As someone who’s worked on similar stuff, I just wanted to pitch in with two thoughts (hopefully helpful!):
On extrusion modeling, I (& others) looked into something similar about 4-5 years ago and I personally I came away quite pessimistic for 2 reasons.i) The search space is practically a lot narrower than it seems (very limited by ingredients accessible and “suppliable” at scale).ii) Furthermore, an extrusion expert I spoke to thought that more of the variability probably came from ingredients rather than actual extrusion parameters so a model alone seemed unlikely to create “big” improvements to the texture / product quality. Somewhat confirming this: I spoke to a senior researcher at a well known plant-based company + and an ex-one from a mainstream CPG company and they both confirmed that they’d tried something like this and came away from it without any real product improvements.
On the second idea, I have less to add but it may be worth checking out AMII’s recent collab with New Harvest. More general hot take though: My (the low confidence) general impression has been that solving “big” fermentation / cultivated problems with an AI-ish approach has been somewhat bottlenecked by the ability to get good data and iterate (in other words, the “sensors” aren’t great and limited). In-silico modeling is seemingly pretty far away too for most cell lines/platforms that I’ve heard folks talk about in recent years but unsure where that will go.
Other interesting things from an ML/AI perspective may be looking at areas like ingredient functionalization & processing for ingredient companies (ADM, Roquette etc.). The idea here would be to tweak processes to get better functionality or lower prices. There was some interest in this work 4-5 years ago but I haven’t followed it much since.
Anyway, hope some of that is useful and that you find an interesting problem & funding!
Thank you for this detailed reply! These are some very relevant facts. I know of some of the problems you mentioned, but the fact that others have tried and fail is new to me—I haven’t been able to find anyone who attempted the same thing up until now.
I would love to talk to you about it f2f and get some more detail, if that’s ok. Would you be willing to have a short call?
Hey Noa,
Thanks for writing this up!
As someone who’s worked on similar stuff, I just wanted to pitch in with two thoughts (hopefully helpful!):
On extrusion modeling, I (& others) looked into something similar about 4-5 years ago and I personally I came away quite pessimistic for 2 reasons.i) The search space is practically a lot narrower than it seems (very limited by ingredients accessible and “suppliable” at scale).ii) Furthermore, an extrusion expert I spoke to thought that more of the variability probably came from ingredients rather than actual extrusion parameters so a model alone seemed unlikely to create “big” improvements to the texture / product quality.
Somewhat confirming this: I spoke to a senior researcher at a well known plant-based company + and an ex-one from a mainstream CPG company and they both confirmed that they’d tried something like this and came away from it without any real product improvements.
On the second idea, I have less to add but it may be worth checking out AMII’s recent collab with New Harvest. More general hot take though: My (the low confidence) general impression has been that solving “big” fermentation / cultivated problems with an AI-ish approach has been somewhat bottlenecked by the ability to get good data and iterate (in other words, the “sensors” aren’t great and limited). In-silico modeling is seemingly pretty far away too for most cell lines/platforms that I’ve heard folks talk about in recent years but unsure where that will go.
Other interesting things from an ML/AI perspective may be looking at areas like ingredient functionalization & processing for ingredient companies (ADM, Roquette etc.). The idea here would be to tweak processes to get better functionality or lower prices. There was some interest in this work 4-5 years ago but I haven’t followed it much since.
Anyway, hope some of that is useful and that you find an interesting problem & funding!
Hi,
Thank you for this detailed reply! These are some very relevant facts. I know of some of the problems you mentioned, but the fact that others have tried and fail is new to me—I haven’t been able to find anyone who attempted the same thing up until now.
I would love to talk to you about it f2f and get some more detail, if that’s ok. Would you be willing to have a short call?