AI Reinventing the Alphabet?

AI Is Not Going to Reinvent the Alphabet Anytime Soon

By Wired, May 1, 2023

Illustration shows a mechanical, metal brain superimposed on a host of letters and typerwriter keys.

Looking at typography developed by AI is like looking at lettering submerged in deep water, warped and fuzzy. The words are barely recognizable, and the original form has been lost. AI typography is, charitably, bad.

Where typography is concerned, it is becoming clear that AI innovations are focusing on the wrong ideas. Some are playing with using this technology to try to redefine visual language—in the case of our Latin letterset, one that has existed for more than 2,000 years.

Ultimately this is an unworkable course. The key to setting AI typography on a better, more accessible path is to think of it as assistive rather than generative.

From Wired     

View Full Article  

Google/Alphabet Cutting Jobs

Tech slipping.  To be replaced by AI? 

Google's parent company Alphabet will cut 12,000 jobs, in the latest staff redundancies to hit the tech industry.   in BBD

Google and Alphabet CEO Sundar Pichai said he took "full responsibility" for the cuts, in an internal email.  The cuts will affect 6% of Alphabet's workforce worldwide, in teams including recruitment and engineering.  This comes days after Microsoft announced 10,000 jobs would be lost, and weeks after Amazon announced 18,000 job cuts.

Mr Pichai thanked staff for "working so hard" in their roles, adding that their "contributions have been invaluable".  He wrote: "While this transition won't be easy, we're going to support employees as they look for their next opportunity.

"Until then, please take good care of yourselves as you absorb this difficult news. As part of that, if you are just starting your work day, please feel free to work from home today."

According to a recent filing with Companies House, Google has more than 5,500 staff in the UK. But it is unclear how many of these will be affected by the cuts.   .... '

Google Tests AI Robotics in a Kitchen

Look forward to seeing this.  Kitchen has many tasks that could be automated, but usually not well positioned for broad application and integration.  Note here a number of hints at Google investment in the topic.  Ready to test.

Google Is Testing Its Latest AI-Powered Robot In a Kitchen  in ExtremeTech   By Adrianna Nine on August 17, 2022

Many robots are created to conduct highly-controlled jobs like frying potatoes, watering plants, or collecting litter. But a truly life-changing robot is one that can adapt to changing—and sometimes hectic—circumstances (ideally without losing its cool). That’s the line of thinking Google is following as it meshes its language-handling AI with a handy robot assistant.

Google’s Pathways Language Model (PaLM) is a relatively new 540-billion parameter network built to complete a variety of complex language-based tasks. It’s said to be intelligent enough to describe how it solved a math problem and annoy you by explaining your own jokes. Rather than focusing on one area of “expertise” and starting fresh every time it learns a new skill, PaLM can “stack” previously-learned knowledge to devise solutions to novel problems, similar to how humans assess new situations. This is important if a robot is meant to help humans in their jobs or day-to-day personal lives.

It just so happens that Google’s parent company, Alphabet, has been working on a new robotics firm called Everyday Robots. As its name suggests, the firm’s goal is to build robots that learn on their own and take care of “time-consuming, everyday tasks.” Combined with PaLM, Everyday Robots’ SayCan robot becomes the PaLM-SayCan, a bot capable of assessing its own capabilities, its environment, and the task at hand, then breaking that task into smaller sub-tasks to achieve the desired goal. ... ' 

DeepMind Makes a Profit

Do we expect a leading edge research lab to make a profit?

AI News

DeepMind sees revenue jump and turns first ever profit  By Fin Strathern | October 7, 2021

Categories: DeepMind, Google,

DeepMind, the Alphabet-owned AI research lab, has turned a profit for the first time since the company was founded in 2010.

The London-based firm recorded a profit of £43.8 million in 2020 with revenue at £826 million, according to its annual results filing with Companies House.

In previous years, DeepMind have posted losses well into the hundreds of millions that Alphabet has heavily subsidised. For example, the company lost £476 million in 2019 and has made cumulative losses of nearly £2 billion since 2014.

How the company has more than tripled its revenue from £265 million in 2019 to £826 million in 2020 remains somewhat of a mystery. No explanation has been provided by DeepMind.

There is of course the host of companies that fall under the Alphabet umbrella which DeepMind sells software and services to, namely YouTube, Google, and X. However, aside from this, the research lab sells no products to consumers and has not announced any partnerships publicly. ... 

GM and Cruise Get California DMV Approval for Waymo

Observing this space I have received many questions about how likely autonomous driving will become, and how soon.  Mainly due to safety considerations.  Yet there still seems  a continuing movement to the approval of autonomous vehicles of  many kinds.   Here an example.   With carefully tailored regulation.  Interesting details below.  In the US this is often by State,  which adds complexity and thus cost.

