Thread by @sh_reya

every morning i wake up with more and more conviction that applied machine learning is turning into enterprise saas. i’m not sure if this is what we want (1/9)

why do i say saas? every ML company is becoming a dashboard and API company, regardless of whether the customer asked for a dashboard or not. there’s this unspoken need to “have a product” that isn’t a serialized list of model weights & mechanisms to trust model outputs (2/9)

why is saas not perfectly analogous? “correctness” at the global scale is not binary for ML, but it is for software. i get the need to package ML into something that sells, but i’m not sure why it needs to replicate the trajectory of enterprise saas (3/9)

what’s the problem? in enterprise saas, providers are responsible for security issues, but customers face consequences of decisions made using the tool. branding ML as enterprise saas silences conversations around the need to hold ML providers accountable for model outputs (4/9)

if big co. ABC sells a ML dashboard to the local courts of Waxahachie, TX: who’s “fault” will it be when people of color are disproportionately sentenced for crimes (5/9)

the problem is pervasive, and many people are a part of it. VCs who write blank $20M+ seed checks to do an “AI startup.” engineers who are complacent building ML tools for mass surveillance companies (6/9)

policymakers who diss US & champion GDPR without noting that the EU is building a monopoly on Europeans’ private info. researchers who think their work is impactful because people forked their GitHub repo, when impact really begins when people stop using your work as a toy (7/9)

managers that expect ML work to be articulated perfectly in Jira tickets with binary “success or failure” criteria. data scientists who refuse to write non-Pandas code or take zero interest in productionizing the model (8/9)

we are building a dangerous culture around ML, where everyone is trying to expedite progress while playing the “blame game.” maybe this works in enterprise saas, but ML outputs have more uncertainty, and the consequences of something “going wrong” are more high-stakes. (9/9)