The Economic Singularity: Artificial intelligence and the death of capitalism .. This new book from best-selling AI writer Calum Chace argues that within a few decades, most humans will not be able to work for money.
Homo economicus won't worrk if unemployment is well paid. Homo sapiens is a more complicated animal. Some people volunteer. Ie work for nothing for instance.
I am less interested in questions of capability, and more in questions of supply and generality.
1. This wave of automation is software rather than hardware; this means simultaneously that there is effectively no limit to the supply and that employment doesn't scale with adoption.
2. The service sector tends to be more general in the skills it requires than agriculture or manufacturing. When software can completely replace a paralegal, the same or similar software is also able to do all the jobs a paralegal could do. This is distinct from robots who do a specific sequence of tasks in a specific setting, and allowed a laborer to move to a different environment and learn a different sequence of tasks for which there was no robot.
So I expect automation to proceed exactly as fast as demand for it does, and also for it to automate whole skill sets instead of individual jobs. This implies a faster process than previous waves of automation. I have low confidence that our methods of discovering new ways to create value and of producing workers to implement them are experiencing a similar adjustment.
Regarding the last quote and economic contraction, there is also a problem with the distinction between greater efficiency and improvements brought about from creative destruction. Growing bread lines might (but doesn't have to) swamp price reductions from efficiency due to diminishing returns, but it is unclear that they must swamp price reductions brought about by innovation. It's a black swan thing. Hard to predict.
The majority of unemployed individuals will only develop an economy if the value they can offer through their labor is worth the unpleasantness of that labor.
As long as (as seems likely) political expediency or moral concern causes society to offer some minimal monetary assistance to the unemployed it could well be that anything most people could make wouldn't be worth the toil it took to make it. After all they would already have a very comfortable life with any items they desire and robots to perform menial tasks.
Prostitution raises it's head as an obvious form of employment but there has to be something of sufficient value to trade sex for...and there could always be good sex robots.
It seems to me that "full human level common sense AI" could be a much higher standard than what Chace/Hinton were originally referring to. And it may well turn out that we don't even need to get to "human level" common sense for AI to make a serious difference. Couldn't a rudimentary level of common sense be enough to allow AIs to make profound dents in our economy?
A large minority have always been out of the workforce but were usually wives. The problem is more that the skilled will be employed and married while the unskilled will tend to be unemployed and unmarried. While a fall in median wages needn't cause market difficulties, they are more likely to when high earners see no reason to spend much being more than satisfied, see no reason to support low earners and the unemployed, and see no reason to invest, the skilled being fully employed and the unskilled without the income. It would be very difficult to keep average demand growing as a result. Growth in non profits and in skilled labor saving would be the main avenues for investment.
Keynes anticipated that industrial/technological progress would ultimately eliminate our need to work. "Economic Possibilities for our Grandchildren (1930)" http://www.econ.yale.edu/sm...
"Will machine intelligence automate most human jobs within the next few decades, and leave a large minority of people – perhaps a majority – unable to gain paid employment? It seems to me that you have to accept that this proposition is at least possible if you admit the following three premises: 1. It is possible to automate the cognitive and manual tasks that we carry out to do our jobs. 2. Machine intelligence is approaching or overtaking our ability to ingest, process and pass on data presented in visual form and in natural language. 3. Machine intelligence is improving at an exponential rate."
A fourth premise, necessary for the stated conclusion yet not mentioned, is that what enables humans to be economically valuable is merely 'our ability to ingest, process and pass on data presented in visual form and in natural language'. In other words, the ability to learn is all that matters. What Hanson calls 'content' - the product of learning, i.e. the ability to actually do a useful task - is not very important, as that's really the easy part. Once you have an abstract intelligent agent you can relatively quickly train it to do whatever task you want done.
I'm happy to trust you, Robin, but maybe we should appoint an impartial judge as to whether common sense has been achieved. Someone we both know and trust. David Wood?
After July 23, 2026, we'll judge if "full human level common sense AI" had been achieved by that date. Seems like that will be clear enough that you and I can bet on an honor system.
Lots to disagree with in this review, Robin, but let's not chase each other down those rabbit holes. Instead, I'll take your odds. After all, it's Geoff Hinton's judgement you're betting 50-1 against, not mine. Where do I sign? :-) Oh, and I know I owe you a review. I will get to it, I promise.
"3. Machine intelligence is improving at an exponential rate."
That's not a meaningful statement. The S&P 500 has been growing exponentially for more than 100 years. But it's been less than 10% per year.
