General Automation Thread

[Exonerate]

Supposedly a case of the pedestrian stepping out before the car could react, not something like the car glitching out or going berserk at a crossing. I'd be interested in the rate of fatal accidents per mile for the self-driving cars compared to human drivers - Google's are supposed to be fantastically safe and haven't been in any accidents in which they were at fault.

Interesting problems of culpability though. I'm sure the companies would love to offload responsibility to the human drivers who are supposed to constantly stay alert and be ready to step in at a moment's notice, but past a certain level of proliferation and sophistication, I don't think that's going to be a realistic expectation. The moral calculus of collision avoidance raises some issues too - I've seen some articles on whether the car should swerve and kill the passenger instead of running down multiple people, but none of them seem to have recognized the potential for moral hazard here.

[GuppyShark]

http://www.abc.net.au/news/2018-03-22/s ... ed/9575624

Looks like the self-driving tech failed to detect the pedestrian as they walked with their bicycle slowly across a clear road. It's not a 'someone jumped out in front of traffic' scenario.

[bilateralrope]

Three things:
- Unless that video is significantly darker than reality, I wouldn't have seen that pedestrian in time. At the very least, it shows the pedestrian crossing the darkest part of the road.
- Self-driving cars should not have their sensors limited to the visible light spectrum. What other sensors did this car have ?
- Waymo's trade secret allegations include lidar designs.

[GuppyShark]

1) Yes, our eyes are better adapted to changing light conditions than most video cameras. It doesn't look like this one was very expensive.
2) LIDAR is what I'm reading.
3) Yes, that's the one.

[bilateralrope]

I see three possibilities regarding any LIDAR in the car:
- They are still using Waymo's LIDAR. Despite agreeing not to do so as part of the settlement that happened early February.
- They are using a LIDAR design that they only started working on after agreeing to stop using the stolen Waymo design. Meaning less than two months of work on their LIDAR system.
- They didn't have any LIDAR system.

[Stormin]

Pic of the car, it looks like it has almost full roof top scanner set, possibly others elsewhere but I'm not bothering with pixel hunting.
https://twitter.com/NTSB_Newsroom/statu ... 6323194880

The car straight up failed to detect the person crossing the road. If that had been a rock that fell off a truck it might have hit that too.

[Exonerate]

Yeah, the guy in the car is clearly paying attention only intermittently. Paid tester on a self-driving car only in a developmental stage and people are already slipping. Still, I don't know if a human who was paying attention would've been able to react in time.

Uber's self-driving car performance is apparently shit. It's behind a paywall, but I found an excerpt:

In one such ride-along Tuesday, a driver took control of the vehicle more than a dozen times in less than 30 minutes. His reasons included: He was worried that the car would get too close to a pedestrian, that the vehicle wouldn’t let another merge and that the car would potentially create gridlock by entering an already crowded intersection. Other reasons were more mysterious. Sometimes the car would simply hand over control to the driver with little explanation. The driver said that the car was probably getting its sensors overloaded. In this one short jaunt around downtown San Francisco, when another car honked at the self-driving car for trying to change lanes, the Uber driver took control of the vehicle.

Google's, in contrast, only needs a driver intervention every like few thousand miles.

[FaxModem1]

Los Angeles Times

Skip to content

Los Angeles Times Logo
SUBSCRIBE
AUTOS
BUSINESS
LA TIMES
Musk has second thoughts on aggressive automation for Tesla Model 3
By RUSS MITCHELL
APR 17, 2018 | 5:05 PM

Musk has second thoughts on aggressive automation for Tesla Model 3
Tesla should be able to sustain producing 2,000 Model 3 sedans a week, CEO Elon Musk said a week ago. (David Zalubowski / Associated Press)

In early 2017, Elon Musk told stock analysts that Tesla Inc.'s goal "is to be the best manufacturer on Earth." He'd get there by inventing a factory so dense with robots and devoid of human beings it would resemble an "alien dreadnaught" video game warship.

Speed, he's said, "is the ultimate weapon when it comes to innovation or production."


ADVERTISEMENT
Instead, Tesla this week was forced to bring the Fremont, Calif., production line of its crucial mass-market Model 3 electric sedan to a complete halt, according to the company. It is at least the second shutdown in the last two months. And in an internal email, Musk said Tesla will add workers and go to round-the-clock production to meet goals.

Officially, the company is describing the weeklong shutdown of the Model 3 line as "planned downtime" to "improve automation and systemically address bottlenecks in order to increase production rates."


Tesla executives hatched the plan weeks ago, a spokesperson said. "This is not unusual and is in fact common in production ramps like this," the spokesperson said.

Auto industry veterans disagree. Several said that stopping an assembly line for a car in commercial production is not only unusual, it's exceptional. And manufacturing experts say the retrofits being made during the shutdown will make Model 3 assembly more closely resemble other automobile plants around the world.

"Periodic shutdowns of hours or a day are not uncommon during pre-launch pilot build. They are unheard of in regular production, where he supposedly is," said Bob Lutz, the former General Motors vice chairman and noted Tesla critic, who has also held top executive positions at Chrysler, Ford and BMW.

