Liveblogging the Bloomberg session on AI

Noah Feldman is moderator, Noah Smith and Justin Fox and Leonid Bershidsky and Shira Ovide and Dina Bass are presenting…you will find my live-blogging in the comments section of this post.  Just keep on refreshing for the updates.

Livestream is here:


The Bloomberg building I always find so beautiful and now we are up on the 28th floor, with wonderful views. But what about AI?
Use Noah Feldman as your moderator, he was great yesterday and they are (wisely) using him again. Some of this is going to be stream of consciousness, in case you hadn't figured that out.

Fraud investigations are one of the areas of biggest impact for AI, says Shira Ovide. The non-exotic uses may prove most important.

And Bloomberg News itself has automated news stories, these seem relatively easy for some kinds of business stories, which fit a common format. But when is the totalitarianism coming some of you are wondering!? Maybe never!

"Businesses move their automated surplus worker to higher level tasks" .. No, not in my experience.

They add lines and move the workers to cover without hiring in an expanding industry / market. In a static market, they start using Temps. Assumably there are less Temps over time.

Nor has anyone lost his or her job yet, at least not in this session. None of the session comments are automatable....Noah Feldman just reminded us there is a lot of advertising in Huxley's *Brave New World*. Noah Smith is now recommending the book *Prediction Machines*, from U. of Toronto.

Tyler left out the setup.

A panel member said (paraphrased), "AI is mostly being used in advertising. I don't recall reading any science fiction that predicted that would be it's use."

"Noah Feldman just reminded us there is a lot of advertising in Huxley's *Brave New World*. "

+1, it was a great comment. Though to be fair, "Brave New World" didn't really have AI.

And the 'advertising' was roughly along the lines of the 'advertising' in 1984.

Brave New World was published in 1932. Huxley had been living in Mussolini's Italy. It was more heavily oriented around brainwashing and corporate advertising. 1984 was published in 1948 and was heavily oriented around explaining the Soviet show-trials to the Stalin supporting Left. Dictatorships tend to have a fairly primitive form of propaganda, as they can mostly just tell people what they have to think. Huxley was more concerned with how propaganda could lead to a totalitarian dictatorship in a bastion of democracy like England. The methods and importance of propaganda leading up to and during WWII became a huge issue in society and academics, leading eventually to both Chomsky and deconstructionism. Compared to Orwell, Huxley presented a perspective much more relevant to more modern developments. Huxley expanded on the propaganda/persuasion methods touched on in Brave New World with the 1958 book/booklet Brave New World Revisited.

Will the robots take "half of your job," or "half of your jobs" -- from Noah S., he is not in fact a man of "no opinion"!

Shira Ovdie came to this panel from jury duty. Noah Smith is continuing to sound fairly optimistic. Who will first scream "Where is the totalitarianism?" I am waiting! And now Noah is saying that the cost disease is not as bad as you think, it also means you can afford the higher cost sectors, Alex T. makes similar points I might add.

"There's not a lot of automation in services..."

Really? Ask the travel agents, video store clerks & operators about that.

And toll booth clerks, bank clerks hand keying checks every afternoon, FedEx package sorters, elevator operators, people taking your money at gas pumps, manual type setters, ... it’s a long list

Not to mention stock brokers! (floor traders)

My gosh, how does TC do it? He's a multitasker. I sent him an email extolling a stock and he replied, within moments! And all the while he's Live blogging something at Bloomberg?

Hot stock tip: buy penny stocks Freddie Mac and Fannie Mae, since some articles online say eventually they'll be allowed to privatize again. Hedge fund guru Bill Ackman has a position. Ten-fold profit, you read it here first.

Wait, Clive Crook is sitting up there and talking! He wasn't on the program...he is stressing the speed of adjustment as a major factor. No hook has pulled him off the stage.

Of course the big paradox is why all these AI innovations do not really show up in the productivity numbers. It's not like free Wikipedia pages that somehow don't get picked up by the extant statistics.

The average person uses a i doing voice input like this, or saying breakfast things to their Alexa, and of course driving. Of those only the last one my potentially impact productivity.

