FatCrab

AI in health and medtech has been around and in the field for ages. However, two persistent challenges make roll out slow-- and they're not going anywhere because of the stakes at hand.

The first is just straight regulatory. Regulators don't have a very good or very consistent working framework to apply to to these technologies, but that's in part due to how vast the field is in terms of application. The second is somewhat related to the first but really is also very market driven, and that is the issue of explainability of outputs. Regulators generally want it of course, but also customers (i.e., doctors) don't just want predictions/detections, but want and need to understand why a model "thinks" what it does. Doing that in a way that does not itself require significant training in the data and computer science underlying the particular model and architecture is often pretty damned hard.

I think it's an enormous oversimplification to say modern AI is just "fancy signal processing" unless all inference, including that done by humans, is also just signal processing. Modern AI applies rules it is given, explicitly or by virtue of complex pattern identification, to inputs to produce outputs according to those "given" rules. Now, what no current AI can really do is synthesize new rules uncoupled from the act of pattern matching. Effectively, a priori reasoning is still out of scope for the most part, but the reality is that that simply is not necessary for an enormous portion of the value proposition of "AI" to be realized.

Summary judgement is not a thing separate from a lawsuit. It's literally a standard filling made in nearly every lawsuit (even if just as a hail mary). You referenced "beyond a reasonable doubt" earlier. This is also not the standard used in (US) civil cases--it's typically a standard consisting of the preponderance of the evidence.

I'm also not sure what you mean by "court approved documentation." Different jurisdictions approach contract law differently, but courts don't "approve" most contracts--parties allege there was a binding and contractual agreement, present their evidence to the court, and a mix of judge and jury determines whether under the jurisdictions laws and enforceable agreement occurred and how it can be enforced (i.e., are the obligations severable, what damages, etc.).

There's plenty you could do if no label was produced with a sufficiently high confidence. These are continuous systems, so the idea of "rerunning" the model isn't that crazy, but you could pair that with an automatic decrease in speed to generate more frames, stop the whole vehicle (safely of course), divert path, and I'm sure plenty more an actual domain and subject matter expert might come up with--or a whole team of them. But while we're on the topic, it's not really right to even label these confidence intervals as such--they're just output weighting associated with respective levels. We've sort of decided they vaguely match up to something kind of sort approximate to confidence values but they aren't based on a ground truth like I'm understanding your comment to imply--they entirely derive out of the trained model weights and their confluence. Don't really have anywhere to go with that thought beyond the observation itself.

Are you under the impression that I think Teslas approach to AI and computer vision is anything but fucking dumb? The person said a stupid and patently incorrect thing. I corrected them. Confidence values being literally baked into how most ML architectures work is unrelated to intentionally depriving your system of one of the most robust ccomputer vision signals we can come up with right now.

All probabilistic models output a confidence value, and it's very common and basic practice to gate downstream processes around that value. This person just doesn't know what they're talking about. Though, that puts them on about the same footing as Elono when it comes to AI/ML.

I've worked on processing submissions for this project. Honestly, it probably ends up just costing them more to do this program, which is mostly just a paid PR activity. The overwhelming majority of submissions, and I mean like 99%, are either not prior art in the sense of patent law or were already retrieved by the law firm on the case.

My z flip is hands down my favorite phone I've ever owned and I didn't get it expecting to like it much. I just needed a new phone and with Samsung's recycling program, my old near-tablet sized phone made the switch like barely 100 bucks.

There are a lot of small advantages it provides that quickly add up to it being an overall superior experience. Now if only Bixby wasn't the worst fucking thing ever.

Their non-profit status had nothing to do with the legality of their training data acquisition methods. Some of it was still legal and some of it was still illegal (torrenting a bunch of books off a piracy site).

Oh wow, this suit is shaping up to be silly. I didn't realize it was filed in Japan, too. That makes the patent aspect even shakier. Japan has no discovery process like in the US, which is generally very necessary for many software-related patents as, assuming they have a strong likelihood of surviving challenge, they are typically drawn to processes that are completely obfuscated from the user and outside observes.

There is an era of patents from the late 90s through the early-mid-00s that were insanely vague and rarely stand up to scrutiny, but most are expiring at this point, if they haven't already. Generally, though, patents are not granted on "concepts" but on implementations. That's a sometimes ambiguous line, but that's a fundamental principle of modern patents.

If it's patented, you can just read the patent to know what else is in it.

My point is just that they're effectively describing a discriminator. Like, yeah, it entails a lot more tough problems to be tackled than that sentence makes it seem, but it's a known and very active area of ML. Sure, there may be other metadata and contextual features to discriminate upon, but eventually those heuristics will inevitably be closed up and we'll just end up with a giant distributed, quasi-federated GAN. Which, setting aside the externalities that I'm skeptical anyone in a position of power to address is equally in an informed position of understanding, is kind of neat in a vacuum.