The former director of the AI Lab at MIT looks into the current hype around AI - worth reading :
https://www.technologyreview.com/s/609048/the-seven-deadly-s…
The former director of the AI Lab at MIT looks into the current hype around AI - worth reading :
https://www.technologyreview.com/s/609048/the-seven-deadly-s…
The former director of the AI Lab at MIT looks into the current hype around AI - worth reading
Thanks free capital, that really was interesting.
Saul
Great article!
Thanks for posting it.
– This…is the MaineReason
That is an interesting article
But I remember with ISRG the older and more esteemed the urologist the more he said they daVinci would never catch on.
No doubt he is right that GAI is a long layoff. But there will be plenty enough specific, non generalized potential AI uses to make investors rich. If they can figure out the right ones.
A great article, thanks for sharing!
I also think this “singularity” idea is too far fetched and is not going to come around anytime soon.
However, from investing point of view, I would like to add a more positive outlook. I have studied information science specializing in optimization techniques and algorithms, some machine learning techniques as well. I work in an IT department of a major industrial company in the field of automation of the maintenance procedures. You might have seen the GE ad with Agent Smith - that’s what I’m doing, albeit for a different company.
I’ll put aside the concept of general purpose AI - that’s not coming anytime soon and I fully agree with the article’s take on it. However, I think it painted the machine learning techniques a bit too bleakly - the are more useful than the article may make you feel.
First of all, the scientific field is a bit older than you may think. The mathematical tools for machine learning and specialized AI go back to the 1940s - neural networks, fuzzy logic and other basic techniques were developed as mathematical models then and in the 50s the first computers implementing these mathematical models were developed. These gradually developed into what we call “expert systems” in the early 1980s (meaning a system which is supposed to replace a human expert). Then in the 90s disulusion came as expert systems didn’t take the world by storm.
So the “story arc” for technologies that the author describes in part 1. on the example of GPS happened with specialized AI as well - just with a 10 years delay. The reason why it failed to impress in the 80s and 90s I think is that we could not supply the algorithms with enough data. The storage mediums were small and slow and data transfer bandwidth was terrible. Well, that is not the case anymore.
Google was the first well known successful and useful machine learning system that could take advantage of rising data transfer speeds and storage solutions. It still needed a huge computation center and deal with a lot of technical hurdles, but even that is now history - that was 15 years ago. You don’t need to maintain a server farm anymore, you can rent as much server space you need in a cloud from AMZN or MSFT and you pay as you go (and you pay very little by 2000’s standards). Even more importantly, computers are powerful (and cheap) enough to deploy parts of the algorithms on the target machines rather than on the servers.
One example of what we are doing - we gather data from all of our machines at the customer sites over a period of time - say, 1 year. 100s of terabytes from 1000s of machines. During this period some will fail. The machine learning algorithms will look at these cases and see if they see any common patterns in the data that would have predicted the failure. These patterns will then be redeployed on the machines so they can predict what failure is about to happen in what time horizon with a degree of confidence. A service technician will show up before that happens with the right spare part and will install it with minimum downtime. This is extremely useful for high availability systems like factory automation or medical devices.
This is already in use, but you wouldn’t see it as a lay person. The net result of this is that a factory can produce more products in the same time or that a radiology can process more patients.
Another example - the same as above can be done with medical imaging for the purpose of correctly identifying a diagnosis in an image (such as an MR or a CT scan). The result is that the radiologist will get an image with higlighted areas showing him what the most probable diagnosis is including what most probably could be a false positive. This is not in the market yet, but it’s coming. There is no technical limitation to implementing it, it’s just a matter of proper medical/regulatory procedure to make it safe and legal.
And these two are just scratching the surface, I could talk about dozens of other examples and there are thousands more that I don’t even know about. All in all specialized AI is on the upswing and we as a society are only starting to use it correctly.