No Corrections, showing a high for what it is.

Why is it a paradox? Most historians say it takes about almost two generations (50 years) for inventions to make their mark on most households. Electricity and electric motors took about that much time, ditto the automobile, telephone and so forth. Even allowing for 'faster stuff happening in the modern era' you're looking at probably 25 years before adoption by the masses. One reason I'm for a longer patent term for pioneer inventions.

The diffusion of technology has been accelerating. Do you think it will take 50 years for the VCR, home computer or cell phone "to make its mark on households"?

@Todd Kreider - it depends on the tech in question. For pioneering tech, yes, for trivial stuff, no. BTW, household uptakes of the VCR have collapsed, the curve is heading back to zero.

Out of curiosity, what technology now takes even 25 years to diffuse to most households? smart phone: 2005 to 2018 and over half the world has one - 13 years. Kindle diffusion about 10 years. Virtual reality really began in 2015 and will be everywhere in 2028, 13 years.

I recently heard that

Because all this stuff is a way of doing things that you didn't do before. Not productive stuff; the manager wants hourly reactive graphs of all kinds of stuff. They couldn't do it before, now they can, but nothing gets sold or value added.

Clive is now wondering when the robots will take over the jobs of the columnists, such as himself...

Noah points out there has not been a decrease in the hiring of radiologists. The food at Bloomberg is always very good by the way -- just don't expect too much candy!

There hasn't been a decrease in radiologists because 1) It is only 2018 2) The AMA exists. Check back in 2023 and 2028 to see how many radiologists there are.

Worth posting again:

"I think if you work as a radiologist, you're like the coyote that's already over the edge of the cliff, but hasn't yet looked down and so doesn't realize that there's no ground underneath him. They should stop training radiologists now. It’s just completely obvious within five years that deep learning is going to do better than radiologists. It might take ten years, but we’ve got plenty of radiologists already. I said this at a hospital, and it didn’t go down to well.”

— deep learning expert Geoff Hinton, 2016

True. I figured that radiology was a declining industry when about 15 years ago there was a dubious albeit legal medical company that was encouraging people to get chest x-rays. The reason? Most x-rays will show some little spot, and a significant percentage of the population will opt for follow up treatment, even exploratory surgery, which brings in big bucks to the doctors. Some moderate doctors pointed out that if you're healthy, probably such an x-ray is overkill, no pun intended.

Cue to human-machine synthesis point...

Cue to Trolley Problem for driverless cars point...

Good one! The Trolley Problem of philosophy no longer becomes an agonizing moral problem but a simple min/max operations research Simplex solution! A trivial one at that, Jeremy Bentham would be pleased!

"Radioligists aren't going..."

Exactly my point above. Automation in an expanding industry, results in the humans being more specialized (fewer humans on more lines). It does not cause a drop, but instead a downward pressure in hiring.

@JWatts- not historically, but in abstract theory you are right. Historically, automation actually increases employment overall (the guys that serviced horse and buggies became automobile mechanics).

Bonus trivia: during the end of the horse and buggy era, carriages became longer and longer, and further away from the team of horses pulling them, some up to sixty feet long! Popular in Cuba, so that rich 1% families (like mine) could stay further away from the horse's ass, and also it was a status symbol when on the road. They also 'paired up' stage coach wagons for efficiency reasons, not unlike 'dual tractor trailers' today, to carry more cargo.

Actually, I intended my comment more in the short to medium term. IE The existing stock of radioligists won't all be fired in the next 15 years as automation increases. Instead new hiring will start to decline and the existing pool of radioligists will be tasked with overseeing a growing pool of work augmented by automation.

"@JWatts- not historically,"

Yes, looked at over the long run, say the next 50 years. There might not even be such a thing as a radioligists.

Leonid says AI art is still crap and the auction houses are selling the hype -- I agree.

"They are trading on a lot of hype in AI"...

Leonid says if AI is so great, why do you sometimes get a tiny item in a huge Amazon box -- I have wondered the same!

