A series of recent studies has shed new light on the complex relationships between human and artificial intelligence. Researchers have explored the geometry of predictive representations, the misalignment between backpropagation and brain responses, and the controllability of human outcomes through causal state intervention. Additionally, studies have examined the illusion of opting in AI-mediated consequential decisions and the potential benefits of vision-language models (VLMs) over large language models (LLMs).

A study published on arXiv, "Exploratory Experience Shapes the Geometry of Predictive Representations," investigates how exploratory and exploitative behavioral strategies shape internal predictive representations. The researchers built an online learning agent in a tree-like maze with a controllable parameter regulating the balance between exploratory and exploitative regimes. The agent updates a predictive-coding-based perception model from experience generated by its own behavior, predicting both future maze states and reward probability.

Another study, "Misalignment Between Backpropagation and the Hierarchy of Brain Responses to Images," addresses the question of whether backpropagated gradients exhibit a similar correspondence to the cortical hierarchy of visual processing as forward activations of pretrained models. Using functional magnetic resonance imaging (fMRI) and magnetoencephalography (MEG) recordings of human brain responses to natural images, the researchers found that backpropagated gradients can reliably predict both fMRI and MEG signals, specifically in higher-level visual cortex and for later latencies.

The study "You Are in Control of Your State: Why Human Outcomes Are Controllable Through Causal State Intervention" argues that human outcomes are controllable in a precise and operational sense through interventions that target the state and its weighting at the moment a decision is being formed. The researchers define a state as the time-indexed weighting vector over the dimensions that govern how an individual's biology, physiology, and neuropsychology process the next event into a decision and an outcome.

A fourth study, "The Illusion of Opting in AI-Mediated Consequential Decisions," highlights the profound ethical problem of the illusion of opting, in which persons and groups encounter the deceptive appearance of meaningful consequential choice while the agency needed to become genuinely capable of choosing is weakened. The researchers argue that AI systems should be evaluated by whether they protect and cultivate meta-capacity against the illusion of opting.

The final study, "VLMs May Not Globally Enhance Human Alignment over LLMs During Natural Reading," compares the performance of VLMs and LLMs in modeling human text processing. The researchers found that multimodal pretraining may not confer a uniform, global advantage in human alignment during natural reading, indicating that language-internal representations remain the key factor for modeling human text processing.

• Who: Researchers from various institutions
• What: Studies on predictive representations, brain responses, and human alignment with AI systems
• Where: International research institutions
• Impact: New insights into human and artificial intelligence

As research in human and artificial intelligence continues to advance, it is essential to consider the implications of these findings on the development of AI systems that align with human values and promote controllable outcomes. Future studies should investigate the potential benefits and limitations of VLMs and LLMs in various applications, as well as the ethical considerations surrounding AI-mediated consequential decisions.