Neuroadaptive Bayesian Optimisation to Identify which Combination of Gaze and Emotion in the Parent Face Maximises Attention in the Individual Infant, 2023-2024

Infants’ motivation to engage with the social world depends on the interplay between individual brain’s characteristics and previous exposure to social cues such as the parent’s smile or eye contact. Different hypotheses about why specific combinations of emotional expressions and gaze direction engage children have been tested with group-level approaches rather than focusing on individual differences in the social brain development. Here, a novel Artificial Intelligence-enhanced brain-imaging approach, Neuroadaptive Bayesian Optimisation (NBO), was applied to infant electro-encephalography (EEG) to understand how selected neural signals encode social cues in individual infants.
EEG data was acquired from 42 6- to 9-month-old infants looking at images of their parent’s face, analysed in real-time and selected by a Bayesian Optimisation algorithm to identify which combination of gaze and emotional expression of the parent’s face produces the strongest brain activation in the child. This individualised approach supported the theory that the infant’s brain is maximally engaged by communicative cues with a negative valence such as direct gaze and angry facial expressions. Moreover, we evaluated whether results also capture individual differences in behaviour. We found that infants attending preferentially to faces with direct gaze had increased positive affectivity and decreased negative affectivity compared to infants preferentially attending to faces with averted gaze.
This work supports the idea that infants’ attentional preferences for social cues are heterogeneous and lays the foundation for the development of neuroimaging-informed personalized experiments to study diversity in neurodevelopmental trajectories of social skills.<p>Babies are born with a drive to interact with other people. Within a year, this drive takes them from a passive newborn to a smiling, talking toddler. Our goals shape how sociable we are and who we socialise with across the lifespan, and are thus fundamental to social psychology (Over, 2016). However, the reasons why babies choose to interact remains a mystery. Measuring motivation is difficult because it is generated by the child, whilst traditional experimental methods measure passive responses to stimuli produced by the experimenter. Our transformative approach to studying infant social motivation is inspired by innovations in advertising. In the last twenty years, advertising has been revolutionised by the use of artificial intelligence (AI). Rather than the traditional model of creating generic campaigns based on what creatives thought consumers wanted, on the internet advertisers can now identify what exactly motivates individual customers by trying out different adverts and measuring an individual customers reaction to them. For example, if you click on an advert for a holiday in Mauritius, you will then see adverts for holiday resorts on other websites that you later visit. We aim to use the principles of this approach to determine what motivates babies to interact with other people. Study 1: Identify brain signals and networks related to social motivation. As a first step, we need to identify readouts of core social reward networks in the brain; measuring the brain (rather than behaviour) allows us to measure social motivation using the same signals across development. We can measure these networks very precisely using functional magnetic resonance imaging (fMRI), but this is not suitable for babies who are awake. Functional Near-Infrared Spectroscopy (fNIRS) is an alternative imaging method that is very similar to fMRI but that can be used with babies who are awake. We will use a combination of fNIRS and fMRI to identify brain signals of the brain networks that are involved in the core social reward networks, which we can then measure with fNIRS alone in Studies 2 and 3. Study 2: Identify the social cues infants find maximally rewarding. We will use social tasks that use eye tracking methodology. This technology follows exactly where infants look at on a screen, with infants looking behaviour even triggering visual events on the screen (e.g., if infants look towards a face, this will trigger a social reward such as a smiling or talking face). As the infant watches the screen and completes the tasks, the algorithm will be able to learn which tasks produce a larger brain signal from the social reward networks. This then allows us to determine which type of social interaction is particularly rewarding for the infants and how this may change as babies grow up. For example, very young babies may be particularly interested in eye gaze and smiling, but as they grow into toddlers and begin to talk, language may be more interesting for them. Study 3: Develop tools for using our approach within real-life interaction. Screen based social tasks are extremely useful, but watching social stimuli on a screen is very different from the dynamic nature of interacting with people. Here, we will measure infants brain responses whilst they interact with a social partner. As the infant interacts with their partner, the algorithm will identify the type of social cues that they find particularly rewarding. The algorithm will then prompt the trained social partner to engage in these maximally rewarding social interactions (such as eye contact, smiling or touch). This will provide a demonstration of how our tools can be used within a custom intervention design for children with conditions that affect social motivation, like autism. Taken together, our work is designed to produce new tools to transform our understanding of why babies socialise with other people, and to help vulnerable children to reach their full potential.</p>

婴儿参与社交世界的动机，取决于个体大脑特征与既往接触社交线索（如父母微笑、眼神交流）之间的交互作用。过往针对“特定情绪表达与注视方向的组合为何能吸引儿童”的相关假设，多采用群体水平研究方法进行验证，而非聚焦于社交脑发育中的个体差异。本研究采用一种新型人工智能（Artificial Intelligence）增强脑成像方法——神经自适应贝叶斯优化（Neuroadaptive Bayesian Optimisation, NBO），对婴儿脑电图（electro-encephalography, EEG）数据进行分析，以探究个体婴儿的选定神经信号如何编码社交线索。

