Round Up Rotating Arrow Heading Up

Tom Austin’s Tweet-Wrangling Newsletter #2

This tweet wrangler is specific to tweets I’ve posted[1] in the past few weeks as @tomaustin. Other collections are in the works as well (see Item7 below.) Why a roundup like this?

Tweets can serve as a lens through which to examine a topic. One can’t say much in a single tweet. Putting them together in a coherent newsletter can be more efficient and effective for you. It also lets the author interweave content from other sources and articles where that extra material further enhances the content. You’ll find examples below.

There are eight items in this issue:

  1. AI patience or outright skepticism?
  2. Is AI smart enough?
  3. A great advance in mapping connections in the brain (of fruit flies)
  4. Facial Recognition Accuracy, Clearview, and Other Privacy Issues
  5. How some of the big commercial AI players are evolving
  6. Financial applications – FinTech
  7. Future Analyst Syndicate Newsletters
  8. My Travel Plans – Meet me in San Jose or San Francisco

Item 1: AI patience or outright skepticism?

A Financial Times opinion piece[2] headlined “AI tech smarts” said, “While few sophisticated commercial applications have yet emerged, patience may be wiser than outright skepticism.” Similarly, a Forbes article on 28 January said, “AI is still a science project at most companies.”[3]

My take: I don’t see this as an either-or issue. You need large dollops of both patience and outright skepticism.

Since mid-2016 – early-2017, I have been tempering my earlier enthusiasm for the great AI research results of this decade because they weren’t translating well into production applications.

Surveys that I trust report a disappointing level of enterprise production investment on AI. The fault is not due to a skills shortage (albeit, there is one). Hype-inspired premature investments are driving a significant part of the skills shortage because there’s an application shortage.

Too much experimenting and pilots. Not enough commercially available applications exploiting AI.

Here’s skill demand data.

USA Job postings by AI related skill

Per “AI Winter Is Coming”[2] in November 2019:

  • The use of products exploiting less visible, ingrained AI techs will grow dramatically. Over 90% of enterprise AI use will be via embedded AI. Embedded techs are ingrained, integral to the products that they’re woven into, part of the products’ fabric. They boost the value the products deliver and don’t require buyers to commit to significant development projects.
  • AI technologies embedded in the latest versions of business application suites and enterprise search are the best examples. Their embedding makes users smarter, more productive.

That’s the good news.

The bad?

Most surveys on AI adoption are distorting reality. They’re biased in a way that makes adoption appear more robust than it is.

There’s a lot of AI experimentation and piloting (E&P) going on, particularly but not exclusively in large organizations. Much of it isn’t turning into production implementations. So why do market researchers mash together E&P investments and production implementation investments? Mashing the two categories together makes it seem like there’s more investment going on than there really is. Consider two different ways I could describe a survey outcome:

    1. 50 percent of respondents say their enterprise is investing in projects that exploit AI.
    2. 50 percent of respondents say their enterprise is experimenting with and piloting projects that exploit AI; only five percent have put these projects into production.

Whether by design or not, the omission of the E&P detail from the first statement is misleading. It speeds up the pace of investment which generates more demand for expertise which further exacerbates the skills shortages in the labor market.

By the way, I was pleasantly blown away by the degree to which IBM’s Rob Thomas very candidly said, in October 2019

Today there is only a 4%-8% adoption rate of AI within the corporate world.

In “Pole Vaulting Unicorn Wins Olympic Gold,”[3] I said I don’t see the likes of Siri, Alexa, Insight Engines, Convolutional Neural Nets, Deep Neural Networks, Generative Adversarial Nets, Word Embeddings and all the rest disappearing.

But the fever is cooling, and the market is waiting. If we’re patient and lucky, we’ll see another AI boom cycle by 2023.

Item 2: Is AI smart enough?

Yann LeCun (Turing Award Winner and Facebook Chief AI scientist) doesn’t think so. He was quoted in the Wall Street Journal[4] as saying no:

I would declare victory if in my professional lifetime we could make machines that are as intelligent as a rat.

To which I opined, I think LeCun is aiming too high!