Alphabet's Waymo, GM's Cruise Get California DMV Approval to Run Commercial Autonomous Car Services By CNBC 

GM-based Cruise and Alphabet's Waymo were granted autonomous vehicle deployment permits from the California Department of Motor Vehicles. This will allow them to charge fees and receive compensation for ride-hailing and other autonomous services offered to the public in certain areas.

However, the companies also need approval from the California Public Utilities Commission before they can offer such services to the general public outside of a testing program.  The new authorization allows Cruise vehicles to operate on public roads in certain areas of San Francisco between 10 p.m. and 6 a.m. at speeds no faster than 30 mph, while Waymo can operate in designated parts of San Francisco and San Mateo counties at speeds no faster than 65 mph.

Both companies are permitted to operate vehicles in light rain or fog. ..... 

Both companies have been testing fleets of autonomous vehicles in California with permits that allow for free driverless rides to passengers in test vehicles.  .... "

DeepMind Learns Football from Motion Capture

Intriguing, with the implication that you can go from direct motion capture to broader contexts.   Team work, which implies both individual skills, planning and team strategy.   Consider it being used for corporate resource use and allocation. 

DeepMind AI Taught Digital People to Play Football from Scratch,  By New Scientist

An artificial intelligence (AI)-trained computer learned to play football from scratch, using digital players.

Researchers at Alphabet's U.K.-based AI subsidiary DeepMind used motion-capture data from real players to teach the digital footballers to get up and run.

They became more skilled in specific training tasks, like dribbling a ball to follow a target, or kicking a ball to a target, via reinforcement learning.

The digital players engaged in a series of 45-second matches, and after 24 hours became quite adept; after a further training period of up to 10 days, they were able to consider future actions and work as a team.

Sebastian Risi at Denmark's IT University Copenhagen said although the system lacks full automation, it is "an exciting open challenge how we can learn complex tasks such as football end-to-end through more open-ended approaches that would discover the necessary stepping stones by themselves."

From New Scientist 

DeepMind Masters Games without the Rules, Just Rewards

MuZero. An intriguing  example, with the right data.

 DeepMind's AI Masters Games Without Even Being Taught the Rules

IEEE Spectrum by   Philip E. Ross

The MuZero artificial intelligence (AI) developed by Alphabet subsidiary DeepMind can master games without learning their rules beforehand. The system attempts first one action, then another, learning what the game rules permit, and concurrently noting the rewards proffered. MuZero then modifies its strategy until it finds a way to collect rewards more readily; this observational learning is perfect for any AI confronting problems that cannot be specified easily. MuZero outperforms previous DeepMind AIs in its economical use of data, because it models only environmental parameters that matter in decision-making. Once trained, MuZero processes so little for decision-making that its entire function might be managed on a smartphone. ... "

Alphabet's Loon Balloons for Internet Connectivity

Application of the general idea is expanding.  Here with autonomous navigation. 

Alphabet's Loon Balloons in Internet Connectivity

Alphabet’s Loon hands the reins of its internet air balloons to self-learning AI.  The company’s new AI flight control system outperforms its human-made one

By Nick Statt@nickstatt

Alphabet’s Loon, the team responsible for beaming internet down to Earth from stratospheric helium balloons, has achieved a new milestone: its navigation system is no longer run by human-designed software. ... " 

Alphabet Buying North Glasses

Somewhat Unexpected perhaps,  but some indication of seriousness on the ultimate wearable. Had heard from others that North Glasses were good.

Report: Alphabet looking to buy smart glasses startup North for $180M   By  Maria Deutscher

Google LLC parent Alphabet Inc. is reportedly in advanced talks to acquire North Inc., a Canadian smart glasses startup, for about $180 million.

The discussions were first reported by Canada’s The Globe and Mail on Thursday evening. The publication’s sources indicated that Ontario-based North is in the “final stages” of hammering out an agreement with Alphabet.

Founded in 2012, North is backed by about $160 million from investors that include Intel Corp.’s venture capital arm and Salesforce.com Inc. founder Marc Benioff. The startup offers a smart glasses product called Focals that’s touted as less bulky than many other products in the category, closely resembling a regular pair of glasses. Users can view mobile notifications without looking at their phone and perform certain other tasks such as ordering an Uber ride.  ..."

Alphabet Sidewalk Labs Abandon's Toronto Smart-Neighborhood

Heard a report early on on this, because it linked to some of our own smart store efforts, details will be interesting to see.   Where will the result be published?

Alphabet's Sidewalk Labs abandons its Toronto smart neighborhood project
Just like that, it's over.

Nick Summers, @nisummers in Engadget

Alphabet subsidiary Sidewalk Labs will no longer pursue its dream of a smart neighbourhood in Toronto.

