> Now we're worried about all the AI jobs disappearing.
I work in the space, and I'm not[1] worried about that at all. For what it's worth, anyone who knows what they're talking about also wasn't worried that all menial jobs would be taken by AI: fine motor control is really difficult, so paradoxically, janitors are safer than many middle-skilled white collar jobs are. More broadly, my model of AI is one of slow but inexorable progress: the dirty secret of tech eating the world is that most industries are run obscenely inefficiently. This doesn't suggest that Google et al can just jump into any random industry with both feet and have immediate success, but this is largely due to inertia (and in some case, regulatory barriers slowing transitions of any kind[2]), both of which are short- and medium-run concerns.
The Cruise announcement in particular is a bit of a strange one to choose to make this point:
> The layoffs represent about 8% of Cruise’s workforce. The losses were mostly in business strategy, product development, design, and recruiting. The company also dismissed staff working on Lidar engineering in Pasadena, Calif.[3]
Cruise has >2 years of runway in the bank right now[3], is a financial entity separate from GM (no longer wholly owned), and is shifting their resources away from "close to market" towards "investing in the research problem", just like other SDC companies like Waymo. The Covid downturn, like most downturns, is just a useful pretext to engage in this restructuring.
I usually tend towards taking the pessimistic view of uncertainty that affects me personally, I guess because I'm relatively risk-averse. But all the signals I've seen from the economic impact of Covid indicate that AI jobs are going to be affected by the general economic downturn but be in much _better_ relative shape than many other jobs: while shelter-in-place won't last forever, the more-lsating effects will likely include things like increased inshoring of manufacturing and an increase in demand for delivery and logistics (both central applications for robotics). The trend of people spending more time online is only increasing too, and the applications of AI there are self-explanatory from the last decade (people really do underestimate what ML specialists do: "decreasing Google's datacenter cooling costs by 40%" isn't the central AI problem most people think of, but it's the kind of things that creates massive amounts of value.
Possible longer-term trends that pose challenges for autonomous vehicle companies in particular (not AI eng in general) potentially include a) some deurbanization and b) less theoretical demand in shared transportation methods; but the latter may be somewhat mitigated by the refugees from transit to robotaxis (esp given that sharing an air pocket is so much more dangerous and hard to control than sharing surfaces).
[1] With caveats, I guess; I and most of my friends in the field have been in or adjacent to AI for a lot longer than the current boom, so we're seeing a very different picture than people who did a Tensorflow tutorial two years ago.
[2] Since I know this will be misinterpreted, I'll clarify: This isn't a suggestion that regulations are bad and should be run rough-shod over, just an acknowledgement that even good regulation tends to slow large shifts in the way industries run, just by introducing complexity and barriers to entry due to tribal knowledge of regulatory compliance. This isn't a criticism of regulation, just a well-known downside.
I work in the space, and I'm not[1] worried about that at all. For what it's worth, anyone who knows what they're talking about also wasn't worried that all menial jobs would be taken by AI: fine motor control is really difficult, so paradoxically, janitors are safer than many middle-skilled white collar jobs are. More broadly, my model of AI is one of slow but inexorable progress: the dirty secret of tech eating the world is that most industries are run obscenely inefficiently. This doesn't suggest that Google et al can just jump into any random industry with both feet and have immediate success, but this is largely due to inertia (and in some case, regulatory barriers slowing transitions of any kind[2]), both of which are short- and medium-run concerns.
The Cruise announcement in particular is a bit of a strange one to choose to make this point:
> The layoffs represent about 8% of Cruise’s workforce. The losses were mostly in business strategy, product development, design, and recruiting. The company also dismissed staff working on Lidar engineering in Pasadena, Calif.[3]
Cruise has >2 years of runway in the bank right now[3], is a financial entity separate from GM (no longer wholly owned), and is shifting their resources away from "close to market" towards "investing in the research problem", just like other SDC companies like Waymo. The Covid downturn, like most downturns, is just a useful pretext to engage in this restructuring.
I usually tend towards taking the pessimistic view of uncertainty that affects me personally, I guess because I'm relatively risk-averse. But all the signals I've seen from the economic impact of Covid indicate that AI jobs are going to be affected by the general economic downturn but be in much _better_ relative shape than many other jobs: while shelter-in-place won't last forever, the more-lsating effects will likely include things like increased inshoring of manufacturing and an increase in demand for delivery and logistics (both central applications for robotics). The trend of people spending more time online is only increasing too, and the applications of AI there are self-explanatory from the last decade (people really do underestimate what ML specialists do: "decreasing Google's datacenter cooling costs by 40%" isn't the central AI problem most people think of, but it's the kind of things that creates massive amounts of value.
Possible longer-term trends that pose challenges for autonomous vehicle companies in particular (not AI eng in general) potentially include a) some deurbanization and b) less theoretical demand in shared transportation methods; but the latter may be somewhat mitigated by the refugees from transit to robotaxis (esp given that sharing an air pocket is so much more dangerous and hard to control than sharing surfaces).
[1] With caveats, I guess; I and most of my friends in the field have been in or adjacent to AI for a lot longer than the current boom, so we're seeing a very different picture than people who did a Tensorflow tutorial two years ago.
[2] Since I know this will be misinterpreted, I'll clarify: This isn't a suggestion that regulations are bad and should be run rough-shod over, just an acknowledgement that even good regulation tends to slow large shifts in the way industries run, just by introducing complexity and barriers to entry due to tribal knowledge of regulatory compliance. This isn't a criticism of regulation, just a well-known downside.
[3] https://electrek.co/2020/05/14/gm-cruise-lays-off-8-of-staff...