"This shutdown is most likely for the purpose of ripping out all the '22nd century' fully-automated assembly systems which were going to 'revolutionize automotive manufacturing' and turned out not to work," Lutz said via email.


Dave Sullivan, an analyst at AutoPacific Inc., said in an email: "Traditional automakers adjust bottlenecks on the fly during a launch. This is totally out of the ordinary."

The news was another blow to Tesla shareholders, who have seen the stock price recover somewhat after plunging 17% in one week in April on news of production problems, a car crash involving the company's Autopilot technology and a big recall. Tesla shares fell sharply at the start of trading Tuesday before finishing down 1.2% at $287.69.

The shutdown caught workers by surprise as well. Over the weekend, Tesla ordered Model 3 assembly line workers to not show up for work Monday through Saturday. Take vacation days, they were told, or use up remaining personal days off, or elect not to be paid at all.

Workers were told at the last minute, a Tesla spokesperson said, because "the exact timing (of planned shutdowns) may not be finalized until closer to when it happens."

But on Tuesday, Tesla said in an internal email it will begin around-the-clock production and add another factory shift in Fremont as it tries to ramp up Model 3 output to 6,000 a week by the end of June. (Musk said the carmaker produced 2,250 of the sedans last week.) Between the Fremont plant and its battery factory, Tesla will be adding about 400 people per week for several weeks, Musk wrote in the email obtained by Electrek.

The company had previously said it was targeting production of 5,000 vehicles a week by around the end of the second quarter.

"The reason that the burst-build target rate is 6,000 and not 5,000 per week in June is that we cannot have a number with no margin for error across thousands of internally and externally produced parts and processes," Musk wrote.

"We are burning the midnight oil to burn the midnight oil," he added.


ADVERTISEMENT
Musk described the 6,000 car last-week-in-June production level as a "burst-build" that will "lay the groundwork" for steadily high production rates a few months from now. That means any given week of production could fall far short, making third quarter production forecasts more uncertain than ever.

The shutdown came just days after Musk appeared on a CBS This Morning segment, admitting that Tesla over-automated the Model 3 assembly line. In the taped segment, CBS This Morning co-host Gayle King was led by Musk on a tour through Tesla's then-bustling Fremont factory.

Musk sought to reassure viewers that lessons have been learned and that the Model 3 project, considered crucial to Tesla's continued existence, was back on track. Musk offered few details about automation problems beyond the admission that too many robots were installed. But he did say that "we have this crazy complex network of conveyor belts and it was not working. We got rid of the whole thing."

The too-many-robots admission appears to be a sobering comedown for Musk, who has said automation would one day become Tesla's primary business. Today, Tesla builds electric cars, solar energy systems and battery storage devices.

Although he's famous for warning that humanity should fear the arrival of artificial intelligence, Musk is eager to see robots replace human workers at Tesla. "Parking is one of my biggest nightmares," he said in 2016, referring to the always-overloaded parking lot at the Fremont plant. "It's like you can't fit everyone."

In the CBS This Morning interview, Musk backpedaled. "Maybe you need more people here working," King told him. "Yeah, we do," Musk said. Later that day he tweeted, "Yes, excessive automation at Tesla was a mistake. To be precise, my mistake. Humans are underrated."

He also copped to over-ambition in the design of the Model 3 in the interview. "We got complacent about some of the things we thought were our core technology, we put too much technology into the Model 3 all at once." Tesla owner forums are ripe with complaints including dead batteries, body panels that don't fit right and malfunctions in the input-output screen. Many customer service calls, owners report, result in instructions to reboot the system.

For years, Musk and his lieutenants have said that the Model 3 would be easier to manufacture than the Model X sport utility vehicle, which remains riddled with quality problems, earning low marks in Consumer Reports. "We're really trying to take a lot of lessons learned from Model X. We put a lot of bells and whistles on Model X and a lot of advanced technologies that weren't necessary for version one of the vehicle," Musk said in 2016.

If the Model 3 assembly line cranks back up as scheduled Monday, Musk will be hoping it marks a fresh start. Progress toward Musk's goal of 10,000 cars a week by the end of December is essential if Tesla is going to fulfill his prediction that the company will generate more cash than it burns in the third and fourth quarters of this year, alleviating the need to raise additional capital.

Analysts such as Efraim Levy at CFRA are skeptical. Barring some sort of "manipulation," he sees no way Tesla can produce positive free cash flow two quarters in a row this year.

In the CBS interview, Musk dismissed stock analysts.


ADVERTISEMENT
"The problem a lot of analysts have is they kind of look in the rearview mirror instead of looking out the front windscreen," he said. "People have underestimated Tesla because they have looked at [what] Tesla's done in the past and used that as a proxy for what we're able to do in the future."

russ.mitchell@latimes.com

Bloomberg contributed to this report.

UPDATES:

7:41 p.m.: This articles was updated with additional details.

5:05 p.m.: This article has been written through by staff with additional detail and commentary.

This article was originally published at 6:50 a.m.

What steps are missing to have fully running automated factories? What would be needed for these to not fail? Or at least speed up time in-between breakdowns or cooling down?

Is this achievable? Or is it a pipe dream?