研究采集了42名6至9个月大婴儿的脑电图数据，这些婴儿在观看父母面部图像时，算法会实时分析数据并通过贝叶斯优化算法筛选，以确定父母面部的注视方向与情绪表达的哪种组合能在婴儿脑中引发最强的激活反应。这种个体化研究方法验证了相关理论：带有负效价的社交线索（如直接注视与愤怒面部表情）能最大程度激活婴儿大脑。此外，本研究还验证了研究结果是否能反映行为层面的个体差异。结果发现，相较于优先关注偏斜注视面部的婴儿，优先关注直接注视面部的婴儿表现出更高的积极情感与更低的消极情感。本研究证实了婴儿对社交线索的注意偏好具有异质性，并为开发基于神经成像的个体化实验奠定了基础，以此探究社交技能神经发育轨迹的多样性。

婴儿天生便具备与他人互动的驱力。在出生后的一年内，这种驱力便会推动他们从被动的新生儿成长为会微笑、会牙牙学语的幼儿。我们的人生目标会塑造我们整个生命周期的社交倾向与社交对象，因此这也是社会心理学的核心议题（Over, 2016）。然而，婴儿为何会选择与他人互动的原因仍是未解之谜。对动机进行测量颇具挑战，因为动机由儿童自身产生，而传统实验方法仅能测量受试者对实验者所呈现刺激的被动反应。

本研究用于探究婴儿社交动机的创新性方法，灵感来源于广告领域的技术革新。在过去二十年中，人工智能的应用彻底革新了广告行业。相较于传统模式——即广告创意人员基于自身对消费者需求的预判制作通用广告投放方案，如今的互联网广告商可通过测试不同广告素材、并监测个体消费者的反应，精准识别出能够驱动个体顾客的具体因素。举例而言，若你点击了一则毛里求斯度假广告，后续浏览其他网站时便会看到度假酒店相关的广告推送。

本研究旨在借鉴此类方法的原理，探明驱动婴儿与他人互动的具体因素。
研究1：识别与社交动机相关的脑信号与脑网络。作为第一步，我们需要识别脑中核心社交奖赏网络的信号读出指标；相较于行为测量，直接测量大脑活动能够让我们在整个发育阶段中，通过统一的信号指标对社交动机进行量化。我们可通过功能磁共振成像（functional magnetic resonance imaging, fMRI）对这些脑网络进行高精度测量，但该技术并不适用于清醒状态下的婴儿。功能近红外光谱（functional Near-Infrared Spectroscopy, fNIRS）是一种替代成像技术，其原理与fMRI类似，但可用于清醒状态下的婴儿。我们将结合fNIRS与fMRI技术，识别核心社交奖赏网络所涉及的脑信号与脑网络，后续研究2与研究3便可仅通过fNIRS技术完成相关测量。

研究2：识别婴儿认为最具奖赏性的社交线索。我们将采用结合眼动追踪技术的社交任务。该技术可精准追踪婴儿在屏幕上的注视位置，甚至可通过婴儿的注视行为触发屏幕上的视觉事件（例如，若婴儿注视某张面部图像，系统便会触发微笑或说话的面部图像这类社交奖赏刺激）。当婴儿观看屏幕并完成任务时，算法将学习哪些任务能够在社交奖赏网络中引发更强的脑信号。借此我们便可明确，哪种类型的社交互动对婴儿而言最具奖赏价值，以及这种偏好会如何随着婴儿的成长而发生变化。例如，极低龄婴儿可能对眼神交流与微笑格外感兴趣，但当他们成长为幼儿并开始学习说话时，语言或许会成为他们更关注的对象。

研究3：开发适用于真实社交互动场景的研究工具。基于屏幕的社交任务虽极具应用价值，但屏幕上呈现的社交刺激与真实人际互动的动态特性存在显著差异。因此，本研究将在婴儿与社交伙伴互动的过程中，测量其脑活动反应。当婴儿与互动伙伴交流时，算法将识别出婴儿认为最具奖赏性的社交线索类型，并提示经过训练的互动伙伴做出此类最能引发婴儿愉悦感的社交行为，例如眼神交流、微笑或肢体接触。这将为我们的工具在针对社交动机受损儿童（如自闭症患儿）的定制化干预方案中的应用提供示范。

综上，本研究旨在开发全新工具，以革新我们对婴儿为何会产生社交行为的认知，并助力存在社交发展障碍的儿童充分发挥其潜能。