Autonomous Learning and Development in Embodied Neuromorphic Systems (ALDENS)

这个项目的目标是为机器人创造一个开放式的人工智能，让它们像孩子一样长大——通过与人类和环境的多模式互动，自主学习和发展新技能。这将使人工智能和机器人技术实现全球期望的范式转变：从执行狭隘的预定义任务到自主的智力开发。为此，ALDENS将开创一种创新的跨学科方法，通过高效的类脑（神经形态）计算，生成具有实时发展的类人学习的交互式机器人模型，这将超越目前主流深度学习方法的可能性。ALDENS项目将建立新的发育型神经形态范式，这是一种超越个体范式局限性的协同组合：发育型机器人将为神经形态尖峰神经网络提供缺失的学习机制；与此同时，神经形态计算将提供高效的类脑资源，并能准确表征现实世界。具体来说，该项目的研究将创建和验证新的突破性方法，以构建自主、灵活和可扩展的人工大脑架构。这将改变交互式机器人认知架构的设计。事实上，计划中的创新发展将为预期的范式转变铺平道路，并为下一代真正自主的机器人奠定基础，这些机器人能够以类似人类的方式进行推理、行为和互动。研究人脑模型的一般风险因素是其功能组织和学习机制尚未完全了解。在跨学科的科学团体中，对于应该在人工代理中建模的基本结构和能力存在分歧。这使得这项研究具有不确定性，因为每个学科对“智力”都有自己的看法；不同的实验程序和方法来解释结果。重要的是，新的发育神经形态模型将成为提高发展心理学和神经科学等生命科学研究能力的有力工具。研究人员将能够使用发育神经形态模型来获取信息，并推进我们对人类学习和智力的理解。这个项目设想的生物学上合理的模拟将允许研究人员在对人类进行测试之前快速收集信息，以支持新的实验预测。有趣的是，病变模型将有可能复制认知功能障碍，以产生大脑内部工作的模拟信息，否则无法发现。该方法将有助于提高对神经发育和学习障碍的认识，以提高其诊断和治疗。伦理问题、缺乏信任和公众的偏见可能导致对自主学习机器人的排斥，并否定这项研究未来的社会经济影响。设想的学习过程的人性化将积极影响人们在生活中对机器人的信任、接受和采用。新的方法将使智能机器人能够像人类一样学习，这是一种通过实现最高程度的个性化来促进社交应用的新能力，即教师（用户）的需求和偏好可以塑造人工思维，使交互更加自然和可接受。为了最大限度地发挥未来的社会和经济影响，ALDENS项目还将定期让人工智能伦理的利益相关者和公众参与讨论研究，并就信任、安全和更广泛接受的伦理和法律界限的定义获得反馈。

This project ambition is to create an open-ended artificial mind for robots that can grow up like a child - autonomously learning and developing new skills via multimodal interaction with humans and the environment. This will enable a globally desired paradigm shift in AI and robotics: from performing narrowly pre-defined tasks to autonomous mental development.To this end, ALDENS will pioneer an innovative cross-disciplinary approach to generate models of interactive robots with real-time developmental human-like learning made possible by efficient brain-like (neuromorphic) computing, which will go above and beyond what is currently possible with the mainstream deep-learning approach.The ALDENS project will establish the new developmental neuromorphic paradigm, a synergic combination that will go beyond the limitations of the individual paradigms: developmental robotics will deliver the missing learning mechanisms for neuromorphic spiking neural networks; meanwhile, neuromorphic computing will provide efficient brain-like resources with an accurate representation of the real world. Specifically, the research in this project will create and validate new ground-breaking methodologies to build an autonomous, flexible, and scalable artificial brain architecture. These will transform the design of interactive robots' cognitive architectures. Indeed, the planned innovative developments will pave the way for the expected paradigm shift and lay the foundations of the next generation of truly autonomous robots able to reason, behave and interact in a human-like fashion.A general risk factor for research modelling of the human brain is that its functional organisation and learning mechanisms are not yet fully understood. There is disagreement within the cross-disciplinary scientific community regarding the fundamental structure and capabilities that should be modelled in artificial agents. This makes this research uncertain, with each discipline having its own view of "intelligence"; different experimental procedures and methodologies to interpret the results.Importantly, the new developmental neuromorphic models will be a powerful tool for increasing the research capacity in life sciences, like developmental psychology and neuroscience. Researchers will be able to use the developmental neuromorphic models to gain information and progress our understanding of human learning and intelligence. Biologically plausible simulations envisioned by this project will allow researchers to quickly collect information in support of novel experimental predictions before being tested on humans. Interestingly, it will be possible to lesion models to replicate cognitive dysfunctions to generate simulated information of the inner workings of the brain that cannot be discovered otherwise. This method would be useful for boosting the understanding of neurodevelopmental and learning disorders for the enhancement of their diagnosis and treatment.Ethical issues, lack of trust and prejudice of the public can result in the rejection of self-learning robots and negate the future socio-economic impact of this research.The envisioned humanization of the learning process will positively impact people's trust, acceptance, and adoption of robots in people lives. The new methodologies will enable intelligent robots to learn like humans, a new capability that will boost social applications by achieving the highest degree of personalisation, i.e. the needs and preferences of the teacher (user) can shape the artificial minds, making the interaction more natural and acceptable.To maximise the future social and economic impact, the ALDENS project will also regularly engage stakeholders in AI ethics and the public to receive discuss the research and get feedback on the definition of the ethical and legal boundaries for trust, safety, and wider acceptance.