Because in packaging, 'one size fits all', more efficient. For example, notice the bottlenecks on big plastic bottles have the same size neck and cap as do the small bottles, usually.

First AI painting sold for $430,000. But that's purely a signalling issue. In the long run, will see the AI painting get a lot better and the costs to plummet.

Will people pay more for a mediocre human painting or will all those painters disappear?

I say there'll always be a market for fine art that's produced by hand: Norman Rockwell, Bob Ross come to mind... and remember, if the artist dies, your painting goes up in value. By contrast, a robot never dies (I guess the art gallery can kill it though to increase the value of their paintings, assuming the robot has unique talents; is that an ethical dilemma?)

Third quarter gdp 3.5% -- pretty, pretty good! But probably not a robots effect.

Market down 2% at the moment however. I hope you bought Fannie Mae and Freddie Mac like I told you to TC. You'll be richer!

Shira says some people are seeing more companies spring up which use humans to collect data in a very readable way. People taking photographs of street signs, etc. Companies believe there is valuable data trapped in their own enterprises, she says.

Sounds bogus to me. But perhaps Shira is a 'good reporter' in that she does not filter what she hears, just reports it. I would have tossed that observation into the mental 'round file' as trash.

Does the data effect mean the rich and big companies get richer and bigger? What will stop that process? Regulation? A new medium for tech communications, creating new sector leaders?

the EU? I dunno…

Still waiting for the totalitarianism to be mentioned...

Someone just tried to smash open the Magna Carta and destroy it with a hammer...will enough AI and surveillance render such attempts futile?

Surveillance? I think you need bulletproof and shatterproof glass, not surveillance. Usually such smash-and-grab jobs are done at night, but in the Netherlands even in broad daylight.

Somebody is resembling a stalker!

No I'm not Viking, it's TC who is the OP in all the comments! Coincidence...

Water out of thin air!:
From California, of course. (That's not from the session, nor is the Magna Carta.)

Noah S. wants to make the driverless car algorithms "public" and open to everybody.

Noah S. is a communist. Algorithms are not easily patentable and hence trade secret. Only Ben Goldacre would disagree with such a reasonable position.

In practice when you force an entity to make their private algorithms public, and that entity doesn't really want to comply, it is worthless. Open source software has to be designed to be interoperable with other open source software to be useful. It takes a decent amount of work to turn closed source software into useful open source software and if you don't do that work then the artifacts you release won't help anyone.

Is it an option for a person to own his or her AI data? What does consent mean when a person doesn't understand what he or she is giving up? Points from Leonid B.

Actually the traditional option is the person collecting the data owns it, but the modern trend is away from that.

A trained AI's internal data is like an unlabeled list of a billion numbers. You can own it, but it's not like you're going to be able to look at it and conclude anything. You could own the raw inputs, but there just won't be anything interesting there - it'll be like owning a recording of everything you've ever said to Alexa.

Xinjiang mentioned!

And why is that significant again? Oh, I see! Xinjiang Uyghur Autonomous Region is a provincial-level autonomous region of China in the northwest of the country. It is the largest Chinese administrative division and the eighth largest country subdivision in the world,

AI everywhere!

"No dispute in China as to who owns the data!" -- good line from Noah F.

Now it's about CCTV as a crime-fighting device, including against terrorists. Many London homeowners petition to have CCTV put in their neighborhoods -- what does that indicate about the future political equilibrium?

Should we just accept all of this?

"The declining marginal utility of Big Brother..." I laughed.

It isn't up to us old guys.

Future, increasingly digital-native, generations will decide.

I have heard that the people in China read Neal Stephenson's Diamond Age and decided to just do it.

I don't think the West has picked their sci-fi model quite yet. Best contenders?

I agree. I think the younger generation will eventually go with just digital cash. Oldster who carry cash will be like the little old lady who takes out a checkbook to write a cheque at the supermarket queue.

Bonus trivia: I can't keep up with the comments! It's not quite verbal diarrhea but man, it's streaming and steaming out fast!

"I have heard that the people in China read Neal Stephenson's Diamond Age and decided to just do it."

