How AI is Transforming Grocery Shopping and Predictive Coding

AI Revolutionizes Grocery Shopping

Dmytro Lylyk and Vladyslav Mehera, co-founders of Neomi, are pioneering a new approach to grocery shopping that focuses on the shopper rather than the product. They argue that AI should consider the shopper's health goals, dietary restrictions, and context to construct personalized shopping baskets. McKinsey's 2026 grocery research supports this shift, noting that nearly 55% of consumers want personalized nutrition recommendations. Lylyk emphasizes that their philosophy is about 'foods for health, not for shelves.' Neomi's AI assistant exemplifies this by tailoring shopping experiences to fit individual needs, such as dietary preferences and personal goals.

The technology behind Neomi's approach involves understanding the emotional and physiological needs of consumers. AI interprets contextual signals, allowing for a more coherent and relevant shopping experience. For instance, if a user selects a 'Romantic Dinner for Two,' the AI might include wine in the basket, even if it wasn't specifically requested. This is a departure from traditional product-led systems that rely heavily on visibility and merchandising strategies.

The implications of this technology are significant for the grocery industry. According to McKinsey, 70% of consumers prefer home delivery, with 67% citing time savings as a primary reason. AI-driven personalized shopping could reduce the need for browsing, offering convenience and efficiency. Retailers are expected to see a rise in fully personalized promotions, which could redefine the relationship between brands and consumers. As Lylyk notes, 'AI could push products into the cart just because they are promoted, but they wouldn’t fit the cart, or they wouldn’t fit the expectation.'

From my perspective, this shift mirrors broader trends in AI, where the focus is increasingly on understanding the user rather than the product. This aligns with the human brain's ability to process complex contextual information, suggesting a future where AI and neuroscience continue to converge.

Augmented Lagrangian Predictive Coding as a Backprop Alternative

In the realm of deep learning, PC-ALM (Augmented Lagrangian Predictive Coding) is emerging as a promising alternative to backpropagation. Developed by researchers at Sakana.ai, PC-ALM trains residual MLPs (multi-layer perceptrons) up to 1000 layers using layer-local dynamics. Unlike traditional backpropagation, which requires a strict sequence of forward and backward passes, PC-ALM employs a feedback control dynamical system at each layer. This system distributes supervision credit throughout the network, offering a more efficient training method.

Technically, PC-ALM is an extension of standard predictive coding, introducing dual neurons per layer as Lagrange multipliers. These neurons create a PI feedback controller, allowing the network to function with only local computation. This method sidesteps the need for synchronized forward and backward passes, making it more akin to the brain's distributed processing. The approach shows promise in energy-efficient deep learning, particularly for neuromorphic hardware, where simulation is cheaper than on GPUs.

The implications of PC-ALM for the deep learning industry are profound. As traditional backpropagation faces scalability challenges, particularly in deep, narrow networks, PC-ALM offers a viable alternative. This method could revolutionize how networks are trained, particularly for tasks where energy efficiency is crucial. By understanding how the brain solves the multilayer credit assignment problem, PC-ALM could inform the development of more biologically plausible AI systems.

In my view, PC-ALM represents a significant step towards integrating principles of neuroscience into AI. By mimicking the brain's approach to processing, we can develop more efficient and scalable AI systems. This aligns with the broader trend of leveraging insights from neuroscience to enhance AI capabilities.

Retraction of AI Study Highlights Data Transparency Issues

The retraction of a paper on the cognitive impact of generative AI, authored by Sarah Baldeo, CEO of ID Quotient Advisory Group, underscores significant challenges in data transparency and ethics. Published in the American Psychological Association's journal, the paper was withdrawn after Baldeo refused to share data with the editors. The study had involved 1,923 adults performing AI-assisted tasks, but inconsistencies in data and lack of ethics approval were flagged by experts.

The technical issues revolved around the inability to share participant-level data due to Canadian Research Ethics Board frameworks. Baldeo argued that the requested data transfer was unauthorized under these frameworks, complicating post-publication verification. Critics like Sandra Grinschgl pointed out that even if participant data can't be shared, demonstrating the platform used or sharing statistical code should be feasible.

This incident has significant implications for AI research and its credibility. The reluctance to share data and the establishment of ethics bodies by researchers themselves raise concerns about accountability and transparency. As Malte Elson, a psychologist involved in assessing research trustworthiness, noted, 'Withholding those is not normal, it is odd to us.' This case highlights the need for robust data-sharing practices and independent oversight in AI research.

From an editorial perspective, the retraction reflects broader issues in the AI industry, where rapid advancements often outpace ethical considerations. Ensuring transparency and accountability is crucial as AI continues to influence various aspects of society. This aligns with the need for a framework that balances innovation with ethical responsibility.

Where Robles.AI Fits

Robles.AI is uniquely positioned to leverage the advancements discussed in these articles. Our expertise in AI-driven personalization can enhance the shopping experiences envisioned by Neomi, tailoring recommendations based on individual health goals and preferences. By integrating principles from PC-ALM, we can develop more efficient training methods for our AI models, ensuring they are both scalable and energy-efficient. Furthermore, our commitment to transparency and ethical AI aligns with the need for accountability in research, addressing the challenges highlighted in the generative AI study retraction.