[SolarpunkFan]

What steps are missing to have fully running automated factories? What would be needed for these to not fail? Or at least speed up time in-between breakdowns or cooling down?

Is this achievable? Or is it a pipe dream?

If I had to guess: on-the-fly object recognition and better ability to deal with ambiguity and unforeseen problems.

This article is probably over-hyped (and the title is a bit misleading, it's an AI that can write code not "its own code"), but interesting.

Futurity: New A.I. application can write its own code

Computer scientists have created a deep-learning, software-coding application that can help human programmers navigate the growing multitude of often-undocumented application programming interfaces, or APIs.

Designing applications that can program computers is a long-sought grail of the branch of computer science called artificial intelligence (AI). The new application, called Bayou, came out of an initiative aimed at extracting knowledge from online source code repositories like GitHub. Users can try it out at askbayou.com.

“People have tried for 60 years to build systems that can write code, but the problem is that these methods aren’t that good with ambiguity,” says Bayou co-creator Swarat Chaudhuri, associate professor of computer science at Rice University. “You usually need to give a lot of details about what the target program does, and writing down these details can be as much work as just writing the code.”

“Bayou is a considerable improvement,” he says. “A developer can give Bayou a very small amount of information—just a few keywords or prompts, really—and Bayou will try to read the programmer’s mind and predict the program they want.”

Chaudhuri says Bayou trained itself by studying millions of lines of human-written Java code. “It’s basically studied everything on GitHub, and it draws on that to write its own code.”

Bayou co-creator Chris Jermaine, a professor of computer science who co-directs Rice’s Intelligent Software Systems Laboratory with Chaudhuri, says Bayou is particularly useful for synthesizing examples of code for specific software APIs.

“Programming today is very different than it was 30 or 40 years ago,” Jermaine says. “Computers today are in our pockets, on our wrists and in billions of home appliances, vehicles, and other devices. The days when a programmer could write code from scratch are long gone.”

Bayou architect Vijay Murali, a research scientist at the lab, says, “Modern software development is all about APls. These are system-specific rules, tools, definitions, and protocols that allow a piece of code to interact with a specific operating system, database, hardware platform, or another software system. There are hundreds of APIs, and navigating them is very difficult for developers. They spend lots of time at question-answer sites like Stack Overflow asking other developers for help.”

Murali says developers can now begin asking some of those questions at Bayou, which will give an immediate answer.

“That immediate feedback could solve the problem right away, and if it doesn’t, Bayou’s example code should lead to a more informed question for their human peers,” Murali says.

Jermaine says the team’s primary goal is to get developers to try to extend Bayou, which has been released under a permissive open-source license.

“The more information we have about what people want from a system like Bayou, the better we can make it,” he says. “We want as many people to use it as we can get.”

The researchers based Bayou on a method called neural sketch learning, which trains an artificial neural network to recognize high-level patterns in hundreds of thousands of Java programs. It does this by creating a “sketch” for each program it reads and then associating this sketch with the “intent” that lies behind the program.

When a user asks Bayou questions, the system makes a judgment call about what program it’s being asked to write. It then creates sketches for several of the most likely candidate programs the user might want.

“Based on that guess, a separate part of Bayou, a module that understands the low-level details of Java and can do automatic logical reasoning, is going to generate four or five different chunks of code,” Jermaine says. “It’s going to present those to the user like hits on a web search. ‘This one is most likely the correct answer, but here are three more that could be what you’re looking for.'”

The Defense Advanced Research Projects Agency funded the initiative that led to the project.

The researchers will present a paper on Bayou on May 1 in Vancouver, British Columbia, at the Sixth International Conference on Learning Representations.

Here's the site itself for any of you here who are Java programmers.

[FaxModem1]

The Wall Street Journal

How the World’s Biggest Companies Are Fine-Tuning the Robot Revolution
Automation is leading to job growth in certain industries where machines take on repetitive tasks, freeing humans for more creative duties
By William Wilkes
May 14, 2018 10:24 a.m. ET
35 COMMENTS
ANSBACH, Germany—A few years ago, Roland Rösch’s job involved grabbing scalding-hot auto parts from an oven and inspecting them for signs they had failed a safety test.

These days he still inspects, but the grabbing is being done by Fritz, a robot that auto-parts manufacturer Robert Bosch GmbH installed three years ago at this German factory as part of an automation effort.

Fritz is more efficient at handling the dangerous and repetitive task of lifting the 8-inch metal-and-circuitry pieces out of the furnace. This leaves Mr. Rösch less exposed to potential accidents and gives him time to test 20% more parts than he did before the robots.


The big question surrounding automation has long been whether robots would compete with workers or help them. Initially, workers feared robots would destroy jobs across the economy. Scholarly research and real-life experience have eased that concern, although some types of workers and industries are ending up on the losing side.

Today, the question is more precise: In which industries does automation help both employer and employee?

The companies that may have cracked the code are those that can assign repetitive, precise tasks to robots, freeing human workers to undertake creative, problem-solving duties that machines aren’t very good at. That’s particularly relevant for manufacturing, the food sector and service sectors such as billing, where timetable spreadsheets can be automated, freeing up workers to do higher-value tasks.