Stenlis, do you know about DataRPM? Sounds like a fit for the kind of requirement you are describing.
https://www.progress.com/predictive-analytics
From RockoYates, over on Free For All Economics board.
http://discussion.fool.com/japanese-co39s-robots-make-everything…
The following link to Fanuc, in part, describes robots and AI using data from their worldwide network of robots, to teach new robots a job. Robots building robots… And training them. No humans required?
https://www.bloomberg.com/news/features/2017-10-18/this-comp…
Ralph who thinks that perhaps the ability of AI/robots to use the “machine learning” of other robots is being underestimated?
But I remember with ISRG the older and more esteemed the urologist the more he said they daVinci would never catch on.
To the point of the OP. How many jobs does ISRG cause to go away?
Let’s see. We call it robotic surgery, but there is still a highly trained doctor operating it, just as with normal surgery. Isn’t the daVinci system just a better set of surgical tools?
And they cost more to make and develop.
It requires many more jobs to make this better tool.
Design, construction, testing, training, tech support from Intuitive’s employees all the way to the chips to make it work and the FABs where the chips are made, etc.
I don;t know how much more productive a surgical team is with daVinci compared to without, but I doubt it is significantly better. The advantage is that the surgery is supposed to be less invasive and have a shorter patient recovery time. Which is good because it potentially leads to better outcomes.
Mike
The DV made plenty of jobs go away. Those urologists who stuck with old methods ,especially open surgery found few patients.
Whether the DV is “better” is not relevant to investors. What is relevant is that early ISRG investors made a lot of money.
The term “robotic” was marketing genius, how many would have been excited by “Waldo Surgery”?
I was a computer science major back in the early 1980’s and took the one course on AI. It was still very archaic back then and computing power was low. (my first year we had punch cards).
It has been over 30 years, and I can clearly see the exponential growth and this is going to make Moore’s law look slow. This is the real thing. This will change the world. This will do many things that are invisible to us, but really change us. If it was minor, Bill Gates, Elon Musk and other very smart people would not be worried about how wrong it could go and how dangerous it could be.
I evaluated an IT tool that used AI to learn and correct problems in the network using virtual engineers. This was very impressive but currently beuyond our budget and complicated to setup. That will change quickly.
Think of the compound interest rate of knowledge we have had as humans, spending centuries and millenia collecting knowledge and building on that faster and faster. Once all the knowledge of the world could be stored in a single library, not it takes the internet. AI is learning from learing and will eclipse our human knowledge quickly, it can save us or ruin us (also see the very cool CRSPR technology for an alternate end of the world).
But back to the short term, it is real and investible. NVDA is the picks-and-shovels play, supply awesome tools for data centers that can be rented out by the minute. That is real. AI will drive our trucks and cars and make them like riding an elevator (which used to require a human to drive). Watson is intering in medicine at UNC and scans every new medical study every day to advise human doctors on new possible approaches to a specific patients specific illness. There are AI chat bots that learn and respond to customers until they need to be handed off to humans, but many issues are resolved before that and people don’t even know it is AI behind the scenes. The SiFi movies all come true. People will fall in love with AI avatars, they may well challenage our race, “freaks” will add more and more pieces of computers and AI to their bodies, bluring the line between humans and terminators. Where will it stop, can it stop? We have always been able to control our technology and I am optomistic, but who knows.
IBM is finally turning around, but I didn’t check the earnnings results, was it Watson? Tesla’s future is autonomous cars powered by machine learning.
But where to invest? Definitely NVDA. Intel bought my MBLY, but will it work and become investable? Google and Microsoft datacenters (too small a piece of their pie?). Tesla on a big dip if their production can meet expections as model 3 comes to market.
The future is nigh.
The DV made plenty of jobs go away. Those urologists who stuck with old methods ,especially open surgery found few patients.
What do you estimate is the net change in jobs?
lost jobs: those urologists who stuck with old methods
new jobs: everyone at ISRG and theirs suppliers
added jobs: new urologists using the DV
Mike
net jobs ?- probably no loss
http://www.onclive.com/publications/obtn/2011/march-2011/rob…
despite the title, the dV has a shorter learning curve than lap type surgery. Both are way longer than open surgery…
As people live longer there are more males having prostate cancer so I suppose it was a growth industry. Until recently that is
https://www.nbcnews.com/health/mens-health/prostate-surgery-…
fewer biopsies fewer prostate surgeries of any kind
Hate to say it guys, but if you live long enough most of you will get prostate cancer. And the feds spend less on researching treatments for those " mostly involving males" type cancers than they spend on researching the more PC " mostly involving females "type cancers.
What do you estimate is the net change in jobs?
lost jobs: those urologists who stuck with old methods
new jobs: everyone at ISRG and theirs suppliers
added jobs: new urologists using the DV
Mike
I think there are very few urologists who lost or added jobs because of ISRG. Those so old that they just didn’t want to learn it probably just kept doing laparoscopy or open surgery until they retired. They more likely quit because of the requirements to use electronic medical records or being fed up with the general health care environment (less pay, more work, more regulations, more administrators). And the number of machines isn’t enough to materially increase the number of doctors needed since it really just provides an alternative to an existing surgery. Hopefully nobody said no to a laparoscopic prostatectomy but jumped at the chance of a robotic prostatectomy–there they’re just not that different.
Over the years things like prostate MRI and prostate artery embolization are more likely to materially reduce the number of biopsies and prostatectomies respectively.
Rodney Brooks is a well known AI sceptic, so its worth noting that in the reading.
His points about robotics are probably spot on. Extrapolating robotics to AI is problematic however, and the article reads mostly as things he is hoping will be true.
He might be right, but simply hoping for a future he wants is not super-compelling.
“Today’s robots and AI systems are incredibly narrow in what they can do. Human-style generalizations do not apply.”
This is true, to a decreasing extent. AlphaGo proved that the concepts of “strategy” and “imagination” and in general, surpassing anything a human had imagined were possible in a given domain, which was a huge shock to just about everyone in the industry. Except the Deepmind team.
If we’re talking AI, the key question is “Whats the timeframe?”. What will change in 5? 10? 20+ years? It’s a technology thats starting to make significant strides now that the world is allocating significant resource.
https://www.ft.com/content/856753d6-8d31-11e7-a352-e46f43c58…
“Headlines trumpet the suitcase word, and warp the general understanding of where AI is and how close it is to accomplishing more.”
This is not really a point. The point we’re concerned about is “how close it is to accomplishing more…”. Essentially, the technology is accomplishing more and more, and importantly, there seem no real limits.
“…the deep-learning success was 30 years in the making, and it was an isolated event.”
True. Sort of. I think he’s saying that because it happened that way in the past, it will happen that way in the future. Whats the stock-related quote? “Past Performance Is Not Indicative Of Future Results”.
“We will change our world along the way, adjusting both the environment for new technologies and the new technologies themselves”
This is Rodney hoping for his future. Theres nothing to guarantee that will be the case.
Enormous resources are now pouring into AI, from Google et al to the Chinese government. If we increase the amount of intelligence in the world, what does that mean? You can either treat AI as a “new improved version of Microsoft Excel”, ie, new tools for humans to use, or, you can recognise that increasing intelligence in general is likely to have very profound impacts on the world.
As investors we should probably consider both camps. Rodney is very firmly in the first, and gives little credence to the second.
My 2c
Greg
It’s a good idea to separate the brains of robots (computers ,AI) from the brawn (metal and plastic moving parts) The brains are getting better very quickly, the brawn incrementally at best. So “robots” will displace humans first where size does not matter. Look at those giant robot welders in a car assembly plant. Or where the task is almost all brains.
We are beginning to see devices designed for robots, not vice versa. The glass top on the 3 and even more on the Y are examples. Eventually I expect things like houses to be built with a series of standardized modules. As robots get better, “stuff” is designed to not only be built by them but to be used and repaired by them.
The tipping point will be a series of small tipping points. From an investors point of view, generalized AI is not needed for lots of opportunities.
But for the moment my handy man household fix it guy has one of the safest jobs in the world.