Appropriate Computing

AI Systems Engineering as Craft

Or

Learning to Take Responsibility

Peter Overholser

• Computing and AI
• Sustainability considerations
• Framing and reframing
• Current work
• Computing projects at Schatz Energy Research Center and elsewhere
• Development of a curriculum in AI systems engineering in the tradition of Appropriate Technology:, “Appropriate Computing”

Current AI trajectory

Rapid buildout

• "We're also building out Hyperion, which will be able to scale up to 5GW [enough to power several million homes] over several years. We're building multiple more titan clusters as well. Just one of these covers a significant part of the footprint of Manhattan” - Mark Zuckerberg, CEO Meta
• "Our vision is simple: we want to create a factory that can produce a gigawatt [one nuclear power plant’s power output] of new AI infrastructure every week…” - Sam Altman, CEO OpenAI
• “A gold rush of intelligence has arrived. The one who goes fastest will win…” - Masayoshi Son, CEO Softbank

Current AI trajectory

Rapid buildout

• Energy implications
• Water implications
• Wealth concentration implications
• Social implications
• All three pillars of sustainability (social, environmental, and economic) are directly threatened

Current AI trajectory

Philosophical Explanations

• Bitter lesson and scaling hypothesis
• Lack of imagination
• Nihilism
• Utopian/Apocalyptic framing
• Lack of compelling alternative narratives

Current AI trajectory

Practical Explanations

• Greed and Power
• Circular investments and perverse incentives
• Demographic stress
• Geopolitical tensions (Is China in an arms race with the US?)
• Debt concerns
• No fundable alternatives

Current AI trajectory

Practical Explanations

Current AI trajectory

Major shortcomings

• Energy inefficiency
• Unsafe output and wide application
• Corrosive to human social structures (job loss, AI psychosis, misinformation)
• Downstream from philosophical and practical drivers

Current AI trajectory

Mitigation Approaches

Architecture and

Design

Sustainable AI

Human Centered AI

Open Source AI

A different approach

Architecture and

Design

Sustainable AI

Human Centered AI

Open Source AI

Appropriate Technology

History

• Often traced back to E. F. Schumacher’s book Small is Beautiful (1973) arguing against scale for its own sake. Promotes distributed, sustainable, and human-centered technologies.
• Influential, but only three universities in the US have maintained explicit support
• Impact
• Global
• Local: CCAT, Appropedia, Schatz Energy Research Center

Appropriate Technology

Examples

Appropriate Technology

Reordering of objectives and aesthetics

• Profit -> Sustainability
• Work as labor -> Work as dignity / vocation
• Gigantism -> Fit to human scale
• Overengineering -> Elegance
• Centralization -> Local control
• Commodification -> Specificity of context
• Planned obsolescence -> Robustness and ease of repair

Appropriate Technology

Aesthetic Signals of System Fit (or not!)

• Human judgment elevated | Human judgment displaced
• Contextual by design | Universal by default
• Explicit tradeoffs | Hidden tradeoffs
• Bounded scope | Indiscriminate scope
• Legible failure modes | Opaque failure modes
• Local control and repair | Centralized control
• Adequate capability | Maximized capability
• Territory-first reasoning | Map-first reasoning

Appropriate Computing

How to apply these ideas to AI and (more broadly) algorithms? Some ideas.

• Understand how model architecture drives efficiency and failure modes.
• Find examples of promising areas for application. One example is “cognitive bottlenecks”— area of high need for attention in complex systems where “almost” is good enough. Many examples in medicine, education, environmental systems, and agriculture.
• Elevate human judgement through rigorous system modeling.
• Study examples of inappropriate computing (e.g. Post scandal, Robodebt)
• High stakes applications should be implemented with clear assumptions, rigorous understanding of failure modes, and clarity on usefulness.
• Local models should be used whenever possible to protect sovereignty and control.
• Understand the distinction between an object and its digital representation.

Appropriate Computing

Pedagogical Cores

• Mathematical foundations of AI models
• AI systems engineering
• Inappropriate computing (failure case studies)
• Human and cultural systems thinking
• Resilience in disruption (Indigenous perspectives on survivance)
• Real world, locally relevant projects

Appropriate Computing

Current research projects

• Efficiency upgrades in LLM architecture
• Curriculum training for alignment and efficiency in LLMs
• Multimodal California forest foundation model
• Acoustic waste contaminant detection
• AI powered medical screening/review of systems
• Compression based Proof of Useful Work

Appropriate Computing

Action!

• Discussion group has been meeting weekly for the past year or so on this theme: Fridays 12-1 in SBS 405
• Rosanna and Peter Overholser were awarded a SHIFT grant to help build local teaching and research capacity around this theme.
• Student market research and community outreach: See posting on Handshake Jobs
• Campus visits this spring by high-profile experts (next slides)
• Interested in the development of a curriculum? Join our team:

Appropriate Computing

Campus Visitors, Spring 2026

• Mark van der Laan
• Director, Center for Targeted Learning at UC Berkeley
• Expert on causal inference. His targeted learning framework combines modern statistical methods with the black box flexibility of modern ML/AI methods and is being adopted by statisticians at the FDA
• Causal analysis is exactly the ingredient needed for rigorous decision support, elevating agency (helping remove bias toward inaction).

Appropriate Computing

Campus Visitors, Spring 2026

• Tajana Šimunić Rosing
• Expert on efficient, robust deployment of computational solutions, integrating hardware and software.
• Co-director of the PRISM Center at UC San Diego
• We have free access to the same computing network!

Appropriate Computing

Campus Visitors, Spring 2026

• Jaron Lanier
• Influential commentator on humane computing, and consistent, strong critic of AI’s totalizing vision
• “Father of virtual reality”

Appropriate Computing

Campus Visitors, Spring 2026

• Advisor on “Resilience in Disruption”
• Invitation in progress

Appropriate Computing

Thank you!