Door assembly work at BMW’s Spartanburg, S.C., automobile plant.
Door assembly work at BMW’s Spartanburg, S.C., automobile plant. PHOTO: FRED ROLLISON/BMW GROUP
With demand for Bosch-built steering controls high, the company has used automation to increase its output, leading it to hire more people to perform the type of checks Mr. Rösch conducts.

“We looked for 20,000 new hires last year,” a mix of new positions and replacement staff, said Stefan Assmann, one of the company’s chief engineers, to join Bosch’s total 400,000 employees. Bosch factories world-wide now make use of 140 robotic arms, up from zero in 2011. “We can’t see robots having a negative impact on our workforce,” Mr. Assmann said.

Computers can zoom through activities humans find difficult, such as playing chess, doing calculus or repeating a set of movements precisely over time. Other, seemingly mundane tasks—brushing your teeth or running through the woods—can overwhelm even complex machines.

Those tasks call on multiple senses, including touch and depth perception, feeding information to a problem-solving brain, which can then finely adjust movements, said Satyandra Gupta, professor of mechanical engineering at the University of Southern California.

For companies, choosing the appropriate tasks to automate is important. Auto maker BMW AG automated some of the physical labor at the Spartanburg plant in South Carolina while retaining tasks involving judgment and quality control for workers.

Adding to the Team
German auto-parts maker Robert Bosch GmbH's profits are up as it has invested in more automation, and it has hired more workers.

Robots fit black, soundproofing rubber tubes to the inner rim of car doors, a task once done entirely by hand, on the more than 5,000 or so car doors that pass through the production line each day. Human workers do final checks on the tube’s placement. The division of labor speeds up the process.


Since BMW introduced this and other automated processes over the past decade, it has more than doubled its annual car production at Spartanburg to more than 400,000. The workforce has risen from 4,200 workers to 10,000, and they handle vastly more complex autos—cars that once had 3,000 parts now have 15,000.

Being spared strenuous activities gives workers the time and energy to tackle more demanding and creative tasks, BMW said in a statement.

James Bessen, an economist who teaches at Boston University School of Law, said automation like that at the Spartanburg plant has enabled a huge increase in the quality and variety of products, which help spur consumer demand. BMW’s share of luxury-car sales in the U.S. has risen sharply, with over 300,000 cars sold last year compared with just over 120,000 in 1997, company figures show.

Tesla Inc., by contrast, has struggled with production of the Model 3 car at its Fremont, Calif., plant after its use of robots got out of balance. Undetected errors in parts built by robots caused bottlenecks in production, meaning it could build only 2,020 cars a week compared with the 5,000 it originally promised, according to the company.

Analysts at investment research firm Bernstein said Tesla automated welding, paint and body work processes, as other manufacturers have done, but also automated final assembly work, in which parts, seats and the engine are installed in the car’s painted shell. Errors in this work caused production bottlenecks. “Automation in final assembly doesn’t work,” said analyst Max Warburton.

“Yes, excessive automation at Tesla was a mistake…Humans are underrated,” wrote Tesla CEO Elon Musk in a tweet last month.

Robots have resulted in pay cuts for low-skilled machine operators, such as those who oversee wood- or leather-cutting machines, who play a diminishing role in production due to automation. And they have eliminated entire occupations, especially in simple manufacturing processes where there aren’t value-added jobs for displaced workers to move to.

Mining, for example, hinges on raw high-volume production—dig more rock, make more money—which is better done by machines that won’t tire or get injured.


Rio Tinto PLC plans to lay off drivers as it introduces self-driving trucks to move iron ore at its mines in Western Australia. The trucks, which follow sensors and maps of the mining site installed in onboard computers, can operate longer than human drivers and are more reliable. Beneath the ground, robotic drilling rigs have taken over the dangerous work of inserting explosives into holes dug in mining shafts.

The automation would improve safety and unlock significant productivity gains, helping generate annual savings of around $500 million beginning in 2021, said Chris Salisbury, the Rio Tinto board member in charge of the firm’s iron-ore mining operations. The company said it would look to retrain or find new roles for the workers affected by the automation.

Jobs in the garment industry are also disappearing as firms automate repetitive, high-volume tasks such as sewing and knitting, where machines can work faster and more accurately than humans.

Technological breakthroughs have enabled robots to take on delicate tasks, such as manipulating pliable fabrics, stitching pockets and attaching belt loops to pants. In the early days of automation, it was thought that humans would be needed for such finishing work.

The International Labor Organization has warned that nearly 90% of garment and footwear workers in Cambodia and Vietnam are at risk from “sewbots.”

At an aggregate level, however, the jobs created by automation outnumber those that are being destroyed, according to analysis by the Massachusetts Institute of Technology’s David Autor and Utrecht University’s Anna Salomons.

People losing jobs, however, may not be the same ones filling newly created ones, since different skills are often required.

A Fiskars worker tests that scissors snip properly.
A Fiskars worker tests that scissors snip properly. PHOTO: HENRI VOGT FOR THE WALL STREET JOURNAL
The Asian Development Bank said in April that automation had created an extra 34 million jobs in its region after price falls and quality improvements spurred demand for Asian factory-made goods.

