Towards conscious artifacts:ChatGPT is the key,but to open the door,one must find the lock  

走向有意识的人工制品:ChatGPT是钥匙,但要打开门,必须找到锁

作　　者：SAMSONOVICH Alexei V SAMSONOVICH Alexei V(俄罗斯国立核研究大学,俄罗斯莫斯科115409;乔治梅森大学,美国弗吉尼亚费尔法克斯22030-4444)

机构地区：[1]National Research Nuclear University MEPhI,Moscow 115409,Russian Federation [2]George Mason University,Fairfax,Virginia 22030-4444,USA

出　　处：《宁波大学学报(理工版)》2025年第2期1-16,共16页Journal of Ningbo University:Natural Science and Engineering Edition

基　　金：Supported by Russian Science Foundation(22-11-00213).

摘　　要：The idea to replicate in a machine human consciousness,understood as an immediate first-person subjective experience,becomes one of the greatest challenges of our time.Scientific understanding of the nature of phenomenal consciousness may require an expansion of the scientific method itself.While this resolution is postponed,the task of reproducing the consciousness phenomenon in a computer should be approached within the framework of functionalism.Proposed so far theories of this sort appear problematic.An approach to implementing humanlike consciousness is proposed here based on BICA(biologically inspired cognitive architectures)combined with LLM(large language models).The concept is illustrated by example of a virtual tutor.It is argued that final solutions can be obtained in a purely neuromorphic form,which will enable their further adaptation,growth,and evolution.We start from discussion of the ontological problem of consciousness and its temporary nonreductive solution,which makes a scientific study possible.The main hypothesis for the proposed study is formulated based on the BICA Challenge in terms of the eBICA cognitive architecture.Only the limited case of socially emotional consciousness is analyzed as example.It is explained how LLM,being not conscious on their own,accessible via ChatGPT,can be connected to eBICA,used to build a conscious prototype and validate its claimed property.Selected example of a virtual tutor further illustrates how this technology will allow the tutor to develop human-level teacher-student relationships that will increase the efficiency of tutoring.Future prospects for the proposed direction of research include a variety of sentient agents to be created for the many practical human needs with the possibility of their further autonomous evolution.在机器中复制人类意识(可被理解为一种直接的第一人称主观体验)的想法是这个时代最大的挑战之一.对现象意识的本质的科学理解可能需要扩展科学方法本身.可以暂时搁置解决该问题,而先尝试在功能主义的框架内在计算机中重现意识现象.迄今为止,为此所提出的相关理论似还存在问题.提出了一种基于BICA(生物启发认知架构)结合LLM(大语言模型)实现类人意识的方法,并以虚拟导师为案例对其进行说明.一般认为,上述问题的最终解决方案可以以纯神经形态的形式获得,且这一形式能使其得以进一步变更、发展和演化.为此,从意识的本体论问题及其暂时的非还原解决方案开始讨论,因为这是科学研究的一条可行路线.主要假设是基于“BICA挑战”中有关eBICA认知架构制定的,作为例子仅分析了有限的社会情感意识案例,解释了本身不具备意识的LLM如何通过ChatGPT连接到eBICA,用于构建有意识的人工制品原型并验证其声称的性能.所选定的虚拟导师案例进一步说明了这项技术将如何使虚拟导师发展出人类级别的师生关系,从而提高辅导效率.未来的研究方向可以包括创建各种各样具有感知能力、且有可能进一步自主进化的代理物,以满足人类的众多实际需求.

关 键 词：machine consciousness the Hard Problem BDI BICA LLM virtual tutor 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]