Diamond Age is about nano tech and the dissolution of government nation states into city states and global ideological tribes. People in China that attempted to just do that would get run over by a tank.

It was also about "A Young Lady's Illustrated Primer" shaping thousands of young women in parallel.

For most of the story we were told the Primer was an AI, but the big reveal was that technology could not quite do that yet.

It's not hard to see Chinese social scoring as a kind of ongoing Primer, and one that will get much better as AIs get better and better at looking over your shoulder.

The Primer was designed to teach young women to buck the status quo. It led to a revolution.

That's completely different that the "Chinese social scoring". Indeed it's nearly the exact opposite.

Do you think the Magna Carta villain was pro- or anti-Brexit? That it is not obvious tells you something about political crimes and why terrorism is often so inefficient.

This is an actual story? Yep: "A man suspected of trying to steal a Magna Carta from its display case at Salisbury Cathedral has been arrested after he was wrestled to the ground by cathedral employees, British authorities said Friday... The dean of the cathedral said the suspect was able to smash through one of two protective glass screens housing the precious document. The Rev. Canon Nicholas Papadopulos praised "courageous" staff members who he said were able to restrain the man for 12 minutes before authorities got to the scene." - note the Greek surname. Why isn't he Greek Orthodox? The heathen...

Dina suggests that surveillance software works much worse on people of darker skin, leading to many mistakes.

I don't think it is a skin tone problem per se, but I have heard that there were some errors because early training sets were overwhelmingly white.

So the AI was like "wait, what?"

Khashoggi talk now -- they screwed up on the shoes!

Will the machines discriminate more or less than the humans do?

remember that movie with dersowitz the lawyer and about a 1/2dozen black fellas who had their rape convictions overturned by the innocence project
after spending like 10 years in jail?
according to the movie almost
80 percent of them were wrongly identified by the accuser

@ .... mr rogers ... and the 20% that were not? A lot of these so called innocent people were at the scene of the crime it seems, and possibly were accomplices or lookouts. Hang 'em high?

the lawyer seemed to think that the innocence project made the case thateyewitness testimony in sexual assault cases was not reliable enough to just automatically believe it

And how much should the human judge defer to the sentencing recommendations of an AI box? More generally, how much should explainability matter? But of course a lot of judges also cannot explain their intuitive judgments.

Why should explicability matter at all? Can anything be explained? What *isn't* a black box, when you get right down to it?

Explain ability matters because without it you don't know when you have reached a corner case relative to the training sets.

Noah F. -- And are judge self-reports actually an account of explicability for their decisions?

The accountability of the machine? Well it's probably re-programmed at a minimum. Isn't that what we try to do when we send prisoners to "Correctional facilities"?

And Sinead O'Connor converts to Islam, also not done by a robot, I strongly suspect.

She better not call the imam or whatever 'the antichrist' as she called the Pope (if I remember the right artist), or she'll be issued a fatwa like S. Rushdie maybe!

"If you can't explain it, people really have trouble accepting it" -- Justin Fox.

Break now!
And a Turing test is coming after the break, I plan to fail it!

Tyler, feel free to delete my comments if you intended this to be just your monologue. I don't want to disrupt the content.

That goes without saying, doesn't it?

What is the psychology of human-machine interaction? How satisfying is it to interact with machines? With voice recognition software?

The new robo-cat Lady.

Or a sex doll.

How bad is it to chat with AI over the phone?

Should you allow yourself to be befriended by a robotic dog? -- Shira

"Might as well talk to the radio!" -- Leonid.

Clive feels better when he has the option of speaking to an agent, but he doesn't always choose that.

Yes, people are getting used to dealing with chatbots on the phone. Once you're familiar with it (and you can reasonably expect it to fix your problem), you won't just immediately bypass it to get to a human.

How many people use a human to find out their bank balance any longer? When's the last time you used a human to book or change a flight (outside of sitting at the gate)? Etc?

Using AI instead of an agent won't go over well here in the Philippines, where call centers are a big business.

