Not just tech. It takes a minimum of a decade for medical best practices to permeate through a country's healthcare system, even when the evidence is clear and funding isn't an issue.
Yes! Everyone wants “plug compatible Ai”, where they just plug in AI module in place of an old system. But with very few exceptions, that’s not how it works!
Many other parts of a production system have to be changed, e.g. training, career paths. In the long run, entire industry structures will change to get the maximum benefit from new capabilities.
Chip design has used “computer intelligence” for decades. The founders of Intel physical drew the layout of their first chips on transparent sheets that the lithography masks. (Analogous to photo negatives - for those who remember them.) Over decades, more and more of chip design was handled by AI.
One of the big revolutions in electronics industry was the foundry/design organization of companies. Qualcomm is the third largest chip company in the world (or was last time I looked), but it physically makes zero of its chips. They are all outsourced to specialists like TSMC.
The same reorganizations will happen due to what we now call AI. But we can’t predict what, where, or when. Meanwhile, the technical capability of the core algorithms will be continually advancing.
You point is spot on but I want to point out a few things missing from your analysis that you might conclude will accelerate the time frame you put forward
1. You left out the “Information Age”. This is a critical step in human progress that is enabling every single person in the world to have access to the entire world’s knowledge. As a result, when problems and opportunities are identified and/ insights gained, this information spreads to all of humanity. As a result, you multiple magnitudes more people working on the innovation at the same time.
When the steam engine was invented in 1712, it took a long time for the knowledge of this invention to spread across the globe. As a result, it took a long time to harness the collective ingenuity and motivation of humanity to leverage this invention.
Today that can happen in hours, if not minutes.
2. The very invention itself will accelerate innovation. Either through direct innovation itself (this is more speculative) or at the very least enabling innovators to innovate more effectively and efficiently.
Similar phenomena occur in several other fields, too. For example, hand washing by doctors took more than 25 years to become a norm, even though Hungarian physician Ignaz Semmelweis identified benefits in the 1840s. His theories were mocked until the 1870s and beyond.
Similarly, it took more than 40 years for James Lind to discover the cure for scurvy and the Royal Navy to begin issuing lemon juice to sailors.
Crop rotation, vaccination, alternate current, and anesthesia are other examples from my readings.
It is mainly due to one or more reasons listed below:
resistance to new ideas;
a lack of understanding of the underlying mechanism;
systematic inertia, including economic and logistic constraints;
influence of a specific individual with competing technology and
sometimes prevailing social and cultural norms and practices.
The above suggests that the widespread adoption of AI, even with its demonstrable benefits, will likely encounter similar obstacles, leading to a slower transition than predicted.
I'd just like to point out that, at present, AI is shockingly bad at using references, or even making lists of references. University librarians across the USA are reporting AI-generated reference lists that are comprised of apparent journal articles that do not exist--they're amalgamations of author names, article titles, and journal names, with totally fabricated DOIs and volume and page numbers. (I have also seen this in my own educational and scholarly work.) This is the case even when, for instance, ChatGPT is given a list or real references formatted in one style, and just asked to reformat the existing list of references to another style. AI can make predictions about what words follow other words to make something that looks like a reference page, but at present it cannot apply a set of formatting rules. On the other hand, purpose-built programs like EndNote have existed for more than 20 years and execute this task flawlessly.
I think that something that could hold AI-related progress back is this expectation that all things AI outstrip all previous technologies right this minute, causing people to try to force AI to do things it can't (and that we have other technologies that can.)
I see this as being roughly analogous to early steam engines that couldn't do rotary motion well. You use them for some things and not others, and in the meantime, the underlying technology improves and becomes suitable for new use cases.
I think that could be the case, but in the meanwhile the overselling of how AI is going to revolutionize everything right now could hurt the development of AI, by causing disappointed adopters to become critics. As someone who is not an expert on the history of steam engines, I suspect that this problem of over enthusiasm leading to a backlash was not one that affected steam engine development, because there was so much less communication about new products and their potential at the time.
Societal transitions led by technological progress take time. But due to accelerated returns this big transition will take way less. The implications of having human level general AI by the end of the decade, as Kurzweil says (and I think he is right), will disrupt everything in much more profound ways. It will take time for massive adoption and acceptability, but I think it will be achieved within a decade, not decades or centuries.
> Not only did the transition take a long time, it produced counterintuitive effects. At first, the use of draft horses did not decline: it increased. Railroads provide long-haul transportation, but not the last mile to farms and houses, so while they substitute for some usage of horses, they are complementary to much of it. An agricultural census from 1860 commented on the “extraordinary increase in the number of horses,” noting that paradoxically “railroads tend to increase their number and value.” A similar story has been told about how computers, at first, increased the demand for paper.
I wonder if the same logic applies to technological unemployment.
Not just tech. It takes a minimum of a decade for medical best practices to permeate through a country's healthcare system, even when the evidence is clear and funding isn't an issue.
Yes! Everyone wants “plug compatible Ai”, where they just plug in AI module in place of an old system. But with very few exceptions, that’s not how it works!