More-developed economies have also seen job growth. Automation in the U.K. over the past 15 years has destroyed 800,000 lower-skilled services jobs—such as call centers—but has created 3.5 million higher-skilled ones in their place, according to a 2017 workforce study by consultancy firm Deloitte. The new jobs paid around $13,500 more than the jobs they replaced, Deloitte said.


Industrial employment in Germany is expected to rise 1.8% by 2021 because robots and automation are making the country’s factories more competitive, according to the Germany-based Centre for European Economic Research in April.

Automation can help feed demand for a product—because quality improves or it becomes less expensive or more available—which can create jobs as a result.

Finnish firm Fiskars AB, manufacturer of iconic and once pricey orange-handled scissors, used automation to reach more customers. Workers at its Helsinki plant formerly forged steel blades by hand in 2,700-degree furnaces, repetitive and dangerous work that was slow and costly.

When robots took over the tasks in 2011, technicians moved to quality control, testing the scissors to make sure the blades made the right “snip” sound as they sliced together, and if they smoothly cut strips of fabric. If necessary, workers could adjust the blades bit by bit, in a process calling on multiple senses that machines couldn’t replicate.

Once the process was partly automated, the company was able to increase production and lower prices, stimulating new demand without sacrificing quality, according to Chief Supply Chain Officer Risto Gaggl.

Employment at Fiskars has soared along with higher production, with the company now employing 8,560 people in its factories and offices compared with 4,515 in 2007.

Machines in the Workforce
Companies are expanding the use of robots, artificial intelligence and other forms of automation.


Source: Source: International Federation of Robotics
In Europe, “we couldn’t find anyone who has been fired because of robots,” said Professor Wolfgang Dauth, leader of a yearslong study into the impact on workers of robotization on the continent by the Bonn-based Institute of Labor Economics. Part of the reason is strong labor unions require retraining for workers when robots take over tasks. Another part is that Europe’s more-complex industries need human thinkers to work in complement with machines.

Electrolux AB, the world’s second-largest appliance maker by units sold after Whirlpool Corp. , has spent millions of euros on automating the production of washing machines and other devices, which are now assembled almost entirely by machine.

The company said robots freed up technicians to spend time on a creative task that is impossible to automate: designing and implementing changes to the factory floor and robot layout to customize procedures and make production more efficient. The constant, incremental improvements make a broader range of production in the same factory space possible, which in turn supports more workers.

The company said it tweaked hiring and training so that its workforce could successfully operate with robots, including a month of robotics training when hired and bimonthly half-day sessions. The company also built robot-testing areas at its factories where technicians can experiment with different robot hardware and software.

Employment at Electrolux has risen to more than 55,000 in 2017 from about 53,000 in 2011, reversing a yearslong trend of shrinking staff numbers after China’s December 2001 entry to the World Trade Organization flooded the market with cheaper washing machines.

“We don’t see automation going over 50%,” said Jan Brockmann, chief operations officer at Electrolux. “There’s not much point.” He said machines would likely take over routine work like assembly, freeing workers to make repairs and improvements to an increasingly efficient production line.

The slow pace of robot rollouts can shield workers, providing time for retraining. Companies rarely automate all of a worker’s tasks in one swoop, and it takes time to work out how best to use robots. The high cost of adding new automation also slows the process.

A Fiskars technician monitors quality control while a robot moves steel blades into a furnace.
A Fiskars technician monitors quality control while a robot moves steel blades into a furnace. PHOTO: HENRI VOGT FOR THE WALL STREET JOURNAL
Bosch developed training courses for workers, teaching once single-skilled welders, joiners and mechanics basic software coding skills to enable them to use robots as tools much like hammers or screwdrivers. “We employ designers, engineers and scientists,” said Mr. Assmann, one of the firm’s chief engineers. “But you still need people who are good with their hands.”

U.K.-based food delivery company Ocado Group has progressively automated work processes and has added workers as demand for its once-exclusive internet-grocery shopping service has surged, in part driven by the efficiency savings that have lowered prices.

The company’s chief innovation is a complex web of grocery-transporting conveyor belts that allow it to process consumers’ online orders. Another set of robots under development will be assistants for its human maintenance staff, allowing them to be more productive in managing the conveyor belts and other machinery. The company shuts down operations for three hours each day for maintenance, and missing that window could mean being unable to process deliveries.

Instead of walking around the factory to collect whatever tools are needed, the robots will anticipate what tools the workers need, and bring them to hand, acting as automated assistants.

“Our business model would just fail if these machines didn’t work,” said Graham Deacon, head of automation at Ocado. “We need humans to make sure they don’t break down.”

Write to William Wilkes at william.wilkes@wsj.com

Charts and graphs at the link showing the math. What are your thoughts?

[bilateralrope]

The article has anecdotes from various companies. Selection bias is a concern. How were those companies selected ?

Those companies say that the number of units they are producing has increased. Which means that the number they are selling has increased. Where are they getting their new customers from ?
One possibility is that they are taking customers from their competitors. Any jobs lost at their competitors due to the competitors loss of customers are still jobs lost to automation.

[SolarpunkFan]

DARPA is getting in on some action.