Bonus trivia: the use voice filters and other techniques to give the call center operator a 'western accent' sometimes. But here in PH their English is pretty western sounding so no need.

"Using AI instead of an agent won't go over well here in the Philippines, where call centers are a big business."

Yep, much like CF bulbs didn't go over well at the GE light bulb plants in Youngstown, Ohio.

Leonid: AI develops its own language when they are talking to each other.

I myself still prefer to book a flight with a human. Not for aesthetic reasons, simply to get the best option.

I suspect that's an age division issue. Below a certain age (mid 40's?), the normal way of booking a flight is via a web page. Booking via a human is an odd case that you'd only do if you have some problem. Furthermore, it seems inefficient, since I can have two or three different web sites pulled up on 2 different monitors and can see the flights available from different air lines.

I assume it's an age thing. It could be a STEM versus non-STEM aspect?

Isn’t the person you are using to make the booking simply using the same software interface to make the booking? So what is gained? I have to book my work flights via a travel agent and they make many more mistakes than I do making my own travel bookings for personal travel. One recent one the agent simply forgot to make the booking after I confirmed everything. I don’t like doing any fiddly detail stuff like travel booking by phone for that reason, you cannot easily visually check things.

Now Q&A, that is it for now!

Shira - "Companies should make sure that their AI is easy to distinguish from humans."

A good point.

No, he's wrong. The US would agree to restrictions on militarized AI if it felt like it would devalue it's nuclear deterrence. He's only thinking of AI, not the broader picture of the entire US military arsenal.

AI security should be the biggest issue of our time, but we have apocalypse fatigue from the global warming lobby. What can the AI safety community learn from them to highlight better this risk and organise the world to better police AI development?

Says the man who thinks Japan could not be blockaded during WWII... ChrisA does not believe in global warming...what else doesn't he know?

-- sotto voce --

I just realized this is an AI panel without practitioners. No wonder it it sounds like reworked new stories for people who didn't see the originals.

Hate to be harsh, but there it is. What you wanted was a debate between Andrew Ng and Rodney Brooks about what is possible in the short term, and what will ever be possible in the long term.

Unless you are just catching up business people on articles they didn't read.

-- sotto voce --

@anonymous - we're just getting old my friend. As some French dude years ago said, 'in every generation, in every instance, man must reinvent man' or something. Don't know what he meant, but to me it's like every generation has to reinvent the wheel and put old ideas in their own words... that's why you hear old songs 'redone' every few years or decades. I did like the Vamp's 2014 remake of "Cecelia" by Simon & Garfunkel.

Yeah, I feel kind of bad. But at least it was sotto voce.

For anyone interested in Ng v Brooks,

Noah - "We are a more mature society. Some would say Complacent ."

Call out for Tyler.

The host, discussing technological diffusion, said that the U.S. was "way behind Europe" until the Ford revolution.

The U.S. GDP per capita was 85% that of the UK in 1850 and equal with Europe then. By 1900, the U.S. GDP per capita was 90% of the U.K. and by 1920, the same

Plumbing skills for AI. Huh? Most house call plumbing is basic tasks and some brute force with a high ick factor. There's no easy AI replacement. AI won't pop the sewer cap and clear a clog, or replace the wax seal under the bowl.

@JWatts - lol! Comedy gold my friend. You do know about robotic 'plumbing pigs' don't you? They've had them now for nearly forty years. At first, they were remotely controlled by humans, but now, AI can do the job of fixing a broken underground pipe without the human even having to do anything (except push a button authorizing the repair).

Someone else asked the person who brought it up if he was referring to large scale sewer systems or to someone that comes to a house. The reply was to a house.

"AI can do the job of fixing a broken underground pipe without the human even having to do anything (except push a button authorizing the repair)."

In the future maybe. Current pigs tend to be remotely controlled.

Talking about US citizens and DNA testing. My wife just ordered a DNA test. She's not naive. She volunteered to share her results with various research groups.

I bet she did that to impress you... my hunch.

No more comments? Good, I will go back to my normal job of goofing around.

Comments for this post are closed