Platystemon posted:

We really can’t.

The U.S. had one point one fatalities per hundred million vehicle miles traveled in 2019. Computer systems cannot be said to be safer than human drivers till they’ve driven at least that far without a single fatality.

D-Pad posted:

https://interestingengineering.com/how-safe-are-self-driving-cars

I don't actually entirely agree with that analysis for the reasons outlined in the 2nd article below which is why I said I don't believe they are ready yet:

https://www.scientificamerican.com/article/are-autonomous-cars-really-safer-than-human-drivers/

This article is a great analysis of how we should approach the problem and make a decision as well as good statistics on human drivers. There is a lot to consider that your average American will never do:

https://www.forbes.com/sites/lancee...sh=7bec605546ed


So like I said I don't think that Tesla has any business releasing this into the wild at this time, but I do feel like for me personally, I will be ok with self driving cars on the road in the future sooner than the average American will. The first time one of them runs over a kid the technology will forever be evil in the eyes of the general public even if studies were to conclusively show that they drastically reduce child pedestrian deaths overall.

The psychology behind that is why I'm not sure if we ever will get full self driving cars because regulators are going to come down hard on them due to outrage from the public most likely in the next couple of years. Especially because a lot of these originate from silicon valley and the public and government is already starting to turn on those companies for lots of other reasons.

Kith posted:

https://i.imgur.com/kx4hz3o.mp4

GotLag posted:

What scale is that?

KoRMaK posted:

https://i.imgur.com/lZQnNWU.mp4
all that exposed skin all i can think about is one thing

BaldDwarfOnPCP posted:

*orders a polonium 210 enema*

Yeah I'm finally gonna poo poo right.

GotLag posted:

That sounds more like multiple cases of cops who don't feel like investigating murder

Kith posted:

https://i.imgur.com/kx4hz3o.mp4

Jabor posted:

one curious fact, that you might or might not find interesting, is that the people most cynical about it are the people that actually know how it (and the industry that produced it) works.

while the people most in favour of it are people that see it as a magical black box that solves their problems.

just something to think about.

Jabor posted:

one curious fact, that you might or might not find interesting, is that the people most cynical about it are the people that actually know how it (and the industry that produced it) works.

while the people most in favour of it are people that see it as a magical black box that solves their problems.

just something to think about.

Jabor posted:

one curious fact, that you might or might not find interesting, is that the people most cynical about it are the people that actually know how it (and the industry that produced it) works.

while the people most in favour of it are people that see it as a magical black box that solves their problems.

just something to think about.

CarForumPoster posted:

Im a mech E undergrad, MS Systems engineering and have developed/deployed a web app that can take a photo of any US coin since the 1860s and classify what it is (eg barber dime) with 97% accuracy using a DL model I gathered the data for and trained using pytorch. I’d say I have a better understanding than most.

CarForumPoster posted:

I find your implication that I’m dumb and don’t understand pretty goon-

Im a mech E undergrad, MS Systems engineering and have developed/deployed a web app that can take a photo of any US coin since the 1860s and classify what it is (eg barber dime) with 97% accuracy using a DL model I gathered the data for and trained using pytorch. I’d say I have a better understanding than most.

CarForumPoster posted:

I find your implication that I’m dumb and don’t understand pretty goon-

Im a mech E undergrad, MS Systems engineering and have developed/deployed a web app that can take a photo of any US coin since the 1860s and classify what it is (eg barber dime) with 97% accuracy using a DL model I gathered the data for and trained using pytorch. I’d say I have a better understanding than most.

CarForumPoster posted:

I find your implication that I’m dumb and don’t understand pretty goon-

Im a mech E undergrad, MS Systems engineering and have developed/deployed a web app that can take a photo of any US coin since the 1860s and classify what it is (eg barber dime) with 97% accuracy using a DL model I gathered the data for and trained using pytorch. I’d say I have a better understanding than most.

quote:

