An AI design tool helped teams spot more possible harms early

A web-based exercise gave technical and non-technical workers shared ways to examine risky design choices before an AI system was built.

Editorial illustration — not from the study.

Responsible AI reviews can be difficult when technical and non-technical workers do not share a common way to discuss design choices. Interviews with eight practitioners found that technical plans were often handed off poorly, while existing tools such as JIRA and Google Docs did not provide enough structure for joint harm identification.

The researchers developed AI LEGO, a web-based prototype that lets technical workers create development plans with interactive blocks. Other team members review those blocks with stage-specific checklists and simulated personas generated by a large language model, helping them consider who might be harmed by a design choice.

What the teams identified

In a study with 18 cross-functional practitioners, AI LEGO increased both the number of potential harms identified and the likelihood that participants would identify them, compared with baseline worksheets. Participants said the tool’s modular structure and persona prompts made harm identification more accessible and supported clearer, more collaborative responsible AI practices during early design. The researchers’ interviews with eight practitioners also identified three obstacles: technical choices were rarely handed off in ways that supported meaningful input from non-technical roles; teams lacked shared visual structures for collaboration; and non-technical practitioners lacked scaffolds for systematic harm evaluation.

Why early review matters

Potential harms can be harder to address once an AI system is already being built. By giving different roles a shared structure for discussing technical plans and considering possible harms, AI LEGO addresses a coordination problem that the researchers found in cross-functional responsible AI work. The study suggests that design tools built for joint review may help teams examine these concerns earlier than ordinary product-tracking or document tools do.

Evidence and caveats

The findings come from a literature review, semi-structured interviews with eight practitioners, and a study involving 18 cross-functional practitioners who used the prototype and baseline worksheets. The abstract does not report the size of the difference between the two approaches, nor does it specify the participants’ industries, tasks or how long the effects lasted. AI LEGO is described as a web-based prototype, so the findings do not establish how well it would work across organizations or in long-term development practice.

In Saudi Arabia, AI-guided building controls cut electricity use in desert heat

An adaptive AI control system cut electricity use by 37.8% compared with conventional thermostats in eight occupied buildings in Hail, Saudi Arabia. It kept indoor conditions within the study’s comfort range for most occupied hours and increased savings during dust storms.

1 day ago·7 views

Engineering & Technology

Ammonia could cut coal-power emissions, but nitrogen oxides remain a hurdle

A review of laboratory, industrial and computer studies finds that co-firing 20%–30% ammonia with coal is feasible in existing coal-fired boilers. Higher ammonia shares may raise nitrogen-oxide emissions and lower boiler efficiency, making fuel-injection design and air staging important.

3 days ago·13 views

Engineering & Technology

Using satellite-based altitude in cruise could burn more fuel, study finds

A simulation study found that using geometric altitude instead of barometric altitude during cruise flight would, on average, increase fuel use. It also found a considerable risk of exceeding aircrafts’ maximum barometric altitude, potentially requiring more step descents.

4 days ago·18 views