作为全国工商联直属的唯一金融属性行业商会，全联并购公会现拥有200余家机构会员和4000余名个人会员，建立了法律、基金、标准、国际、并购维权、数字经济以及信用管理、金融文化、金融科技、中小企业投融资等 20个委员会，为规范并购行业发展、促进产业资本与金融资本的深度结合、提升中国企业竞争力、促进中国企业参与全球并购做出了积极贡献。
电话：010-65171198
网址：www.ma-china.com
邮箱：cmaa@mergers-china.com