In order to have a meaningful assessment, it's necessary to say what the doubling time is, and how far behind machine intelligence is.
I would like to live in a world where homo economicus is extinct. But I keep seeing evidence to the contrary.
Homo economicus won't worrk if unemployment is well paid. Homo sapiens is a more complicated animal. Some people volunteer. Ie work for nothing for instance.
What exactly is special about a majority? [What's the highest percentage that could be unemployed?]
I am less interested in questions of capability, and more in questions of supply and generality.
1. This wave of automation is software rather than hardware; this means simultaneously that there is effectively no limit to the supply and that employment doesn't scale with adoption.
2. The service sector tends to be more general in the skills it requires than agriculture or manufacturing. When software can completely replace a paralegal, the same or similar software is also able to do all the jobs a paralegal could do. This is distinct from robots who do a specific sequence of tasks in a specific setting, and allowed a laborer to move to a different environment and learn a different sequence of tasks for which there was no robot.
So I expect automation to proceed exactly as fast as demand for it does, and also for it to automate whole skill sets instead of individual jobs. This implies a faster process than previous waves of automation. I have low confidence that our methods of discovering new ways to create value and of producing workers to implement them are experiencing a similar adjustment.
Regarding the last quote and economic contraction, there is also a problem with the distinction between greater efficiency and improvements brought about from creative destruction. Growing bread lines might (but doesn't have to) swamp price reductions from efficiency due to diminishing returns, but it is unclear that they must swamp price reductions brought about by innovation. It's a black swan thing. Hard to predict.
The majority of unemployed individuals will only develop an economy if the value they can offer through their labor is worth the unpleasantness of that labor.
As long as (as seems likely) political expediency or moral concern causes society to offer some minimal monetary assistance to the unemployed it could well be that anything most people could make wouldn't be worth the toil it took to make it. After all they would already have a very comfortable life with any items they desire and robots to perform menial tasks.
Prostitution raises it's head as an obvious form of employment but there has to be something of sufficient value to trade sex for...and there could always be good sex robots.
I agree that full common sense is different criteria from "serious difference."
It seems to me that "full human level common sense AI" could be a much higher standard than what Chace/Hinton were originally referring to. And it may well turn out that we don't even need to get to "human level" common sense for AI to make a serious difference. Couldn't a rudimentary level of common sense be enough to allow AIs to make profound dents in our economy?
A large minority have always been out of the workforce but were usually wives. The problem is more that the skilled will be employed and married while the unskilled will tend to be unemployed and unmarried. While a fall in median wages needn't cause market difficulties, they are more likely to when high earners see no reason to spend much being more than satisfied, see no reason to support low earners and the unemployed, and see no reason to invest, the skilled being fully employed and the unskilled without the income. It would be very difficult to keep average demand growing as a result. Growth in non profits and in skilled labor saving would be the main avenues for investment.
I'm happy to accept David Wood as judge.
Keynes anticipated that industrial/technological progress would ultimately eliminate our need to work. "Economic Possibilities for our Grandchildren (1930)" http://www.econ.yale.edu/sm...
"Will machine intelligence automate most human jobs within the next few decades, and leave a large minority of people – perhaps a majority – unable to gain paid employment? It seems to me that you have to accept that this proposition is at least possible if you admit the following three premises: 1. It is possible to automate the cognitive and manual tasks that we carry out to do our jobs. 2. Machine intelligence is approaching or overtaking our ability to ingest, process and pass on data presented in visual form and in natural language. 3. Machine intelligence is improving at an exponential rate."
A fourth premise, necessary for the stated conclusion yet not mentioned, is that what enables humans to be economically valuable is merely 'our ability to ingest, process and pass on data presented in visual form and in natural language'. In other words, the ability to learn is all that matters. What Hanson calls 'content' - the product of learning, i.e. the ability to actually do a useful task - is not very important, as that's really the easy part. Once you have an abstract intelligent agent you can relatively quickly train it to do whatever task you want done.
I'm happy to trust you, Robin, but maybe we should appoint an impartial judge as to whether common sense has been achieved. Someone we both know and trust. David Wood?
After July 23, 2026, we'll judge if "full human level common sense AI" had been achieved by that date. Seems like that will be clear enough that you and I can bet on an honor system.
Lots to disagree with in this review, Robin, but let's not chase each other down those rabbit holes. Instead, I'll take your odds. After all, it's Geoff Hinton's judgement you're betting 50-1 against, not mine. Where do I sign? :-) Oh, and I know I owe you a review. I will get to it, I promise.