In a Medium blog post, CEO Daniel Doctoroff said “unprecedented economic uncertainty” meant it was “too difficult” to achieve its dreams for Quayside, a proposed redevelopment on the city’s waterfront. If the company pushed forward with its vision — which was yet to receive sign-off from the Canadian government — “core parts of the plan” would need to be scarified, he said. “And so, after a great deal of deliberation, we concluded that it no longer made sense to proceed with the Quayside project,” Doctoroff added. ... "

Reinforcement Learning Improvements

Very interesting item.   Quite technical.  I like the hint of using pools or teams of agents to solve reinforcement problems.   Brings to mind the idea of process design and bringing multiple resources to bear.    Can it be more directly linked to process optimization processes?  Checking it out.

Google Brain and DeepMind researchers attack reinforcement learning efficiency   In VenturebeatBy Kyle Wiggers

Reinforcement learning, which spurs AI to complete goals using rewards or punishments, is a form of training that’s led to gains in robotics, speech synthesis, and more. Unfortunately, it’s data-intensive, which motivated research teams — one from Google Brain (one of Google’s AI research divisions) and the other from Alphabet’s DeepMind — to prototype more efficient means of executing it. In a pair of preprint papers, the researchers propose  (technical paper) Adaptive Behavior Policy Sharing (ABPS)  , an algorithm that allows the sharing of experience adaptively selected from a pool of AI agents, and a framework — Universal Value Function Approximators (UVFA) — that simultaneously learns directed exploration policies with the same AI, with different trade-offs between exploration and exploitation.  .... "

Alphabet X Everyday Robot

Early tasks will have these robots sorting trash.   In practice a tough job, with vision and other sensory aspects.   Learning also means adapting, so the hardest thing might be to adapt to a changing stream.    Adapting is a form of maintenance, and every ML project I have worked has under designed that part of the problem.   So looking forward to more from this effort.

Alphabet X’s “Everyday Robot” project is making machines that learn as they go in Technology Review

The news: Alphabet X, the company’s early research and development division, has unveiled the Everyday Robot project, whose aim is to develop a “general-purpose learning robot.” The idea is to equip robots with cameras and complex machine-learning software, letting them observe the world around them and learn from it without needing to be taught every potential situation they may encounter.   ....  "

Drones are Filling the Skies

A broad piece in the WSJ on how major players are examining, testing and using drones.  And the methods they are using to make this all work.

The Drones Are Coming! How Amazon, Alphabet, Uber Are Taking to the Skies
The Wall Street Journal
By Sebastian Herrera; Alberto Cervantes
October 25, 2019

Companies including Amazon, Alphabet's Wing, and Uber, are launching more advanced trials of drone delivery. Wing started tests in Christiansburg, VA, this month, while Uber will set up experiments in San Diego before the end of the year. Amazon said last June it would begin delivering packages to consumers via drone "within months." The companies have to overcome a number of obstructions and concerns before drone delivery can become widespread. Amazon uses machine learning algorithms and infrared sensors to detect obstacles like birds and wires, and programs its drones with scenarios (such as when a delivery location cannot be detected), and commands to follow in such scenarios. Wing, meanwhile, has tested its drone north of Helsinki under snowy and windy conditions; its drone has built-in wind sensors and is waterproof. A challenge that remains is that no standard exists on how drones can identify and communicate with each other while in flight, so drone delivery by multiple companies in the same area is not currently possible. ... ' 

New Reinforcement Training Frameworks

Good to see the kind of open source development being done.

DeepMind Has Quietly Open Sourced Three New Impressive Reinforcement Learning Frameworks
Three new releases that will help researchers streamline the implementation of reinforcement learning programs   By Jesus Rodriguez

Deep reinforcement learning(DRL) has been at the center of some of the biggest breakthroughs of artificial intelligence(AI) in the last few years. However, despite all its progress, DRL methods remain incredibly difficult to apply in mainstream solutions given the lack of tooling and libraries. 

Consequently, DRL remains mostly a research activity that hasn’t seen a lot of adoption into real world machine learning solutions. Addressing that problem requires better tools and frameworks. Among the current generation of artificial intelligence(AI) leaders, DeepMind stands alone as the company that has done the most to advance DRL research and development. Recently, the Alphabet subsidiary has been releasing a series of new open source technologies that can help to streamline the adoption of DRL methods.  ... " 

AI Drug Hunting for Pharma

Note the mention of simulations as a means of determining the effectiveness of prospective drugs.

AI Drug Hunters Could Give Big Pharma a Run for Its Money 
Bloomberg
By Robert Langreth
July 15, 2019

Using the latest neural-network algorithms, DeepMind, the artificial intelligence (AI) arm of Alphabet, beat seasoned biologists at 50 top labs from around the world in predicting the shapes of proteins. The company's win at the CASP13 meeting in Mexico in December has serious implications, as a tool able to accurately model protein structures could speed up the development of new drugs. Although DeepMind's simulation was unable to produce the atomic-level resolution necessary for drug discovery, its victory points to the potential for practical application of AI in one of the most expensive and failure-prone parts of the pharmaceutical business. AI could be used, for example, to scan millions of high-resolution cellular images to identify therapies researchers might otherwise have missed. In the short term, experts say AI-based simulations likely will be used to determine whether prospective drugs will be effective before proceeding to a full clinical trial. .... "

Wing Delivers from Drones in Australia

Have not seen it in the US except for specialty examples, but Alphabet is doing it in Australia.