Researchers Selected to Develop Novel Approaches to Lifelong Machine Learning

Machine learning (ML) and artificial intelligence (AI) systems have significantly advanced in recent years. However, they are currently limited to executing only those tasks they are specifically designed to perform and are unable to adapt when encountering situations outside their programming or training. DARPA’s Lifelong Learning Machines (L2M) program, drawing inspiration from biological systems, seeks to develop fundamentally new ML approaches that allow systems to adapt continually to new circumstances without forgetting previous learning.

First announced in 2017, DARPA’s L2M program has selected the research teams who will work under its two technical areas. The first technical area focuses on the development of complete systems and their components, and the second will explore learning mechanisms in biological organisms with the goal of translating them into computational processes. Discoveries in both technical areas are expected to generate new methodologies that will allow AI systems to learn and improve during tasks, apply previous skills and knowledge to new situations, incorporate innate system limits, and enhance safety in automated assignments.

The L2M research teams are now focusing their diverse expertise on understanding how a computational system can adapt to new circumstances in real time and without losing its previous knowledge. One group, the team at University of California, Irvine plans to study the dual memory architecture of the hippocampus and cortex. The team seeks to create an ML system capable of predicting potential outcomes by comparing inputs to existing memories, which should allow the system to become more adaptable while retaining previous learnings. The Tufts University team is examining a regeneration mechanism observed in animals like salamanders to create flexible robots that are capable of altering their structure and function on the fly to adapt to changes in their environment. Adapting methods from biological memory reconsolidation, a team from University of Wyoming will work on developing a computational system that uses context to identify appropriate modular memories that can be reassembled with new sensory input to rapidly form behaviors to suit novel circumstances.

“With the L2M program, we are not looking for incremental improvements in state-of-the-art AI and neural networks, but rather paradigm-changing approaches to machine learning that will enable systems to continuously improve based on experience,” said Dr. Hava Siegelmann, the program manager leading L2M. “Teams selected to take on this novel research are comprised of a cross-section of some of the world’s top researchers in a variety of scientific disciplines, and their approaches are equally diverse.”

While still in its early stages, the L2M program has already seen results from a team led by Dr. Hod Lipson at Columbia University’s Engineering School. Dr. Lipson and his team recently identified and solved challenges associated with building and training a self-reproducing neural network, publishing their findings in Arvix Sanity. While neural networks are trainable to produce almost any kind of pattern, training a network to reproduce its own structure is paradoxically difficult. As the network learns, it changes, and therefore the goal continuously shifts. The continued efforts of the team will focus on developing a system that can adapt and improve by using knowledge of its own structure. “The research team’s work with self-replicating neural networks is just one of many possible approaches that will lead to breakthroughs in lifelong learning,” said Siegelmann.

“We are on the threshold of a major jump in AI technology,” stated Siegelmann. “The L2M program will require significantly more ingenuity and effort than incremental changes to current systems. L2M seeks to enable AI systems to learn from experience and become smarter, safer, and more reliable than existing AI.”

Image Caption Today's machine learning and AI systems are limited to executing only tasks they are specifically programmed to perform, without being able to adapt to new situations outside of their training. DARPA's L2M program aims to generate new methodologies that will allow these systems to learn and improve during tasks, apply previous skills and knowledge to new situations, incorporate innate system limits, and enhance safety in automated assignments.

[FaxModem1]

CNN

shopping
Kroger's automation deal sends Ocado shares up 50%
By Alanna Petroff May 17, 2018: 6:52 AM ET
Kroger has signed an exclusive deal with UK online supermarket Ocado to use its technology in the United States.

Shares in Ocado shot up by more than 50% to a record high in London after the deal was announced Thursday.

Ocado has built a reputation for fast deliveries and high-quality customer service thanks to its complex web of logistics systems and warehouses run by smart robots.

The Kroger (KR) deal will see the American grocery giant take a 5% stake in Ocado (OCDGF) with an investment worth about £183 million ($247 million).

Ocado said it will begin setting up Kroger with various systems to help it manage warehouse operations, automation, logistics and delivery route planning in the US.

The companies will identify three sites in 2018 for development of new "automated warehouse facilities" in the United States, they said in a statement. A total of 20 will be identified in the first three years of the deal.

"Ocado believes Kroger to be the grocer best-positioned to win in US grocery and will discontinue discussions with other US-based retailers," the companies said in a statement.

The deal should help Kroger compete with Amazon (AMZN), which has entered the grocery delivery market in a big way.

Ocado CEO Tim Steiner said he expected the "transformative relationship" to reshape the food retailing industry in the United States.

Ocado, which has no brick-and-mortar stores, has warehouses that "are capable of collaborating to pick a typical 50-item order in a matter of minutes," its chief technology officer, Paul Clarke, wrote this week.

"Applications of [artificial intelligence] and machine learning pervade this platform," he said, noting the system helps predict changes in customer demand for its 50,000 items.

Ocado has signed a string of similar technology deals with major supermarkets in France, Canada and Sweden.

The Kroger deal is "an unmitigated positive in our eyes," Barclays analysts said in a research note.


Ocado is considered a leader in online shopping and delivery in the United Kingdom.
Ocado's stock market surge pushed the value of the company above £5 billion ($6.8 billion) on Thursday.