We’re really come a long way in the last decade or two but none of the AI technologies around are really smart or intelligent in a human way – or in the way rats, goldfish or, perhaps, fire ants are.  As LeCun has been repeating in all his major presentations, AI has no common sense. (And the research community agrees pretty well with that sentiment.)

So watch your assumptions!

Since we’re talking about animal smarts, let’s turn to the next item.

Item 3: A great advance in mapping connections in the brain (of fruit flies)

I wrote a Tweet[5], a LinkedIn post[6], and an in-depth article “The rest of the story: fruit fly brain research challenges singularity predictions” about the import of the brain connections research.

When I first saw the story, I opined

Brilliant work! Still eons before scientists build complete connectomes for a human brain. It will still not be enough to reconstruct how human behavior functions…

Too many confusing metaphors. Too little understanding of brain sciences or AI.

A connection map depicts anatomical features, not functions. It’s neuroanatomy, not neurophysiology. And it doesn’t add any credence to the Kurzweil singularity hypothesis either. Read more of my analysis in my new article, The rest of the story: fruit fly brain research challenges singularity predictions. It discusses some of the flaws in the singularity brain-computer metaphor. [7]

Item 4: Facial Recognition Accuracy, Clearview[8],[9] and Other Privacy Issues[10]

NIST (the US National Institute of Standards and Technology) published in hard data[11] on commercially available facial recognition technologies. It documents and quantifies major improvements in the technology.

More ominously, under a headline reading “The Secretive Company That Might End Privacy As We Know It[12]” The New York Times broke the story on Clearview’s facial recognition system. They wrote:

You take a picture of a person, upload it and get to see public photos of that person, along with links to where those photos appeared. The system — whose backbone is a database of more than three billion images that Clearview claims to have scraped from Facebook, YouTube, Venmo, and millions of other websites — goes far beyond anything ever constructed by the United States government or Silicon Valley giants.

Federal and state law enforcement officers said that while they had only limited knowledge of how Clearview works and who is behind it, they had used its app to help solve shoplifting, identity theft, credit card fraud, murder and child sexual exploitation cases.

There’s reason to believe that Clearview’s scraping images this way violates the terms of service of all their primary image sources. Citing another New York Times article[13]:

Twitter sent a letter this week to … Clearview AI, demanding that it stop taking photos and any other data from the social media website “for any reason” and delete any data that it previously collected … The cease-and-desist letter, sent on Tuesday, accused Clearview of violating Twitter’s policies.

I’m still waiting to see Facebook take a similar move.

But what problems does this present for most enterprises?

Are any of your lines of business using Clearview? Has it been cleared by purchasing? Legal? senior executives?

More broadly, what are your policies on employees taking pictures at work-related events? Posting those pictures on Facebook and other sites? Including pictures of facilities, documents, customers, suppliers, and others in their photostream?

In the same vein, we learned via Reuters that Apple dropped a plan for encrypting backups after the FBI complained[14]. What does that do for Apple’s claim that user privacy is paramount?

These are messy times!

    • Clean up your policies.
    • Ensure your employees know what should be off-limits.
    • Enforce the policies.
    • Add governance checks.

As a segue to the next big item, Amazon has an alternative to facial recognition for payment in retail and other locations: Paying with a hand wave[15]. Your handprint is less visible in public and potentially less troubling than facial recognition.

Item 5: How some of the big commercial AI players are evolving

Jeff Dean SVP and Senior Fellow at Google published an extensive article entitled “Google Research: Looking Back at 2019, and Forward to 2020 and Beyond[16]

It’s a valuable narrative. Google submits more work to the leading AI conferences than anyone else, at least in the US. And it winds up with more acceptances as well. So look at Jeff’s article. And take a look at the ongoing AI research at others, as well, including Microsoft[17], Amazon[18], IBM[19]. And let’s not forget Facebook AI Research (FAIR)[20].

Item 6: financial applications – FinTech

The Financial Times ran a story, “Fund managers must embrace AI disruption[21]” the day after the Wall Street Journal went with “Use AI for picking stocks? Not so fast.[22]

My reaction to the WSJ article was, “Why isn’t this common wisdom yet? Perhaps too much hype, not enough data transparency.”

So which way should you (and others) gravitate? Was I prematurely agreeing with the WSJ author’s take?