Wing Officially Launches Australian Drone Delivery Service in IEEE Spectrum

After years of testing, Wing is now offering consumer drone delivery to select Australian suburbs   By Evan Ackerman

This drone can deliver your morning coffee directly to your house. But is that something people really want?

Alphabet’s subsidiary Wing announced this week that it has officially launched a commercial drone delivery service “to a limited set of eligible homes in the suburbs of Crace, Palmerston and Franklin,” which are just north of Canberra, in Australia. Wing’s drones are able to drop a variety of small products, including coffee, food, and pharmacy items, shuttling them from local stores to customers’ backyards within minutes.    ..... "

Protein Folding with Alphabet Deep Mind

A big, big deal we took a look at for industry, even suggested a neural net possibility, but methods were still too primitive at the time. This still not a compete solution,  but looks to be a step forward.

Alphabet's DeepMind AI Algorithm Wins Protein-Folding Contest 
V3.co.uk   By Dev Kundaliya

DeepMind's latest artificial intelligence (AI) software won the Protein Structure Prediction Center's Critical Assessment of Structure Prediction contest by accurately predicting the three-dimensional structures into which proteins can be folded. The AlphaFold algorithm predicted the configurations of 25 out of 43 proteins, making it far more accurate than any other software. AlphaFold was designed and taught to model target shapes from scratch, without using previously solved proteins as templates. The DeepMind team used two distinct neural networks to predict the proteins' structures. DeepMind's Demis Hassabis said, "We've not solved the protein folding problem, this is just a first step. It's a hugely challenging problem, but we have a good system and we have a ton of ideas we haven't implemented yet."    ... " 

Match Retail Search via Card Transactions

Remember discussing how such a thing could be done, and how the approach could bring along lots of additional information.  Probably a privacy violation, we concluded it was not possible.  But then again Amazon now has a Prime Card,  which is busily racking up purchasing data.   See much more below.

Google and Mastercard cut a secret deal to track retail sales data      By Mark Bergen and Jennifer Surane  in Information Management 

(Bloomberg) -- For the past year, select Google advertisers have had access to a potent new tool to track whether the ads they ran online led to a sale at a physical store in the U.S. That insight came thanks in part to a stockpile of Mastercard transactions that Google paid for.

But most of the two billion Mastercard holders aren’t aware of this behind-the-scenes tracking. That’s because the companies never told the public about the arrangement.

Alphabet Inc.’s Google and Mastercard Inc. brokered a business partnership during about four years of negotiations, according to four people with knowledge of the deal, three of whom worked on it directly. The alliance gave Google an unprecedented asset for measuring retail spending, part of the search giant’s strategy to fortify its primary business against onslaughts from Amazon.com Inc. and others.

But the deal, which has not been previously reported, could raise broader privacy concerns about how much consumer data technology companies like Google quietly absorb...."

Waymo Training Cars

Inside Waymo's Secret World for Training Self-Driving Cars

An exclusive look at how Alphabet understands its most ambitious artificial intelligence project

Alexis C. Madrigal ....

Google Going AI First

All the indications are there, the data, especially dynamic data, is broad and deep.  What should business reactions be?  Open source capabilities, such as TensorFlow,  could lead to better sharing of AI engines.

Just Leaked: Google Preps to Invest Boldly in Artificial Intelligence   
As Google goes AI-first, here's the latest development.    By Lisa Calhoun ,  In Inc. 

As reported first by Axios and then Tech Crunch, Google is doubling down on its ability to invest in artificial intelligence. Alphabet recently announced an AI-first strategy at Google I/O. Now, it's reportedly building out a group of investors that will invest exclusively in AI. It's the first time Google has made a move to invest in an exclusive branch of technology. Google hasn't confirmed or denied the news, but a business unit that invests exclusively in AI makes sense for Google right now for a number of reasons:

· A lean team of engineers can invest in the venture frontier more effectively than a larger, classic corporate venture unit. GV, Alphabet's corporate venture arm, has over 300 investments like Nest, Ionic Security, Uber, and 23&Me. That's not only a large portfolio to manage, it's also a broad mandate that spans consumer goods to life sciences. Artificial intelligence is a hot space for the entire venture capital community. Over a billion dollars were invested in AI last year. Just take a look at key artificial intelligence deals to get a sense of the scope. So having a group that focuses on it specifically? Smart. .... "