Hacker News Reader

Comment by christkv | Hacker News Reader

ha ha, no. Considering the difficulty of humans equipped with the most advanced pattern matching hardware ever devised by evolution to discern nuances of meaning in spoken language I would say it's a rehash of the AI promises of the 50's that feel short of any practical implementation. We still got a long way to go before a skynet shows up and fires us.

View on HN

Skynet has already fired many of us. We just don't notice because it happened incrementally.

To see an example of this on the macroscale, look at the current recession. AD (Aggregate Demand) and production tanked, leading to unemployment. AD and production have recovered, but employment has not. There is a very plausible case to be made that productivity gains made many workers obsolete, and employers took advantage of the recession to cut them loose.

[edit: forgot that not everyone is an amateur economist. AD = Aggregate Demand.]

Sorry, what is AD?

Aggregate Demand, see my edit.

That is the thesis of Jeremy Rifkin exposed in the "The End of Work"

Which jobs were made obsolete? If unemployment were technologically driven, you would expect IT to have less unemployment relative to other industries. But that's not the case. Paul Krugman had a pretty good post on the topic:

http://krugman.blogs.nytimes.com/2011/02/13/whos-unemployed/

Which jobs were made obsolete?

The data is insufficiently granular to determine that. But as you can see, production has more or less completely recovered: http://research.stlouisfed.org/fred2/series/GDP http://research.stlouisfed.org/fred2/series/INDPRO

Employment has not recovered comparably: http://research.stlouisfed.org/fred2/series/PAYEMS

This means we are producing more now than ever before, and doing it with fewer people. This tells us that many of the people who were laid off are obsolete.

Now, as for Krugman's chart, it's somewhat tangential. Krugman as arguing (misleadingly, BTW) against the recalculation hypothesis, which proposes that we have a recession because the economy misallocated people into the wrong sectors. His chart is misleading since it focuses on an irrelevant ratio of two other irrelevant ratios. To determine if the recession is sectorial, one must look at employment [1].

If we do this, we find that construction employment is down 27% (from Jan 2008 to Jan 2011). Information services employment (I think this includes IT) is down 11%. Finance is down 8%, as is Retail. Durable goods manufacturing is down 22%.

http://research.stlouisfed.org/fred2/series/USFIRE http://research.stlouisfed.org/fred2/series/USCONS http://research.stlouisfed.org/fred2/series/USINFO http://research.stlouisfed.org/fred2/data/USTRADE.txt http://research.stlouisfed.org/fred2/series/DMANEMP

This clearly shows some sectors hit much harder than others, contrary to Krugman's claims. But Krugman's claims are tangential to the main point anyway.

[1] Unemployment is a skewed indicator because it excludes people not seeking work and people who find work in other sectors. I.e., a construction worker who finds a new job in retail lowers both the unemployment rate in construction (smaller numerator) and the unemployment rate in retail (bigger denominator). As the employment numbers show, this is a rather large set of people.

Ah, good point on why we should look at employment vs. unemployment. Makes sense.

I was tempted to argue that the decline in construction doesn't support the "obsolete jobs" hypothesis, because the jobs were created by an overheated housing market and shouldn't have been there in the first place. But regardless of whether the jobs were justified in the first place, they're not coming back. A technological shift isn't the only thing that can make a job obsolete; a one-time speculative bubble can do the same thing.

No, you are quite right. The decline in construction jobs doesn't support the "jobs obsolete due to technology" hypothesis. It supports the recalculation hypothesis, but that's a separate issue.

It's the increase in production combined with the lack of increase in employment that is evidence in favor of technology making jobs obsolete.

It doesn't happen overnight. I'll go out on a limb here and say that we are only 10 years away from the first completely autonomous delivery trucks.

Google is already doing this with human co-pilots. Once there is solid statistical data showing that a computer driver is safer and cheaper than a human, the phase-out of the human drivers will be pretty swift. After all, imagine the competitive advantage you'd have if your trucking company can replace a human driver with a computer. One that can drive 24 hours a day while making fewer mistakes. Even if this technology cost half a million dollars (and it won't), doing so would be a no-brainer.

Once there is solid statistical data showing that a computer driver is safer and cheaper than a human, the phase-out of the human drivers will be pretty swift

I'm not so sure. Solid statistical data won't deter lawyers filing frivolous suits over every accident. "If only Big Faceless Co. didn't employ faceless robots, this sympathetic human might have lived! You should award lots of punitive damages."

Not to mention the teamsters would probably have a law against robot drivers passed...

People will say the same things 50 years later!

However, it is a fact that AI are pervasive in our world today. We just don't recognize it as artificial intelligence.

True but this kind of stories conjure up the classic idea of the human computer :). Practical use is cool like the google self-driven car I'm not sure if it's a learning system (does anyone know ?). I figure deduction to be the truly hard part to do. Like deducting the laws of geometry by observation or something similar. But hey I'm an amateur on this :)

Researchers have built AI programs that were claimed to have rediscovered some basic results in maths:

http://en.wikipedia.org/wiki/Automated_Mathematician

That line of symbolic/formal AI seems to have pretty much died out (or at least not going anywhere very quickly), consider:

http://en.wikipedia.org/wiki/Cyc

Reply on news.ycombinator.com