How will this affect the grocery chains in the US, both long term and short term? How will this affect employment? Is this just Kroger moving to the online market against Amazon in addition to it's physical stores, or is this the future of grocery shopping?

And guys, I remember the Flippy thread, please no flames about people's shopping habits via age.

[Gandalf]

It's certainly a growing trend here. If grocery warehouses replace a lot of these horrid supermarkets, then bring them on.

[FaxModem1]

It's certainly a growing trend here. If grocery warehouses replace a lot of these horrid supermarkets, then bring them on.

Off-topic, but why do you hate supermarkets?

[Gandalf]

Similar reasons to why I don't like Walmarts, but also because of their massive contributions to food waste.

Necessities like food are horrible when combined with capitalism.

[bilateralrope]

https://www.ntsb.gov/investigations/Acc ... prelim.pdf

The report about the Uber car that killed someone is out. The reason for the crash turns out to be stupider than I predicted:

According to Uber, the developmental self-driving system relies on an attentive operator to intervene if the system fails to perform appropriately during testing. In addition, the operator is responsible for monitoring diagnostic messages that appear on an interface in the center stack of the vehicle dash and tagging events of interest for subsequent review.

The driver was not watching the road was because he was doing his job.

According to data obtained from the self-driving system, the system first registered radar and LIDAR observations of the pedestrian about 6 seconds before impact, when the vehicle was traveling at 43 mph. As the vehicle and pedestrian paths converged, the self-driving system software classified the pedestrian as an unknown object, as a vehicle, and then as a bicycle with varying expectations of future travel path. At 1.3 seconds before impact, the self-driving system determined that an emergency braking maneuver was needed to mitigate a collision (see figure 2). According to Uber, emergency braking maneuvers are not enabled while the vehicle is under computer control, to reduce the potential for erratic vehicle behavior. The vehicle operator is relied on to intervene and take action. The system is not designed to alert the operator.

So whenever the car computer decides that an emergency braking maneuver is required its options are:
- Perform the emergency braking maneuver. Except Uber decided that they couldn't allow this option.
- Alert the operator. Uber provided no way to to do this.
- Rely on the operator to take action. Except the car was designed to leave him paying less attention to the road than any human driver.

I've also got some complaints about the car not taking precautionary action when it detected an unknown object crossing the road and the design of the brick path by the road.

[Sea Skimmer]

This basically is what I expected to see and in line with unofficial reporting on the matter prior. Uber's entire self driving effort was borderline pathetic, they were getting 10-12 miles between driver interventions to avoid an accident at at time when other companies are into the low to mid single digit thousands of miles. They wanted a system you could install as software in a cellphone...

Not that long before this crash uber's test cars had two operators, one to like, be the driver, one to take care of the diagnostic issue. Uber cut the second person to save money and get more miles out of cars within the budget they had.

[The Romulan Republic]

And they got someone killed with this shit?

I hope the victim's family squeezes every penny from them that they can.

[bilateralrope]

I think there should be manslaughter charges against everyone who knew that emergency braking was disabled.

[FaxModem1]

Well, it's Uber, the company that has employees living in their cars because they can't afford otherwise and they're cheap because they undercut the taxi companies, leading to the problems that passengers have had with criminal drivers.

That makes me want to bring up another topic. How important is cheap development of automation as compared to expensive automation? Clearly, in the case of transportation, money must be spent so that human lives aren't taken, but in how many areas can they, scrimp on research and development? Or scrimp before something goes horribly wrong?

Is this why quite a few automated devices break down so easily, in that the money wasn't put into them to make a lasting machine?

[FaxModem1]

The Verge

TECH
SCIENCE
CULTURE
MORE

Search
US & WORLD

TECH

ARTIFICIAL INTELLIGENCE
Economists worry we aren’t prepared for the fallout from automation
29
Too much time discussing whether robots can take your job; not enough time discussing what happens next
By James Vincent@jjvincent Jul 2, 2018, 11:38am EDT
SHARE

Photo by Bill Pugliano/Getty Images
Are we focusing too much on analyzing exactly how many jobs could be destroyed by the coming wave of automation, and not enough on how to actually fix the problem? That’s one conclusion in a new paper on the potential effects of robotics and AI on global labor markets from US think tank, the Center for Global Development (CGD).


The paper’s authors, Lukas Schlogl and Andy Sumner, say it’s impossible to know exactly how many jobs will be destroyed or disrupted by new technology. But, they add, it’s fairly certain there are going to be significant effects — especially in developing economies, where the labor market is skewed toward work that requires the sort of routine, manual labor that’s so susceptible to automation. Think unskilled jobs in factories or agriculture.

AUTOMATION PROBABLY WON’T KILL JOBS, BUT IT’LL CREATE MORE BAD ONES
As earlier studies have also suggested, Schlogl and Sumner think the effects of automation on these and other nations is not likely to be mass unemployment, but the stagnation of wages and polarization of the labor market. In other words, there will still be work for most people, but it’ll be increasingly low-paid and unstable; without benefits such as paid vacation, health insurance, or pensions. On the other end of the employment spectrum, meanwhile, there will continue to be a small number of rich and super-rich individuals who reap the benefits of increased productivity created by technology.


