Homo Ratiocinator (Reckoning Human)
After 40,000 years of making tools for computing and reasoning, it is time for Homo Ratiocinator to live up to its traditional name, Homo Sapiens.
March 2025 - Vol. 68 No. 3
Features
Putting the Smarts into Robot Bodies
Delving into the key tradeoffs of building foundation models for embodied AI systems.
Feedback Loops Guide AI to Proof Checking
Recent developments suggest artificial intelligence could play a role in helping computers test and verify smaller, more manageable portions of mathematical proofs.
How Software Bugs led to ‘One of the Greatest Miscarriages of Justice’ in British History
Bad coding and bad testing characterize the software that led to wrongful convictions, financial ruin, and four suicides.
Controlling AI’s Growing Energy Needs
Lower-energy options are being sought to reduce the energy footprint of AI training.
Should governments decide what regulations are necessary to ensure safe development and deployment of AI technologies?
Abstraction is a by-product of computer science's central purpose, understanding information processes.
A fundamental paradox riddles much of today’s mainstream AI alignment research, and the community needs to find ways to mitigate it.
Artificial Intelligence as Catalyst for Biodiversity Understanding
AI should be a bridge rather than a barrier to the biological sciences' pursuit of understanding and preserving the natural world.
A Glimpse Into the Pandora’s Box
A combination of safety measures and safety labels should be developed and employed on how AI models in applications analyze camera frames in real time.
Program Merge: What’s Deep Learning Got to Do with It?
Leading figures of Microsoft Research's DeepMerge project discuss their efforts to apply machine learning to complicated program merges.
Prevalence and Prevention of Large Language Model Use in Crowd Work
LLMs may be particularly harmful in crowd work when the goal is to capture the diversity of human preferences, behaviors, or opinions.
Exploiting Cross-Layer Vulnerabilities: Off-Path Attacks on the TCP/IP Protocol Suite
An investigation of vulnerabilities within the TCP/IP protocol suite that can be exploited by forged ICMP errors.
Molecular Communications in Blood Vessels: Models, Analysis, and Enabling Technologies
Considering models and technologies to implement proposed alternatives to molecular communications in the bloodstream.
The Sustainability Gap for Computing: Quo Vadis?
A reformulated IPAT model provides insight for computer system engineers to consider computing's environmental impact.
Technical Perspective: The Surprising Power of Spectral Refutation
Could an algorithm that seems unrelated to SAT serve as a reliable refutation heuristic? Perhaps surprisingly, the answer is yes.
New Spectral Algorithms for Refuting Smoothed k-SAT
Our methods yield new algorithms for smoothed k-SAT instances with guarantees that match those for the significantly simpler and well-studied model of random formulas.
Technical Perspective: Toward Building a Differentially Private DBMS
The paper is an important step toward automatically ensuring privacy for arbitrary computations.
R2T: Instance-Optimal Truncation for Differentially Private Query Evaluation with Foreign Keys
The first DP mechanism for answering arbitrary SPJA queries in a database with foreign-key constraints.
"LLMs might get it sort of right some of the time, but nowhere near as reliably as the average person."