Many other parts of a production system have to be changed, e.g. training, career paths. In the long run, entire industry structures will change to get the maximum benefit from new capabilities.
Chip design has used “computer intelligence” for decades. The founders of Intel physical drew the layout of their first chips on transparent sheets that the lithography masks. (Analogous to photo negatives - for those who remember them.) Over decades, more and more of chip design was handled by AI.
One of the big revolutions in electronics industry was the foundry/design organization of companies. Qualcomm is the third largest chip company in the world (or was last time I looked), but it physically makes zero of its chips. They are all outsourced to specialists like TSMC.
The same reorganizations will happen due to what we now call AI. But we can’t predict what, where, or when. Meanwhile, the technical capability of the core algorithms will be continually advancing.
You point is spot on but I want to point out a few things missing from your analysis that you might conclude will accelerate the time frame you put forward
1. You left out the “Information Age”. This is a critical step in human progress that is enabling every single person in the world to have access to the entire world’s knowledge. As a result, when problems and opportunities are identified and/ insights gained, this information spreads to all of humanity. As a result, you multiple magnitudes more people working on the innovation at the same time.
When the steam engine was invented in 1712, it took a long time for the knowledge of this invention to spread across the globe. As a result, it took a long time to harness the collective ingenuity and motivation of humanity to leverage this invention.
Today that can happen in hours, if not minutes.
2. The very invention itself will accelerate innovation. Either through direct innovation itself (this is more speculative) or at the very least enabling innovators to innovate more effectively and efficiently.
Thanks. Those factors are real, and they're why I think the transition can happen faster this time. But again, I still think that's decades.
Similar phenomena occur in several other fields, too. For example, hand washing by doctors took more than 25 years to become a norm, even though Hungarian physician Ignaz Semmelweis identified benefits in the 1840s. His theories were mocked until the 1870s and beyond.
Similarly, it took more than 40 years for James Lind to discover the cure for scurvy and the Royal Navy to begin issuing lemon juice to sailors.
Crop rotation, vaccination, alternate current, and anesthesia are other examples from my readings.
It is mainly due to one or more reasons listed below:
resistance to new ideas;
a lack of understanding of the underlying mechanism;
systematic inertia, including economic and logistic constraints;
influence of a specific individual with competing technology and
sometimes prevailing social and cultural norms and practices.
The above suggests that the widespread adoption of AI, even with its demonstrable benefits, will likely encounter similar obstacles, leading to a slower transition than predicted.
I'd just like to point out that, at present, AI is shockingly bad at using references, or even making lists of references. University librarians across the USA are reporting AI-generated reference lists that are comprised of apparent journal articles that do not exist--they're amalgamations of author names, article titles, and journal names, with totally fabricated DOIs and volume and page numbers. (I have also seen this in my own educational and scholarly work.) This is the case even when, for instance, ChatGPT is given a list or real references formatted in one style, and just asked to reformat the existing list of references to another style. AI can make predictions about what words follow other words to make something that looks like a reference page, but at present it cannot apply a set of formatting rules. On the other hand, purpose-built programs like EndNote have existed for more than 20 years and execute this task flawlessly.
I think that something that could hold AI-related progress back is this expectation that all things AI outstrip all previous technologies right this minute, causing people to try to force AI to do things it can't (and that we have other technologies that can.)
I see this as being roughly analogous to early steam engines that couldn't do rotary motion well. You use them for some things and not others, and in the meantime, the underlying technology improves and becomes suitable for new use cases.
I think that could be the case, but in the meanwhile the overselling of how AI is going to revolutionize everything right now could hurt the development of AI, by causing disappointed adopters to become critics. As someone who is not an expert on the history of steam engines, I suspect that this problem of over enthusiasm leading to a backlash was not one that affected steam engine development, because there was so much less communication about new products and their potential at the time.
"My guess is the intelligence transition will take decades."
It would be useful to specify what constitutes the endpoint of such transition.
Societal transitions led by technological progress take time. But due to accelerated returns this big transition will take way less. The implications of having human level general AI by the end of the decade, as Kurzweil says (and I think he is right), will disrupt everything in much more profound ways. It will take time for massive adoption and acceptability, but I think it will be achieved within a decade, not decades or centuries.
I wonder how this big transition relates to another, more biology-based, driver of home human progress... https://nichabuffona.substack.com/p/the-hormonal-novel-behind-human-progress
> Not only did the transition take a long time, it produced counterintuitive effects. At first, the use of draft horses did not decline: it increased. Railroads provide long-haul transportation, but not the last mile to farms and houses, so while they substitute for some usage of horses, they are complementary to much of it. An agricultural census from 1860 commented on the “extraordinary increase in the number of horses,” noting that paradoxically “railroads tend to increase their number and value.” A similar story has been told about how computers, at first, increased the demand for paper.
I wonder if the same logic applies to technological unemployment.
You should read up about the Diffusion of Innovations https://en.m.wikipedia.org/wiki/Diffusion_of_innovations