These changes will likely mean a decline in job security and standards of living for many, which in turn could lead to political dissatisfaction. (Some suggest we’ve already seen the early impact of this, with US cities where jobs are at risk of automation more likely to vote Republican.) Schlogl and Sumner give an overview of proposed solutions to these challenges, but seem skeptical that any go far enough.

One class of solution they call “quasi-Luddite” — measures that try to stall or reverse the trend of automation. These include taxes on goods made with robots (or taxes on the robots themselves) and regulations that make it difficult to automate existing jobs. They suggest that these measures are challenging to implement in “an open economy,” because if automation makes for cheaper goods or services, then customers will naturally look for them elsewhere; i.e. outside the area covered by such regulations.


A related strategy is to reduce the cost of human labor, by driving down wages or cutting benefits, for example. “The question is how desirable and politically feasible such strategies are,” say Schlogl and Sumner, which is a nice way of saying “it’s not clear how much you can hurt people before they riot in the streets.”

CeBIT Technology Trade Fair 2018
Taxes on robots or goods produced by robots are “quasi-Luddite,” say Schlogl and Sumner. Photo by Alexander Koerner/Getty Images
The other class of solution they call “coping strategies,” which tend to focus on one of two things: re-skilling workers whose jobs are threatened by automation or providing economic safety nets to those affected (for example, a universal basic income or UBI).

RETRAINING WORKERS IS EXPENSIVE, AND SOMETIMES NOT POSSIBLE
Schlogl and Sumner suggest that the problem with retraining workers is that it’s not clear what new skills will be “automation-resistant for a sufficient time” or whether it’s even worth the money to retrain someone in the middle of their working life. (Retraining is also more expensive and challenging for developing countries where there’s less infrastructure for tertiary education.) As for economic safety nets like UBI, they suggest these might not even be possible in developing countries. That’s because they presuppose the existence of prosperous jobs somewhere in the economy from which profits can be skimmed and redistributed. They also note that such UBI-related schemes might raise the cost of labor, which in turn would encourage more jobs to be substituted with technology.


All this leads the pair to conclude that there’s simply not enough work being done researching the political and economic solutions to what could be a growing global crisis. “Questions like profitability, labor regulations, unionization, and corporate-social expectations will be at least as important as technical constraints in determining which jobs get automated,” they write.

And do Schlogl and Sumner propose any of their own solutions? They write: “In the long term, utopian as it may seem now, [there is a] moral case for a global UBI-style redistribution framework financed by profits from ... high-income countries.” Now that would certainly get the anti-globalist crowd incensed, and the pair admit that it’s “difficult to see how such a framework would be politically enacted.” Back to the drawing board then.

So, instead of jobs just disappearing, the quality and security of the remaining jobs would go down. That's an alarming possibility.

I agree with the statement that it's going to be affecting developing nation's first, which primarily rely on manufacturing and resource collection for their economies. I'm just not sure what the future holds, because I don't see international UBI becoming a thing.

[Gandalf]

That makes me want to bring up another topic. How important is cheap development of automation as compared to expensive automation? Clearly, in the case of transportation, money must be spent so that human lives aren't taken, but in how many areas can they, scrimp on research and development? Or scrimp before something goes horribly wrong?

Is this why quite a few automated devices break down so easily, in that the money wasn't put into them to make a lasting machine?

For the whole savings set against lives thing, I think of this bit from Fight Club;

"I'm a recall coordinator. My job is to apply the formula. It's a story problem. A new car built by my company leaves somewhere traveling at 60 miles per hour. The rear differential locks up. The car crashes and burns with everyone trapped inside. Now: do we initiate a recall? Take the number of vehicles in the field, (A), and multiply it by the probable rate of failure, (B), then multiply the result by the average out-of-court settlement, (C). A times B times C equals X. If X is less than the cost of a recall, we don't do one."

Now apply the rationale to an automated taxi fleet.

[FaxModem1]

That makes me want to bring up another topic. How important is cheap development of automation as compared to expensive automation? Clearly, in the case of transportation, money must be spent so that human lives aren't taken, but in how many areas can they, scrimp on research and development? Or scrimp before something goes horribly wrong?

Is this why quite a few automated devices break down so easily, in that the money wasn't put into them to make a lasting machine?

For the whole savings set against lives thing, I think of this bit from Fight Club;

"I'm a recall coordinator. My job is to apply the formula. It's a story problem. A new car built by my company leaves somewhere traveling at 60 miles per hour. The rear differential locks up. The car crashes and burns with everyone trapped inside. Now: do we initiate a recall? Take the number of vehicles in the field, (A), and multiply it by the probable rate of failure, (B), then multiply the result by the average out-of-court settlement, (C). A times B times C equals X. If X is less than the cost of a recall, we don't do one."

Now apply the rationale to an automated taxi fleet.

Wouldn't bad press hurt for such new technologies?

[Gandalf]

Not necessarily. There's a lot of little variables, like the ability to PR something away, and the promise of what's coming.

If you told everyone that in five years time, taxis would be automated and fares would halve, I wager a lot of stuff would be accepted for sheer convenience. As a society, we're pretty ready to ignore a lot if it means things are cheaper and easier.