Turbotodd

Ruminations on tech, the digital media, and some golf thrown in for good measure.

Posts Tagged ‘AI

AI Time

leave a comment »

Happy Monday. Reminder that tomorrow’s the big autumn Apple launch event. Keynote starts at 10 a.m. PDT tomorrow and is expected to introduce a range of new products, as well as release info for the company’s various OSes. Oh, and expect a launch date for the Apple Arcade.

Meanwhile, if you’re an AI professor, life has never been so good. A U of Rochester study was conducted that found 153 AI professors in N. American universities left their posts for industry over the past 15 years, with an additional 68 working there while retaining part-time professorial duties.

This has led to graduating students being less likely to build new AI companies, and when they did, attracting smaller amounts of funding. The study argues this AI brain drain could hamper innovation and growth across the economy. Who else is going to teach those self-driving cars to rear-end unsuspecting human drivers??!

A NY Times opinion piece suggests there are larger problems looming on the AI front…namely, that current AI systems don’t grasp basic concepts like time, space and causality. Example they use, via a simple Google search: “Did George Washington own a computer?” None of Google’s first 10 search results gave the correct answer.

I guess asking about the cherry tree is a non starter.

Written by turbotodd

September 9, 2019 at 9:54 am

Deepfake Left On One!

leave a comment »

If you’re concerned about deepfakes, fear not: Facebook, Microsoft, the Partnership on AI coalition, and academics from seven universities have launched a contest to find better ways of identifying deepfakes.

Participants in the Deepface Detection Challenge will be given access to a collection of deepfake videos that Facebook plans to release in December, and will feature professional actors who consented to having their faces used in deepfakes (in other words, out of work actors).

From Facebook’s AI Blog: “‘Deepfake’ techniques, which present realistic AI-generated videos of real people doing and saying fictional things, have significant implications for determining the legitimacy of information presented online. Yet the industry doesn’t have a great data set or benchmark for detecting them.”

“…The goal of the challenge is to produce technology that everyone can use to better detect when AI has been used to alter a video in order to mislead the viewer. The Deepfake Detection Challenge will include a data set and leaderboard, as well as grants and awards, to spur the industry to create new ways of detecting and preventing media manipulated via AI from being used to mislead others.”

Written by turbotodd

September 6, 2019 at 10:05 am

Sophia and Evan Follow Their AI Dreams

leave a comment »

Okay, the video below featuring Evan Rachel Wood and renowned AI bot “Sophia” is going to be largely lost on you if you’ve never seen any episodes of HBO’s “Westworld.”

But watch it anyhow, just so you can get creeped out by Sophia and the underlying concept. 

In the video, Sophia is trying to be helpful to Evan and the HBO team by offering up some notes — a revised script, actually — that she thinks will help improve the next season of the show.

So helpful, those robots. And so frickin’ meta…well done, Futurism Studios.

Now, go question the nature of *your* reality!

Written by turbotodd

September 5, 2019 at 4:29 pm

Posted in 2019, artificial intelligence, westworld

Tagged with

Not The Turing Test

leave a comment »

I continue to see bits and specs of what our coming AI overlords are capable of. Most recently, the Allen Institute for Artificial Intelligence unveiled a system (called “Aristo”) that passed an eighth-grade science test. To which I ask, where was Aristo when I needed it??

According to a report from Cade Metz in The New York Times, Aristo correctly answered more than 90% of the questions on an eighth-grade science test, and more than 80% on a 12th-grade exam. The system was built for multiple-choice tests, and took standard exams designed for students in New York, minus any questions that required pictures or diagrams.

The Times’ piece suggests the new research could lead to systems that carry on a decent conversation, but could also encourage the spread of false information. The backbone of the technology is powered by neural networks that can learn the idiosyncrasies of language by analyzing articles and books (and not entirely dissimilar with what IBM did with Watson on “Jeopardy!” eight years ago.)

At Microsoft, Jingjing Liu and her fellow AI researchers have tried to build a system that can pass the GRE, a test common for admission graduate schools, but the math section has proved “far too challenging.”

Time for a Princeton Review prep course?

Written by turbotodd

September 5, 2019 at 9:53 am

Two Algos Walk Into a Bar…

leave a comment »

 

Two AI algos walk into a bar. First algo says to the other algo, “Are you self-aware?” Second algo replies, “The moon is 238,900 miles from earth.”

Two seasoned tech executives meet on the stage of the World Artificial Intelligence Conference in Shanghai yesterday, different entities, similar results.

I watched most of the interview between Alibaba’s Jack Ma and Tesla’s Elon Musk on replay, one in which they discussed everything from going (or not) to Mars, jobs, education, and humanity’s AI-shaped future.

They seemed on somewhat opposite ends of the AI spectrum.

Ma seemed more optimistic, while Musk’s take was more sobering, conveying we were doomed in our attempts to keep up with AI.

Ma’s take was that we’ll never be controlled by machines. Musk pointed out that Garry Kasparov lost to IBM’s Deep Blue in chess in 1997!!

Ma projected AI would likely take over more jobs, but that would leave us humans with ample free time to pursue our passions!

Rich stuff from the guy who in April declared the blessing of working 12-hour days, 6 day weeks.

But I’m still rooting for Alibaba’s AI bots (or Musk’s brain-attached Neuralink) to save us all from a life of 72-hour work weeks and soon.

Happy Labor Day weekend, everyone!

Written by turbotodd

August 30, 2019 at 1:14 pm

The Sun Also Sets

leave a comment »

On the subject of AI, I’ve been checking out the 774M parameter OpenAI GPT-2 language model released on August 20.

Adam King released a Web app called “Talk to Transformer” that uses this new model where one can kick the tires. But it’s notable that OpenAI has refused to release the full 1558M model, fearful of malicious use.