Literal-mindedness creates the risk that a system can’t tell if its model of the world is the world it is actually in (Wiener, 1950). As a result, the system will do the right thing [in the sense that the actions are appropriate given its model of the world], when it is in a different world [producing quite unintended and potentially harmful effects]. This pattern underlies all of the coordination breakdowns between people and automation.
[…]
The context gap is the need to test whether the situation assumed in the model underlying the algorithm to be deployed matches the actual situation in the world. Monitoring this gap is fundamental to avoiding the error of the third kind (solving the wrong problem), and to the demand for revision and re-framing in JCSs [Joint Cognitive Systems].
[…]
When a field of practice is about to experience an expanding role for automata, we can predict quite confidently that practitioners and organizations will adapt to develop means to align, monitor, and repair the context gap between the automata’s model of the world and the world. Limits to their ability to carry out these functions will mark potential paths to failure—breakdowns in resilience. In addition, we note that developers’ beliefs about the relationship of people and automation in complex and high consequence systems (substitution myth and other over-simplifications) lead designers to miss the need to provide this support and even to deny that such a role exists (but see Roth et al., 1987, Smith et al., 1997 and other studies of the brittleness of automata).

Now people, as a cognitive system, also are vulnerable to being trapped in literal-mindedness, where they correctly deploy a routine given their model of the world, but are in fact facing a different situation (Weick et al., 1999). When anomaly response breaks down, it is often associated with an inability to revise plans and assessments as new evidence arrives and as situations change.
[…]
But in observing work, we usually see practitioners probing and testing whether the routine, plan or assessment fits the actual situation they are facing (e.g., how mission control, as a JCS, works). This points to a fundamental difference between people and automata—people have the capability to repair the context gap for themselves (to some degree), but more powerfully, for and through others as part of a process of cross-checking and broadening checks across different perspectives.
[…]
    Norbert’s Contrast
   Artificial agents are literal minded and disconnected from the world while human agents are context sensitive and have a stake in outcomes.

The key is that people and computer automata start from opposite points—the former as context-bound agent and the latter as literal-minded agent—and tend to fall back or default to those points without the continued investment of effort and energy from outside.
Automata start: literal --> developers exert effort and inventiveness to move these computer systems to be more --> adaptive, situated, contextualized --> but there are always limits in this process requiring human mediation to maintain or repair the link between model and actual situation (the fundamental potential for surprise that drives the need for resilience).
On the other hand, people start contextualized --> developers exert effort and inventiveness to move these human systems toward --> more abstract, more encompassing models for effective analysis/action and away from local, narrow, surface models.

Norbert Weiner’s (1950) original analysis revealed that literal-minded agents are paradoxically unpredictable because they are not sensitive to context. The now classic clumsy automation quotes from studies of cockpit automation—“What’s it doing now? Why is it doing that? What will it do next?“—arise from the gap between the human flight crew as context sensitive agents grounded in the world, and the literal mindedness of the computer systems that fly the aircraft.

The computer starts from and defaults back to the position of a literal-minded agent. Being literal-minded, a computer can’t tell if its model of the world is the world it is in. This is a by-product of the limits of any model to capture the full range of factors and variations in the world. A model or representation, as an abstraction, corresponds to the referent processes in the world only in some ways. Good models capture the essential and leave out the irrelevant; the catch is that knowing what is essential and irrelevant depends on the goal and task context.
  As Ackoff (1979, p. 97) put it,
The optimal solution of a model is not an optimal solution of a problem unless the model is a perfect representation of the problem,  which it never is.

This “catch” on models captures the need to re-visit the connection between the model being deployed in the form of algorithms, plans, procedures, routines or skills and the actual conditions being faced, given change and the changing flow of evidence. This is the context gap that needs to be closed, as we saw in previous discussions of revising assessments, reframing conceptualizations, modifying plans in progress.
[…]
It is up to people situated in context to ground computer processing in the world, given that particular situations can arise to challenge the boundary conditions of the model behind the algorithm (the potential for surprise). For people in context there is an open future, while literal-minded agents are stuck within the walls of the model underlying their operation. […]

Closing the context gap is about knowing and testing what “rules” apply in what kind of situation. The key is to determine the kind of situation faced, and to recognize how situations can change. People can be sensitive to cues that enable them to switch rules or routines as they test whether the situation is different from what was originally construed. And despite not always performing well at this function (getting stuck in one view; pp. 76-77 and 104-105), people provide the only model of competence at re-framing or re-conceptualizing that we can study for clues about what contributes to expert performance and how to support it.
[…]
With advances (developer effort), computer agents can be made more situated in the sense of taking more factors into account in their own processing. These automata still function well only within artificially bounded arenas (although the size and position of these arenas grow and shift as people learn to produce new capabilities). Literal-minded automata are always limited by the brittleness problem—in other words, however capable, developer effort stretches automata away from their literal-minded origin, but without chronic effort to ground these systems in context, the risk of literal-mindedness re-emerges.