In a recent article entitled “Anyone can make a prediction but are they any good?[23]” I looked briefly at the predictive abilities of AI and concluded:

AI is very good at accurately predicting near-term outcomes (as in High-Frequency-Trading), but its predictive capabilities deteriorate as the future time-period gets further out.  After all, AI is trained on historical data. There are no future data to train it with!

You might say the same thing about the weather. The further in the future you look, more the predictive capabilities of weather models deteriorate too.

Item 7: Future Analyst Syndicate Newsletters

We pride ourselves on being independent analysts – we value diversity in opinion. We provide feedback to each other to help everyone sharpen their perspectives but we don’t try to homogenize our positions.

There’s far more to our work than AI! Expect more:

    • Tweet RoundUp Newsletters (Analyst Driven)
    • Topical or Event-driven RoundUp Newsletters
    • Single-author Newsletters
    • Community Newsletters
    • State of the Syndicate Updates

In a little more detail, expect

    • Tweet RoundUp Newsletters, like this one, not only on AI (there’s far more to our work than AI) but also many coverage areas across a broad range of topics[24] and, at times, different opinions on the same subject matter.
    • Topical RoundUp Newsletters like last month’s Predictions 2020 Round Up [25]
    • And more – what would you like to see?

Item 8: My Travel Plans – Meet me in San Jose or San Francisco

I am planning to be at Nvidia’s GTC event in San Jose, CA in March and IBM’s Think 2020 in San Francisco in May. If you want to talk with me at either event, reach me at tom@thansyn.com or, better yet, use https://www.calendly.com/tom-austin to book a 15-minute slot on my event calendar.

In anticipating of GTC, I commented on Twitter[26] on a promo for “Left Hand Robotics SnowBot Pro Self-Driving Snow Clearing Robot” saying “Commercial Snow Job! Where’s the vision system? How does it see my dog before running it over?” I’m hoping to see it at GTC.

 

[1] https://twitter.com/tomaustin

[2] https://twitter.com/tomaustin/status/1221164601955536897

[3] https://twitter.com/tomaustin/status/1222615616064049152

[4] https://twitter.com/tomaustin/status/1217092698794053632

[5] https://twitter.com/tomaustin/status/1222610364598308864

[6] https://www.linkedin.com/feed/update/urn:li:activity:6628375990970175488/

[7] https://thansyn.com/the-rest-of-the-story-fruit-fly-brain-research-challenges-singularity-predictions/

[8] https://twitter.com/tomaustin/status/1218915654755143680

[9] https://twitter.com/tomaustin/status/1220362224394477569

[10] https://twitter.com/tomaustin/status/1219637874028634118

[11] https://twitter.com/tomaustin/status/1216674567495675904

[12] https://www.nytimes.com/2020/01/18/technology/clearview-privacy-facial-recognition.html

[13] https://www.nytimes.com/2020/01/22/technology/clearview-ai-twitter-letter.html?smid=nytcore-ios-share

[14] https://www.reuters.com/article/us-apple-fbi-icloud-exclusive/exclusive-apple-dropped-plan-for-encrypting-backups-after-fbi-complained-sources-idUSKBN1ZK1CT

[15] https://twitter.com/tomaustin/status/1219351462897160198

[16] https://twitter.com/tomaustin/status/1219340314529452039

[17] https://www.microsoft.com/en-us/research/research-area/artificial-intelligence/?facet%5Btax%5D%5Bmsr-research-area%5D%5B0%5D=13556&sort_by=most-recent

[18] https://blog.aboutamazon.com/amazon-ai

[19] https://www.ibm.com/blogs/research/category/ai/

[20] https://ai.facebook.com/blog/facebook-ai-year-in-review-2019/

[21] https://twitter.com/tomaustin/status/1217457135086440449

[22] https://twitter.com/tomaustin/status/1217139425613963265

[23] https://thansyn.com/anyone-can-make-a-prediction-but-how-good-are-they/

[24] https://thansyn.com/featured-topics/

[25] https://thansyn.com/predictions-2020-round-up/

[26] https://twitter.com/tomaustin/status/1220770210745147394

 

Disclosure

The views and opinions in this analysis are my own and do not represent positions or opinions of The Analyst Syndicate. Read more on the Disclosure Policy.

Leave a Reply