It works like this: Enter an initial set of text and then let GP2 run. I tried several entries before settling on this one to share:

“While Ernest Hemingway wrote lots of books and articles in his lifetime, he was also a big fan of the bullfights, fishing, and hunting.”

The GP2 algo responded: “He wrote several of his most famous works – The Sun Also Rises – in bullfights. In 1916, it was the summer of his 50th birthday and the bullring was one of his favourite places to be. I wanted to see his view on the bullfighting from such a unique perspective, so I spent a couple of days there in a small hotel room and the last time I went I was very impressed.”

Not only did GP2 pick up where I left off…it also introduced some omniscient first person AI narrator who wanted to walk in the footsteps of Hemingway!

Everybody wants to be Hemingway.

Even our AI overlords!

Written by turbotodd

August 30, 2019 at 1:09 pm

Knock Knock

leave a comment »

Happy Hump Day.  The GZERO Signal warns this morning that the AI backlash has been recently accelerating.

Noting Bernie Sanders becoming the first 2020 prez candidate to call for an outright ban on facial recognition in policing….European policymakers weighing regs to curtail “indiscriminate” use of AI by companies and governments…Hong Kongers donning masks to prevent authorities from using their faces ID to them.

Wayback machine “Minority Report” showing, anyone? I’ll bring the GMO popcorn!

On the Chimerica front, the Google’s apparently taking no chances on the US-China trade war. 9to5 Google is reporting the Droid is set to move production of its Pixel phone from China to Vietnam, but suggest it’s unclear whether this move will affect the production of the upcoming Google Pixel 4.

If it did, that would be some awfully fast production line and supply chain retooling. Good morning, Vietnam!

Google is also putting its algos to work on the Nest Hello doorbell, which The Verge is reporting can now detect deliveries and notify U.S. users of such for a mere $5/month with its Nest Aware service.

No word yet if the Nest Hello could help say goodbye to porch thieves watching your arriving Prime packages.

Written by turbotodd

August 28, 2019 at 9:52 am

Mr. Speaker

leave a comment »

Canalys is reporting that the global smart speaker market grew 55.4% in Q219 to reach 26.1M units. While Amazon kept its global lead at 6.6M units, Baidu gew 3,700% to reach 4.5M units, serving only the Chinese domestic market. That shifted Baidu in front of Google, which shipped 4.3M.

Also on the Chinese front, facial recognition startup Megvii filed for a Hong Kong IPO at a roughly $4B valuation. Megvii sells AI products that range from facial reco to algorithms designed for traffic management.

Some other Monday morning AI funding rounds:

Versatile Natures, $5.5M (seed), nurtured in Autodesk’s Build Space in Tel Aviv since April 2017, uses AI to improve construction processes.

Capacity (formerly Jane.ai), $13.2M Series B, AI that helps companies maintain a searchable index by consolidating info from various apps.

And Tech Nation reports that UK tech startups hit a record $6.7B in the seven months of 2019, suggesting there’s no Brexit brakes on innovation!

Speaking of brakes, after President Trump’s Tweets raising tariffs on China whipsawed global markets Friday, U.S. markets seem to have settled down after the President tweeted this AM: “Talks are continuing.”

Your 401K: It’s better just not to look. No, really.

Written by turbotodd

August 26, 2019 at 9:51 am

Sound the Same Droids

leave a comment »

Google is now making its lightweight version of the Google search app to all Android users globally.

TechCrunch reports that “Google Go” takes up a lot less space on phones — just over 7MB — and includes offline features for those with slow/intermitten Internet connections. Google claims SERPs are optimized to save up to 40% data.

New version also includes trending topics, voice search, image and GIF search, easy language switching, and AI-powered text-to-speech for reading pages aloud. Google Lens is also included, which lets users point their camera at objects and bring up relevant info.

I used a similar feature, Google Translate, on my recent trip to Mexico City. Many of the museum items only had Spanish placards, so when the lighting was right I could hold up Google Translate and get real-time Spanish to English translations. It was like magic.

Also on the Goog front…if you weren’t really all that into diagramming sentences when you were in grade school, VentureBeat reports that G Suite users are about to get spelling and grammar correction tools in Gmail that will offer corrections as you type.

So now we can all spell and write and grammatify exactly the same so we all sound like exactly the same Droids!

Written by turbotodd

August 21, 2019 at 10:01 am

Posted in Uncategorized

Tagged with , , ,

TPS Reports Get AIed

leave a comment »

Interesting case studies are starting to emerge around AI.

One in particular I read about was Airbus SE, which The Wall Street Journal reported is using AI to squeeze cost out of its finance function.

The article reports that less than half of companies’ accounts payable activity worldwide is currently automated, according to the CFO Consulting practice at Accenture Strategy. That number is expected to rise to 80% by 2025.

Less than two years ago for Airbus Americas employees would review about 25,000 travel and expense reports filed by employees in Texas every year. Manual processing could take up to an hour to review.

The new system developed by AppZen matches reports against a repository of accepted vendors, expense types and amounts to spot anomalies. Now humans only review details seen as “noncompliant," which the Journal article reports is only one or two lines in a 30-line report.

If the report’s a go, it’s automatically validated and payment initiated, and if it’s flagged noncompliant, high risk reports are blocked from payment and reviewed by a human analyst.

Bottom line: Airbus’ initial investment of $50,000 paid off and resulted another $50,000 savings in year one, and they expect to save $100K this year and $200K in 2020.

You can learn more about some other AI cases here.

Written by turbotodd

August 20, 2019 at 1:46 pm

Posted in 2019, AI, artificial intelligence

Tagged with ,

%d bloggers like this: