A lot of this is vagueness and hand-waving, with few specific numbers. Solar technology, for example, has so consistently outpaced forecasts that it's almost a joke:
but this is handwaved away as "ignorant of underlying economics", even though the economics themselves have changed massively and continue to do so. (Anyone even vaguely familiar with the industry knows about the difference between baseload and peak power; a local newspaper might mess this up in their solar story, but it's hard to believe that any serious analyst would.)
The world's fossil fuel dependency is actually up from the early 2000s, and about 3% lower than in the 90s. (from roughly 91% to 88%).
So even though Solarpower is growing relatively fast, it's not having a big impact on the structure of the energy sector.
If you want more hard data you should actually pick up the book mentioned in the article, Gordon's excellent
The Rise and Fall of American Growth. Otherwise, anemic economic growth around the world and diminishing returns on science actually very intuitively show that the author is right. Total factor productivity in the UK for example has not gone up at all in over 10 years.
Yes, it's like rating a hedge fund's performance -- doesn't matter how much its portfolio grew by but how much MORE it grew relative to the growth rate of the market. If solar is not outpacing the growth of coal, oil, and gas, then it's not helping.
True for the timeframe under-analysis, but not necessarily for long-term trends if there are changing fundamentals.
One of the interesting things that has happened with solar is that subsidy has decreased. So even if the composition is remaining static for new development, if we now have sustainable solar compared the the historically (economically) unsustainable version that required government intervention, that was an accomplishment. The counterfactual is not stable composition of generative sources but one with renewables decreasing because the political will waned to continue subsidy.
Good article. I feel like there are a few unstated "elephants" in the room that the author briefly outlined but could be explored more:
quote:
"A comprehensive study published by the National Bureau of Economic Research found that the number of researchers needed to develop new drugs, improved crop yields, and better microprocessors has risen substantially over the past 50 years. Other studies have found that research and development (R&D) productivity has fallen across a wide variety of industries, with lower growth in corporate revenues per research dollar than in the past, and that the impact of Nobel Prize–winning research has also declined. From a given amount of effort (or dollars), less knowledge and innovation arise. Our optimism about the economic prospects of new technologies and innovation should be going down, not up."
Is this, perhaps, also a product of system complexity? maturity and limits of certain processes? guidance and management seeking goals that are de-coupled from reality?
Declining VC investments in, what I read as, CORE TECH.
We are so busy funding some website social convenience service that we forgot about the chips powering it, the power powering it, etc. and rapidly hit EOL on processes.
Perhaps real innovation requires
1) actual major efforts, efforts that our economic system doesn't prioritize enough. there's a million monkeys happy to use your platform for trivial mc-innovations, but you have to slog it out and build said platform first.
2) a deliberate plan, path, trajectory. (I hate to use the manhattan project as a good example for anything, perhaps Arpanet is a little less ugly of an example, in terms of impact.)
Favorite Quote:
"The strategic management discipline places far more emphasis on capturing value through managing revenue streams (e.g., advertising versus subscriptions or product sales) and the scope of activities (e.g., the degree of vertical integration) than on creating value through high-performing, low-cost products that improve productivity in industries. Currently the most popular method of capturing value is with so-called platforms that organize an industry around a single product or service (think of Airbnb, for example, which offers a platform by which homeowners can rent out rooms), often enabling the platform provider to capture most of an industry’s profits. Not only is value creation clearly more important for increasing productivity than is value capture, it is not a coincidence that most of the current money-losing start-ups (e.g., Uber and Lyft) emphasize platforms in their announcements and IPO filings, building from the hype that business schools created."
Which of course leads to me saying:
"there's very little incentive to build a platform that's useful for other actors. there's very little incentive to respect and treat users like humans"
Hype comes from a promise to increase subjective well-being, while ultimately failing to do so, at least in the short term. But this definition should not discount the long term potential.
What is needed is an actual increase in subjective well-being, which is frustratingly diverse, subjective, and what you touched on in your last sentence. Hype does the opposite, it works to decouple capital from labor.
Unfortunately, until western society can find a better balance between capital, democracy, and global relations, entropy will outpace innovation, always.
Simple enough... because hype is what gets funded. You tell people about the vast engineering challenges that lie ahead, and they'll term you a downer and not fund you. Ironically, this policy ends people like working for such "haloed" geniuses.
I fear the article suffers from the myopia of pre-conceived notions of what constitutes innovation. Every generation imagines the future to be mere extension of the present. But innovation doesn't happen that way. Technological development is inherently fractal, the obvious directions of progress are not necessarily the most interesting ones. In the 1950s, the perceived future was all about nuclear power, and supersonic flight, turned out that while we didn't get supersonic passenger planes, air travel became affordable to a large fraction of the planet because of amazing advances in power efficiency of jet engines. In the late 60s the future was going to be in space. We didn't get a moon base, but we got a global computer network that changed how people lived and worked. We got containerized transport, that dramatically improved global trade and pulled millions out of poverty. Yes, the world as a whole is wealthier per capita than it was in the 70s. So innovation not just what some government think tank or corporate strategy group think it should be. It may well be that future developments in computation come from breakthroughs in mathematics rather than improvements in semi-conductors. Innovation comes from comes from many places and its a mugs game to predict how it will evolve. The best that any society can do is to encourage as much of it as can be sustained by its level of economic development. Let chaos and lady luck determine what comes next.
I like your idea of "technological development being fractal". Fractals get more complex by zooming in. On a high level we sat in front of a screen 20 years ago and probably will sit in front of a similar screen 20 years from now. But zooming in we have so much more content and abilities on a computer.
And I bet we will sit on chairs even in 200 years. ;)
I mean I can list a 100 points to prove what a stupid article this is but I am too busy picking out parts to build this super computer for like $1200...
People with the mindset of this article need to get some antidepressants and try working out. You are obviously not putting things in the proper context.
I'm somewhat qualified to talk about AI and to call it hype is itself hype. As in, it's looking at outliers and trying to understand the market based on them alone. It's deciding that because there's no glamorous reality shifting changes there must be no impact at all. For example, IBM failed but every pharma company has a decent bunch of PhDs building machine learning models for a variety of reasons. Those models aren't as headline catching but rather they improve certain unglamorous business processes by X%. Taken as a whole that adds up to a lot of value for those companies. Facebook probably makes tens of billions per year due to AI. In a way AI has become mundane and doesn't require specialized centralized companies to be implemented which makes fewer headlines.
edit: I also suspect it's a bias due to people being used to traditional industries. To make a new drug you need a lab, infrastructure, equipment, well trained experts and so on. Lot's of investment so it's centralized and visible. To make an AI model requires some data, a laptop and an internet connection.
> every pharma company has a decent bunch of PhDs building machine learning models for a variety of reasons. Those models aren't as headline catching but rather they improve certain unglamorous business processes by X%. Taken as a whole that adds up to a lot of value for those companies.
This!
Although I am not an expert in AI field or AI-based methods, I have been working as an s/w engineer / plumber / fireman in projects deploying some AI-based methods and models in production. These methods are nothing catchy, that would unlikely get into headlines, nothing like self-driving cars, etc. But they are just pretty effective applied in a quite well defined scope improving parts of existing processes. These would be considered by many as 'boring' and are related e.g. with improving the data acquisition processes, such as, improving quality of text extraction from images or improving recognition of named entities from free text by using a tailored neural network model in the system. Nothing big nor fancy, but for the business it makes a significant difference.
So, on one hand, yes, there's a lot of hype for applying AI-based methods for just anything and wanting to show to investors using the keywords for getting more $$$. But on the other hand, having a well scoped use-case with clear understanding on what method and why, it can bring pretty good results. However, as a community, we are still trying to catch-up with all the recent advancements in AI-based field, and trying to understand when to use XYZ and why over well established 'classical' approaches. These are / will be another tools in our toolbox and this fact cannot be ignored.
I like this article very much. As someone who makes a living from R&D funding into hyped technologies, this made me do some amount of introspection.
AI, however much we love it, and however good computers are at chess and go, has done relatively little to increase productivity in already existing processes. Yes, it's nice to tell apart cats from dogs with F1=0.98.. but, what bushiness actually added image classification into their processes and made money out of it? Furthermore, as any one searching for specific stuff on Google and being frustrated by synonyms being substituted or search terms being ranked in importance, can tell you, F1=0.98 isn't enough to really outsource human "intelligent" work onto computers.
The same goes for big data.. yes, we love our fancy Sparks and Airflows and Kafkas, yes, I've pushed to build "big data" pipelines several times.. yes, fancy dashboards can come up. But any increase in productivity or quality of life could have also been achieved with shell scripts and a bit of patience. Uber your say? they use big data to foo and to bar...yes, but not in real time, for real money making purposes: even if you ignore the fact that they are still in the negatives, the computations that have to be done real fast for their service to be usable, are all "small" as in, they-can-be-done-in-a-mobile-phone small.
> AI, however much we love it, and however good computers are at chess and go, has done relatively little to increase productivity in already existing processes.
Is that true?
Uber's ETA prediction is, to me, a very impactful and technologically impressive application of ML.
Recommendation engines--from Netflix/Spotify to paid advertising platforms--seem to be pretty powerful.
Image classifiers, from automated tagging to monitoring solutions, seem to have made a big impact. This isn't just Facebook suggested tags or whatever, Wildlife Protection Services recently doubled their detection rates of poachers throughout nature reserves by introducing ML.
And what about smartphones? Speech interfaces and conversational assistants are a standard part of every flagship phone now. Mobile cameras use ML to improve images without upgrading hardware.
I agree that there is a lot of snake oil AI nonsense out there, but I think it's a bit of a false dilemma to deciding whether or not machine learning is "valuable" by evaluating the over-promises of sales teams. The reality is that almost all of the most popular apps on your phone (Uber, Lyft, Maps, Gmail, Spotify, Netflix, Facebook, Instagram, Snapchat, etc.) all incorporate ML.
Indeed, ML is vastly deployed, I do not doubt that. And maybe those "AI-powered" companies are making money, or making lives better.
But I was referring explicitly to existing processes. Your poacher detection example is a good one, there was a process (based on some other tech/skill/k9 units?)and then it was improved with AI. My assertion is that there are very few such examples.
Contrast that to new enterprises that use AI/ML. Of course, if you create a whole new market and corner it, whatever technology you use might seem like the recipe of your success. But in the case of ride hailing apps, they created a brand new market of taxis-called-directly-by-the-user (and still not making money), Facebook created a whole new market of personalized matchmaking between people and ads, etc. Yes, they are kings in their markets, yes, they use "AI", but what portion of their market dominance is due to the use of that specific technology? I would say close to 0.
19 comments
[ 3.8 ms ] story [ 63.3 ms ] threadhttps://16iwyl195vvfgoqu3136p2ly-wpengine.netdna-ssl.com/wp-...
and costs continue to fall steadily:
https://www.thesolarnerd.com/blog/solar-pricing-2019/images/...
but this is handwaved away as "ignorant of underlying economics", even though the economics themselves have changed massively and continue to do so. (Anyone even vaguely familiar with the industry knows about the difference between baseload and peak power; a local newspaper might mess this up in their solar story, but it's hard to believe that any serious analyst would.)
So even though Solarpower is growing relatively fast, it's not having a big impact on the structure of the energy sector.
If you want more hard data you should actually pick up the book mentioned in the article, Gordon's excellent The Rise and Fall of American Growth. Otherwise, anemic economic growth around the world and diminishing returns on science actually very intuitively show that the author is right. Total factor productivity in the UK for example has not gone up at all in over 10 years.
One of the interesting things that has happened with solar is that subsidy has decreased. So even if the composition is remaining static for new development, if we now have sustainable solar compared the the historically (economically) unsustainable version that required government intervention, that was an accomplishment. The counterfactual is not stable composition of generative sources but one with renewables decreasing because the political will waned to continue subsidy.
Seems like renewable are having a big impact.
From: https://www.eia.gov/todayinenergy/detail.php?id=37952
quote: "A comprehensive study published by the National Bureau of Economic Research found that the number of researchers needed to develop new drugs, improved crop yields, and better microprocessors has risen substantially over the past 50 years. Other studies have found that research and development (R&D) productivity has fallen across a wide variety of industries, with lower growth in corporate revenues per research dollar than in the past, and that the impact of Nobel Prize–winning research has also declined. From a given amount of effort (or dollars), less knowledge and innovation arise. Our optimism about the economic prospects of new technologies and innovation should be going down, not up."
Is this, perhaps, also a product of system complexity? maturity and limits of certain processes? guidance and management seeking goals that are de-coupled from reality?
Declining VC investments in, what I read as, CORE TECH. We are so busy funding some website social convenience service that we forgot about the chips powering it, the power powering it, etc. and rapidly hit EOL on processes.
Perhaps real innovation requires
1) actual major efforts, efforts that our economic system doesn't prioritize enough. there's a million monkeys happy to use your platform for trivial mc-innovations, but you have to slog it out and build said platform first.
2) a deliberate plan, path, trajectory. (I hate to use the manhattan project as a good example for anything, perhaps Arpanet is a little less ugly of an example, in terms of impact.)
Favorite Quote: "The strategic management discipline places far more emphasis on capturing value through managing revenue streams (e.g., advertising versus subscriptions or product sales) and the scope of activities (e.g., the degree of vertical integration) than on creating value through high-performing, low-cost products that improve productivity in industries. Currently the most popular method of capturing value is with so-called platforms that organize an industry around a single product or service (think of Airbnb, for example, which offers a platform by which homeowners can rent out rooms), often enabling the platform provider to capture most of an industry’s profits. Not only is value creation clearly more important for increasing productivity than is value capture, it is not a coincidence that most of the current money-losing start-ups (e.g., Uber and Lyft) emphasize platforms in their announcements and IPO filings, building from the hype that business schools created."
Which of course leads to me saying: "there's very little incentive to build a platform that's useful for other actors. there's very little incentive to respect and treat users like humans"
What is needed is an actual increase in subjective well-being, which is frustratingly diverse, subjective, and what you touched on in your last sentence. Hype does the opposite, it works to decouple capital from labor.
Unfortunately, until western society can find a better balance between capital, democracy, and global relations, entropy will outpace innovation, always.
And I bet we will sit on chairs even in 200 years. ;)
Well that's something to think about. I do wonder how all the tech "trends" that i read here begin
I mean I can list a 100 points to prove what a stupid article this is but I am too busy picking out parts to build this super computer for like $1200...
People with the mindset of this article need to get some antidepressants and try working out. You are obviously not putting things in the proper context.
edit: I also suspect it's a bias due to people being used to traditional industries. To make a new drug you need a lab, infrastructure, equipment, well trained experts and so on. Lot's of investment so it's centralized and visible. To make an AI model requires some data, a laptop and an internet connection.
This!
Although I am not an expert in AI field or AI-based methods, I have been working as an s/w engineer / plumber / fireman in projects deploying some AI-based methods and models in production. These methods are nothing catchy, that would unlikely get into headlines, nothing like self-driving cars, etc. But they are just pretty effective applied in a quite well defined scope improving parts of existing processes. These would be considered by many as 'boring' and are related e.g. with improving the data acquisition processes, such as, improving quality of text extraction from images or improving recognition of named entities from free text by using a tailored neural network model in the system. Nothing big nor fancy, but for the business it makes a significant difference.
So, on one hand, yes, there's a lot of hype for applying AI-based methods for just anything and wanting to show to investors using the keywords for getting more $$$. But on the other hand, having a well scoped use-case with clear understanding on what method and why, it can bring pretty good results. However, as a community, we are still trying to catch-up with all the recent advancements in AI-based field, and trying to understand when to use XYZ and why over well established 'classical' approaches. These are / will be another tools in our toolbox and this fact cannot be ignored.
AI, however much we love it, and however good computers are at chess and go, has done relatively little to increase productivity in already existing processes. Yes, it's nice to tell apart cats from dogs with F1=0.98.. but, what bushiness actually added image classification into their processes and made money out of it? Furthermore, as any one searching for specific stuff on Google and being frustrated by synonyms being substituted or search terms being ranked in importance, can tell you, F1=0.98 isn't enough to really outsource human "intelligent" work onto computers.
The same goes for big data.. yes, we love our fancy Sparks and Airflows and Kafkas, yes, I've pushed to build "big data" pipelines several times.. yes, fancy dashboards can come up. But any increase in productivity or quality of life could have also been achieved with shell scripts and a bit of patience. Uber your say? they use big data to foo and to bar...yes, but not in real time, for real money making purposes: even if you ignore the fact that they are still in the negatives, the computations that have to be done real fast for their service to be usable, are all "small" as in, they-can-be-done-in-a-mobile-phone small.
Is that true?
Uber's ETA prediction is, to me, a very impactful and technologically impressive application of ML.
Recommendation engines--from Netflix/Spotify to paid advertising platforms--seem to be pretty powerful.
Image classifiers, from automated tagging to monitoring solutions, seem to have made a big impact. This isn't just Facebook suggested tags or whatever, Wildlife Protection Services recently doubled their detection rates of poachers throughout nature reserves by introducing ML.
And what about smartphones? Speech interfaces and conversational assistants are a standard part of every flagship phone now. Mobile cameras use ML to improve images without upgrading hardware.
I agree that there is a lot of snake oil AI nonsense out there, but I think it's a bit of a false dilemma to deciding whether or not machine learning is "valuable" by evaluating the over-promises of sales teams. The reality is that almost all of the most popular apps on your phone (Uber, Lyft, Maps, Gmail, Spotify, Netflix, Facebook, Instagram, Snapchat, etc.) all incorporate ML.
But I was referring explicitly to existing processes. Your poacher detection example is a good one, there was a process (based on some other tech/skill/k9 units?)and then it was improved with AI. My assertion is that there are very few such examples.
Contrast that to new enterprises that use AI/ML. Of course, if you create a whole new market and corner it, whatever technology you use might seem like the recipe of your success. But in the case of ride hailing apps, they created a brand new market of taxis-called-directly-by-the-user (and still not making money), Facebook created a whole new market of personalized matchmaking between people and ads, etc. Yes, they are kings in their markets, yes, they use "AI", but what portion of their market dominance is due to the use of that specific technology? I would say close to 0.