Amanda
@amandagci.bsky.social
Sr. Program Manager at the SFPE Foundation https://www.sfpe.org/foundation/home
Engineering a fire safe world for all through research and education
Engineering a fire safe world for all through research and education
"Just because we can, does that mean we should?" As a society we don't have an answer to that, but we seem to go for more data even when it's not necessary. Does it make sense? We will one day run out of finite resources.
May 30, 2025 at 1:30 AM
"Just because we can, does that mean we should?" As a society we don't have an answer to that, but we seem to go for more data even when it's not necessary. Does it make sense? We will one day run out of finite resources.
Environmental considerations. AI is not currently environmentally sustainable, which outweighs the benefits of its use. Yes, Gen AI is overkill in some situations. Need to educate the gen population on using the right tool for the job.
May 30, 2025 at 1:30 AM
Environmental considerations. AI is not currently environmentally sustainable, which outweighs the benefits of its use. Yes, Gen AI is overkill in some situations. Need to educate the gen population on using the right tool for the job.
Trust adapts to the context. If an AI decision kills people, hopefully we would stop using that AI, just like an engineer would get fired. If the error is making a mistake in a recipe, the consequences aren't so serious.
May 30, 2025 at 1:30 AM
Trust adapts to the context. If an AI decision kills people, hopefully we would stop using that AI, just like an engineer would get fired. If the error is making a mistake in a recipe, the consequences aren't so serious.
Explainability is key to trust. e.g. knowing where the training data came from.
May 30, 2025 at 1:30 AM
Explainability is key to trust. e.g. knowing where the training data came from.
Xinyan: AI can't be legally responsible for anything. Maybe one day it will reach that level. Trust now is often about AI as an assistant, not the decision maker with signing authority.
May 30, 2025 at 1:30 AM
Xinyan: AI can't be legally responsible for anything. Maybe one day it will reach that level. Trust now is often about AI as an assistant, not the decision maker with signing authority.
Robin: Trust is dependent on our tolerance for error.
May 30, 2025 at 1:30 AM
Robin: Trust is dependent on our tolerance for error.
Kevin: The fire service is extremely slow to change, but even they are adopting new tech. Trust is evolving faster than we think.
May 30, 2025 at 1:30 AM
Kevin: The fire service is extremely slow to change, but even they are adopting new tech. Trust is evolving faster than we think.
Ben: Psychology is a large aspect. You may not use a tool made by someone you don't trust. Likewise, people can be swayed by marketing and advertising.
May 30, 2025 at 1:30 AM
Ben: Psychology is a large aspect. You may not use a tool made by someone you don't trust. Likewise, people can be swayed by marketing and advertising.
Robin: We may never understand the intricacies of how AI works (black box), but we need to be able to trust it or at least trust the people behind it. We went through this with CFD models as well. People were skeptical. Now they're the standard to check AI against.
May 30, 2025 at 1:30 AM
Robin: We may never understand the intricacies of how AI works (black box), but we need to be able to trust it or at least trust the people behind it. We went through this with CFD models as well. People were skeptical. Now they're the standard to check AI against.
Ben: When it comes to tech, we shouldn't do nothing, but we also shouldn't be crazy foolhardy in adopting it. Aim for the middle.
May 30, 2025 at 1:30 AM
Ben: When it comes to tech, we shouldn't do nothing, but we also shouldn't be crazy foolhardy in adopting it. Aim for the middle.
Ben: SaaS has to meet state and federal security requirements to be considered for purchase. Policy and controls can help with trust.
May 30, 2025 at 1:30 AM
Ben: SaaS has to meet state and federal security requirements to be considered for purchase. Policy and controls can help with trust.
Xinyan: The tools have to be useful before even considering trust.
May 30, 2025 at 1:30 AM
Xinyan: The tools have to be useful before even considering trust.
Kevin, who is a retired fire chief/incident commander, says that he is excited about AI, but we need to figure out how to get it to "street level" to win over the fire service.
May 30, 2025 at 1:30 AM
Kevin, who is a retired fire chief/incident commander, says that he is excited about AI, but we need to figure out how to get it to "street level" to win over the fire service.
Version 4.0 used a recurrent neural network. Used SwinLSTM. Prediction has limitations. AI will forget if prediction time is too long.
Not bad for a pilot study, though! He and colleagues are continuing to work on the model. Need more data and input from firefighters on AI needs.
Not bad for a pilot study, though! He and colleagues are continuing to work on the model. Need more data and input from firefighters on AI needs.
May 30, 2025 at 12:28 AM
Version 4.0 used a recurrent neural network. Used SwinLSTM. Prediction has limitations. AI will forget if prediction time is too long.
Not bad for a pilot study, though! He and colleagues are continuing to work on the model. Need more data and input from firefighters on AI needs.
Not bad for a pilot study, though! He and colleagues are continuing to work on the model. Need more data and input from firefighters on AI needs.
Collected more data with newer equipment to develop version 3.0. Simpler FlashoverNet detected the flashover 31 seconds in advance with 94% accuracy on test data. Fewer false alarms. Still need more data for less noise in predictions.
May 30, 2025 at 12:28 AM
Collected more data with newer equipment to develop version 3.0. Simpler FlashoverNet detected the flashover 31 seconds in advance with 94% accuracy on test data. Fewer false alarms. Still need more data for less noise in predictions.
For their flashover prediction tool, thermal data was not enough. Version 2.0, used generative AI to convert RGB in the videos to thermal infrared data. 2.0 can predict flashover earlier, but is very complicated.
May 30, 2025 at 12:28 AM
For their flashover prediction tool, thermal data was not enough. Version 2.0, used generative AI to convert RGB in the videos to thermal infrared data. 2.0 can predict flashover earlier, but is very complicated.
NRC is working on a ton of projects in their fire lab like fire forensics data AI processing, remote sensing data collection, digitalized evacuation simulation, and digitalized fire dynamic simulation using BIM models.
May 30, 2025 at 12:28 AM
NRC is working on a ton of projects in their fire lab like fire forensics data AI processing, remote sensing data collection, digitalized evacuation simulation, and digitalized fire dynamic simulation using BIM models.
They developed a LLM Agent called FireSeek to interact with the fire service, and are working on intelligent and autonomous robots to aid in firefighting operations. Very cool tech with a ton of potential!
May 30, 2025 at 12:09 AM
They developed a LLM Agent called FireSeek to interact with the fire service, and are working on intelligent and autonomous robots to aid in firefighting operations. Very cool tech with a ton of potential!
For fire research, Huang's lab has used AI to predict heat release rate from videos of fires. Videos are from the NIST database. (Sidenote, the project was partially funded by an SFPE Foundation student research grant!)
May 30, 2025 at 12:09 AM
For fire research, Huang's lab has used AI to predict heat release rate from videos of fires. Videos are from the NIST database. (Sidenote, the project was partially funded by an SFPE Foundation student research grant!)
Have used AI for RSET/ASET, Virtual Reality Systems, and more! All of the AI tools are available as IFETool at ifetool.firelabxy.com
IFETool v1.1
ifetool.firelabxy.com
May 30, 2025 at 12:09 AM
Have used AI for RSET/ASET, Virtual Reality Systems, and more! All of the AI tools are available as IFETool at ifetool.firelabxy.com
AI Smart Firefighting is actually a combination of #FireEngineering, Fire Research, and Firefighting. Sought to train AI model to aid in fire safety design.
May 30, 2025 at 12:09 AM
AI Smart Firefighting is actually a combination of #FireEngineering, Fire Research, and Firefighting. Sought to train AI model to aid in fire safety design.