This work to close the context gap tries to keep the environment where the automata are placed in alignment with the assumptions underlying them. However, the effort to ground literal-minded automata in context often occurs in the background, hidden. Nevertheless, it takes continued effort to maintain this match of model-environment in the face of the eroding forces of variability and change inherent in our physical world (the potential for surprise inherent in the Law of Requisite Variety). And, as the envelope of competence of automata shifts, the resulting new capability will afford adaptation by stakeholders in the pursuit of goals. This creates new contexts of use that may challenge the boundary conditions of the underlying algorithms (as captured in the Law of Stretched Systems, p. 18, and as has been experienced in software failures that arose from use migration and requirements change).

Thus, Norbert’s Contrast specifies two complementary vectors:
(1) The capabilities of automata grow as (some) people learn to create and instantiate new algorithms, plans and routines;
(2) In parallel, people at the sharp end adapt to the brittleness of literal-minded agencies to monitor, align and repair the context gap omnipresent in a changing uncertain world (and the change results, in part, from how people adapt to exploit the new capabilities.)

The contrast captures a basic tradeoff that must be balanced in the design of any JCS because of the basic constraint of bounded rationality of finite resources and irreducible uncertainty (p. 2). Literal-minded and context-bound agents represent different responses to the tradeoff, each with different vulnerabilities. Norbert’s Contrast points out that literal, disconnected agents can be powerful help for context bound agents, but the literal-minded agents in turn need help to be grounded in context (Clancey, 1997). Coordinating or complementing different kinds of cognitive agents (literal-minded and context-bound) in a joint cognitive system is the best strategy for handling the multiple demands for performance in a finite-resource, uncertain, and changing universe.

Jabor posted:

Imagine a car with a 97% chance of not driving you straight into oncoming traffic each time you took it out on the road.

Hexyflexy posted:

*slams down her mathematician card* I've worked on and with the maths behind modern pattern recognition systems (and sold some!), if you think a set of dumb rear end classifiers with some fuzzy logic on top can provide human like control impulses to a car without a bit of the old motor-slaughtering going on I don't know what to tell you. The problem is inherently non-computational, god knows why we're investing so much money in this poo poo as a species.

quote:

They’re familiar with the tech industry. That’s why they’re worried about what the self-driving revolution will entail.

quote:

This paper discusses the ways in which automation of industrial processes may expand rather than eliminate problems with the human operator. Some comments will be made on methods of alleviating these problems within the "classic' approach of leaving the operator with responsibility for abnormal conditions, and on the potential for continued use of the human operator for on-line decision-making within human-computer collaboration.

ultrafilter posted:

This paper discusses the ways in which automation of industrial processes may expand rather than eliminate problems with the human operator. Some comments will be made on methods of alleviating these problems within the "classic' approach of leaving the operator with responsibility for abnormal conditions, and on the potential for continued use of the human operator for on-line decision-making within human-computer collaboration.

ncumbered_by_idgits posted:

I read this as " the human operator will do everything is his/her power to gently caress the thing up."

ncumbered_by_idgits posted:

I read this as " the human operator will do everything is his/her power to gently caress the thing up."

MononcQc posted:

The problem is one where we will collectively rely on people continuously breaking rules to make things functional, but when a failure happens you end up blaming the human for doing that very same thing.

KoRMaK posted:

D-Pad posted:

I'd rather a Tesla drive itself over some human drivers I know getting behind it's wheel.

CarForumPoster posted:

I find your implication that I’m dumb and don’t understand pretty goon-

Im a mech E undergrad, MS Systems engineering and have developed/deployed a web app that can take a photo of any US coin since the 1860s and classify what it is (eg barber dime) with 97% accuracy using a DL model I gathered the data for and trained using pytorch. I’d say I have a better understanding than most.